1482238985

Metasolutions of
Parabolic Equations
in Population
Dynamics
2016 by Taylor & Francis Group, LLC
Metasolutions of
Parabolic Equations
in Population
Dynamics
Julin Lpez-Gmez
Universidad Complutense de Madrid
Spain
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Version Date: 20150909
International Standard Book Number-13: 978-1-4822-3899-0 (eBook - PDF)
This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
To Rosa G
omez-Gonz
alez
my mother,
with love and gratitude
Contents
List of Figures
xi
Preface
xiii
Large solutions and metasolutions: Dynamics
1 Introduction: Preliminaries
1.1
1.2
1.3
1.4
1.5
1.6
1.7
The meaning of the KellerOsserman condition . .

Model in population dynamics . . . . . . . . . . .
Characterization of the maximum principle . . . .
Existence through subsolutions and supersolutions
Some abstract pivotal results . . . . . . . . . . . .
Logistic equation in population dynamics . . . . .
Comments on Chapter 1 . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2 Classical diffusive logistic equation

2.1
2.2
2.3
2.4
2.5
33
Unperturbed logistic problem . . . . . . . . . . . .

Solution set for the unperturbed problem . . . . .
Perturbed logistic problem . . . . . . . . . . . . .
Structural stability as M > 0 perturbs from M = 0
. . . .
. . . .
. . . .
. . .
. . . .
.
.
.
.
.
.
.
.
.
.
3 A priori bounds in
3.1
3.2
3.3
3.4
3.5
Singular problem in a ball BR (x0 ) b .

Singular problem in a general D . .
Existence of minimal and maximal solutions
Some sufficient conditions for (KO) . . . .
Comments on Chapter 3 . . . . . . . . . .
34
37
41
42
47
49
. . .
. . .
. .
. . .
. . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4 Generalized diffusive logistic equation

4.1
4.2
4.3
8
11
15
17
20
27
30
Classical positive solutions in . . . . . . . . . . . . . . . .

Associated inhomogeneous problems . . . . . . . . . . . . . .
Hierarchic chain of inhomogeneous problems . . . . . . . . .
50
57
59
61
64
67
69
76
80
vii
viii
Contents
4.4
4.5
4.6
4.7
4.8
4.9
Large solutions of arbitrary order j {1, ..., q0 } . . . . . . .

Limiting behavior of the positive solution as d1 . . . . .
Direct proof of Theorem 4.8 when a C 1 () . . . . . . . . .
Limiting behavior of the large solutions of order 1 j q0 1
as dj+1 . . . . . . . . . . . . . . . . . . . . . . . . . . .
Limiting behavior as of the large solutions . . . . . .
Comments on Chapter 4 . . . . . . . . . . . . . . . . . . . .
5 Dynamics: Metasolutions
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
II
Concept of metasolution: The main theorem . . .

Paradigmatic bifurcation diagram with q0 = 2 . .
Numerical example with q0 = 2 . . . . . . . . . . .
Proof of Theorem 5.2 . . . . . . . . . . . . . . . .
Approximating metasolutions by classical solutions
Pattern formation in classical logistic problems . .
Biological discussion . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
. .
. .
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
137
Existence and uniqueness . . . . . . . . . . . . . . .

Auxiliary function b . . . . . . . . . . . . . . . . . .
Getting sharp estimates for `(x) at x = 0 through b
The exact blow-up rate of `(x) at x = 0 . . . . . . .
Comments on Chapter 6 . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7 Uniqueness of the large solution under radial symmetry
7.3
7.4
7.5
The main uniqueness result . . . . . . . . .

Proof of Theorem 7.1 . . . . . . . . . . . .
7.2.1 The case when = BR (x0 ) . . . . .
7.2.2 The case when = AR1 ,R2 (x0 ) . . .
Exact blow-up rates on the boundary . . .
Simple application in population dynamics
Comments on Chapter 7 . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Boundary normal section of a . . . . . . . .

Boundary blow-up rate of the large solutions
Proof of Theorem 8.2 . . . . . . . . . . . . .
Special case when a(x) > 0 for some x
Uniqueness of the large solution . . . . . . .
139
148
159
164
165
169
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8 General uniqueness results

8.1
8.2
8.3
8.4
8.5
106
108
111
117
125
128
129
131
135
6 A canonical one-dimensional problem
7.1
7.2
96
98
101
105
Uniqueness of the large solution
6.1
6.2
6.3
6.4
6.5
83
86
93
170
172
172
174
177
183
184
189
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
190
192
193
201
202
Contents
8.6
III
Comments on Chapter 8
ix
. . . . . . . . . . . . . . . . . . . .
Metasolutions do arise everywhere
207
9 A paradigmatic superlinear indefinite problem

9.1
9.2
9.3
9.4
9.5
9.6
Components of positive steady states . . . . . . . .

Local structure at stable positive solutions . . . . .
Existence of stable positive solutions . . . . . . . . .
Uniqueness of the stable positive solution . . . . . .
Curve of stable positive solutions . . . . . . . . . . .
Dynamics in the presence of a stable steady state .
9.6.1 Global existence versus blow-up in finite time
9.6.2 Complete blow-up in + . . . . . . . . . . . .
9.7 Dynamics for [d1 , d2 ) and a+ small . . . . . .
9.8 Dynamics for [d1 , d2 ) and a+ large . . . . . .
9.9 Dynamics for d2 . . . . . . . . . . . . . . . . .
9.10 Comments on Chapter 9 . . . . . . . . . . . . . . .
9.10.1 Abiotic stress hypothesis . . . . . . . . . . . .
204
209
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
10 Spatially heterogeneous competitions

10.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . .
10.1.1 Dynamics of the semi-trivial positive solutions .
10.1.2 Dynamics when d1 0 or d2 0 . . . . . .
10.2 Dynamics of the model when c = 0 . . . . . . . . . . . .
10.3 A priori bounds for the coexistence states . . . . . . . .
10.4 Global continua of coexistence states . . . . . . . . . .
10.4.1 Regarding as the main bifurcation parameter .
10.4.2 Regarding as the main bifurcation parameter .
10.5 Strong maximum principle for quasi-cooperative systems
10.6 Multiplicity of coexistence states . . . . . . . . . . . . .
10.7 Dynamics of (1.1) when b = 0 . . . . . . . . . . . . . .
10.8 Existence of meta-coexistence states . . . . . . . . . . .
10.9 Comments on Chapter 10 . . . . . . . . . . . . . . . . .
212
224
230
233
237
240
240
245
252
254
256
257
265
267
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
272
272
273
277
288
293
293
302
308
311
319
325
328
Bibliography
331
Index
351
List of Figures
P.1
An admissible nodal configuration for a(x) . . . . . . . . . .
xv
1.1
1.2
1.3
1.4
1.5
An admissible nodal configuration for a(x) . . . .

The phase plane of equation (1.14) for u > 0 . . .
The dynamics of u according to the Malthus law
The dynamics of u according to the logistic law .
Brownian motion . . . . . . . . . . . . . . . . . .
.
.
.
.
.
5
10
13
14
30
2.1
2.2
The dynamics of (2.1) for M = 0 . . . . . . . . . . . . . . .

The dynamics of (2.1) in case M > 0 . . . . . . . . . . . . .
40
44
3.1
Scheme of the construction of L[,D] and L[,D] . . . . . . .
59
4.1
4.2
4.3
4.4
4.5
An intricate nodal configuration for a(x) .

The map 7 [,] (x) for x 0,1 . . . .
The -neighborhoods ,1 and ,2 . . . .
The profile of the supersolution element
The ball BR (Y (x )) . . . . . . . . . . . .
.
.
.
.
.
68
70
72
73
89
5.1
5.2
5.3
5.4
The dynamics of (1.1). . . . . . . . . . . . . . . . . . . . . .

A plot of a(x) . . . . . . . . . . . . . . . . . . . . . . . . .
The curve 7 k[,] k2 . . . . . . . . . . . . . . . . . . . .
The classical positive steady states [,] for = 101.229218,
= 191.845246, = 217.71643, and = 232.193199 . . . . .
Classical solutions and stable metasolutions supported in 1
The large solutions L[,1 ] for = 300, = 450, = 500, and
= 525 < 2 . . . . . . . . . . . . . . . . . . . . . . . . . .
109
112
113
5.5
5.6
min
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
max
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
114
115
116
6.1
The graph of `(x) . . . . . . . . . . . . . . . . . . . . . . . .
139
7.1
and Lmin
The functions L
[,BR (x0 )] . . . . . . . . . . . . . .
173
8.1
8.2
,x, are supported . .

The balls where the supersolutions L
The annuli where the subsolutions are supported . . . . . .
195
199
9.1
9.2
An admissible nodal configuration of a(x) . . . . . . . . . .

Bifurcation diagram and solution plots for the choice (9.13)
211
218
xi
xii
List of Figures
9.3
9.4
9.5
9.11
9.12
Bifurcation diagram for the choice (9.14) . . . . . . . . . . .

Plots of a series of solutions on C+ for the choice (9.14) . . .
Plots of a series of positive solutions of (9.3) for the choice
(9.14) on the components F1 , F2 , and F3 . . . . . . . . . . .
Local structure of the set of positive solutions of (9.3) around
a linearly stable solution and a neutrally stable solution . .
The asymptotic profiles of u
in cases (9.60) and (9.61). . . .
Global bifurcation diagram for = 5 and plots of some
solutions along it . . . . . . . . . . . . . . . . . . . . . . . .
Global bifurcation diagram for = 300, magnification of
the turning point along the primary curve exhibiting the subcritical bifurcation of the first closed loop, and plots of ut and
two solutions of type (3, s) . . . . . . . . . . . . . . . . . . .
Global bifurcation diagram for = 750, = 760.3, =
800, and = 1300 . . . . . . . . . . . . . . . . . . . . . .
Global bifurcation diagram for = 2000 . . . . . . . . . .
A genuine symmetry breaking of a bifurcation diagram . . .
10.1
10.2
10.3
10.4
10.5
10.6
10.7
The curves arising in Theorem 10.5 . . . . . . . . . . . . .

The component C+
(,0,,v ) in case c = 0 with d1 1 . . .
The component C+
(,0,,v ) for c 0 and satisfying (10.73)
Some significant curves in the (, ) plane . . . . . . . . . .
An admissible bifurcation diagram for ' 0 , < 0 . . . .
An admissible bifurcation diagram for 0 < < 1d1 2 . . .
A possible bifurcation diagram for > 1d1 2 . . . . . . . . .
9.6
9.7
9.8
9.9
9.10
220
221
223
227
251
259
260
261
263
264
286
299
301
304
305
306
307
Preface
This book studies the dynamics of a generalized

parabolic logistic problem
u
in
t du = u + a(x)u2
u=0
on
u(, 0) = u0 > 0
in
prototype of the semilinear

(0, ),
(0, ),
,
(P.1)
where RN , with N 1, is a smooth bounded domain, t > 0 stands for

the time, and a(x) is an arbitrary continuous function such that a(x) 0 for
all x but a =
6 0. So, (P.1) is a parabolic boundary value problem for the
degenerate diffusive logistic equation
u
du = u + a(x)u2
t
(P.2)
in . It is said to be degenerate because a(x) can vanish on some patches of
, in contrast to the classical case when a(x) < 0 for all x .

In the context of population dynamics, N 3, is the inhabiting area
where the individuals of a species, u, disperse randomly at a constant rate
measured by d > 0; u(x, t) is the density of the individuals of the species at
the location x after time t > 0; is the intrinsic rate of natural increase
of the species; u0 is the initial distribution of the species in ; and
K(x)
,
a(x)
x ,
(P.3)
is the carrying capacity of at each location x . As we are imposing

homogeneous Dirichlet boundary conditions on , the surroundings of are
assumed to be hostile for the species u. So, no individual of the species can
survive on the habitat edges. However, this assumption is far from necessary
for the validity of most of the results discussed in this book. Two classic books
on population dynamics from the perspective of reaction diffusion equations
are by J. D. Murray [197] and A. Okubo and S. A. Levin [200].
Although it is folklore that the classical non-spatial logistic equation
u0 (t) = u(t) + au2 (t)
where a is a negative constant goes back to P. F. Verhulst [230] (1838), it is
less known that the diffusive logistic equation (P.2) was introduced by A. N.
xiii
xiv
Preface
Kolmogorov I. G. Petrovsky and N. S. Piskunov [124], and independently by

R. A. Fisher [88], in 1937, to study some problems of a biological nature. In
the classical context, a(x) is a continuous function such that a(x) < 0 for all
The analysis of the degenerate parabolic problem when a 0 in but
x .
a 0 on some subset of with non-empty interior goes back to J. M. Fraile
et al. [90] (1996). An elliptic counterpart of these degenerate models had been
previously analyzed by H. Brezis and L. Oswald [31], and by T. Ouyang [203],
[204], as part of his PhD thesis under the supervision of W. M. Ni.
Naturally, in spatially heterogeneous environments, the carrying capacity,
K(x), might suffer dramatic variations according to the location of the individuals of the species on the territory, x . Indeed, although K(x) might
be very small on some patches of the territory as an effect of harsh environmental conditions or abiotic stress, in benign areas, natural refuges or special
protected zones, K(x) might reach huge values.
From the mathematical point of view, a rather reasonable methodology
to deal with huge variations of the carrying capacity K(x) in the territory
is assuming that K = , or equivalently a 0, in the protected areas,
while it is finite in less favorable zones. This strategy also makes sense from the
biological point of view, as it is equivalent to combining, simultaneously, within
the same territory, the Mathus and the Verhulst laws regulating the growth
of the species. A further perturbation analysis should reveal the complete list
of admissible limiting distribution patterns of the population as time passes
in general diffusive spatially heterogeneous logistic problems.
In the region where a(x) < 0 the temporal evolution of species u is assumed
to be governed by a logistic growth, while in the region where a(x) 0 the
species u increases according to an exponential growth. The main goal of
this book is predicting the time evolution of the species u in under such
circumstances. Should the species exhibit a genuine logistic behavior in ,
or, on the contrary, should it exhibit an exponential growth? There is the
possibility that u grows according to the Malthus law on some areas of ,
while it simultaneously inherits a limited growth on others.
In an effort to summarize the contents of this book in this short general
preliminary presentation, suppose a(x) has a nodal behavior of the type described in Figure P.1, where the territory contains ten protected zones, 10,1 ,
20,1 , which are two balls, or discs if N = 2, with the same radius R1 , 10,2 ,
1 i 4, which are four balls with radius R2 < R1 , and i0,3 , 1 i 4,
which are four balls with radius R3 < R2 . The weight function a(x) is assumed
to vanish in all these refuges, or protected zones, while it is negative on their
complement, the shadow region of Figure P.1, denoted by .
To describe the main findings of this book for this special configuration
of the territory we need to introduce some notation. Given any nice open
connected subset, D, of we will denote by 1 [, D] the lowest eigenvalue
of the linear eigenvalue problem

u = u
in D,
(P.4)
u=0
on D.
Preface
xv
As will be discussed in Chapter 1, from a biological perspective, d1 [, D]

measures the critical size of the rate of natural increase, , so that the inhabited area D can maintain the species u dispersing at the rate d in the patch
D, in the sense that u is driven to extinction if < d1 [, D], while it is
permanent if > d1 [, D]. So, the condition d1 [, D] < measures
the necessary geometrical properties and size of the patch D to maintain the
species dispersing at the rate d in D with an intrinsic rate of natural increase
. When D is a ball of radius R, a simple change of scale reveals that
1 [, D] =
1 [, B1 ]
R2
FIGURE P.1: An admissible nodal configuration for a(x).
xvi
Preface
where B1 is the unit ball of RN . In particular, the larger the protected zone
the smaller the principal eigenvalue. Consequently, setting
0 = 1 [, ],
j := 1 [, 10,j ],
1 j 3,
it becomes apparent that

d1 [, B1 ]
(1 i 2)
R12
d0 = d1 [, ] < d1 = d1 [, i0,1 ] =
d1 [, B1 ]
R22
d1 [, B1 ]
< d3 = d1 [, i0,3 ] =
R32
< d2 = d1 [, i0,2 ] =
(1 i 4)
(1 i 4)
Naturally, for each j = 1, 2, 3, dj measures the critical size of so that the

protected zone i0,j can maintain the species u in isolation. In other words,
i0,j has sufficient resources to maintain u at the increase rate if, and only
if, > dj . The main results of the first five chapters of this book, which
constitute Part I, can be summarized as follows:
If d0 , u is driven to extinction in .
If d0 < < d1 , i.e., can maintain the species at the increase rate ,
but the larger refuges, 10,1 and 20,1 , are unable to maintain it, by e.g.,
a shortage of resources, then the species u grows according to a logistic
law everywhere in .
If d1 < d2 , i.e., the larger protected zones, 10,1 and 20,1 , can
maintain u at the increase rate , but i0,2 , 1 i 4, are unable to do
so, then u grows up exponentially in the largest protected areas
0,1 10,1 20,1 ,
according to a genuine Malthusian growth, but according to the logistic
law in the remaining areas of the territory
0,1 .
1 \
If d2 < d3 , i.e., the refuges i0,2 , 1 i 4, can maintain u at
the increase rate , but i0,3 , 1 i 4, cannot do so, then u grows
exponentially in the protected areas
0,1 10,1 20,1
and
0,2
4
[
i=1
whereas it has a limited logistic increase in

0,1
0,2 ).
2 \ (
i0,1
Preface
xvii
If d3 , i.e., any refuge is able to maintain u at the increase rate ,

then u grows exponentially in all the protected zones
0,1 10,1 20,1 ,
0,2
4
[
i0,2
and
0,3
i=1
4
[
i0,3
i=1
but according to a logistic law in the region

0,1
0,2
0,3 ) = .
3 \ (
Consequently, if, for instance, d1 < < d2 , and we denote by u(x, t) the
unique solution of (P.1), then, as a consequence of the analysis in Part I, we
find that
lim u(x, t) = for all x 0,1 = 10,1 20,1
t
0,1 ,
whereas, in the region \
Lmin
lim inf u(, t) lim sup u(, t) Lmax
(P.5)
where Lmin
and Lmax
stand for the minimal and the maximal positive solutions
of the singular problem
0,1 ,
in \
dL = L + a(x)L2
L=
on 10,1 20,1 ,
(P.6)
L=0
on .
Therefore, the limiting profile of u(x, t) as time t becomes infinity in the
1 and
2 , while it remains bounded in the complement.
larger refuges,
0,1
0,1
These limiting profiles are referred to in this book as metasolutions supported
in the complement of the largest protected zones, 1 , because 1 is the portion
of the inhabiting area where the growth of u inherits a genuine logistic character and hence it is limited. It should be noted that the smaller refuges cannot
support the species u in isolation if < d2 . The formal concept metasolution was coined in [109], submitted for publication in September 1998. Then,
it was incorporated into the PhD thesis of R. Gomez-Re
nasco [105], under
the supervision of the author and defended at the University of La Laguna
(Tenerife, Spain) in early May 1999.
For those readers not familiarized yet with the most recent advances in
the theory of nonlinear parabolic problems, possibly under the influence of
the established (wrong) paradigm that the Harnack inequality is one of the
driving forces of the theory of nonlinear partial differential equations, the
emergence of such metasolutions in the context of population dynamics might
be slightly shocking, as large solutions and metasolutions provide us with
uncontestable evidence that the Harnack inequality is a technical device of a
linear nature of doubtful interest in analyzing global nonlinear problems, as
will become apparent in Section 4.9.
xviii
Preface
This might possibly explain the reaction of an anonymous reviewer of [195]

who noted that a series of classical solutions and metasolutions were computed
in the disc of radius 1 centered at the origin, B1 , with the choices
= B1 (0) = {x R2 : |x| < 1},
0,1 = A(0.5, 1) = {x R2 : 0.5 < |x| < 1},
0,2 = B0.3 (0) = {x R2 : |x| < 0.3},
= A(0.3, 0.5) = {x R2 : 0.3 < |x| < 0.5},
by using pseudo-spectral methods.
What the heck is a metasolution? Please provide a formal definition. Okay, one
is provided in (4.8), but metasolutions is used in the abstract and intro; definition
needs to be earlier. The definition puzzles me. The large solution would seem to
be very difficult to compute because of the singularity on the boundary. And what
is the use or point of a solution that is infinite everywhere on another subdomain?
Metasolutions are wierd...
I am alarmed by the references to blow upand approach infinity on the boundary. Spectral methods are notoriously sensitive to singularities of the solution including singularities on the boundaries...
The serious problem with the paper is that the discontinuities of slope in coefficients of partial differential equation and the infinities on the boundary both makes
the solution of partial differential equation singular within the domain.
I hate their Br (0), A(R0 , R1 ) notation for what are simply the disk of radius R
and the annulus bounded in radius by R0 and R1 . For goodness sake, use conventional notation and wording: disk of radius R, r [0, R],...
I am further bothered that their coefficient function a(x) is nonzero only for
r [0.3, 0.5] for a problem in the unit disk. The PDE thus has a coefficient with
a slope discontinuity. The function u(r, ) will be singular on the lines r = 0.3 and
r = 0.5. The usual spectral strategy would be to split the domain into three and
solve the linear Helmholtz equation on r [0, 0.3] and r [0.5, 1], the nonlinear
PDE on [0.3, 0.5] and carefully match the pieces taking account of the singularities.
Instead the authors blithely ignored the singularities entirely...
Although the strategy proposed by the reviewer in the previous paragraphs

is the most natural one when dealing with linear problems where the Harnack
inequality applies, it is of no help in dealing with singular boundary value
problems such as those treated in [195] and in this book. Contrary to what
happens in most academic problems, real problems might be highly nonlinear
and hence can develop internal interfaces whose numerical treatment is a top
level challenge.
It is the hope of the author that the readers of this book will not be
alarmed by the large solutions and the metasolutions as much as the reviewer of [195] was. Although, at first glance, metasolutions might be slightly
hard to digest because of the number of technicalities involved in their study,
during the last two decades they have proven to be categorical imperatives to
Preface
xix
describe the dynamics of wide classes of parabolic equations and systems in

the presence of spatial heterogeneities.
It should be noted that (P.5) does not fully characterize the asymptotic
behavior of the solutions of (P.1) unless,
Lmin
= Lmax
.
Consequently, to characterize the exact asymptotic profiles as t of the

solutions of (P.1), one must face the problem of the uniqueness of the solutions to the singular problem (P.6) and some other closely related singular
problems that the reader will find in Chapter 4. This is the main bulk of Part
II, consisting of Chapters 6, 7 and 8, where a series of very sharp optimal
uniqueness results found by the author and his coworkers will be analyzed in
a self-contained way.
Finally, the main goal of Part III, formed by the last two chapters, is to
reinforce the evidence that metasolutions also are categorical imperatives to
describe the dynamics of huge classes of spatially heterogeneous semilinear
parabolic problems. Precisely, Chapter 9 analyzes (P.1) in the more general
case when a(x) changes sign, giving a rather complete account of some of
the most relevant recent advances in the theory of superlinear indefinite problems, and Chapter 10 studies a paradigmatic competing species model with
a protected zone for one of the species to illustrate how large solutions and
metasolutions play a pivotal role in describing the dynamics of spatially heterogeneous systems.
This book grew from the monograph [160] and the lecture notes of the
Metasolutions course delivered by the author at the National Center for Theoretical Sciences, Tsing Hua University, Hsinchu (Taiwan), during July and
August of 2009. The author is delighted to thank Professor Sze-Bi Hsu for his
kind invitation to deliver it, as well as for his brilliant questions and sharp
comments during these lectures. The time spent in Taiwan by the author was
certainly unforgettable, both personally and professionally.
To complete this book, the author has been supported by Research Grant
Ref: MTM2012-30669 of the Ministry of Economy and Competitiveness of
Spain.
Madrid
J. L
opez-G
omez
Part I
Large solutions and

metasolutions: Dynamics
Chapter 1
Introduction: Preliminaries
1.1
1.2
1.3
1.4
1.5
1.6
1.7
The meaning of the KellerOsserman condition . . . . . . . . . . . . . . . . .

Model in population dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Characterization of the maximum principle . . . . . . . . . . . . . . . . . . . . .
Existence through subsolutions and supersolutions . . . . . . . . . . . . .
Some abstract pivotal results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Logistic equation in population dynamics . . . . . . . . . . . . . . . . . . . . . . .
Comments on Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8
11
15
17
20
27
30
This book focuses attention into the problem of ascertaining the asymptotic
behavior of the solutions of the parabolic problem
u
in (0, ),
t du = u + a(x)f (x, u)u
(1.1)
u=0
on (0, ),
u(, 0) = u0 > 0
in ,
where d > 0 is a constant, stands for the Laplace operator in RN ,
N
X
2
,
:=
x2j
j=1
x = (x1 , ..., xN ) RN ,
is a bounded domain of RN , N 1, with smooth boundary of class

C 2+ , for some (0, 1], R is regarded as a parameter, and
a C (),
[0, )),
f C ,1+ (
although these regularity requirements can be substantially relaxed to assume

[0, )). However the first eight
is of class C 2 , a L () and f C 0,1 (
chapters of this book will study the special situation when
a<0
(a 0, a 6= 0),
which is usually referred to as the sublinear case if f 0, because in such

circumstances we have
u + a(x)f (x, u)u u
for all u 0,
the general case when a(x) changes sign will be dealt with in Chapter 9.
3
Metasolutions of Parabolic Equations in Population Dynamics
Throughout this book, given a C 2 -subdomain D of and V L (D),

we will denote by 1 [d + V, D] the principal eigenvalue of d + V in D
under homogeneous Dirichlet boundary conditions on , i.e.,

R
d||2 + V 2
D
R
.
(1.2)
1 [d + V, D] =
inf
2
C01 (D)\{0}
D
As the nodal behavior of a(x) might be rather involved without further assumptions, in most of this book we will assume that a(x) satisfies the next
structural hypothesis:
Hypothesis (Ha)
The open sets
,
0 := \
:= { x : a(x) < 0 },
are of class C 2+ and consist of finitely many components

,j ,
i0,j 0 ,
1 j q ,
1 j q0 ,
1 i mj ,
such that
q
[
,j
=
,j
,j ,
if j 6= j,
j=1
0 =
mj
q0 [
[
i0,j ,
i0,j
i =
0,j
if (j, i) 6= (j, i),
j=1 i=1
for some integers q 1, q0 1 and mj 1, 1 j q0 . Moreover,

If 0 is a component of 0 such that 0 6= , then 0
is a component of .
If is a component of such that 6= , then
is a component of .
Without loss of generality, throughout this book we will assume that the
labeling of these components has been already carried out so that
j := 1 [, i0,j ],
j < j+1 ,
1 i mj , 1 j q0 ,
1 j q0 1.
(1.3)
Also, for every 1 j q0 , we will set

0,j :=
mj
[
i0,j ,
i=1
1 [, 0,j ] := j = 1 [, i0,j ], 1 i mj .
Figure 1.1 illustrates a typical situation where assumption (Ha) is fulfilled

with q0 = 2 and q = 1, where we have denoted
= ,
1 = 0,1 \ ,
2 := 0,2 ,
= 1 2 .
In this example, (1.3) becomes

1 := 1 [, 0,1 ] < 2 := 1 [, 0,2 ].
(1.4)
Thanks to the FaberKrahn inequality (cf. C. Faber [83] and E. Krahn [126]),
(1.4) holds if 0,2 has sufficiently small Lebesgue measure (e.g., [144, Section
5]). Indeed, according to these results, among all domains D of RN with a
fixed Lebesgue measure, |D|, the ball has the smallest principal eigenvalue.
Consequently, setting
B% := {x RN : |x| < %},
% > 0,
and
1/N
R := (|D|/|B1 |)
we have that |D| = |BR | and hence,

1 [, D] 1 [, BR ] = 1 [, B1 ]R2 = 1 [, B1 ]|B1 |2/N |D|2/N .
FIGURE 1.1: An admissible nodal configuration for a(x).
Therefore,
lim 1 [, D] = .
|D|0
Although one might think of (1.3) as a sort of hierarchical ordering between

the components of 0 establishing that 0,j is larger than 0,j+1 for all 1
j q0 1, one should not forget that 1 [, D] also depends on certain
(hidden) geometrical properties of D. Setting
0 := 1 [, ],
(1.3) implies that
0 < 1 < 2 < < q0 ,
(1.5)
by the monotonicity of the principal eigenvalue with respect to the domain.

Although most of the properties of 1 [, D] invoked in this book can be
easily inferred from the variational characterization (1.2), the reader is sent to
Chapters 8 and 9 of [163], if necessary, where all the properties are collected
in a much more general setting.
As for the function f (x, u), in most of this book we will assume that it
satisfies the following structural assumption:
Hypothesis (Hf )
f C
( [0, )) satisfies f (x, 0) = 0 and u f (x, u) > 0 for all x
and u > 0.
,1+
Obviously, the function

f (x, u) = b(x)up ,
[0, ),
(x, u)
(1.6)
such that b(x) > 0 for all x .

In
satisfies (Hf) for all p 1 and b C ()
case (1.6), the function b(x) can be glued by a(x) as soon as b(x) > 0 for all
x . In many circumstances, f will be also required to satisfy
Hypothesis (Hg)
There exists g C
[0, ) with f (, u) g(u) for all u 0 such that
g(0) = 0, g 0 (u) > 0 for all u > 0 and limu g(u) = , where 0 = d/du.
1+
Note that (Hg) implies
lim f (x, u) = uniformly in x ,
and that the special choice (1.6) satisfies (Hg) with

g(u) = up ,
u 0.
Under these structural assumptions on a(x) and f (x, u), it will be established
in Chapter 4 that the set of positive steady states of (1.1), which are the
positive solutions of

du = u + a(x)f (x, u)u
in ,
(1.7)
u=0
on ,
consist of a curve of class C 1 , as a function of the parameter , that bifurcates
from u = 0 at = d0 and blows up to infinity in 0,1 as d1 . It turns
0,1 as
out that whether the positive solutions of (1.7) stay bounded in \
d1 depends on whether f , or the function g of (Hg), satisfies a further
condition reminiscent of J. B. Keller [123] and R. Osserman [202], which can
be stated as follows:
Hypothesis (KO)
f satisfies (Hf), (Hg), and, for every > 0, there exists u > g 1 (1/)
such that
Z
d
qR
I(u) :=
<
(1.8)
1
[ g(ut) 1] t dt
1
for every u u , and
lim I(u) = 0.
(1.9)
As
I(u) I(v)
if u > v > g 1 (1/),
it is apparent that I(u ) < implies (1.8) for all u u and, consequently,
the hypothesis (KO) can be shortly expressed as follows
Abbreviated hypothesis (KO)
f satisfies (Hf), (Hg), and limu I(u) = 0 for all > 0.
Throughout this book, (KO) will be referred to as the KellerOsserman
condition. It is an imperative condition to get uniform a priori estimates in
for the positive solutions of the equation
du = u + a(x)f (x, u)u.
(1.10)
It should be noted that (KO) holds if there exist p > 0 and C > 0 such that
g(u) Cup
for all
u 0.
(1.11)
Indeed, (1.11) implies that
[ g(ut) 1] t dt
( Cup tp 1) t dt
1 2

Cup p+2
1
1
p+2
2
2
1
> (p+2 2 ) = (p 1) > 0

2
2
=
provided

> 0,
u>
p+2
2C
p1
,
> 1.
Thus,
Z
I(u)
1

1 2
1/2
Cup p+2
1
1
d < ,
p+2
2
(1.12)
because the function

R() :=
1 2

Cup p+2
1
1 ,
p+2
2
1,
satisfies
R(1) = 0,
R0 (1) = Cup 1 > 0,
and, since p > 0,

Z
p+2
2
d =
R()
Cup
=
> 0,
p+2
p+2
lim
2
< .
p
Moreover, letting u in (1.12) yields (1.9). Therefore, (KO) indeed holds

under condition (1.11).
1.1
The meaning of the KellerOsserman condition
The condition (KO) is imposed so that, for every > 0, A > 0 and L > 0,
there is a unique value of
x := x(, A, L) > 0
for which the singular one-dimensional problem

du00 = u Ag(u)u
in (0, L),
u(0) = x, u0 (0) = 0,
u(L) = ,
(1.13)
possesses a positive solution. Consequently, by reflection about 0, the diffusive

logistic equation
du00 = u Ag(u)u
(1.14)
has a large, or explosive, solution in (L, L), i.e., a positive solution u such
that
lim u(t) = lim u(t) = +.
tL
tL
In such case, the large solution is unique.

As will become apparent in Chapter 3, under conditions (Hf) and (Hg)
the existence of these one-dimensional solutions allows the construction of
uniform a priori bounds for all the positive solutions of (1.14), even in the
general multidimensional problem, because large solutions are above any other
positive solution.
Multiplying the differential equation by v = u0 yields
dvv 0 + uu0 Ag(u)uu0 = 0,
or, equivalently,
d 2 2
v + u A
2
2
g(s)s ds =
0
2
x A
2
g(s)s ds.
(1.15)
The potential energy of the associated (u, v)system is given by

Z u
g(s)s ds,
u R.
P (u) := u2 A
2
0
As under condition (Hg) we have that g(0) = 0, g 0 (u) > 0 for all u > 0 and
limu g(u) = , there exists a unique > 0 such that g() = /A. Thus, 0
and are the unique equilibria of (1.14). Obviously,
P 0 (u) = u Ag(u)u,
P 00 (u) = Ag 0 (u)u Ag(u),
and hence,
P 00 (0) = > 0,
P 00 () = Ag 0 () < 0.
Thus, 0 is a center and is a saddle point. Moreover, P 0 (u) > 0 for all
u (0, ) and
g(u) > g() = /A for all u > .
Consequently, Ag(u) < 0 and
P 00 (u) < Ag 0 (u)u < 0
for all u > .
Therefore, there exists a unique u0 > such that P (u0 ) = 0, and

lim P (u) = .
10
So, the underlying phase portrait of the positive solutions of (1.14) looks like
Figure 1.2
In order to solve (1.13) we should find x > for which the half-upper
trajectory is run exactly in a time L. Rearranging (1.15) we find that

Z u
Z x
2
2
2
g(s)s ds
g(s)s ds
dv = (x u ) + 2A
0
0

Z u
Z u
Z u
Z u
A
= 2
s ds + 2A
g(s)s ds = 2
g(s)s ds
s ds
x
x
x
x
Z u
A
= 2
g(s) 1 s ds.
x
Consequently, performing the change of variable s = xt in the previous integral
yields

Z
2 2 u/x A
x
g(xt) 1 t dt.
v2 =
d
FIGURE 1.2: The phase plane of equation (1.14) for u > 0.
11
Therefore, the solution of the singular problem (1.13) is characterized through

Z
du
qR
L=
,
q

u/x
2
A
x
d x
g(xt) 1 t dt
1
or, equivalently, after performing the change of variable u = x,
r
Z
2
d
qR
L=
= I(x),
d
A
1
g(xt)
1
t
dt
1
where I(x) is the function defined in (1.8) with = A/. As g is increasing, I is
decreasing. Moreover, since (, 0) is an equilibrium, by continuous dependence
lim I(x) = .
x
Thus, if we impose (1.9), i.e.,

lim I(x) = 0,
then there is a unique x > 0 such that

r
I(x) =
2
L.
d
Therefore, (1.13) admits a unique positive solution. Consequently, in this context, the condition (KO) is equivalent to the existence of a unique x > 0
for which (1.13) admits a positive solution, independently of the size of the
positive constants , A and L.
1.2
Model in population dynamics
Throughout this book, for a given D of class C 1 and a function h C(D),

it is said that h is positive in D, h > 0, if h 0 but h =
6 0. Obviously, h < 0
it is said that h is strongly positive
if h > 0. Similarly, given h C 1 (D),
h
in D, h 0, if h(x) > 0 for all x D and n
(x) < 0 for all x D with
h(x) = 0, where n stands for the outward unit normal vector field along D.
we write
Finally, it is said that h 0 if h 0, and, given h1 , h2 C(D),
h1 > h2 if h1 h2 > 0 and h1 h2 if h1 h2 0.
In the context of population dynamics, the parabolic problem (1.1) models
the evolution of a single species u(x, t) randomly dispersed in the inhabiting
area , with constant dispersion rate d > 0. In such models, stands for
the intrinsic growth rate of u and a(x) measures the crowding effects of the
12
population in , while in 0 = int a1 (0) the species increases according to

the Malthus law of population dynamics, because a = 0 there in. In our setting,
the territory is fully surrounded by completely hostile regions, because we
u0 > 0 represents the
are imposing u = 0 on . The function u0 C(),
initial population distribution. So, (1.1) may be regarded as a prototype model
linking the Malthus and the logistic laws of population dynamics within the
same territory. Indeed, if f (x, u) = u, = and n := a satisfies n(x) > 0
then (1.1) provides us with the classical diffusive logistic problem
for all x ,
u
in (0, ),
t du = u n(x)u2
(1.16)
u=0
on (0, ),
u(, 0) = u0 > 0
in ,
whereas it provides us with the (linear)
u
t du = u
u=0
u(, 0) = u0 > 0
diffusive Malthusian problem

in (0, ),
on (0, ),
in ,
(1.17)
when 0 = . The main goal of this book is ascertaining the interplay between
these two angular laws of population dynamics when they arise simultaneously
in a heterogeneous environment.
As the unique solution of (1.17) is given through the heat semigroup by
the formula
u(, t; u0 ) := et(+d) u0 ,
it is apparent that
lim u(, t; u0 ) = 0
uniformly in
if < d0 , while
lim u(, t; u0 ) =
uniformly exponentially in compact subsets of , if > d0 . Indeed, let 0

be a principal eigenfunction associated with 0 . By the parabolic maximum
principle (e.g., L. Nirenberg [199]), we have that
u(, t; u0 ) := et(+d) u0 0
for all t > 0 and hence, there exists 0 < < C such that
< e+d u0 < C.
Thus,
u(, t; ) e(t+1)(+d) u0 u(, t; C)
for all t > 0. On the other hand, for every R, we have that
u(, t; ) = et(+d) = e(d0 )t
13
and hence,
e(d0 )t e(t+1)(+d) u0 Ce(d0 )t .
Consequently,
lim u = 0
with decay rate e(d0 )t
if < d0 ,
while
lim u = with exponential growth rate e(d0 )t
if > d0 .
In the intermediate case when = d0 , (1.17) possesses an entire half line of

positive steady-state solutions. Namely,
E+ := { : 0}.
Figure 1.3 represents the corresponding dynamics. The arrows provide us with
a scheme of the flow of (1.17). The thick half line at = d0 represents the
set of positive equilibria E+ .
The dynamics of the classical diffusive logistic problem (1.16) change drastically because, as a result of the theory developed in Chapter 2, for every
> d0 the problem (1.16) possesses a unique positive steady state, which
FIGURE 1.3: The dynamics of u according to the Malthus law.
14
will be denoted by [,] throughout this book, while zero is the unique nonnegative steady state for each d0 . Moreover, 0 is a global attractor for
all the positive solutions of (1.16) if d0 , while [,] is a global attractor
for the positive solutions if > d0 . Figure 1.4 represents the corresponding
dynamics. As a consequence from the results of Chapter 2, it will become
apparent that the curve 7 [,] is point-wise increasing and smooth, as
sketched in Figure 1.4. As in all the forthcoming dynamical schemes, the sense
of the arrows provides us with the direction of the flow of (1.16).
In the context of population dynamics, the previous results can be stated
as follows. Suppose the species has an intrinsic growth rate > 0 and the territory is sufficiently large, or enjoys the appropriate geometric properties,
so that d0 < . Then, u increases to infinity, exponentially, if it has Malthusian growth, while it approximates the limiting profile [,] if it governed by
the logistic law. Moreover, independently of the governing growth law, if the
territory is sufficiently small so that d0 , then the species is driven to
extinction inexorably.
Although in the classical cases that we have just described the dynamics
of (1.1) are governed by the non-negative steady states, or by if a = 0
and > d0 , in the general setting of this book, under Hypothesis (Ha) the
classical steady states of (1.1), given by the elliptic problem (1.7), cannot
describe the dynamics of (1.1) if d1 , because in such regimes u(, t; u0 )
must approximate a metasolution. The metasolutions of (1.1) are the extensions by infinity of the large (or explosive) solutions of equation (1.10) in the
FIGURE 1.4: The dynamics of u according to the logistic law.
15
components of

0,1
0,j
\

1 j q0 .
Note that

0,1
0,q .
= \
0
Under Hypotheses (Ha), (Hf), (Hg) and (KO), the main findings of the first
five chapters of this book, which constitute Part I, can be summarized as
follows:
If d0 , then the inhabiting region cannot support the species u.
If d0 < < d1 , then the species u exhibits a logistic growth in .
0,1 \;
If d1 < d2 , then the species u has Malthusian growth in
0,1 is of logistic type.
however its growth in \
If dj < dj+1 for some j {2, ..., q0 1}, then u has Malthusian
growth in

0,1
0,j \ ,

0,1
0,j .
but logistic growth in \
,
If dq0 , then the species u exhibits Malthusian growth in \
but logistic growth in .
In particular, both growth laws can coexist within a spatially heterogeneous
territory, where the asymptotic behavior of the species can be seriously affected by the geometric and the distribution of the several protection zones of
0 , whose components might be viewed as sort of natural protected areas for
the species u.
The next sections of this chapter contain some mathematical preliminaries
necessary to reach the main goals of Part I.
1.3
Characterization of the maximum principle
Subsequently, for a sufficiently smooth subdomain D of and a given bounded

and measurable potential V L (D), we consider the linear eigenvalue problem

(d + V )u = u
in D,
(1.18)
u=0
on D.
It is a very classical result (e.g., R. Courant and D. Hilbert [61]) that any
eigenvalue of (1.18) must be real and that (1.18) has an unbounded sequence
of eigenvalues
1 < 2 n . . .
16
whose associated eigenfunctions are total in L2 (D). Throughout this book,

the lowest, or principal, eigenvalue, 1 , will be denoted by 1 [d + V, D]. It
is well known that it satisfies the variational characterization (1.2). Moreover,
from (1.2) one can easily infer the following monotonicity properties:
If V1 < V2 , then 1 [d + V1 , D] < 1 [d + V2 , D].
If D0
D, then 1 [d + V, D] < 1 [d + V, D0 ].
The first property is usually referred to as the monotonicity of the principal

eigenvalue with respect to the potential, and the second one is known as the
monotonicity with respect to the domain. From the first one, the continuity of
the map
L (D)
R
V
7 1 [d + V, D]
readily follows, which will be refereed to as the continuity of the principal
eigenvalue with respect to the potential. In particular, the real function
7 1 [d + V, D]
is continuous.
A most sophisticated property, going back to R. Courant and D. Hilbert
[61] for general self-adjoint operators, establishes that if {Dn }n1 is a sequence
of smooth subdomains of D approximating a nice subdomain D0 D as
n , in the appropriate sense, then
lim 1 [d + V, Dn ] = 1 [d + V, D0 ].
(1.19)
This property will be called the continuity of the principal eigenvalue with
respect to the domain. This result was later refined by E. N. Dancer [64], J.
L
opez-G
omez [144] and S. Cano-Casanova and J. Lopez-Gomez [35, 37, 38]
to cover a general class of operators under rather general mixed boundary
conditions (see [163] for further details).
An important property of 1 [d + V, D] is the fact that it admits a
positive eigenfunction, , unique up to a multiplicative constant, and that it
is algebraically simple. Moreover, 0 as discussed in Section 1.2, and, since
d + V defines a self-adjoint operator in L2 (D), it is easy to see that
Z
n = 0,
n 2,
D
for any eigenfunction n associated to n . Consequently, the principal eigenvalue is the unique one which admits a positive eigenfunction. The associated
eigenfunction, , will be referred to as the principal eigenfunction of (1.18).
These results admit very general counterparts valid for wide classes of
linear second order elliptic operators, not necessarily selfadjoint, under general
mixed boundary conditions (see J. Lopez-Gomez [163]).
17
The next characterization of the maximum principle provides us with a pivotal property of the principle eigenvalue. It will be used very often throughout
the remainder of this book.
Theorem 1.1 Suppose D is an open subdomain of RN , N 1, of class C 2+ ,
Then, for every d > 0, the following
for some (0, 1], and V C (D).
assertions are equivalent.
(a) 1 [d + V, D] > 0.
such that h > 0 in D,
(b) There exists a function h C 2 (D) C 1 (D)
(d + V )h 0
in
D,
and either h|D > 0, or (d + V )h > 0 in D. Such a function is called

a positive strict supersolution of d + V in D (under Dirichlet
boundary conditions).
(c) The operator d+V satisfies the strong maximum principle in D,
and b C 2+ (D), with f 0,
in the sense that, for every f C (D)
satisfying
b 0 and (f, b) =
6 (0, 0), and any u C 2+ (D)

(d + V )u = f
in D,
u=b
on D,
necessarily u 0 in D, i.e.,
u(x) > 0
xD
and
u
(x) < 0
n
x u1 (0) D.
Theorem 1.1 goes back to J. Lopez-Gomez and M. Molina-Meyer [165]. Completely self-contained proofs for the scalar problem can be found in [144], [152].
The monograph of J. L
opez-Gomez [163] focuses on a refinement of this result
going back to H. Amann and J. Lopez-Gomez [13].
1.4
Existence through subsolutions and supersolutions
Throughout this section, D stands for a subdomain of class C 2+ of such

that a < 0 in D, i.e.,
D =
6 .
(1.20)
As an immediate consequence, from H. Amann [11] the next result holds.
[0, )), b C 2+ (D), b 0, and the
Theorem 1.2 Suppose f C , (
problem

du = u + af (, u)u
in D,
(1.21)
u=b
on D,
18
and a supersolution u C 2+ (D)

with
possesses a subsolution u C 2+ (D)
2+
0 u u in D. Then, (1.21) possesses a solution u C
(D) such that
0 u u u.
(1.22)
Moreover, (1.21) has a minimal and a maximal solution in [u, u], in the sense
such that:
that there exist u , u C 2+ (D)
(i) u and u solve (1.21) and satisfy u u u u.
of (1.21) with u u u satisfies
(ii) Any solution u C 2+ (D)
u u u .
Remark 1.3 If u (resp. u) is a strict subsolution (resp. supersolution) of
(1.21), then any solution u [u, u] satisfies u < u u (resp. u u < u).
As a consequence of Theorem 1.2, the next result holds.
[0, )) and u, u C 2+ (D) satisfy
Theorem 1.4 Suppose f C , (
0 u u in D and

du u + af (, u)u
in D,
lim
u(x) = .
du u + af (, u)u
dist (x,D)0
Then, the singular boundary value problem

du = u + af (, u)u
u=
in D,
on D,
(1.23)
possesses a solution u C 2+ (D) such that

uuu
.
Remark 1.5 By a solution of (1.23) it is meant any function u C 2+ (D)
solving (1.10) in D with
lim
u(x) = .
dist (x,D)0
In this book, these solutions are called large, or explosive, solutions of (1.10)
in D. Naturally, u (resp. u) is said to be a large subsolution (resp. large
supersolution) of (1.10) in D.
Proof of Theorem 1.4: For sufficiently large n 1, say n n0 , consider
Dn := {x D : dist (x, D) > 1/n} .
The integer n0 1 should be chosen so that Dn inherits the same regularity
properties as D. Since
u+u
u
u,
2
19
we find from Theorem 1.2 that, for every n n0 , the auxiliary problem

du = u + af (, u)u
in Dn ,
u = (u + u
)/2
on Dn ,
n ) such that
possesses a solution un C 2+ (D
u|Dn un u|Dn
in Dn .
(1.24)
Now, pick an integer k n0 + 1. As Dk Dn for all n k, (1.24) implies

u|Dk un u|Dk
in Dk
for all n k.
Consequently, thanks to the Schauder interior estimates (see Section 6.1 of

D. Gilbarg and N. S. Trudinger [103]), there exists a constant C = C(k) such
that
kun kC 2+ (D k1 ) C(k)
for all n k.
Thus, since the injection
k1 ) , C 2 (D
k1 )
C 2+ (D
is compact (e.g., J. L
opez-Gomez [163, p. 197]), there exists a subsequence,
{unm }m1 , of {un }nk such that
lim kunm u
k1 kC 2 (D k1 ) = 0
k1 ) such that
for some function u
k1 C 2+ (D
d
uk1 =
uk1 + af (, u
k1 )
uk1
in Dk1 .
Next, consider the sequence

{unm |Dk+1 }mm0
(1.25)
for a sufficiently large m0 1 such that nm0 k + 1. Arguing as above, there

exists a subsequence of (1.25), relabeled by nm , such that
lim kunm u
k kC 2 (D k ) = 0
k ) such that
for some function u
k C 2+ (D
d
uk =
uk + af (, u
k )
uk
in Dk .
By construction,
u
k |Dk1 = u
k1
in Dk1
for all
k n0 + 1.
(1.26)
Initializing this iterative scheme at k = n0 + 1, it is apparent that the limit

u := lim u
k
k
20
provides us with the desired solution of (1.23). Note that (1.24) implies
uu
k u
in Dk
for all k n0 + 1 and, consequently,
uuu
The proof is complete.
1.5
in D.
Some abstract pivotal results
The next result collects some important properties that are going to be used
throughout this book. As in the previous section, D is a subdomain of of
class C 2+ satisfying (1.20). So, a < 0 in D in the sense that a 0 but a =
6 0.
Lemma 1.6 Suppose f satisfies (Hf), b C 2+ (D), b 0, and u

u
C 2+ (D),
> 0, is a supersolution of (1.21). Then, u
0. In particular,
any positive solution u of (1.21) satisfies u 0. Moreover,
1 [d af (, u
), D] 0
and, actually,
= 1 [d af (, u
), D]
(1.27)
if, in addition, b = 0 and u

solves (1.21). Furthermore,
u is a supersolution
of (1.21) for all 1.
Proof: Since u
|D b 0 and
(d af (, u
)) u
0
in D,
u
> 0 is a positive supersolution of d af (, u
) in D under Dirichlet
boundary conditions. Two different situations may arise.
(i) If either u
> 0 on D, or
d
u
u af (, u
)
u>0
in D,
then, it follows from Theorem 1.1 that

1 [d af (, u
), D] > 0
and that u
0.
(1.28)
21
(ii) If u
= 0 on D and
d
u
u af (, u
)
u=0
then u
> 0 solves

du = u + af (, u)u
u=0
in D,
in D,
on D.
(1.29)
Thus, (1.27) holds, because

(d af (, u
))
u =
u,
and hence, must be the principal eigenvalue of d af (, u
). Moreover, since u
is a principal eigenfunction, necessarily u
0. Indeed, let
> 0 sufficiently large so that + > 0. Then,
(d af (, u
) + )
u = ( + )
u>0
in D
and u
= 0 on D. Consequently, by Theorem 1.1(c), u
0.
Finally, pick 1. Then,
bu
|D
u|D ,
whereas in D, we have
d
u
u + af (, u
)
u
u + af (,
u)
u,
because a < 0 in D and, owing to (Hf),
f (,
u) f (, u
)
This concludes the proof.
for all 1.
The next theorem will simplify extraordinarily the mathematical analysis

of this book.
Theorem 1.7 Suppose f satisfies (Hf), b C 2+ (D) satisfies b 0, (1.21)
possesses a supersolution u
> 0 and, in addition, > 1 [d, D] if b = 0.
Then, (1.21) has a unique positive solution, which will be throughout denoted
by [,D,b] . Moreover, the following properties are satisfied:
(a) For every positive subsolution (resp. supersolution) u (resp. u
) of (1.21),
u [,D,b]
(resp.
[,D,b] u
).
u0 > 0,
(b) For every u0 C(D),
lim ku[,D,b] (, t; u0 ) [,D,b] kC(D)
= 0,
where u[,D,b] (x, t; u0 ) stands for the unique solution of

u
in D (0, ),
t du = u + af (, u)u
u=b
on D (0, ),
u(, 0) = u0
in D.
(1.30)
(1.31)
22
Furthermore, if f satisfies (Hf), (1.21) admits a supersolution u

> 0, b = 0
and 1 [d, D], then (1.21) cannot admit a positive subsolution and
lim ku[,D,b] (, t; u0 )kC(D)
=0
for all u0 > 0.
(1.32)
In particular, (1.21) cannot possess a positive solution.

Proof: Suppose b > 0. Then, u := 0 is a strict subsolution of (1.21) and hence,
(0, u
) is an ordered sub-supersolution pair. Thus, by Theorem 1.2 and Remark
1.3, (1.21) possesses a solution, u, such that 0 1 [d, D]. Let 0 be a principal eigenfunction
associated with 1 [d, D]. Then, for sufficiently small > 0, u := is a
positive strict subsolution of (1.21). Indeed, |D = 0 and
d() = 1 [d, D] < + af (, )
in D
for sufficiently small > 0, because 1 [d, D] < and, thanks to (Hf),
lim kaf (, )kC(D)
= 0.
0
Fix one of these s and observe that u = 0. Since u

0, there exists
> 1 such that <
u. Thus, (,
u) provides us with an ordered subsupersolution pair of (1.21). Consequently, by Theorem 1.2 and Remark 1.3,
(1.21) possesses a solution, u, such that
< u
u.
Moreover, by Lemma 1.6, u 0. This completes the proof of the existence of
a positive solution for (1.21).
To prove the uniqueness we proceed by contradiction. Suppose (1.21) has
two different positive solutions
u1 =
6 u2 .
Then, by Lemma 1.6, u1 0, u2 0 and, thanks to the previous analysis,
there exist > 0, > 1 and a strict subsolution
u {0, }
such that
u < min{u1 , u2 } < max{u1 , u2 }
u.
According to Theorem 1.2, the problem (1.21) possesses a minimal positive
solution, u , and a maximal positive solution, u , in the order interval [u,
u].
Necessarily,
u 0.
By construction, w|D = 0 and
(d )w = af (, u )u af (, u )u
in D.
Moreover, setting
(t) := f (, tu + (1 t)u )(tu + (1 t)u ),
t [0, 1],
we have that
f (, u )u f (, u )u = (1) (0) =
0
1
Z
=
0
d
(t) dt
dt
f
(, tu + (1 t)u )(tu + (1 t)u ) dt w
u
Z 1
+
f (, tu + (1 t)u ) dt w.
0
Consequently, w solves the problem

(d + V )w = 0
w=0
in D,
on D,
where V is the potential defined by

Z 1
f
(, tu + (1 t)u )(tu + (1 t)u ) dt
V := a
0 u
Z 1
a
f (, tu + (1 t)u ) dt .
(1.33)
(1.34)
Necessarily w 0, as it is a principal eigenfunction associated to

1 [d + V, D] = 0.
Thus, since u u 0 and a > 0 in D, we find from (Hf) that
Z
V > a
f (, tu + (1 t)u ) dt > af (, u ).
Therefore, by the monotonicity of the principal eigenvalue with respect to the

potential, we find from (1.27) that
1 [d + V, D] > 1 [d af (, u ), D] = 0,
24
which is impossible. This contradiction ends the proof of the uniqueness.

Throughout this book, we will denote by [,D,b] the unique positive solution
of (1.21).
Suppose u > 0 is a subsolution of (1.21). Then, since u
0, for sufficiently
large > 1 we have that u <
u, and hence, by Lemma 1.6 and Theorem 1.2,
u]. By the uniqueness,
(1.21) has a positive solution in the order interval [u,
u.
u [,D,b]
Now, suppose b > 0. Then, u = 0 is a strict subsolution of (1.21) and, due to
Theorem 1.2, (1.21) has a positive solution u in the order interval [0, u
]. By
uniqueness,
[,D,b] u
,
(1.35)
as claimed by Part (a). Similarly, when b = 0, the function u := is a strict
subsolution of (1.21) for sufficiently small > 0. Thus, if is chosen so that
< u
, by Theorem 1.2, (1.21) has a positive solution, u, in [, u
]. Therefore,
again by uniqueness, (1.35) holds, which ends the proof of Part (a).
Next, we will prove (1.30). First, we assume that
b=0
and
> 1 [d, D].
Since u0 > 0, by the parabolic maximum principle, we have that

u[,D,b] (, t; u0 ) 0
for all t > 0.
Now, pick a sufficiently small > 0 and a sufficiently large > 1 such that
(,
u) is an ordered sub-supersolution pair of (1.21) with
< u[,D,b] (, 1; u0 ) <
u.
Enlarging > 1, if necessary, we can also assume that
u > [,D,b] .
By Lemma 1.6 and the uniqueness of the positive solution,
u must be a
positive strict supersolution of (1.21). Thanks again to the parabolic maximum
principle, we find from the semigroup property that
u[,D,b] (, t; ) u[,D,b] (, t; u[,D,b] (, 1; u0 ))
= u[,D,b] (, t + 1; u0 ) u[,D,b] (, t;
u)
(1.36)
for all t > 0. Moreover, according to D. Sattinger [220], since is a strict

subsolution of (1.21), the solution u[,D,b] (, t; ) increases approximating in
the minimal positive solution of (1.21) in [,
C(D)
u] as t . Similarly,
since
u is a strict supersolution of (1.21), u[,D,b] (, t;
u) decreases approx the maximal positive solution of (1.21) in [,
imating in C(D)
u] as t .
25
Since [,D,b] is the unique positive solution of (1.21), letting t in (1.36)

yields
lim u[,D,b] (, t; u0 ) = [,D,b]

in C(D),
t
which ends the proof of (1.30) when b = 0.

Now, suppose b > 0 and pick a sufficiently large > 1 such that
u > [,D,b]
and u := 0 < u[,D,b] (, 1; u0 ) <

u.
Then, arguing as above, we find that

u[,D,b] (, t; 0) u[,D,b] (, t + 1; u0 ) u[,D,b] (, t;
u)
(1.37)
for all t > 0. Thus, letting t in (1.37), (1.30) holds.

Finally, suppose
b=0
and 1 [d, D].
(1.38)
To prove that under condition (1.38) the problem (1.21) cannot admit a positive subsolution and hence, cannot admit a positive solution neither, we proceed by contradiction. So, suppose (1.21) has a positive subsolution u > 0. As
u
0, there exists > 1 such that u 0
in D
and hence, by the monotonicity of the principal eigenvalue with respect to the
potential,
= 1 [d af (, u), D] > 1 [d, D],
which contradicts (1.38). Therefore, (1.21) cannot admit a positive subsolution
under condition (1.38).
To complete the proof, it remains to show that (1.38) implies (1.32). Let
> 1 be such that
0 < u[,D,b] (, 1; u0 ) <
u.
Then, for every t > 0,
0 < u[,D,b] (, t + 1; u0 ) < u[,D,b] (, t;
u).
(1.39)
Since
u is a supersolution of (1.21), u[,D,b] (, t;
u) decreases approximating
the maximal non-negative solution of (1.21) in [0,
u] as t . As 0 is the
maximal non-negative solution of (1.21) in [0,
u], letting t in (1.39)
shows (1.32) and completes the proof.
The following strong comparison holds from Theorem 1.7.
26
Lemma 1.8 Suppose f satisfies (Hf), b C 2+ (D) satisfies b 0, (1.21)

possesses a supersolution u
> 0 and > 1 [d, D] if b = 0. Then, for every
positive strict subsolution (resp. supersolution) u (resp. u
) of (1.21), the next
estimate holds
(resp. [,D,b] u
),
u [,D,b]
where [,D,b] is the unique positive solution of (1.21).
Proof: Suppose u > 0 is a strict subsolution of (1.21). Then, by Theorem
1.7(a), u [,D,b] . Therefore,
u < [,D,b] ,
because u is not a solution of (1.21). Consequently,
w := [,D,b] u > 0.
(1.40)
On the other hand, by adapting the uniqueness argument of the proof of

Theorem 1.7, it becomes apparent that

(d + V )w = 0
in D,
(1.41)
w0
on D,
where V is the potential defined by
Z 1
f
V := a
(, t[,D,b] + (1 t)u)(t[,D,b] + (1 t)u) dt
0 u
Z 1
a
f (, t[,D,b] + (1 t)u) dt.
0
Subsequently, we distinguish two different situations. If w|D > 0, then w > 0

is a positive strict supersolution of d + V in D under homogeneous
Dirichlet boundary conditions. Consequently, by Theorem 1.1, w 0 in D
and therefore,
u [,D,b] .
If w|D = 0, then w > 0 provides us with a principal eigenfunction of
1 [d + V, D] = 0
and hence, w 0. Indeed, since

(d + V + 1)w = w > 0
w=0
in D,
on D,
and
1 [d + V + 1, D] = 1 > 0,
it follows from Theorem 1.1 that w 0. This ends the proof.
Finally, we conclude this section by providing an extremely useful property.
27
Lemma 1.9 Suppose f satisfies (Hf), D is a subdomain of of class C 2+

such that a < 0 in D, V L (D), and the problem

(d + V )u = u + af (, u)u
in D,
(1.42)
u=0
on D,
possesses a subsolution, u, such that u = 0 on D and u 0. Then,
> 1 [d + V, D].
(1.43)
Proof: Let 0 be an eigenfunction associated with 1 [d + V, D] and

choose a sufficiently large constant > 0 such that > u. Then,
(d + V )( u) = 1 [d + V, D] (d + V )u
1 [d + V, D] u af (, u)u
> 1 [d + V, D] u
= 1 [d + V, D]( u) + (1 [d + V, D] )u.
Consequently,
(d + V 1 [d + V, D])( u) > (1 [d + V, D] )u.
(1.44)
Suppose (1.43) fails. Then, it follows from (1.44) that u is a positive

strict supersolution of
L := d + V 1 [d + V, D].
Hence, according to Theorem 1.1, 1 [L, D] > 0. But
1 [L, D] = 1 [d + V 1 [d + V, D], D]
= 1 [d + V, D] 1 [d + V, D] = 0,
which is a contradiction. This ends the proof.
1.6
Logistic equation in population dynamics
Let us denote by p(t) the number of individuals of a population at time t 0

and suppose the instantaneous rate of variation of the population is proportional to the total population p(t) through some constant m R measuring
either the growth of the population, if m > 0, or its decline, if m < 0. Then,
the equation governing the evolution of the population is
p0 (t) = mp(t).
(1.45)
28
Thus, if the initial population equals p0 := p(0) > 0, then

p(t) = emt p0
for all t 0.
Hence, the time the population needs to double itself is given by the value of
T for which emT = 2. In other words,
T =
log 2
.
m
According to the celebrated essay on the Principle of Population of the political economist and Church of England priest, Thomas R. Malthus [187],
In the United States of America, where the means of subsistence have been
more ample, the manners of the people more pure, and consequently the checks to
early marriages fewer, than in any of the modern states of Europe, the population
has been found to double itself in twenty-five years. This ratio of increase, though
short of the utmost power of population, yet as the result of actual experience, we
will take as our rule, and say, that population, when unchecked, goes on doubling
itself every twenty-five years or increases in a geometrical ratio.
It becomes apparent that, around 1798, the neat birth rate of the population
of the United States of America was
m=
log 2
.
25
Actually, Th. R. Malthus [187] established a more general principle:

Assuming then my postulata as granted, I say, that the power of population
is indefinitely greater than the power in the earth to produce subsistence for man.
Population, when unchecked, increases in a geometrical ratio. Subsistence increases
only in an arithmetical ratio. A slight acquaintance with numbers will shew the
immensity of the first power in comparison of the second.
It was 36 years later, in 1835, when the astronomer and statistician A. Quetelet
[212] went back to the essay of Th. R. Malthus in his extremely pioneering
Essay on Social Physics, to complete it with the following, rather independent,
general principle:
The resistance, or the sum of the obstacles to the development of a population,
is like the square of the speed of variation of the population.
Nevertheless, neither the moral father of the population dynamics nor the
father of the scientific sociology translated his laws into mathematical terms.
Instead, it was the mathematician P. F. Verhulst [228], who, inspired by the
these pioneering ideas, added (translated by J. L. Mawhin [192]):
I have tried since a long time to determine, with the help of analysis, the credible
law of the population; but I have abandoned this type of research, because the
observation data are too scarce to allow the verification of the formulas. [...] However,
29
as the methodology I have followed seems to lead necessarily to the knowledge of the
true law, [...] I have thought that I have to comply the invitation of Mr. Quatelet of
making it public.
Then, denoting by p(t) the size at time t of the population, P. F. Verhulst

translated mathematically the Malthus law as (1.45) for some constant m,
and, crucially, observed that
As the speed of the increase of the population is diminished by its very increase, we must subtract from mp an unknown function (p) of p. [...] The simplest
hypothesis [...] consists in taking (p) = np2 .
This led him to the famous differential equation

p0 (t) = mp(t) np2 (t),
(1.46)
often known as the logistic equation, undoubtedly, the most paradigmatic one
of population dynamics. But, after P. F. Verhulst died, his contributions remained in almost complete oblivion for almost 80 years until the biologists
R. Pearl and L. L. Reed [207] rediscovered them in the same demographical
context and compared them with a series of field data. Since then, (1.46) is
also known as Verhulst equation, or Verhulst-Pearl equation.
Incidentally, perhaps by the lack of any mathematical background, the
essay of Th. R. Malthus [187] had a very strong influence on the work of
Ch. Darwin [69], who recognized that in October 1838 he had read it just for
entertainment. As he was very well prepared to appreciate the struggles for
life of animals and plants, he realized that the favorable variations should be
maintained, while the unfavorable ones should be destroyed; the final result
being the formation of new species. The deep influence of the celebrated essay
by Th. R. Malthus [187] on Ch. Darwin [69] can be easily documented by
simply reading the famous essay of Ch. Darwin [69] on The Origin of the
Species by Means of Natural Selection, where the genius recognized that
A struggle for existence inevitably follows from the high rate at which all organic beings tend to increase. Every being, which during its natural lifetime produces several eggs or seeds, must suffer destruction during some period of its life,
and during some season or occasional year, otherwise, on the principle of geometrical
increase, its numbers would quickly become so inordinately great that no country
could support the product. Hence, as more individuals are produced than can possibly survive, there must in every case be a struggle for existence, either one individual
with another of the same species, or with the individuals of distinct species, or with
the physical conditions of life. It is the doctrine of Malthus applied with manifold
force to the whole animal and vegetable kingdoms; for in this case there can be no
artificial increase of food, and no prudential restraint from marriage. Although some
species may be now increasing, more or less rapidly, in numbers, all cannot do so,
for the world would not hold them.
Seventeen years after the relevance of the work of P. F. Verhulst [228] was
30
recognized by R. Pearl and L. L. Reed [207], in 1937, the reaction diffusion

equation
p
dp = mp np2
(1.47)
t
with d > 0, was introduced by A. N. Kolmogorov, I. G. Petrovsky and N.
S. Piskunov [124], and independently by R. A. Fisher [88], to study some
problems of biological nature.
While in the non-spatial model (1.46), the evolution of the species does not
depend on the spatial location of the individuals in the inhabiting region in
the spatial model (1.47) the individuals are assumed to disperse randomly in
, like the particles of an ideal gas governed by a Brownian motion. So, each
individual moves around with no preference for a particular direction.
Rather astonishingly, random motion provokes a regular migration from
highly populated areas to less populated areas in complete agreement with the
heat transport law of the mathematician J. B. J. Fourier [89] and the mass
transfer laws governing most of chemical diffusion processes of the physiologist
A. Fick [87].
FIGURE 1.5: Brownian motion.

The influential essay of E. Shrodinger [223], What is Life? The Physical Aspect of the Living Cell, contains a very illuminating introduction to the role
played by the spatial dispersion in population dynamics, as well some rather
pioneering ideas which strongly influenced the evolution of biological sciences
during the twentieth century.
1.7
A function u : R is said to be (globally) Holder continuous in , with

exponent (0, 1], if there exists a constant C 0 such that
|u(x) u(y)| C|x y|
for all x, y .
31
Obviously, u is H
older continuous with exponent = 1 if, and only if, u is
(globally) Lipschitz continuous. Throughout this book, for any given (0, 1],
the Banach space of all continuous functions u :
R
we denote by C ()
that are H
older continuous in with exponent , endowed with the norm
|u(x) u(y)|
,
|x y|
x,y
kukC ()
:= kukC()
+ sup
u C (),
x6=y
where
kukC()
:= max ku(x)k.
More generally, for every integer k 1 and (0, 1], we denote by C k+ ()

k
k
the Banach subspace of C () consisting of all functions u C () such that
for all multi-index = (1 , ..., N ) NN with
D u C ()
|| = 1 + 2 + + N ,
where we are denoting
D u =
|| u
.
1 x1 N xN
is equipped with the norm

Naturally, the Banach space C k+ ()
kukC k+ ()
:= kukC k ()
+
X
||=k
|D u(x) D u(y)|
,
|x y|
x,y
sup
u C k+ (),
x6=y
where
kukC k ()
=
kD ukC()
.
||k
[0, ) R, two integers k1 , k2 0 and

Similarly, for a given function f :
k1 +1 ,k2 +2
1 , 2 (0, 1), it is said that f C

([0, )) if f (, u) C k1 +1 ()
k2 +2
for all u [0, ) and f (x, ) C

[0, M ] for all M > 0 and x .
In the one-dimensional problem, N = 1, one does not need to use the
spaces of H
older continuous functions for the validity of the results. Indeed,
in such case it suffices to impose
a C(),
[0, )),
f C 1 (
to get classical solutions of class C 2 everywhere.

Most of the results of this chapter are valid for a general linear second
order elliptic operator of the form
L = div (A(x)) + hb, i + c
where
A(x) = (aij (x))1i,jN
(1.48)
32
is a symmetric matrix with
aij = aji C 1+ (),
1 i, j N,
with bj C (),
b = (bj )1jN
Moreover, instead of Dirichlet boundary conditions, given

and c C ().
two open and closed pieces of the boundary, 0 and 1 , of class C 2+ with
= 0 1 , one can work with a general boundary operator
C 1+ ()
B : C 2+ ()
of mixed type

B :=
on 0 ,
on 1 ,
C 2+ (),
(1.49)
where = An is the co-normal vector field and C 1+ (1 ).

Actually, it is far from necessary to work within the setting of the Schauder
theory neither. Indeed, one can also work in a general domain of class C 2
with a general second order elliptic operators of the form (1.48) such that
aij = aji W 1, (),
bj , c L (),
1 i, j N,
as those introduced in Chapter 4 of [163]. Similarly, instead of Dirichlet boundary conditions, one can also consider a boundary operator B like those introduced on page 92 of [163]. Naturally, under these general conditions the
solutions will be strong, in the sense that they satisfy
\
u
W02,p (),
p>N
with all the pertinent implications of this feature (see [163, Ch. 5]).
This chapter has been elaborated from Section 3 of [160].
Chapter 2
Classical diffusive logistic equation
2.1
2.2
2.3
2.4
2.5
Unperturbed logistic problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Solution set for the unperturbed problem . . . . . . . . . . . . . . . . . . . . . . .
Perturbed logistic problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Structural stability as M > 0 perturbs from M = 0 . . . . . . . . . . . .
This chapter studies the dynamics of

u
t du = u + a(x)f (x, u)u
u=M
u(, 0) = u0 > 0
in D (0, ),
on D (0, ),
in D,
34
37
41
42
47
(2.1)
where M [0, ) is a constant, D is a subdomain of of class C 2+ such

that D , and f satisfies (Hf) and (Hg). In particular,
a(x) < 0
for all x D.
Under these conditions, (2.1) is a classical (diffusive) logistic problem, however

a(x) might vanish on some piece of D. The problem (2.1) is said to be
unperturbed if M = 0, and perturbed (from M = 0) if M > 0.
The main result of this chapter establishes that the dynamic of (2.1) is
governed by its maximal non-negative steady state, i.e., by the maximal nonnegative solution of the semilinear elliptic problem

in D,
(2.2)
u=M
on D.
In particular, it will be shown that, for every M > 0, (2.2) possesses a
unique positive solution, which is a global attractor for the solutions of (2.1).
Throughout this book, to be consistent with the notation introduced in Theorem 1.7, this solution will be denoted by [,D,M ] . One of the main results of
this chapter establishes that

0
if 1 [d, D],
lim [,D,M ] =
[,D]
if > 1 [d, D],
M 0
where
[,D] := [,D,0]
33
34
stands for the unique positive solution of

u=0
in D,
on D.
(2.3)
The problem (2.3) admits a (unique) positive solution if, and only if,
> 1 [d, D].
Moreover, the maximal non-negative solution of (2.3) is a global attractor for
(2.1) if M = 0. As a byproduct, there is a continuous transition between the
dynamics of (2.1) as M > 0 perturbs from M = 0.
, i.e., when
The analysis of (2.1) in the special case when D
a(x) < 0
for all x D,
is substantially easier than the analysis of the general case when D ,

because sufficiently large positive constants are supersolutions of (2.2) if D
, while no positive constant can be a supersolution of (2.2) if > 0 and
a(x) vanishes somewhere on D.
This chapter has been distributed in three sections. Section 2.1 studies the
unperturbed problem, Section 2.2 studies the perturbed problem, and Section
2.3 establishes the structural stability of (2.1) as M > 0 perturbs from M = 0.
The following corollary of Lemma 1.6 is very useful.
u > 0, is a solution of (2.2) for some
Lemma 2.1 Suppose u C 2+ (D),
M [0, ). Then, u 0 and
1 [d af (, u), D].
If, in addition, M = 0, then,
= 1 [d af (, u), D].
2.1
(2.4)
Unperturbed logistic problem
Throughout this section we assume that

M = 0.
The following result establishes the existence and the uniqueness of the positive solution of (2.3).
35
Theorem 2.2 Suppose M = 0 and f satisfies (Hf) and (Hg). Then, (2.3)
possesses a positive solution if, and only if,
> 1 [d, D].
Moreover, it is unique if it exists, and if we denote it by [,D] , then [,D] 0
and
if 1 [d, D],
0
lim u[,D] (, t; u0 ) =
in C(D),
t
[,D]
if > 1 [d, D],
where u[,D] stands for the unique solution of (2.1).
Proof: The existence and the uniqueness of u[,D] can be derived, e.g., from
D. Daners and P. Koch-Medina [68]. According to Theorem 1.7, to prove this
theorem it suffices to construct a positive supersolution of (2.3) for every
> 1 [d, D]. By (Hg), in the special case when
,
D
(2.5)
sufficiently large positive constants are positive supersolutions of (2.3). Consequently, for the rest of this proof we will assume that
D
and fix > 1 [d, D]. As D is of class C 2+ , it possesses a finite number of
components, say n 1. Let D,j , 1 j n, denote them, and, for sufficiently
small > 0 and j {1, ..., n}, consider the open subset of D defined by
D,j := {x D : dist (x, D,j ) < }.
For sufficiently small > 0, we have that
,i D
,j =
D
if i =
6 j,
n
[
D,j D.
j=1
Moreover, the open set D,j also is of class C 2+ for all j {1, ..., n} and,
since
lim |D,j | = 0,
1 j n,
0
we already know, by the analysis carried out at the beginning of Chapter 1,

that
lim 1 [d, D,j ] = ,
1 j n.
0
Thus, can be shortened, if necessary, so that

min 1 [d, D,j ] > .
1jn
(2.6)
36
For every 1 j n, let ,j 0 be a principal eigenfunction associated to

1 [d, D,j ]. As ,j (x) > 0 for all x D,j , it is apparent that
min
min
1jn DD/2,j
,j > 0.
Subsequently, we consider the auxiliary function defined through
/2,j ,
,j
in D
1 j n,
n
:=
[
/2,j ,
in Dint := D \
D
(2.7)
(2.8)
j=1
where is any C 2+ extension of the function

,1 ,n
to the open set
Dint = {x D : dist (x, D) > /2} b D
with the special requirement that
inf > 0.
(2.9)
Dint
Such a function exists because of (2.7).

We claim that provides us with a supersolution of (2.3) for sufficiently
large > 1. Indeed, by construction,
=0
on D =
n
[
D,j
(2.10)
j=1
and hence, = 0 on D for all > 0. Moreover, for every j {1, ..., n} and
> 0, we find from (2.6) that, in D/2,j ,
d() = d,j = 1 [d, D,j ],j
> ,j = + af (, ),
because
af (, ) 0.
Finally, in Dint , we have that
d() = d + af (, ) = + af (, )
for sufficiently large > 1, because a(x) is negative and bounded away from
zero in Dint and, owing to (2.9),
lim f (, ) =
uniformly in Dint .
37
Therefore, for sufficiently large > 1,

af (, )
By Theorem 1.7, the proof is complete.
2.2
in Dint .
Solution set for the unperturbed problem
The next result establishes that the set of positive solutions (, u) of (2.3)
consists of a differentiable curve emanating from u = 0 at = 1 [d, D],
where the former attractive character of u = 0 as a steady-state solution of
(2.1) is lost.
Theorem 2.3 Suppose M = 0 and f satisfies (Hf) and (Hg). Then, the solution operator
(1 [d, D], )
C(D)
() := [,D]
(2.11)
is of class C 1 and strongly increasing, in the sense that

() ()
if
> > 1 [d, D].
Moreover, () bifurcates from (, u) = (, 0) at = 1 [d, D], i.e.,

lim
() = 0
in
C(D).
(2.12)
1 [d,D]
the closed subspace of C(D)

formed
Proof: Subsequently, we denote by C0 (D)
with u = 0 on D. Then, the solutions of
by all continuous functions u in D
(2.3) are the zeroes of the nonlinear operator
C0 (D)
F : R C0 (D)
defined by
F(, u) := u (d)1 [u + af (, u)u] ,
where (d)1 stands for the resolvent operator of d in D under homogeneous Dirichlet boundary conditions. In other words, if G(x, y) stands for
the Green function of d in D, [110], whose existence is guaranteed by the
Perron theorem, [208], then
Z
F(, u) := u
G(, y)[u(y) + a(y)f (y, u(y))u(y)] dy
D
for all u C0 (D).
38
The operator F is of class C 1 and, by elliptic regularity, F(, ) is a nonlinear

compact perturbation of the identity map for every R. Moreover,
F(, 0) = 0
for all R
and
Du F(, 0)u = u (d)1 u
for all (, u) R C0 (D).
Thus, Du F(, 0) is a Fredholm analytic pencil of index zero whose spectrum

consists of the eigenvalues of d in D under homogeneous Dirichlet boundary conditions. In particular,
N [Du F(1 [d, D], 0)] = span[],
where 0 is any principal eigenfunction of 1 [d, D]. We claim that
D Du F(1 [d, D], 0) 6 R[Du F(1 [d, D], 0)].
(2.13)
Consequently, the transversality condition of M. G. Crandall and P. H. Rabinowitz [59] holds. The proof of (2.13) proceeds by contradiction. Suppose
D Du F(1 [d, D], 0) = (d)1 R[Du F(1 [d, D], 0)].
such that
Then, there exists u C0 (D)
u 1 [d, D](d)1 u = (d)1 .
and
By elliptic regularity, u C02+ (D)
(d 1 [d, D])u = .
Multiplying this equation by , integrating in D and applying the formula of
integration by parts gives
Z
2 = 0,
D
which is impossible. This contradiction shows (2.13). Therefore, by the theorem of M. G. Crandall and P. H. Rabinowitz [59], (, u) = (1 [d, D], 0)
is a bifurcation point from (, u) = (, 0) to a C 1 -curve of positive solutions
of (2.3). By the uniqueness of the positive solution, already established by
Theorem 2.2, (2.12) holds.
Now, let (, u) = (0 , u0 ) be a positive solution of (2.3). Then,
F(0 , u0 ) = 0
and, according to Lemma 2.1, u0 0 and
0 = 1 [d af (, u0 ), D].
(2.14)
39
Differentiating F with respect to u,

1
Du F(0 , u0 )u = u (d)

f
0 u + a (, u0 )u0 u + af (, u0 )u
u
Hence, Du F(0 , u0 ) is a Fredholm operator of index zero.

for all u C0 (D).
Moreover, as it is injective, it is a linear topological isomorphism. Indeed, if

f
u (d)1 0 u + a (, u0 )u0 u + af (, u0 )u = 0
u
then, by elliptic regularity, u C 2+ (D)
and
for some u C0 (D),
0
L(0 , u0 )u = 0
in D,
(2.15)
where
f
(, u0 )u0 af (, u0 ).
(2.16)
u
On the other hand, since D , we have a(x) < 0 for all x D. Thus, by
the monotonicity of the principal eigenvalue with respect to the potential, it
follows from (2.14) and (Hf) that
L(0 , u0 ) := d 0 a
1 [L(0 , u0 ), D] > 1 [d 0 af (, u0 ), D] = 0.
Hence, (2.15) implies u = 0. Consequently, Du F(0 , u0 ) is a linear topological
isomorphism. Therefore, combining the uniqueness of the positive solution of
(2.3), as a consequence from Theorem 2.2, with the implicit function theorem,
establishes the regularity of the solution operator (2.11). Finally, differentiating the identity
F(, ()) = 0,
> 1 [d, D],
with respect to yields
D = (d)

f
+ D + a (, )D + af (, )D ,
u
where = (), or, equivalently,

L(, ())D () = (),
> 1 [d, D].
As () 0 and
1 [L(, ()), D] > 0
for all > 1 [d, D],
we find from Theorem 1.1 that

1

f
() 0,
D () = d a (, ())() af (, ())
u
40
which concludes the proof. As [,D] is a strict subsolution of (2.3) if

1 [d, D] < < ,
the estimate
[,D] [,D]
can be also derived from Lemma 1.8.
Figure 2.1 illustrates the results established by Theorems 2.2 and 2.3. For
a given value of x D, it shows the curve
7 [,D] (x).
It bifurcates from the horizontal axis, u = 0, at = 1 [d, D] and it
increases for all further values of . The direction of the arrows represents the
flow of (2.1).
According to Theorem 2.2, u[,D] (x, t; u0 ) decays to zero as t if
1 [d, D], while it approximates [,D] (x) for all > 1 [d, D]. Thus,
the trivial equilibrium u = 0 of (2.1) is a global attractor if 1 [d, D],
while it is unstable for > 1 [d, D]. The stability lost by u = 0 as
FIGURE 2.1: The dynamics of (2.1) for M = 0.
41
crosses the critical threshold 1 [d, D] is gained by the positive solution

[,D] bifurcating from u = 0 at 1 [d, D], in complete agreement with the
exchange stability principle of M. G. Crandall and P. H. Rabinowitz [60].
2.3
Perturbed logistic problem
Throughout this section we assume that

M > 0.
In this case, the next result holds.
Theorem 2.4 Suppose M > 0 and f satisfies (Hf) and (Hg). Then, (2.2)
possesses a unique positive solution, [,D,M ] , for each R. Moreover,
lim u[,D,M ] (, t; u0 ) = [,D,M ]
in
C(D),
(2.17)
where u[,D,M ] (x, t; u0 ) stands for the solution of (2.1).

If, in addition, u > 0 (resp. u
> 0) is a strict subsolution (resp. supersolution) of (2.2), then
u [,D,M ]
(resp.
[,D,M ] u
).
Consequently, the estimates

0 < M1 M2 < ,
< 1 2 < ,
M2 M1 + 2 1 > 0,
imply
[1 ,D,M1 ] [2 ,D,M2 ] .
Moreover,
[,D] [,D,M ]
(2.18)
for all > 1 [d, D] and M > 0.

Proof: By Theorem 1.7, it suffices to construct a positive supersolution of
(2.2) for each R. As in the proof of Theorem 2.2, sufficiently large positive
. It remains to
constants provide us with such a supersolution if D
construct a supersolution in the general case when
D .
The construction is a variant of the one already carried out in the proof of
Theorem 2.2. Let D,j , 1 j n, denote the components of D and, for
sufficiently small > 0 and each j {1, ..., n}, consider the open subset
D,j := {x RN : dist (x, D,j ) < }.
42
For sufficiently small > 0, we have that

,i D
,j =
D
if i =
6 j,
D,j D,j
1 j n,
and the open set D,j also is of class C 2+ for all j {1, ..., n}. Moreover,
lim |D,j | = 0,
1 j n,
implies that
lim 1 [d, D,j ] = ,
0
1 j n.
Thus, as in the proof of Theorem 2.2, can be shortened to get (2.6).

For each 1 j n, let ,j 0 be a principal eigenfunction associated
to 1 [d, D,j ]. As ,j (x) > 0 for all x D,j , it is apparent that
min min ,j > 0
1jn D
and
min
min
1jn DD/2,j
,j > 0.
(2.19)
Thus, there exists 0 > 0 such that

> M
on D
for all
0 ,
where is given by (2.8). Reasoning as in the proof of Theorem 2.2, it is

apparent that, for sufficiently large > 1, the function u := provides us
with a strict supersolution of (2.2). This ends the proof.
As a byproduct of Theorem 2.4, the point-wise limit
[,D,] := lim [,D,M ]
M
in D
(2.20)
is well defined, though it might equal in some subregion of D if no further

restriction on f is imposed. The main goal of Chapter 3 is to establish how
(2.20) equals the minimal positive solution of the singular problem

in D,
u=
on D,
when, in addition, f satisfies Hypothesis (KO).
2.4
Structural stability as M > 0 perturbs from M = 0
The next result establishes the structural stability of (2.1) when M > 0 perturbs from M = 0.
43
Proposition 2.5 Suppose f satisfies (Hf) and (Hg). Then,
if 1 [d, D],
0,
lim [,D,M ] =
M 0
[,D] ,
if > 1 [d, D].
Proof: According to Theorem 2.4, the point-wise limit
0 0 := lim [,D,M ]
(2.21)
M 0
is well defined, because

0 [,D,M ] [,D,M ]
if
0<M <M
(2.22)
for all R. We claim that 0 provides us with a (classical) solution of (2.3).

> 0, there exists a constant C := C(M
) > 0
Indeed, by (2.22), for every M
such that
).
k[,D,M ] kC(D)
for all M (0, M
C
Thus, by the Schauder estimates (e.g., D. Gilbarg and N. S. Trudinger [103,
Th. 6.6]), there exists a positive constant C1 > 0 such that
k[,D,M ] kC 2+ (D)
C1
for all
).
M (0, M
As the injection
, C 2 (D)
C 2+ (D)
is compact (see J. L
opez-Gomez [163, p. 197]), and the point-wise limit (2.21)
and
is unique, necessarily 0 C 2 (D)
lim k[,D,M ] 0 kC 2 (D)
= 0.
M 0
Consequently, 0 0 solves (2.3) in D. Actually, by elliptic regularity, 0
C 2+ (D).
Thanks to Theorem 2.2, 0 = 0 if 1 [d, D]. Suppose
> 1 [d, D].
Then, letting M 0 in (2.18) yields
0 [,D] 0
and therefore 0 0. By the uniqueness of the positive solution of (2.3), we
conclude that 0 = [,D] . This ends the proof.
According to Proposition 2.5, the curve of maximal non-negative solutions
of (2.3) perturbs into the curve of positive solutions of (2.2) as the parameter
44
M > 0 leaves the level M = 0. In Figure 2.2, fixing M > 0 and x D, we

have represented the curve
7 [,D,M ] (x).
Thanks to Proposition 2.5, it approximates 0, as M 0, for all
1 [d, D], while it approximates [,D] (x) > 0 for all > 1 [d, D]. The
dashed curves in Figure 2.2 represent the non-negative solutions of (2.3). It
should be compared with Figure 2.1.
The next result provides us with an alternative proof of Proposition 2.5
through the implicit function theorem.
Proposition 2.6 Suppose f satisfies (Hf) and (Hg), and
R \ {1 [d, D]}.
Then, the solution operator
S
(0, )
M
7
C(D)
S(M ) := [,D,M ]
is of class C 1 . Moreover,

lim S(M ) =
M 0
0,
[,D] ,
if < 1 [d, D],

if > 1 [d, D].
FIGURE 2.2: The dynamics of (2.1) in case M > 0.
45
Proof: The change of variable

M R,
u = v + M,
transforms (2.3) into

dv = (v + M ) + af (, v+M )(v+M )
v=0
in D,
on D,
(2.23)
whose solutions are the zeroes of the nonlinear operator

C0 (D)
G : R C0 (D)
defined by
G(M, v) := v (d)1 [(v + M ) + af (, v + M )(v + M )]
= F(, v + M ) M
where F is the operator introduced in the proof of
for all (M, v) R C0 (D),
Theorem 2.3. The operator G is of class C 1 and, by elliptic regularity, G(M, ) is
a nonlinear compact perturbation of the identity map for all M R. Moreover,
G(0, 0)
G(0, [,D] M )
for all R,
for all > 1 [d, D],
=0
=0
for all (, v) R C0 (D),
Dv G(0, 0)v = v (d)1 v

and
Dv G(0, [,D] M )v = Du F(, [,D] )v,

Suppose
for all > 1 [d, D] and v C0 (D).
(2.24)
< 1 [d, D].

Then, Dv G(0, 0) is an isomorphism and, thanks to the implicit function theorem, there exist > 0, 0 < , and a map of class C 1 ,
v
[, ]
M
7
C0 (D)
v(M )
such that v(0) = 0,

G(M, v(M )) = 0
and
G(M, v) = 0
|M | + kvkC(D)

for all M [, ]

=
v = v(M ).
Differentiating (2.25) with respect to M at (M, v) = (0, 0) yields

DM G(0, 0) + Dv G(0, 0)DM v(0) = 0,
(2.25)
46
and hence,
DM v(0) (d)1 DM v(0) = (d)1 1.
is the unique solution of
Thus, by elliptic regularity, DM v(0) C 2+ (D)

(d )DM v(0) =
in D,
DM v(0) = 0
on D.
So,
DM v(0) = (d )1 1.
Thanks to Theorem 1.1, (d )1 is well defined because < 1 [d, D].
Consequently, setting
M [, ],
u(M ) := v(M ) + M,
(2.26)

DM u(0) = DM v(0) + 1 = 1 + (d )1 1
is the unique classical solution of

(d )DM u(0) = 0
DM u(0) = 1
in D
in D,
on D.
Therefore, by Theorem 1.1,

DM u(0) 0
in D.
(2.27)
As u(0) = 0, (2.27) implies u(M ) > 0 for sufficiently small M > 0. Thus,
by the uniqueness of the positive solution of (2.3), already established by
Theorem 2.4, we have
u(M ) = [,D,M ] = S(M )
for sufficiently small M > 0. Consequently, as u(M ) is of class C 1 in M , also
is S(M ) and the proof is complete in this case.
Now, suppose
> 1 [d, D].
In this case, in the proof of Theorem 2.3 we have already shown that
Du F(, [,D] ) is an isomorphism. Thus, thanks to (2.24), Dv G(M, [,D] M )
also is an isomorphism. Hence, by the implicit function theorem, there exist
> 0, 0 < , and a map of class C 1 ,
v
[, ]
M
7
C0 (D)
v(M )
such that v(0) = [,D] ,

G(M, v(M ) M ) = 0
for all M [, ],

and
G(M, v M ) = 0
|M | + kv [,D] kC(D)

47

=
v = v(M ).
As v(0) = [,D] 0, we have that

u(M ) = M + v(M ) 0
for sufficiently small M > 0. Therefore, by Theorem 2.4,
u(M ) = [,D,M ] = S(M ).
This completes the proof.
2.5
The classical diffusive logistic equation arises in the very special case when
, i.e.,
D
a(x) < 0
for all x D.
(2.28)
In such case, sufficiently large positive constants provide us with supersolutions of (2.2) for all M 0 and consequently, Theorem 1.7 applies to the
problems (2.2) and (2.3). If (2.28) fails, the large positive constants do not
provide positive supersolutions of (2.2).
The construction of the supersolutions in these degenerated situations in
order to prove Theorems 2.2 and 2.4, goes back to J. Lopez-Gomez [141].
Originally, they were introduced to prove the next property
lim 1 [d + V, D] = 1 [d, D0 ]
(2.29)
for all potential V > 0 such that D0 = V 1 (0) is a nice region. The identity
(2.29) has a number of rather striking applications in population dynamics. For
instance, it allowed us to establish that the principle of competitive exclusion
is false in the presence of refuges or protection zones because the species
can segregate to each of them as the intensity of the aggressions from the
competitors increases. Actually, we will use this property in Chapter 10.
The reader interested in this particular issue might wish to have a look
at J. L
opez-G
omez [143], J. Lopez-Gomez and J. C. Sabina de Lis [180], S.
Cano-Casanova and J. L
opez-Gomez [36], S. Cano-Casanova, J. Lopez-Gomez
and M. Molina-Meyer [41], and J. Lopez-Gomez and M. Molina-Meyer [168].
Originally, (2.29) was used the characterize the existence of principal eigenvalues for some general classes of linear weighted boundary value problems
(see J. L
opez-G
omez [141, 144, 163]).
48
The first time that these supersolutions were used to characterize the existence of positive solutions in a generalized logistic equation of degenerate type
was in J. M. Fraile et al. [90]. Some degenerated logistic equations, in the sense
that a(x) is allowed to vanish within the support domain, had been previously
studied by H. Brezis and L. Oswald [31] by means of some classical minimization techniques and by T. Ouyang [203] through global continuation, but J.
M. Fraile et al. [90] wrote the first paper in which the method of subsolutions
and supersolutions was incorporated to dealt with a very general class of degenerate logistic equations of diffusive type, and it was the first work where
the problem of analyzing the dynamics of the associated parabolic problem
was addressed successfully, as will become apparent in the next chapters.
This chapter has completed and refined the former results of Section 4 of
[160]. As in Chapter 1, all the results of this chapter are valid for wide classes
of linear second order elliptic operators like (1.48) under mixed boundary
conditions.
Chapter 3
A priori bounds in
3.1
3.2
3.3
3.4
3.5
Singular problem in a ball BR (x0 ) b . . . . . . . . . . . . . . . . . . . . . . . .

Singular problem in a general D . . . . . . . . . . . . . . . . . . . . . . . . . .
Existence of minimal and maximal solutions . . . . . . . . . . . . . . . . . . . .
Some sufficient conditions for (KO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
57
59
61
64
This chapter refines the classical a priori bounds of J. B. Keller [123] and R.
Osserman [202] to establish the existence of a minimal and a maximal positive
solution for the singular boundary value problem

in D,
(3.1)
u=
on D,
under condition (KO). As a byproduct, it will become apparent that
[,D,] := lim [,D,M ]
(3.2)
provides us with the minimal positive solution of (3.1), where [,D,M ] stands
for the unique positive solution of (2.2).
As in Chapter 2, throughout this chapter we assume that
D .
The next result establishes the strong positivity of any positive solution of
(3.1). By a solution of (3.1) we mean a function u C 2+ (D) such that
lim u(x) = ,
xD
d(x)0
where d(x) := dist (x, D)
for all
x RN .
These solutions are usually referred to as large or explosive solutions of

du = u + a(x)f (x, u)u.
Lemma 3.1 Suppose f satisfies (Hf) and u C 2+ (D), u > 0, solves (3.1).
Then, u(x) > 0 for all x D.
Proof: For sufficiently large n, the function u provides us with a positive
supersolution of d af (, u) in
Dn := {x D : dist(x, D) > 1/n}
49
50
such that u > 0 on Dn . Thus, thanks to Lemma 1.6, u 0 in Dn , which

concludes the proof.
Section 3.1 studies the problem (3.1) in the special, but pivotal case, when

,
D = BR (x0 ) := x RN : |x x0 | < R ,
D
(3.3)
for some x0 . The main result establishes the existence of a minimal
large solution of the problem in an arbitrary ball. Section 3.2 establishes the
existence of a minimal positive solution of (3.1) in the general case when D
. Section 3.3 shows the existence of a maximal solution and characterizes
it as the point-wise limit of the minimal solutions in smaller domains, Dn ,
approximating D from the interior as n . Finally, Section 3.4 gives a
sufficient condition for f to satisfy the KellerOsserman condition (KO).
3.1
Singular problem in a ball BR (x0 ) b
Throughout this section we impose (3.3). So,

max a < 0.
Its main result can be summarized as follows.

Theorem 3.2 Suppose (3.3) and assume f satisfies Hypothesis (KO). Then,
the point-wise limit (3.2) is finite in D and it provides us with the minimal
positive solution of (3.1) for all R.
Proof: As f satisfies (KO), it also satisfies (Hf) and (Hg). By (Hg), we have
d[,D,M ] = [,D,M ] + a(x)f (x, [,D,M ] )[,D,M ]
[,D,M ] + a(x)g([,D,M ] )[,D,M ]

max{1, }[,D,M ] + max a g([,D,M ] )[,D,M ]
D

maxD a
= max{1, } 1 +
g([,D,M ] ) [,D,M ]
max{1, }

maxD a
= max{1, }
g([,D,M ] ) 1 [,D,M ] .
max{1, }
Hence, [,D,M ] is a positive subsolution of the problem

du = h(u)
in D = BR (x0 ),
u=M
on D,
(3.4)
A priori bounds in
51
where
h(u) := (g(u) 1)u,
u 0,
(3.5)
with
:= max{1, } > 0
and := max a/ > 0.
By Theorem 2.4, (3.4) possesses a unique positive solution, M . Moreover,

[,D,M ] M
for all
M >0
(3.6)
and
M1 M2
if
0 < M1 < M2 .
Therefore, the point-wise limit

:= lim M
(3.7)
is well defined in D. Thanks to (3.6), the point-wise limit (3.2) is finite in D

provided is finite in D.
As (3.4) is invariant by rotations, M must be radially symmetric for each
M > 0, by uniqueness. Hence,
M (x) = M (r),
r := |x x0 |,
x D = BR (x0 ),
where M stands for the unique positive solution of

0 < r < R,
d 00 (r) + N r1 0 (r) = h((r)),
0 (0) = 0,
(3.8)
(R) = M.
The existence of a solution of (3.8) can be inferred by taking into account

that := 0 is a subsolution and that := C, with C > 0 constant, is a
supersolution of (3.8) provided
= C max{M, g 1 (1/)} > 0 = .
Necessarily, it is unique, as in the contrary case the problem (3.4) should
admit two radially symmetric solutions, which is impossible.
Subsequently, without loss of generality, we can assume that
M > u > g 1 (1/),
(3.9)
where u = u () is the constant arising in Hypothesis (KO). By (3.9), the

constant
u := g 1 (1/)
is a positive strict subsolution of (3.4), because h(u) = 0 and u < M on D,
by (3.9). As every constant u
= C > M provides us with a strict supersolution
of (3.4), it follows from Theorem 2.4 that
g 1 (1/) < M (x) = M (r)
for all x D.
(3.10)
52
Consequently, since h(z) > 0 for all z > g 1 (1/),

for all r [0, R].
h(M (r)) > 0
(3.11)
On the other hand, since for every z g 1 (1/)

h0 (z) = (g(z) + zg 0 (z) 1) zg 0 (z) > 0,
it is apparent that
h is increasing in [g 1 (1/), ).
Multiplying the differential equation of (3.8) by rN 1 and rearranging terms
yields
0
d rN 1 0M (r) = rN 1 h(M (r)),
0 < r < R.
(3.12)
Thus, integrating (3.12) in (0, r), shows that
Z r
d0M (r) = r1N
sN 1 h(M (s)) ds,
0 < r < R.
(3.13)
Consequently, according to (3.11) and (3.13) we find that

0M (r) > 0
for all r (0, R)
and hence, the map r 7 M (r) is increasing, as well as

r 7 h(M (r)),
0 r R,
because h is increasing in [g 1 (1/), ) and, thanks to (3.10),

M (r) > g 1 (1/)
for all r [0, R].
Therefore, it follows from (3.13) that

d0M (r)
1N
h(M (r))
0
sN 1 ds =
r
h(M (r))
N
(3.14)
for all r (0, R). Thus, going back to (3.8), we find from (3.14) that

N 1
N 1 0
00
M (r) d00M (r) +
h(M (r)).
h(M (r)) = d M (r) +
r
N
So,
h(M (r))
,
0 < r < R.
N
0, (3.8) also implies that
d00M (r)
Similarly, since 0M
d00M (r) h(M (r)),
0 < r < R.
A priori bounds in
53
Consequently,
h(M (r)) d00M (r)
h(M (r))
,
N
0 < r < R.
(3.15)
Multiplying (3.15) by 0M (r) and integrating in (0, r) gives

Z r
Z
Z r
1 r
0
00
0
h(M (s))M (s) ds d
h(M (s))0M (s) ds
M (s)M (s) ds
N 0
0
0
for all r [0, R). Thus, performing the change of variable
z = M (s),
0 s r,
and taking into account that 0M (0) = 0, it becomes apparent that

Z
Z M (r)
2 M (r)
2
h(z) dz
2
h(z) dz d (0M (r))
N M (0)
M (0)
(3.16)
for all r [0, R). Hence, taking the square root of the reciprocal of (3.16), we
find that
r
Z M (r) ! 12
Z M (r) ! 12
N
1
1
0
h
2
2
d M (r)
M (0)
M (0)
for all r (0, R). So, multiplying these inequalities by 0M (r), and integrating
in (r, R) yields
r Z R
Z R
1
N
0M (s)
Rr
0 (s)
qR
qR M
ds
ds
M (s)
M (s)
2 r
2 r
d
h
h
M (0)
M (0)
for all r [0, R). Consequently, the change of variable

u = M (s),
r s R,
transforms the previous inequalities into

r Z M
Z M
du
Rr
N
du
1
qR
qR

u
u
2 M (r)
2 M (r)
d
h
M (0)
M (0)
for all r [0, R).

Pick a r [0, R). As M is increasing, we have
M (r) M (0) > g 1 (1/),
which implies
Z
h
M (r)
h
M (0)
for all u M (r),
(3.17)
h
54
because h 0 in [g 1 (1/), ). Consequently, the second inequality of (3.17)

provides us with
r Z M
r Z
Rr
N
du
N
du
qR
qR
0<
du
u
u
2 M (r)
2 M (r)
d
h
h
M (0)
M (r)
for all r [0, R). In other words,

r
Rr
N
0<
J(M (r)),
2
d
0 r < R,
(3.18)
z > g 1 (1/).
(3.19)

Z
J(z) :=
z
du
qR
u
z
h(s) ds
For any given z > g 1 (1/), the successive changes of variable

u = z,
s = zt,
transform (3.19) into

Z
Z
Z
z d
d
z d
qR
qR
qR
.
=
=
J(z) =
z
h(zt)
1
1
1
dt
h(s)
ds
h(zt)z
dt
z
z
1
1
Hence, according to (3.5) and (1.8),
Z
1
d
I(z)
qR
J(z) =
= ,
( g(zt) 1) t dt
1
and therefore, (3.18) can be equivalently written as

s
N
Rr
I(M (r)),
0 r < R.
0<
2
d
Fix r [0, R) and consider the values M (r) for M > u . As the map
M 7 M (r)
is increasing, the limit
(r) := lim M (r)
M
is well defined. Moreover, by construction,

(x) = (|x x0 |)
where is the point-wise limit (3.7).
for all
x D = BR (x0 ),
(3.20)
A priori bounds in
55
Suppose (r) = . Then, letting M in (3.20), it becomes apparent

from the KellerOsserman condition (1.9) that
Rr
0 < 0,
d
which is impossible. Therefore,
(r) <
for all r [0, R).
Hence, < in D, as required.

On the other hand, since
(0) = lim M (0),
M
by continuous dependence with respect to the initial values, (r) must be

the unique solution of the Cauchy problem

0 < r < R,
d 00 (r) + N r1 0 (r) = h((r)),
(0) = (0),
0 (0) = 0.
Multiplying the differential equation by rN 1 , rearranging terms and integrating in (0, r) yields
Z r
d0 (r) = r1N
sN 1 h( (s)) ds > 0,
0 < r < R.
0
Thus, r 7 (r), 0 r < R, is increasing. So, the limit

(R) := lim (r)
rR
is well defined. By continuous dependence, M (R) = M , M > u should be

bounded if (R) < , which is impossible. Consequently,
lim (r) =
rR
and provides us with a radially symmetric positive solution of

du = h(u)
in D = BR (x0 ),
u=
on D.
(3.21)
Moreover, since is finite in D, due to (3.6) and (3.7), the function [,D,]
defined by (3.2) is finite in D. Fix (0, R/2). Then, thanks to (3.6),
[,D,M ]
in BR (x0 )
56
for all M > 0. Hence, by the Schauder interior estimates, there is a constant
C = C() > 0 such that
k[,D,M ] kC 2+ (BR2 (x0 )) C()
for all M > 0.
Thus, since the injection

R2 (x0 )) , C 2 (B
R2 (x0 ))
C 2+ (B
is compact and the limit (3.2) is uniquely determined, we find that
lim k[,D,M ] [,D,] kC 2 (BR2 (x0 )) = 0.
Consequently, [,D,] solves

(3.22)
R2 (x0 ). By elliptic regularity, we actually have that

in B
R2 (x0 )).
[,D,] C 2+ (B
As this holds for sufficiently small > 0, [,D,] must solve the singular
boundary value problem (3.1).
Lastly, we will prove that [,D,] is the minimal positive solution of (3.1).
In particular, must be the minimal positive solution of (3.21). Indeed, let
L be any positive solution of (3.1). Then, for every M > 0, there exists a
constant C > M and a sufficiently small > 0 such that
[,D,M ] C L
in the region R < |x x0 | < R.
(3.23)
By Theorem 2.4, it follows from (3.23) that also

[,D,M ] [,D,C] L
in BR (x0 )
for sufficiently large n 1. Consequently,

[,D,M ] L
in BR (x0 )
for all M > 0.
(3.24)
Therefore, letting M in (3.24) yields

[,D,] L
in BR (x0 ),
which shows the minimality of [,D,] and ends the proof.
A priori bounds in
3.2
57
Singular problem in a general D
As a consequence of Theorem 3.2 the next result holds.

Proposition 3.3 Suppose D is a subdomain of class C 2+ and f satisfies Hypothesis (KO). Then, for every R, the point-wise limit
[,D,] := lim [,D,M ]
M
in D
(3.25)
is finite and it provides us with the minimal positive solution of the singular
of (3.22)
problem (3.1). Moreover, for any positive solution u C 2 (D) C(D)
in D, necessarily
u [,D,]
in D.
(3.26)
R (x0 ) D, and set
Proof: Let x0 D be and R > 0 such that B
M := max [,D,M ] ,
M > 0.
BR (x0 )
According to Theorem 2.4,

[,D,M ] [,BR (x0 ),M +1]
in BR (x0 )
for all M > 0. Thus, due to Theorem 3.2, we have that

[,D,M ] [,BR (x0 ),] <
in BR (x0 ).
Hence, letting M , we find that

[,D,] [,BR (x0 ),] <
in BR (x0 ).
Consequently, since x0 and R > 0 are arbitrary, [,D,] must be finite in D

R (x0 ) D. Moreover, for each compact subset K D there
as soon as B
exists a constant C(K) > 0 such that
[,D,M ] C(K)
in K
for all
M > 0.
(3.27)
Subsequently, for every n N, we consider the open subset

Dn := {x D : dist (x, D) > 1/n} .
(3.28)
By construction,
n Dn+1 D ,
D
max a < 0,
n
D
D=
[
nm
Dn ,
(3.29)
58
for all m 1. Moreover, there exists n0 N such that Dn is a subdomain

n+1 is a compact
of D of class C 2+ for all n n0 . Pick n n0 . As K := D
subset of D, by (3.27), there is a constant C = C(n) > 0 such that
n+1
in D
[,D,M ] C(n)
for all
M > 0.
n Dn+1 , by the Schauder interior estimates, there exists a

Thus, since D
constant C1 = C1 (n) > 0 such that
k[,D,M ] kC 2+ (D n ) C1 (n)
for all M > 0.
Consequently, as the injection

n ) , C 2 (D
n)
C 2+ (D
is compact and the point-wise limit (3.25) is unique, it becomes apparent that
lim k[,D,M ] [,D,] kC 2 (D n ) = 0
for all n n0 .
Therefore, by (3.29), [,D,] solves (3.22) in D. Actually, due to (3.25),

[,D,] solves the singular problem (3.1). Moreover, by elliptic regularity,
[,D,] C 2+ (D).
Next, we will prove that [,D,] is the minimal positive solution of (3.1).
Indeed, let L be any positive solution of (3.1). Then, for every M > 0, there
exists a constant C > M such that, for sufficiently large n N,
[,D,M ] C L
in D \ Dn .
Thanks to Theorem 2.4, this estimate implies

[,D,M ] [,Dn ,C] L
in Dn
and hence,
[,D,M ] L
in D
for all M > 0.
Therefore, letting M yields

[,D,] L,
which shows the minimality of [,D,] .
be a positive solution of (3.22) and
To prove (3.26), let u C 2 (D) C(D)
denote
:= max u.
D
Then, for every M > , we have

u < M = [,D,M ]
on D,
and hence, due to Theorem 2.4,

u [,D,M ] [,D,]
in D.
A priori bounds in
3.3
59
Existence of minimal and maximal solutions
Proposition 3.3 can be sharpened to obtain the existence of minimal and

maximal solutions.
Theorem 3.4 Suppose D is a subdomain of class C 2+ and f satisfies
Hypothesis (KO). Then, (3.1) possesses minimal and maximal positive solumin
max
tions, denoted by L[,D] and L[,D] , respectively, in the sense that any other
positive solution L of (3.1) satisfies
min
max
L[,D] L L[,D] .
Moreover,
min
L[,D] = [,D,] := lim [,D,M ]
(3.30)
and
max
min
L[,D] = lim L[,Dn ] ,
(3.31)
where Dn , for sufficiently large n, are those defined by (3.28). See Figure 3.1.
min
max
FIGURE 3.1: Scheme of the construction of L[,D] and L[,D] .
60
Proof: The fact that (3.30) is the minimal positive solution of (3.1) has been
already established by Proposition 3.3. It remains to prove that (3.31) is the
maximal one.
Subsequently, for any M > 0 and sufficiently large n, say n n0 , we set
:= max [,Dn+1 ,M ] .
n
D
Then, by Theorem 2.4 and Proposition 3.3,

[,Dn+1 ,M ] [,Dn ,+1] [,Dn ,]
in Dn ,
n n0 ,
for all M > 0. Thus, letting M yields

[,Dn+1 ,] [,Dn ,]
in Dn ,
n n0 .
According to Proposition 3.3, this estimate can be equivalently written as

min
min
L[,Dn+1 ] L[,Dn ]
in Dn
for all n n0 .
(3.32)
Therefore, by (3.29), the point-wise limit

min
L := lim L[,Dn ]
n
in D
(3.33)
is well defined.
n Dn+1 for all n n0 , thanks to the Schauder
On the other hand, as D
interior estimates, it follows from (3.32) that, for every n n0 , there exists a
constant C(n) > 0 such that
min
kL[,Dm ] kC 2+ (D n ) C(n)
for all m n + 1.
Consequently, as the injection

n ) , C 2 (D
n)
C 2+ (D
is compact and the point-wise limit (3.33) unique, we find that
min
lim kL[,Dn ] LkC 2 (K) = 0
for all compact subset K D. Consequently, L C 2 (D) must be a solution

of (3.22) in D. By elliptic regularity, L C 2+ (D).
be an arbitrary positive solution of (3.1) and set
Let L
n := max L,
n
D
n n0 .
D provides us with a subsolution of the problem

Then, L|
n

in Dn ,
u = n
on Dn ,
A priori bounds in
61
for all n n0 . Hence, it follows from Theorem 2.4 and Proposition 3.3 that
[,D , ] Lmin
L
[,Dn ]
n n
in Dn
for all n n0 .
Consequently, letting n , we obtain from (3.33) that

L
L
in D.
Therefore, L must be the maximal positive solution of (3.1).
Under the general assumptions of this section, we conjecture that

min
max
L[,D] = L[,D] .
(3.34)
Part II of this book is devoted to the proof of (3.34), which entails the uniqueness of the solution of (3.1) in some special cases. The proof in the general
case remains an open problem.
3.4
Some sufficient conditions for (KO)
Suppose f satisfies (Hf) and (Hg). As, according to (Hg), g(u) is increasing,
the integral function I(u) defined by (1.8) is decreasing. Thus, if there exists
u > g 1 (1/) such that I(u ) < , the limit
I := lim I(u)
u
is well defined and it satisfies I [0, ). Consequently, if I(u ) < for

some u > g 1 (1/), the condition (KO) can be equivalently expressed as
I = 0.
(3.35)
The next result provides us with a simple condition ensuring (3.35).

Lemma 3.5 Suppose f satisfies (Hf) and (Hg), I(u ) < for some u >
g 1 (1/) and there exists u
> u such that
Q(u) := min
t1
g(ut)
,
g(
ut)
uu
,
(3.36)
satisfies
lim Q(u) = .
Then, (3.35) (and hence, (KO)) holds.
(3.37)
62
Proof: For every t 1 and u u

, we have that

g(ut)
g(ut)
g(
ut)
(g(ut) 1)t = g(
ut)
1 t=
g(
ut)
t.
g(
ut)
g(
ut)
g(ut)
According to (Ag),
g(
ut)
1
g(ut)
for all t 1 and u u

.
Thus, since u
> u > g 1 (1/),
g(
ut)
g(
ut)
g(
ut) 1 > 0
g(ut)
for all t 1.
Hence, for every t 1 and u u

,
(g(ut) 1)t Q(u)(g(
ut) 1)t
and therefore,
1
I(u) p
I(
u)
Q(u)
for all u u
.
(3.38)
As
I(
u) < I(u ) < ,
letting u in (3.38) yields (3.35).
The next result provides us with a very simple sufficient condition for
(3.37).
Lemma 3.6 Suppose f satisfies (Hf) and (Hg), I(u ) < for some u >
g 1 (1/) and there exists s0 > 0 such that
G(s) :=
g 0 (s)
s,
g(s)
s s0 ,
(3.39)
is non-decreasing in [s0 , ). Then, there exists u

> u such that the quotient
function Q(u) defined through (3.36) satisfies (3.37).
Proof: The result holds with the choice
u
:= max{u , s0 } + 1.
Indeed, for every u u
, set
(t) :=
g(ut)
,
g(
ut)
t [1, ).
Since g C 1 [0, ) and g(z) > 0 for all z > 0, C 1 [1, ) and
0 (t) = [g(
ut)]2 [g 0 (ut)g(
ut)u g 0 (
ut)g(ut)
u]
A priori bounds in
63
for all t 1. Moreover, since

ut u
t u
> s0 ,
and G is non-decreasing, we find that
G(ut) G(
ut)
Equivalently,
for all t 1.
g 0 (ut)
g 0 (
ut)
ut
u
t
g(ut)
g(
ut)
and hence,
0 (t) 0
for all t 1.
In particular,
(t) (1)
for all t 1,
and so,
Q(u) = min (t) = (1) =
t1
g(u)
.
g(
u)
Therefore, by (Hg),
lim Q(u) = .
In the special case when g(s) = sp , s 0, for some p > 0, we have that
G(s) :=
g 0 (s)
s=p>0
g(s)
for all s 0. Thus, G is increasing. Therefore, owing to (1.12), we find from

Lemmas 3.5 and 3.6 that condition (KO) holds.
Obviously, the function G is non-decreasing if
g0
= (log g)0
g
is non-decreasing, i.e., if g(s) is logarithmically convex for s s0 . When
g C 2 [0, ), this occurs if
g 00 g (g 0 )2
0
g2
in [s0 , ).
(3.40)
As g 0 (s) > 0 for all s > 0, (3.40) can be equivalently written in the form
gg 00
2
(g 0 )
in [s0 , ).
(3.41)
Logarithmically convex functions will arise again throughout Part II in connection with the problem of the uniqueness for the singular problem (3.1).
Although the function g(u) = eu satisfies (3.41), the function g(u) = up ,
u 0, p > 0, cannot satisfy it.
64
3.5
The first results concerning the existence of large solutions of

du = h(u)
(3.42)
in a bounded domain of RN with N 1 go back L. Bieberbach [28] and H.

Rademacher [216], who considered the equation
du = eu
in two and three dimensions, respectively. Later, these results were generalized
by J. B. Keller [123] and, independently, by R. Osserman [202] to any spatial
dimension and any increasing function h(u) satisfying
Z
dx
qR
< ,
(3.43)
x
0
h
0
which explains why these types of conditions are referred to in the specialized
literature as KellerOsserman conditions.
Essentially, in these classical papers it was established that if h C 1 [0, )
satisfies h(u) > 0 and h0 (u) 0 for all u 0, then (3.42) possesses a radially
symmetric solution globally defined in RN if, and only if,
Z
dx
qR
= .
x
0
h
0
As a consequence, it was inferred that if a simply-connected surface S has
a Riemannian metric with negative Gauss curvature everywhere, then S is
conformally equivalent to the interior of the unit circle (see J. B. Keller [123]
and R. Osserman [202]).
In the context of population dynamics, when dealing with the diffusive
logistic equation, the nonlinearity h(u) is given by
h(u) = Ag(u)u u
(3.44)
for some positive constants A, > 0, and a certain function g(u) satisfying
(Hg). So, the nonlinearity h(u) changes sign at
u = g 1 (/A) > 0.
As a result, the classical assumptions of J. B. Keller [123] and R. Osserman
[202] are not satisfied.
The most pioneering results for changing sign nonlinearities were those of
A. C. Lazer and P. J. McKenna [129, 130]. For the choice (3.44) it was assumed
A priori bounds in
65
in [130] that h C 1 [g 1 (/A), ), with h0 0 and h0 (u) non-decreasing for

sufficiently large u and that, in addition,
h0 (u)
lim inf qR u
u
> 0.
g 1 (/A)
(3.45)
Instead of these conditions, J. Lopez-Gomez [147] only imposed the existence

of some u > g 1 (/A) for which (1.8) and (1.9) hold, which is much weaker
than (3.45) for the special, but very important, choice
g(u) = up ,
u 0.
Indeed, we already know that (KO) holds for all p > 0, while a direct calculation shows that (3.45) fails if 0 g 1 (/A) and all u u (see [147]). Although at first glance
(3.46) might look different from (KO), they are equivalent. Indeed, the successive changes of variable x = u and z = ut show that
Z
Z
u d
u d
qR
qR
I(u) =
=
u
1
1
h(z) dz
h(ut)u dt
u
1
Z
I(u)
d
qR
=
= ,
1
(Ag(ut) )t dt
1
where I(u) is given by (1.8) with = A/. Consequently, (KO) is indeed
equivalent to (3.46).
In radical contrast with (3.43), for the choice (3.44) one has that
Z
dx
qR
I(g 1 (/A)) :=
= .
(3.47)
x
1
g (/A)
h
1
g (/A)
Consequently, the classical theory of J. B. Keller [123] and R. Osserman [202]
cannot be applied mutatis mutandis to deal with changing sign nonlinearities.
In order to prove (3.47), one should take into account that
h(z) = h0 (g 1 (/A))(z g 1 (/A)) + o(z g 1 (/A))
and that
h0 (z) = Ag 0 (z)z + Ag(z) ,
z 0,
as z g 1 (/A),
66
which implies
C := h0 (g 1 (/A)) = Ag 0 (g 1 (/A))g 1 (/A) > 0.
Thus, the integrand of (3.47) is given by
r
1
2
1
qR
1
x
C x g (/A)
h
as x g 1 (/A),
g 1 (/A)
which indeed entails (3.47).

It should be noted that all the results of this chapter still remain valid if
we substitute the differential operator d by d + V with V L (D).
Chapter 4
Generalized diffusive logistic
equation
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
Classical positive solutions in . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Associated inhomogeneous problems . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hierarchic chain of inhomogeneous problems . . . . . . . . . . . . . . . . . . . .
Large solutions of arbitrary order j {1, ..., q0 } . . . . . . . . . . . . . . . . .
Limiting behavior of the positive solution as d1 . . . . . . . . . . .
Direct proof of Theorem 4.8 when a C 1 () . . . . . . . . . . . . . . . . . . . .
Limiting behavior of the large solutions of order 1 j q0 1
as dj+1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Limiting behavior as of the large solutions . . . . . . . . . . . . .
This chapter characterizes the existence of positive solutions of

in ,
u=0
on ,
69
76
80
83
86
93
96
98
101
(4.1)
when a(x) and f (x, u) satisfy Hypotheses (Ha) and (Hf)(Hg), respectively, as
well as the existence of positive solutions for the following families of singular
boundary value problems
in j ,
u=0
on j ,
(4.2)
u=
on j \ ,
where f satisfies (KO) and

0,1
0,j ,
j := \
1 j q0 .
(4.3)
It should be remembered that

0,j =
mj
[
i0,j ,
1 j q0 ,
i=1
where, according to the notations of Hypothesis (Ha), i0,j , 1 j q0 ,

1 i mj , are the components of
0 = int a1 (0).
67
68
Those components were labeled to satisfy (1.3), i.e.,

j := 1 [, i0,j ],
j < j+1 ,
1 i mj , 1 j q0 ,
1 j q0 1.
(4.4)
By construction,
0 = .
q0 = \
(4.5)
Throughout the rest of this book, we will adopt the notation

q0 +1 := .
By the results of Chapter 3, f should satisfy (KO) so that (4.2) can admit a
positive solution. These solutions will regulate the dynamics of the parabolic
model (1.1) according to the different ranges of values of the parameter R,
which explains our interest in analyzing (4.2). Precisely, the solutions of (4.1)
regulate the dynamics of (1.1) for < d1 , while the solutions of (4.2) regulate
the dynamics of (1.1) for dj < dj+1 , if 1 j q0 1, and for dq0
if j = q0 . The solutions of (4.2) will be referred to as the large solutions of
order j of
in j .
FIGURE 4.1: An intricate nodal configuration for a(x).
Generalized diffusive logistic equation
69
Although in the special case when a(x) has the nodal behavior described by
Figure 1.1, we see that
0,1 =
0,2 ,
1 = \
2 = ,
and hence, 1 and 2 are connected, in general, some or several of the j s,

1 j q0 1, might not be connected. Indeed, suppose a(x) has the nodal
configuration sketched in Figure 4.1. In this example, has two components,
named ,1 and ,2 ; 0 consists of five components, named 0,1 , 0,2 , 10,3 ,
20,3 and 30,3 , and consists of two components, 1 and 2 . According to
Hypothesis (Ha), we are assuming that
1 = 1 [, 0,1 ] < 2 = 1 [, 0,2 ] < 3 = 1 [, i0,3 ],
1 {1, 2, 3}.
0,1 possesses two components.

In this example, 1 := \
The organization of this chapter is the following. Section 4.1 characterizes the existence of positive solutions of (4.1) and Section 4.2 studies the
inhomogeneous counterpart of (4.1). Section 4.3 studies a hierarchic chain of
classical inhomogeneous problems closely related to (4.2) and Section 4.4 uses
this analysis to establish the existence of minimal and maximal large solutions supported in j for each 1 j q0 . Sections 4.5 and 4.6 ascertain the
point-wise behavior of the positive solution of (4.1) as d0 by means of
two complementary technical devices. Section 4.7 characterizes the limiting
behavior of the minimal large solution of (4.2), 1 j q0 , as dj+1 . Finally, Section 4.8 studies the limiting behavior of the large positive solutions
of (4.2) as .
4.1
Classical positive solutions in
The next result characterizes the existence of positive solutions for (4.1). These
solutions regulate the dynamics of (1.1) within the range
d0 < < d1 .
It should be remembered that 0 := 1 [, ].
Theorem 4.1 Suppose f satisfies (Hf) and (Hg). Then, (4.1) possesses a
positive solution if, and only if,

d0 < < d1
<d<
.
(4.6)
1
0
Moreover, it is unique if it exists and if we denote it by [,] then the solution
operator
(d0 , d1 )
C()
(4.7)
7 () := [,]
70
is point-wise increasing and of class C 1 . Furthermore,

lim [,] = 0
d0
in
C(),
(4.8)
and
lim [,] =
d1
uniformly in compact subsets of
0,1 .
(4.9)
Thus, [,] bifurcates from u = 0 at = d0 , and blows up to infinity in 0,1

as d1 .
Figure 4.2 shows the map 7 [,] (x) for an arbitrary x 0,1 . According to Theorem 4.1, it bifurcates from 0 at = d0 , increases and blows up
as d1 . The dashed line emphasizes the loss of the stability of (, 0) as
crosses d0 .
Proof of Theorem 4.1: Suppose f satisfies (Hf) and (Hg), and u is a positive
solution of (4.1). Then, by Lemma 1.6, u 0 and
= 1 [d af (, u), ].
FIGURE 4.2: The map 7 [,] (x) for x 0,1 .
(4.10)
71
As af (, u) < 0 in , by the monotonicity of the principal eigenvalue with

respect to the potential, (4.10) implies
= 1 [d af (, u), ] > 1 [d, ] = d0 .
Similarly, by the monotonicity with respect to the domain, we find from (4.10)
that
< 1 [d af (, u), 10,1 ] = 1 [d, 10,1 ] = d1
because a = 0 in 10,1 . Therefore, (4.6) is necessary for the existence of a
positive solution of (4.1).
Now, we will prove that (4.6) is also sufficient for the existence of a positive
solution. Suppose (4.6). According to Theorem 1.7, it suffices to construct
a positive supersolution of (4.1). It should be noted that, thanks again to
Theorem 1.7, (4.1) cannot admit a positive supersolution for d1 . To
construct the supersolution we proceed as follows. For each 1 j q0 and
sufficiently small > 0 consider the open neighborhoods

i,j := x : dist (x, i0,j ) < ,
1 j q0 , 1 i mj ,
which have been represented in Figure 4.3 in the special case when a(x) has
the nodal configuration sketched in Figure 1.1. In Figure 4.3, ,1 consists of
0,1 1 and the set of points x such that dist (x, 1 ) < , and ,2
0,2 and the set of points x with dist (x, 2 ) < .
consists of
By the continuous dependence of the principal eigenvalues with respect to
the domain,
lim 1 [, i,j ] = 1 [, i0,j ] = j ,
0
1 j q0 ,
1 i mj .
Thus, by their monotonicity properties, it becomes apparent from (4.4) that

there is 0 > 0 such that, for every (0, 0 ),

d0 < < 1 [d, i,1 ] < d1 ,
1 i m1 ,
(4.11)
i
dj1 < 1 [d, ,j ] < dj ,
2 j q0 , 1 i mj .
By Hypothesis (Ha), consists of a finite number of components of
that simultaneously are components of , say j , 1 j n, if it is
non-empty. In the situation described by Figure 1.1, = , while in
the context described by Figure 4.1, 1 is the unique component of .
Suppose is non-empty and consider the open neighborhoods of
j defined by
N,j, := {x : dist (x, j ) < },
1 j n.
For sufficiently small > 0, N,j, is a C 2+ subdomain of such that

lim |N,j, | = 0
0
for all 1 j n.
72
Thus, 0 can be shortened, if necessary, so that

min 1 [d, N,j, ] >
1jn
for all 0 < < 0 .
(4.12)
Pick (d0 , d1 ), (0, 0 ), and, for every 1 j q0 and 1 i mj ,

let i,j 0 be a principal eigenfunction of 1 [d, i,j ]; it is unique up
to a multiplicative constant. Similarly, for every 1 k n, let ,k, be
a principal eigenfunction of 1 [d, N,k, ]. Then, consider the function
defined through
i
i
,j
in
1 j q0 , 1 i mj ,
/2,j ,
,k,/2 , 1 k n,
:=
(4.13)
,k, in N
in K := {x : dist (x, ) /2},

where is any smooth extension, positive and bounded away from zero, of
the function
mj
q0 O
n
O
O
i
,j
,k,
j=1 i=1
k=1
FIGURE 4.3: The -neighborhoods ,1 and ,2 .
73
to the compact set K . Note that exists because, for every 1 j q0

and 1 i mj , the function i,j is positive and bounded away from zero
on i/2,j , and, similarly, for every 1 k n, the function ,k, is
positive and bounded away from zero on N,k,/2 . Figure 4.4 shows a
genuine profile of when a(x) has the nodal configuration of Figure 1.1.
We claim that the function
u :=
(4.14)
is a supersolution of (4.1) for sufficiently large > 1. Indeed, by construction,

= 0
on ,
because, thanks to (Ha), any component of must be either a component

of , say k , where = ,k, = 0, or a component of 0 and, in this
case it is a component of i,j for some 1 j q0 and 1 i mj , where
also = i,j = 0.
Moreover, thanks to (4.11), for every 1 j q0 and 1 i mj , in i/2,j
we have that,
d() = di,j = 1 [d, i,j ]i,j > i,j
i,j + af (, i,j )i,j = + af (, )
for all > 0. Similarly, thanks to (4.12), for every 1 k n, in N,k,/2 we
have that
d() = d,k, = 1 [d, N,k, ],k, > ,k,
,k, + af (, ,k, ),k, = + af (, ).
Finally, note that
d() + af (, )
in K
FIGURE 4.4: The profile of the supersolution element .
74
if and only if
d
+ af (, )
in K .
(4.15)
As a(x) and (x) are positive and bounded away from zero in K b ,
by Hypothesis (Hg), (4.15) holds for sufficiently large > 1. Therefore, (4.14)
indeed provides us with a supersolution of (4.1) for sufficiently large . Consequently, by Theorem 1.7, (4.1) possesses a positive solution if, and only if, (4.6)
holds. Moreover, it is strongly positive and unique if it exists. Throughout the
rest of this book, we will denote it by
() = [,] = [,,0] .
The proof of Theorem 2.3 can be adapted mutatis mutandis to show (4.8) and
to establish the monotonicity of the solution operator (4.7). Actually, since
[,] is a strict positive supersolution of (4.1) for all (, d1 ), it follows
from Lemma 1.8 that
[,] [,]
if d0 < < < d1 .
So, the technical details of these proofs are omitted here.

To complete the proof of the theorem it remains to show (4.9). Fix
(d0 , d1 ) and 1 i m1 , and let > 0 be such that
[,]
> i0,1
in
i0,1 ,
(4.16)
where i0,1 0 is a principal eigenfunction associated with

1 = 1 [, i0,1 ].
Differentiating with respect to the realization of (4.1) at () and rearranging terms yields

L(, ())D () = ()
in ,
(4.17)
D () = 0
on ,
where
L(, ()) := d a
f
(, ())() af (, ()) .
u
(4.18)
Indeed, (4.17) follows straight away from the differentiability of the solution
operator 7 () = [,] , by adapting the proof of Theorem 2.3. Since
a
f
(, ())() > 0
u
in
= 1 [d af (, ()), ],
from the monotonicity of the principal eigenvalue with respect to the potential,
we can infer that
1 [L(, ()), ] > 1 [d af (, ()) , ] = 0.
75
Therefore, by Theorem 1.1, L(, ()) satisfies the strong maximum principle
in under Dirichlet boundary conditions. In particular, since () > 0 in ,
it follows from (4.17) that
D () = L1 (, ())() 0
(4.19)
and, consequently, the map 7 () is strongly increasing.

d1 ),
On the other hand, by (4.16), we find that, for every (,
[,] [,]
> i0,1
in
Hence, owing to (4.17),

(d ) D () = () > i0,1
D () 0
i0,1 .
in i0,1 ,
on i0,1 ,
(4.20)
Moreover,
because a = 0 in i0,1 and D () 0 in .
1 [d , i0,1 ] = d1 > 0
d1 ). Therefore, thanks again to Theorem 1.1, (4.20) implies
for all (,
D ()
in
where is the unique solution of

(d ) = i0,1
= 0
A direct calculation shows that
i
=
d1 0,1
i0,1 ,
(4.21)
in i0,1 ,
on i0,1 .
d1 )
for all (,
and hence, it follows from (4.21) that

+ log
() > ()
d1
i
d1 0,1
in i0,1 .
Consequently,
lim () = uniformly in compact subsets of i0,1 .
d1
As this occurs for all i {1, ..., m1 }, (4.9) holds, which ends the proof.
Remark 4.2 When either = , or =

6 but a(x) > 0 for
all x , the construction of the supersolution is easier, as in such
case in the construction of one does not need to invoke the eigenfunctions
,k, , 1 k n.
76
4.2
Associated inhomogeneous problems
The main result of this section is the next one.

Theorem 4.3 Let M > 0 and suppose a and f satisfy (Ha) and (Hf)(Hg),
respectively. Then, the problem

in ,
(4.22)
u=M
on ,

<d .
< d1
(4.23)
Moreover, it is unique if it exists, and if we denote it by [,,M ] , then

(a) For every < d1 and any positive strict subsolution (resp. supersolution) u (resp. u
) of (4.22),
u [,D,M ]
(resp.
u
[,D,M ] ).
(b) 0 [,,M1 ] [,,M2 ] in if

< < d1 ,
0 < M1 M2 ,
+ M2 M1 > 0.
(c) limd1 [,,M ] = uniformly in compact subsets of 0,1 .

(d) Naturally,

lim [,,M ] =
M 0
0,
[,] ,
if d0 ,
if d0 < < d1 ,
where [,] is the unique positive solution of (4.1).

(e) Furthermore,
lim [,,M ] = 0
uniformly in compact subsets of .
(4.24)
Proof: Since the solutions of (4.22) are strict supersolutions of (4.1), by

Theorems 1.7 and 4.1, the problem (4.22) cannot admit a positive solution
for d1 . Thus, < d1 is necessary for its existence. Suppose < d1 .
Then, again by Theorem 1.7, to establish the existence and the uniqueness of
a positive solution of (4.22), it suffices to construct a positive supersolution.
As the supersolution already constructed in the proof of Theorem 4.1, u = ,
with given by (4.13), for sufficiently large > 1, satisfies u = 0 on , we
77
should modify on a neighborhood of in order to get > 0 on . This

can be easily accomplished by taking the extended neighborhoods

i,j := x RN : dist (x, i0,j ) < ,
1 j q0 , 1 i mj ,
and
,k, := {x RN : dist (x, j ) < },
N
1 k n,
i,j
for sufficiently small > 0, instead of

and N,k, , respectively. As the
components of must lie in the interior of these open neighborhoods, and,
i ] and
consequently, the principal eigenfunctions associated with 1 [d,
,j
,k, ] are positive on them, the corresponding must be positive on
1 [d, N
and hence, for sufficiently large > 1, it becomes apparent that > M
on , which ends the proof of the existence and the uniqueness.
Part (a) is a straightforward consequence of Lemma 1.8, and Part (b) is a
corollary of Part (a). As an easy consequence, [,] [,,M ] for all M > 0.
Therefore, Part (c) follows from (4.9). The proof of Proposition 2.5 can be
adapted mutatis mutandis to show Part (d). It remains to prove Part (e).
Suppose < 0. Then, since a 0 and f 0, we have that
d[,,M ] 0
in
and hence, [,,M ] is subharmonic in . Thus, by the maximum principle,

M = max [,,M ] = max [,,M ]
and so, [,,M ] provides us with a positive strict subsolution of the linear
problem

du = u
in ,
(4.25)
u=M
on .
Since
1 [d ] = d0 > 0,
it follows from Theorem 1.1 that [,,M ] u , where u 0 stands for the
unique solution of (4.25). Consequently, it suffices to show that
lim u = 0
(4.26)
As the change of variable u = v + M transforms (4.25) in

d

in ,
+ 1 v = M
v=0
on ,
(4.26) follows from the fact that the unique solution of

( + 1) w = M
in ,
w=0
on ,
(4.27)
78
denoted by w , satisfies
lim w = M
(4.28)
Indeed, let K be compact and

:= dist (K, ) > 0.
Then, setting R := /2, for every x K we have that BR (x) . Let
0 be the principal eigenfunction associated to in BR := BR (0),
subject to homogeneous Dirichlet boundary conditions, normalized so that
(0) = kkC(BR ) = 1. Then, for every x K, x := ( x) provides us with
the principal eigenfunction of in BR (x) normalized so that
x (x) = kx kC(BR (x)) = (0) = 1.
Note that
for all x K.
1 [, BR (x)] = 1 [, BR ]
We claim that, for every x K,
w :=
is a supersolution of

M
x
1 [, BR ] + 1
( + 1) w = M
w=0
in BR (x),
on BR (x).
(4.29)
Indeed, by construction,
( + 1)w = M
in BR (x),
and w = 0 > M on BR (x). Moreover, w := M is a subsolution of (4.29)

for all x K. Therefore, since
1 [ + 1, BR (x)] = 1 [, BR ] + 1 > 0
for all x K and > 0, it becomes apparent (e.g., from Theorem 1.1) that
w = M w w =
M
x
1 [, BR ] + 1
in BR (x)
for all x K. Consequently,

M w (x)
M
1 [, BR ] + 1
for all x K.
As these global estimates, for every compact subset K , provide us with

(4.28), the proof is complete.
As a byproduct of Theorem 4.3, the next result holds.
79
Corollary 4.4 Suppose M > 0, a satisfies (Ha), and f satisfies (Hf) and
(Hg). Let j , 1 j m, be the components of and consider any nonempty proper subset J {1, ..., m}. Then, the problem
in ,
u=M
on jJ j ,
(4.30)
u=0
on j J
,
j
/
possesses a positive solution if, and only if, < d1 . Moreover, it is unique if
it exists and if we denote it by [,,M,J] , then:
(a) For every < d1 and any positive strict subsolution (resp. supersolution) u (resp. u) of (4.30),
u [,,M,J]
(resp.
u [,,M,J] ).
(b) 0 [,,M1 ,J] [,,M2 ,J] in if

< < d1 ,
0 < M1 M2 ,
+ M2 M1 > 0.
(c) limd1 [,,M,J] = uniformly in compact subsets of 0,1 .

(d) Naturally,

lim [,,M,J] =
M 0
0,
[,] ,
if d0 ,
if d0 < < d1 ,
where [,] is the unique positive solution of (4.1).

(e) Furthermore,
lim [,,M,J] = 0
Proof: Suppose (4.30) has a positive solution, , and > d0 . Then, u :=

provides us with a positive supersolution of (4.1) and hence, thanks to
Theorem 1.7, (4.1) possesses a positive solution. Thus, by Theorem 4.1, <
d1 . Therefore, < d1 is necessary for the existence of a positive solution of
(4.30). Suppose < d1 . Then, the unique positive solution of (4.22), [,,M ] ,
given by Theorem 4.3, provides us with a positive supersolution of (4.30).
Consequently, by Theorem 1.7, (4.30) possesses a unique positive solution,
denoted by [,,M,J] .
Part (a) follows from Lemma 1.8 and Part (b) is a byproduct of Part (a).
When d1 , (4.6) holds and hence, by Theorem 4.1, (4.1) has a unique
positive solution, [,] , which is a strict subsolution of (4.30). Thus, by Part
(a), we find that
[,] [,,M,J]
in .
(4.31)
80
Part (c) is a direct consequence from (4.9) and (4.31). As J is a proper subset
of {1, ..., m}, [,,M,J] is a strict subsolution of (4.22) and hence, due to
Theorem 4.3(a), we obtain that
[,,M,J] [,,M ]
in ,
which, combined with (4.31), yields

[,] [,,M,J] [,,M ]
in .
(4.32)
Therefore, Part (d) is a consequence from (4.32) and Theorem 4.3(d). Similarly, Part (e) follows from (4.32) and Theorem 4.3(e).
4.3
Hierarchic chain of inhomogeneous problems
The main goal of this section is analyzing the

elliptic boundary value problems

u=0
u=M
following family of semilinear

in j ,
on j ,
on j \ ,
(4.33)
where 1 j q0 and M > 0. The solutions of (4.2) will be constructed

from these solutions by letting M . The main result of this section can be
stated as follows.
Theorem 4.5 Suppose M > 0, a(x) satisfies (Ha), f satisfies (Hf)(Hg),
and 1 j q0 . Then, (4.33) possesses a positive solution, [,j ,M ] , if, and
only if,
< dj+1 .
(4.34)
Moreover, it is unique if it exists and
(a) For every < dj+1 and any positive strict subsolution (resp. supersolution) u (resp. u) of (4.33),
u [,j ,M ]
(resp.
u [,j ,M ] ).
(b) 0 [,j ,M1 ] [,j ,M2 ] in j if

< < dj+1 ,
0 < M1 M2 ,
+ M2 M1 > 0.
(c) limdj+1 [,j ,M ] = uniformly in compact subsets of 0,j+1 .
81
By (4.5), j = if j = q0 . In such case, since q0 +1 := , according

to Theorem 4.5, (4.33) possesses a unique positive solution for each R,
[, ,M ] .
Proof: Suppose j q0 1. Then, the lower order refuges of j , ordered by
the size of the principal eigenvalue of , are i0,j+1 , 1 i mj+1 . As
1 [d, i0,j+1 ] = dj+1 ,
1 i mj ,
Parts (a), (b) and (c) are immediate consequences from Corollary 4.4.
Suppose j = q0 . Then, the construction of a supersolution of (4.33) can
be easily accomplished by adapting the proofs of Theorems 4.1 and 4.3. Indeed, let j , 1 j n, be the components of and consider the open
neighborhoods of j , 1 j n, defined by
N,j, := {x : dist (x, j ) < },
1 j n.
For sufficiently small > 0, N,j, is a C 2+ subdomain of such that

lim |N,j, | = 0
0
for all 1 j n.
Thus, 0 can be shortened, if necessary, so that

min 1 [d, N,j, ] >
1jn
for all 0 < < 0 .
Now, for every 1 j n, let ,j, be a principal eigenfunction associated to

1 [d, N,j, ] and consider
,j,/2 , 1 j n,
,j, in N
:=
in K := {x : dist (x, ) /2},

where is any smooth extension, positive and bounded away from zero, of
the function
n
O
,j,
j=1
to the compact set K . Note that exists because, for every 1 j n,

,j, is positive and bounded away from zero on N,j,/2 . Arguing
as in the proofs of Theorems 4.1 and 4.3, it is easily seen that, under these
conditions, for sufficiently large > 1, the function u := provides us with
a supersolution of (4.33) for j = q0 . Therefore, when j = q0 , Theorem 1.7 and
Lemma 1.8 end the proof.
By letting M , the problems (4.33) provide us with the minimal positive solutions of (4.2). To construct the maximal ones we must shorten the j s
in order to construct the minimal large solutions supported in these shortened
82
open subsets. As in Theorem 3.4, the maximal solution of (4.2) is the limit of
these minimal large solutions.
It should be noted that, thanks to (Ha), for every 1 j q0 , the components of j either are components of or are common components of
and i0,k for some j + 1 k q0 and 1 i mk . Consequently,
j \ . Moreover, although j is not necessarily connected, it consists of a finite number of (disjoint) components of class C 2+ .
Subsequently, for sufficiently large n N, say n n0 , we also consider the
open subsets of j defined through
j,n := {x j : dist (x, j \ ) > 1/n}.
(4.35)
Also, we suppose that n0 has been chosen sufficiently large so that j,n is of
class C 2+ for all n n0 . By construction,
[
j,n j,n+1 j ,
j =
j,n .
nn0
Lastly, we also set

,n := q0 ,n ,
n n0 .
The next result is a direct consequence of Theorem 4.5.

Theorem 4.6 Suppose M > 0, a(x) satisfies (Ha), f satisfies (Hf)(Hg),
n n0 , and 1 j q0 . Then,
in j,n ,
u=0
on j,n ,
(4.36)
u=M
on j,n \ ,
< dj+1 ,
(4.37)
[,j,n ,M ] . Moreover, it is unique if it exists and

(a) For every < dj+1 and any positive strict subsolution (resp. supersolution) u (resp. u) of (4.36),
u [,j,n ,M ]
(resp.
u [,j,n ,M ] ).
(b) 0 [,j,n ,M1 ] [,j,n ,M2 ] in j if

< < dj+1 ,
0 < M1 M2 ,
+ M2 M1 > 0.
(c) If, in addition, j q0 1, then

lim [,j,n ,M ] =
dj+1
uniformly in compact subsets of 0,j+1 .
4.4
83
Large solutions of arbitrary order j {1, ..., q0 }
Throughout this section, the notation introduced in the previous ones will be
kept. Its main result can be stated as follows.
Theorem 4.7 Suppose a(x) satisfies (Ha), f satisfies (KO), and 1 j q0 .
Then, (4.2) possesses a positive solution if and only if < dj+1 . Moreover,
in such case, the point-wise limit
min
L[,j ] := lim [,j ,M ]
(4.38)
provides us with the minimal positive large solution of (4.2). Similarly, the
point-wise limit
max
min
L[,j ] := lim L[,j,n ]
(4.39)
n
is the maximal positive large solution of (4.2).

Furthermore, in case 1 j q0 1,
min
lim L[,j ] =
dj+1
uniformly in compact subsets of 0,j+1 .
(4.40)
Proof: Suppose 1 j q0 1 and < dj+1 , or j = q0 and R. Then, by

Theorem 4.5, [,j ,M ] is well defined. Moreover, M 7 [,j ,M ] is increasing
and hence, the point-wise limit (4.38) is well defined in j . To show that it is
the minimal positive large solution of (4.2), we proceed as follows. First, we
will prove that it is finite in . Indeed, set
bM := max [,j ,M ] M.
Since j , [,j ,M ] is a subsolution of

u = bM
in ,
on .
(4.41)
By Theorem 2.4, (4.41) has a unique positive solution, [, ,bM ] , and

[,j ,M ] [, ,bM ]
in .
(4.42)
According to Theorem 3.4, we already know that

min
L[, ] := lim [, ,b]

b
is the minimal positive large solution of

(4.43)
84
in . Thus, letting M in (4.41) yields

min
min
L[,j ] L[, ]
in ,
and hence,
min
L[,j ] <
in ,
(4.44)
min
as claimed above. Note that (4.44) establishes that L[,j ] < in j if j = q0 .

Suppose j q0 1 and < dj+1 , fix n n0 and set
bn :=
min
max L[,j ] .
j,n \
Then, since j,n \ , we find from (4.44) that bn < . Moreover,

thanks to (4.38),
bn
max [,j ,M ]
for all M > 0
j,n \
and so, [,j ,M ] is a subsolution of

u=0
u = bn
in j,n ,
on j,n ,
on j,n \ ,
(4.45)
for all M > 0. Since < dj+1 , by Theorem 4.6, (4.45) has a unique positive
solution, [,j,n ,bn ] , and
[,j ,M ] [,j,n ,bn ]
in j,n
(4.46)
for all M > 0. Consequently, letting M in (4.46) yields

min
L[,j ] [,j,n ,bn ]
in j,n
for all n n0 .
min
Therefore, L[,j ] < in j , as claimed above.

Furthermore, owing to (4.46), one can easily adapt the proof of Proposition
min
3.3 to prove that L[,j ] C 2+ (j ) provides us with a positive solution of
min
(4.2). Actually, L[,j ] is the minimal positive solution. Indeed, let L be any
positive solution of (4.2). Then, for every M > 0, there exist a constant C > 0
and an integer n N such that
[,j ,M ] C L
in
j,n .
j \
By Theorem 4.6, this estimate implies

[,j ,M ] [,j,n ,C] L
in
and consequently,
[,j ,M ] L
in
j,n ,
85
for all M > 0. Therefore, letting M yields

min
L[,j ] L,
min
which establishes the minimality of L[,j ] .

Next, we will show that the point-wise limit (4.39) is well defined and that
it provides us with the maximal positive solution of (4.2). As in the proof of
Theorem 3.4, for every M > 0 and n n0 , we set
b := max [,j,n+1 ,M ] .
j,n
Then, owing to Theorem 4.6, it follows from (4.38) that

min
[,j,n+1 ,M ] [,j,n ,b+1] L[,j,n ]
in
j,n
for all n n0 and M > 0. Thus, letting M yields

min
min
L[,j,n+1 ] L[,j,n ]
in
n n0 .
j,n
(4.47)
j,n
Hence, the point-wise limit (4.39) is well defined in j . Moreover, since
j,n+1 for all n n0 , by the Schauder interior estimates, it follows from (4.47)
that, for every n n0 , there exists a constant C(n) > 0 such that
min
kL[,j,m ] kC 2+ ( j,n ) C(n)
for all m n + 1.
Consequently, since the injection

j,n ) , C 2 (
j,n )
C 2+ (
is compact and the point-wise limit (4.39) unique, we find that
min
max
lim kL[,j,n ] L[,j ] kC 2 (K) = 0
max
for all compact subsets K j . Therefore, L[,j ] C 2 (j ) solves (4.43) in

max
j . By elliptic regularity, L[,j ] C 2+ (j ).
Let L be an arbitrary positive solution of (4.2) and set
bn := max L,
j,n
n n0 .
Then, L|j,n is a subsolution of

u = bn
in j,n ,
on j,n ,
for all n n0 . Hence, by Theorem 4.6 and (4.38), we find that

min
L [,j,n ,bn ] L[,j,n ]
in
j,n
n n0 .
86
Consequently, letting n yields

max
L L[,j ]
in j .
max
Therefore, L[,j ] indeed is the maximal positive solution of (4.2).

To conclude the proof of the existence, it remains to show that < dj+1
is necessary for the existence of a positive solution of (4.2) if 1 j q0 1.
Suppose 1 j q0 1 and let L be a positive solution of (4.2) for some R.
Then, for every n n0 , L|j,n provides us with a positive strict supersolution
of daf (, L) in j,n under homogeneous Dirichlet boundary conditions
and hence, by Theorem 1.1,
1 [d af (, L) , j,n ] > 0.
Therefore, by the monotonicity of the principal eigenvalue with respect to the
domain,
< 1 [ af (, L), j,n ] < 1 [d af (, L), 10,j+1 ] = dj+1 ,
because a = 0 in 10,j+1 . Consequently, (4.2) admits a positive solution if, and
only if, < dj+1 .
To complete the proof, it remains to show (4.40). Suppose j q0 1.
Then, due to Theorem 4.1, for every (d1 [, j ], dj+1 ), the problem

du = u + af (, u)u
in j ,
u=0
on j ,
has a unique positive solution, [,j ] . As [,j ] provides us with a positive
strict subsolution of (4.33) for all M > 0, by Theorem 4.5, it is apparent that
min
[,j ] [,j ,M ] L[,j ]
in
Therefore, (4.40) is a direct consequence from (4.9).
4.5
j .
Limiting behavior of the positive solution as d1
The next result ascertains the point-wise behavior of the unique positive solution, [,] , of (4.1) as d1 . It sharpens (4.9) substantially.
Theorem 4.8 Suppose a(x) satisfies (Ha) and f satisfies (KO). Then,
(
0,1 \ ,
in
min
lim [,] =
(4.48)
0,1 .
L[d1 ,1 ]
in 1 = \
d1
0,1 \ .
Moreover, [,] approximates uniformly in compact subsets of
87
Proof: Thanks to Theorem 4.5 with j = 1, [d1 ,1 ,M ] is well defined for all
M > 0. Set
M := max [,] ,
(d0 , d1 ).
1
Then, according to Theorems 4.1, 4.5 and 4.7,

min
[,] [d1 ,1 ,M ] lim [d1 ,1 ,M ] = L[d1 ,1 ]

M
in 1 .
(4.49)
On the other hand, since 7 [,] is point-wise increasing, the limit

L := lim [,]
in
d1
(4.50)
is well defined and, thanks to (4.9),

L=
in
0,1 .
Moreover, letting d1 in (4.49) yields

min
L L[d1 ,1 ]
in
and therefore, L is finite in 1 . Let 1,n , n

defined by (4.35). Due to (4.49), for every
Cn > 0 such that
[,] Cn
in
(4.51)
n0 , be the open subsets of 1

n n0 , there exists a constant
1,n+1
for all (d0 , d1 ). Thus, by the Schauder interior estimates, there exists a
constant Cn > 0 such that
k[,] kC 2+ ( 1,n ) Cn
for all
(d0 , d1 ).
Hence, as the injection

1,n ) , C 2 (
1,n )
C 2+ (
is compact and the point-wise limit (4.50) unique, necessarily
lim k[,1 ] LkC 2 ( 1,n ) = 0
d1
for all n n0 . Consequently, L C 2 (1 ) and it solves (4.43) in 1 . By elliptic

regularity, we actually have that L C 2+ (1 ).
According to (4.51), to prove the identity
min
L = L[d1 ,1 ]
(4.52)
it suffices to show that L = on 1 \ . This property follows easily from

lim
min [,] = .
d1 1 \
(4.53)
88
Indeed, suppose (4.53) has been proven. Then, setting

(d0 , d1 ),
m := min [,] ,
1 \
it follows from Theorems 4.1 and 4.6 that

[,] [,1 ,m ] [d0 ,1 ,m ]
in
for all (d0 , d1 ). Thus, letting d1 , we find from (4.38), (4.50) and
(4.53) that
min
L L[d0 ,1 ]
in 1 .
Therefore, L must solve (4.33) with j = 1 and = d1 . Consequently, (4.53)
implies (4.52). Note that (4.9) and (4.53) imply
lim [,] =
d1
0,1 \ .
in
Thus, 1/[,] , (d0 , d1 ), provides us with a decreasing family of contin 0,1 \ . Therefore, by Dinis
uous functions point-wise converging to zero in
0,1 \ . As a
theorem, it approximates zero uniformly in compact subsets of
0,1 \
byproduct, [,] approximates uniformly in compact subsets of
as d1 . Consequently, to conclude the proof of the theorem, it suffices to
prove (4.53). The proof will proceed by contradiction. Suppose (4.53) is not
true. As 1 \ consists of the components of i0,1 , 1 i m1 , which
are not components of , there exist some of these components, e.g., , such
that
b := min L = min L (0, ).
(4.54)
As is of class C
for every x ,
2+
1 \
, there exist R > 0 and a map Y : 0,1 such that,
BR (Y (x)) 0,1 ,
R (Y (x)) = ,
B
R (Y (x)) = {x}.
B
It should be noted that this can be done because, thanks to Theorem 1.9 of
[163], the open set 1 satisfies the uniform exterior sphere property in the
strong sense on , because is of class C 2 .
By (4.50) and (4.54), for each (d0 , d1 ), there exists x such that
[,] (x ) = min [,] min L = b.
Figure 4.5 sketches the construction of BR (Y (x )) for a(x) with the nodal
configuration of Figure 1.1. Note that, in that case, = 1 .
By construction,
dist (x , Y (x )) = R
for all
(d0 , d1 ).
89
Moreover, the manifold R defined by

R := { y 0,1 : dist (y, ) = 2R }
is a compact subset of 0,1 and hence, thanks to (4.9),
lim [,] = uniformly in R .
(4.55)
d1
Setting
0,1,R := {y 0,1 : dist (y, ) < 2R },
0,1,R
it turns out that 0,1,R consists of two components: and R . Let x

be such that
[,] (
x ) = min [,] min [,] = [,] (x ) min L = b.
0,1,R
We claim that x
= x for sufficiently close to d1 . Indeed, suppose x

0,1,R . Then,
[,] (
x ) = 0 ,
[,] (
x ) 0.
FIGURE 4.5: The ball BR (Y (x )).
90
But, since 0,1,R 0,1 and a = 0 in 0,1 , we have that

d[,] (
x ) = [,] (
x ) > 0
for all (d0 , d1 ), because > 0 in such range, which is impossible. Thus,
x
0,1,R = R
for all (d0 , d1 ).
Therefore, by (4.55), there exists

1 (0 , 1 ) such that
min [,] = min [,] = [,] (x ) b for each [d
1 , d1 ).
0,1,R
Consequently, x
= x and
[,] (x) [,] (x )
for all
R (Y (x )).
xB
(4.56)
Subsequently, for every > 0 and [d

1 , d1 ), we will consider the barrier
function defined by
2
(x) := e|xY (x )| eR ,
R (Y (x )).
xB
A direct calculation shows that, for every x BR (Y (x )),

2
2
(d ) (x) = 2N d 42 d|x Y (x )|2 e|xY (x )| + eR
and hence, for every
R/2 (Y (x )),
x AR := BR (Y (x )) \ B
we have that

2
2
(d ) (x) 2N d42 d|x Y (x )|2 e(R/2) + eR

2
2N d 42 d|x Y (x )|2 e(R/2)
for sufficiently large > 0. Consequently, there exist > 0 and > 0 such
that
(d )
in AR
(4.57)
for all [d
1 , d1 ). Throughout the rest of the proof, we will assume that
has been chosen to satisfy (4.57).
R/2 (Y (x )) is a compact subset of 0,1 , (4.9) guarantees that
As B
lim
min
R/2 (Y (x ))
d1 B
[,] = .
Thus, setting
c :=
minBR/2 (Y (x )) [,] [,] (x )

eR2 /4 eR2
91
we have that
lim c = ,
(4.58)
d1
because
[,] (x ) b for all
[d
1 , d1 ).
Moreover, by the definition of c , for every [d

1 , d1 ), we have that

2
2
R/2 (Y (x )). (4.59)
[,] (x) [,] (x ) + c eR /4 eR
xB
Subsequently, for every [d
1 , d1 ), we consider the auxiliary function
v := [,] [,] (x ) c
in AR .
According to (4.59),
v 0
on BR/2 (Y (x )).
Moreover, as = 0 on BR (Y (x )), it follows from (4.56) that

v = [,] [,] (x ) 0
on BR (Y (x )).
Consequently,
v 0
on AR
for all [d
1 , d1 ).
(4.60)
On the other side, by (4.57), in the annular region AR we have that

(d ) v = [,] (x ) c (d ) [,] (x ) + c .
Hence, thanks to (4.58), it becomes apparent that
(d ) v > 0
in AR
(4.61)
for < d1 sufficiently close to d1 . Therefore, since

< d1 = 1 [d, i0,1 ] < 1 [d, AR ],
1 i m1 ,
owing to Theorem 1.1, we can infer from (4.60) and (4.61) that
v (x) > 0
if R/2 < |x Y (x )| < R.
Consequently,
[,] (x) [,] (x ) + c (x)
if < d1 is sufficiently close to d1 .
Subsequently, we set
Y (x ) x
.
n :=
R
for all x AR
(4.62)
92
By definition,
[,]
[,] (x + t n ) [,] (x )
(x ) = lim
.
t0
n
t
Moreover, by (4.62), we find that, for every t (0, R/2),
[,] (x + t n ) [,] (x )
c (x + t n )
t
t

2
|x +t n Y (x )|2
c e
eR
=
t

2
|t n R n |2
c e
eR
=
t

2
2
c e(Rt) eR
=
.
t
Hence, since
2
lim
t0
we obtain that
2
e(Rt) eR
= 2ReR ,
t
[,]
2
(x ) 2ReR c .
n
Therefore, thanks to (4.58), we find that

lim
d1
[,]
(x ) = .
n
(4.63)
Now, for each < d1 , d1 , consider the boundary value problem
0,1 ,
in 1 = \
du = u + af (, u)u
u = [,] (x )
on ,
(4.64)
u=0
on 1 \ .
As < d2 , applying Corollary 4.4 with = 1 , it is apparent that (4.64)
possesses a unique positive solution, denoted by
:= [,1 ,[,] (x ),] .
Moreover, since [,] |1 provides us with a positive supersolution of (4.64),
we find that
1.
[,]
in
Therefore, since
(x ) = [,] (x ),

we find that
93
[,]
(x )
(x ).
n
n
Consequently, by (4.63), we also have that

lim
d1
(x ) = ,
n
as d1 , the
which is impossible, because must approximate in C 1 (D),
unique positive solution
d1 := [d1 ,1 ,b,]
of the problem
du = d1 u + af (, u)u
u=b
u=0
in 1 ,
on ,
on 1 \ .
This contradiction shows (4.53) and completes the proof.
4.6
Direct proof of Theorem 4.8 when a C 1 ()
Throughout this section, as in the proof of Theorem 4.1, we will denote

i,1 := x : dist (x, i0,1 ) < ,
1 i m1 ,
for sufficiently small > 0. The main goal of this section is to give a direct
proof of Theorem 4.8 when
lim max
0 1im1
supi,1 \ i0,1 |a(x)|
= 0.
(4.65)
This condition holds if, for instance, a C 1 in a neighborhood of 0,1 \ ,

because, in such case,
a(x) = 0
and
a(x) = 0
for all x i0,1 \ ,
1 i m1 .
Also, as in the proof of Theorem 4.1, for sufficiently small 0, we denote by

i,1 0 the principal eigenfunction associated to 1 [, i,1 ], normalized
so that
ki,1 kC( i,1 ) = 1,
1 i m1 .
According to J. L
opez-G
omez and J. C. Sabina de Lis [181, Th. 3.2], for every
1 i m1 , we have that
1 [, i,1 ] = 1 [, i0,1 ] + i + O( 2 )
as 0
(4.66)
94
where
Z
i :=
i0,1 \
i0,1
ni
!2
dS < 0,
because i,1 is a holomorphic perturbation of i0,1 ; ni stands for the outward

unit normal to i0,1 along its boundary.
defined by
Subsequently, we consider the function u C()
Ci,1 (x)
if x i,1 , 1 i m1 ,
m1
u (x) :=
(4.67)
[
0
if
x
,1
i=1
where C() > 0 is a constant to be chosen later so that u becomes a weak

subsolution of (4.1). Pick satisfying
1 [d, i,1 ] < 1 [d, i ,1 ] < < 1 [d, i0,1 ] = d1 ,
(4.68)
for sufficiently small > 0 and all 1 i m1 . According to a classical

result of H. Berestycki and P. L. Lions [25], it is easily seen that u is a weak
subsolution of (4.1) if, and only if, for every 1 i m1 ,
a(x)f (x, Ci,1 (x)) 1 [d, i,1 ]
for all x i,1 ,
i
1
because u (x) = 0 and, so, f (x, u (x)) = f (x, 0) = 0 for all x \ m
i=1 ,1
and, according to (4.68), > 1 [d, i,1 ] for all 1 i m1 . Thus, by
(4.68) and (Ha), a sufficient condition is the following
i .
a(x)f (x, Ci,1 (x)) 1 [d, i ,1 ] 1 [d, i,1 ] x i,1 \
0,1
2
The appropriate choice of C requires analyzing the decay rate as 0 of the

several quantities arising in these estimates. First, note that, thanks to (4.66),
we have that
i
as 0.
(4.69)
1 [d, i ,1 ] 1 [d, i,1 ] = d + O( 2 )
2
2
Moreover, by (Hf) and (Hg), the function
F () := max f (x, ),
0,
is non-decreasing and it satisfies

lim F () = .
(4.70)
Also, since the eigenfunctions i,1 , 1 i m1 , have linear decay towards

zero on i,1 , reasoning as in J. Lopez-Gomez and J. C. Sabina de Lis [181],
95
it becomes apparent that there are two positive constants C2 > C1 > 0,
i \ ,
independent of 0, such that, for any compact subset Ki
0,1
C1 inf i,1 ,
Ki
sup
i
i,1 \
0,1
i,1 C2 ,
1 i m1 .
(4.71)
By the previous construction, it becomes apparent from (4.69) that u is a

weak subsolution of (4.1) provided
F (CC2 )
d2 i + O( 2 )
.
supi,1 \ i0,1 a
Consequently, the optimal choice of the value of C is the following
di
F 1 ( sup2 i
C = C() := max
+O( 2 )
a)
i
\
0,1
,1
C2
1im1
Note that, thanks to (4.65) and (4.70),

lim(C()) = .
(4.72)
Moreover, owing to (4.71), we find that

u (x) C1 C()
for all x
m1
[
Ki
i=1
and therefore, thanks to (4.72),

lim u = uniformly in
0
m1
[
Ki .
(4.73)
i=1
As, according to the proof of Theorem 4.1, the problem (4.1) possesses arbitrarily large supersolutions in the interior of the cone of positive functions in
if < d1 , by the weak counterpart of Theorem 1.2, we find that
C01 ()
in
u [,]
and consequently, by (4.73),
lim [,] =
d1
uniformly in
m1
[
Ki .
i=1
As a byproduct,
lim [,] (x) =
d1
for all x
m1
[
i0,1 \ =
0,1 \ .
i=1
Finally, the proof of the first part of Theorem 4.8 shows (4.48).
(4.74)
96
4.7
Limiting behavior of the large solutions of order 1

j q0 1 as dj+1
The main result of this section can be stated as follows.

Theorem 4.9 Suppose a(x) and f satisfy (Ha) and (KO), respectively. Then,
for every 1 j q0 1,
(
0,j+1 \ ,
in
min
min
lim L[,j ] =
(4.75)
0,j+1 .
L[dj+1 ,j+1 ]
in j+1 = j \
dj+1
Proof: Fix 1 j q0 1. By Theorem 4.5, for every M > 0 and , R
with < < dj+1 , we have that
[,j ,M ] [,j ,M ]
in j
and hence, letting M , we find from Theorem 4.7 that

min
min
L[,j ] L[,j ]
in
j .
Consequently, the point-wise limit

L :=
min
lim L[,j ]
in
dj+1
(4.76)
is well defined. By construction, we find from Theorem 4.5 that

min
[,j ] [,j ,M ] L[,j ]
for all M > 0.
(4.77)
As the lower order refuges of j , according to the size of the principal eigenvalues of , are the components of 0,j+1 , Theorem 4.8 guarantees that
lim [,j ] =
dj+1
0,j+1 \ .
in
(4.78)
Thus, letting dj+1 in (4.77), it follows from (4.76) and (4.78) that
L=
0,j+1 \ .
in
(4.79)
Actually, according to (4.53), it becomes apparent that

lim m = ,
dj+1
m :=
min
0,j+1 \
Subsequently, for every M > 0 and < dj+1 , we set

bM := max [,j ,M ] .
min
L[,j ] .
(4.80)
97
Then, by Theorems 4.5 and 4.7,

min
[,j ,M ] [dj+1 , ,bM +1] [dj+1 , ,] = L[dj+1 , ]
in
and hence, letting M , we find from Theorem 4.7 that

min
min
L[,j ] L[dj+1 , ]
in for all < dj+1 .
Therefore, letting dj+1 , (4.76) implies that

min
L L[dj+1 , ]
in .
(4.81)
In particular, L is finite in .
Next, we will consider, for sufficiently large n N, say n n0 , the open
sets
,n := { x : dist (x, ) > 1/n };
n0 should be chosen so that ,n is of class C 2+ for all n n0 . By construction, it follows from (4.81) that, for every n n0 , there exists a constant
Cn > 0 such that
min
,n+1
L[,j ] Cn
in
for all < dj+1 . Thus, by the Schauder interior estimates, there exists a
constant Cn > 0 such that
min
kL[,j ] kC 2+ ( ,n ) Cn
for all
< dj+1 .
Hence, as the injection

,n ) , C 2 (
,n )
C 2+ (
is compact and the point-wise limit (4.76) unique, we obtain that
min
lim kL[,j ] LkC 2 ( ,n ) = 0
dj+1
for all n n0 . As
=
,n ,
nn0
it becomes apparent that L C 2 ( ) solves (4.43) in . By elliptic regularity, we actually have that L C 2+ ( ).
Now, for every R and M > m > 0, we consider the singular boundary
value problem

in j+1 ,
u=M
on (j+1 \ ) \ 0,j+1 ,
(4.82)
u
=
m
on 0,j+1 \ ,
u=0
on j+1 .
98
By adapting the proof of Corollary 4.4 with = j+1 , it becomes apparent

that (4.82) possesses a positive solution if and only if < dj+1 . Moreover,
it is unique if it exists. Let us denote it by [,j+1 ,M,m] . Then, adapting the
proof of Theorem 4.7, it is apparent that
min
L[,j+1 ] := lim [,j+1 ,M,m] .

m
Fix < dj+1 . Then, according to (4.38) and (4.80), there exists M () > m
such that, for every M M (),
min
L[,j ] [,j ,M ()] [,j+1 ,M (),m ]
in
j+1 .
(4.83)
By (4.80), letting dj+1 , it is easy to see that (4.83) implies

min
L L[dj+1 ,j+1 ]
in
j+1 .
min
Consequently, by the minimality of L[dj+1 ,j+1 ] ,

min
L = L[dj+1 ,j+1 ]
which ends the proof.
4.8
in
j+1 ,
Limiting behavior as of the large solutions

Theorem 4.10 Suppose a(x) and f satisfy (Ha) and (KO), respectively.
Then,
lim Lmin
[,j ] = 0
uniformly in compact subsets of j
(4.84)
for every j {1, ..., q0 }.

Proof: Fix j {1, ..., q0 }. We begin by establishing that
lim Lmin
[,j ] = 0
uniformly in compact subsets of j .
(4.85)
R (x0 ) , and set

Let x0 and R > 0 be such that B
:= B R (x0 ), A := max a < 0, := kLmin kC(B (x )) .
D := B R (x0 ), D
[,j ]
0
R
2
R (x0 )
B
Then, in BR (x0 ), we have that

min
min
min
min
min
min
dLmin
[,j ] = L[,j ] + af (, L[,j ] )L[,j ] L[,j ] + Ag(L[,j ] )L[,j ]
99
and hence, for every < dj+1 , with q0 +1 := , Lmin

[,j ] provides us with a
positive subsolution of

du = u + Ag(u)u
in D,
(4.86)
u=
on D.
By Theorem 2.4, (4.86) possesses a unique positive solution, [,D,] . Moreover,
for every < dj+1 ,
Lmin
in D.
(4.87)
[,j ] [,D,]
According to the proof of Theorem 3.2, the function [,D,] is radially symmetric and, for each R, the (increasing) point-wise limit
L := lim [,D,]
provides us with the minimal large positive solution of

du = u + Ag(u)u
(4.88)
in D. Moreover, L is radially symmetric and

L (x0 ) = min L .
(4.89)
On the other hand, by Theorem 2.4, we have that, for every < ,
[,D,] [,D,]
for all > 0
and hence, letting , Theorem 3.4 guarantees that

L L
in D
if < .
(4.90)
Multiplying
Let 0 be any principal eigenfunction associated to 1 [, D].
the differential equation
dL = L + Ag(L )L
in D,
yields
by and integrating by parts in D
Z
Z
Z
)
(1 [d, D]
L dx A
g(L )L dx = d
dS.
n
Thus, since A < 0 and g 0, we find from (4.90) that

Z
Z
Z
0 < (1 [d, D] )
L dx d
L
dS d
L0
dS
n
n
D
D
D
for all < 0, because
Therefore, letting yields

< 0 on D.
Z
lim
L dx = 0.
(4.91)
100
As (4.89) implies
Z
L (x0 )
Z
dx
L dx,
it becomes apparent from (4.91) that

lim L (x0 ) = 0.
(4.92)
By (4.87), we have that

Lmin
[,j ] [,D,] L
in D.
Thus, we find from (4.92) that

lim Lmin
[,j ] (x0 ) = 0.
Similarly, for every x D = BR/2 (x0 ), we have that

Dx := BR/2 (x) BR (x0 )
and that Lmin
[,j ] is a positive subsolution of

du = u + Ag(u)u
u=
in Dx ,
on Dx ,
(4.93)
for all < dj+1 . Consequently, reasoning as above yields

Lmin
[,j ] [,Dx ,] L,x
in Dx ,
where L,x stands for the minimal positive large solution of (4.88) in Dx .
On the other hand, by the radial symmetry of the minimal large solutions
of (4.88) in these balls, necessarily
L,x (y) = L (x0 x + y)
for all y Dx ,
and hence,
Lmin
[,j ] (x) L,x (x) = L (x0 )
for all x D.
Therefore, according to (4.92),

lim Lmin
[,j ] = 0
uniformly in D.
By compactness, (4.85) holds.

As q0 = , (4.85) is (4.84) when j = q0 . Hence, the proof is completed
in this case. So, suppose j q0 1. Let i0,k be, with j + 1 k q0 and
1 i mk , an arbitrary component of 0 j . For sufficiently small > 0,
consider
i,k := {x : dist (x, i0,k ) < }.
101
The boundary i,k consists of the components of i0,k plus the components of i,k \ , which are compact subsets of . As a 0 and f 0,
for each < 0 we have that
dLmin
[,j ] 0
in i,k
i
and hence, Lmin
[,j ] is subharmonic in ,k . Thus, by the maximum principle,
max Lmin
[,j ] =
i
,k
max Lmin
[,j ]
(4.94)
i,k \
i
because Lmin
[,j ] = 0 on 0,k . As, according to (4.85),
lim
max Lmin
[,j ] = 0,
i,k \
the identity (4.94) ends the proof.
4.9
The most pioneering result characterizing the existence of positive solutions of

(4.1) is Theorem 1 of H. Brezis and L. Oswald [31], received by the editors of
Nonlinear Analysis on November 7, 1984. By using some classical minimization methods, it was shown that (4.1) possesses a positive solution if, and only
if,
1 [d V0 (x), ] < 0 < 1 [d V (x), ],
(4.95)
where
V0 (x) := lim( + a(x)f (x, u)) =
u0
and

V (x) := lim ( + a(x)f (x, u)) =
u
if x ,
if x 0 .
Consequently, though the first estimate of (4.95) becomes > d0 , it is far

from evident that the second one should become < d1 , as required by
Theorem 4.1. Actually, to get the equivalence between Theorem 4.1 and [31,
Th. 1] one must invoke the fundamental property that
lim 1 [d a(x)f (x, u), ] = d1 ,
(4.96)
going back to J. L
opez-G
omez [141]. Indeed, thanks to (4.96), the estimate
(4.95) can be equivalently written as
d0 < < d1 .
102
Another significant pioneering contribution was made by T. Ouyang [203], on

this occasion working with Neumann boundary conditions, as a part of his
PhD thesis on Yamabes problem [240] under the supervision of W. M. Ni.
By combining a global continuation argument with the existence of a priori
bounds for the positive solutions of (4.1) in compact subsets of (, d1 ),
T. Ouyang [203] established Theorem 4.1 for the special case f (x, u) = up1 .
Moreover, he also established that
lim k[,] kL2 () = .
d1
(4.97)
A. Ambrosetti and J. L. Gamez [14], and M. A. del Pino [70] slightly refined
some of the previous findings of T. Ouyang [203]. However, none of these works
stressed the tremendous importance of ordering the several components of 0
according to the size of the principal eigenvalues 1 [, i0,j ], 1 j q0 ,
1 i mj , in order to describe the dynamics of (1.1), as was done in [109].
The proof of Theorem 4.1 given here goes back to J. M. Fraile et al. [90],
where the method of subsolutions and supersolutions was incorporated for the
first time to the world of the degenerate diffusive logistic equations. The reader
should be aware that prior to the publication of J. Lopez-Gomez [141, 144]
and J. M. Fraile et al. [90], the only available supersolutions were the large
positive constants, which fail to be supersolutions if a1 (0) =
6 . Probably, this
might explain why H. Brezis and L. Oswald [25] and T. Ouyang [203] did not
use the method of subsolutions and supersolutions to obtain their pioneering
findings. It is worth emphasizing that, according to J. M. Fraile et al. [90],
Theorem 4.1 is valid for general semilinear elliptic equations of the form
Lu = u + a(x)f (x, u)u
under general mixed boundary conditions, with L of type (1.48).
The point-wise blow up in 0,1 of [,] as d1 was observed by the
first time in J. L
opez-G
omez [148, Th. 2.4], received for publication on August
26, 1996, though it was not published electronically until October 29, 1999.
As a consequence of this delay, it appeared first in J. Lopez-Gomez and J.
C. Sabina de Lis [181, Th. 4.2], which was received for publication on May 8,
1997. Undoubtedly, the point-wise blow-up in all the components of 0,1 is
substantially sharper than the L2 estimate (4.97).
The results of Sections 4.1, 4.2 and 4.3 are attributable to R. GomezRe
nasco and J. L
opez-G
omez [109]. Later, they were substantially refined by
J. L
opez-G
omez [147]. Many of these results were part of the PhD thesis of R.
G
omez-Re
nasco [105], completed in February 1999 under the supervision of
the author. Most of these results were found during the summer of 1998. The
resulting monograph [109] was submitted to H. Brezis for publication in the
Archive of Rational Mechanics and Analysis in September 1988. But it was
rejected on March 4, 1999. Then, on April 5, 1999, it was submitted to P. H.
Rabinowitz for publication in Nonlinear Analysis, where it appeared in 2002,
almost four years after it was written. The results of [109] were communicated
103
personally by R. G
omez-Re
nasco at the Conference on Operator Theory and
Its Applications held in Winnipeg (Canada) on October 1998, and by J. LopezG
omez at the Conference on Differential Equations honoring A. C. Lazer held
in Miami (Florida, USA) on January 8 and 9, 1999.
The stabilization of [,] to the minimal large solution in
0,1
1 = \
established by Theorem 4.8 goes back to J. Garca-Melian et al. [100], whose
work was accepted on February 17, 1998, by Professor P. H. Rabinowitz. The
divergence to infinity of [,] on 0,1 \ as d1 goes back to Theorem
4.3 of J. L
opez-G
omez and J. C. Sabina de Lis [181] when a(x) is a function of
class C 1 around 0,1 \ , which was received by the editors of the Journal
of Differential Equations on May 8, 1997. The proof given in Section 4.6 is
the original one of [181]. The proof of this property in the general case when
a(x) is continuous, which is the one adopted in the proof of Theorem 4.8,
goes back to Y. Du and Q. Huang [80], which was received by the editors of
SIAM Journal of Mathematical Analysis on February 22, 1999, and published
electronically on November 4, 1999. The proof of Y. Du and Q. Huang [80]
uses a very classical device originated by the proof of the boundary lemma
of E. Hopf [113] and O. A. Oleinik [201] which was later refined by M. W.
Protter and H. F. Weinberger [210] and J. Lopez-Gomez [163]. Naturally, the
list of references of [80] includes [90], [100] and [181].
Recently, J. L
opez-G
omez and P. H. Rabinowitz [178] have given a short
elementary proof of Theorem 4.8 in the one-dimensional setting and have extended a number of results of this chapter to cover the case of nodal solutions.
Theorem 4.8 is an extremely sharp result establishing that the Harnack
inequality is an utterly linear property, because it can fail in nonlinear problems when the parameters involved in their setting change. Indeed, let K be
the compact subset of defined by
K := {x : dist (x, ) }
for sufficiently small > 0. Thanks to the Harnack inequality, for every
(d0 , d1 ), there exists a positive constant H > 0 such that
max [,] H min [,] .
K
Thus, if there are > 0 and H > 0 such that

sup
H H,
[d1 ,d1 )
then, according to (4.98), we should have

lim [,] = uniformly in K,
d1
(4.98)
104
because
0,1 \ ,
lim [,] = uniformly in compact subsets of
d1
which contradicts the stabilization of [,] to the minimal large solution in

1 established by Theorem 4.8. Therefore,
lim H = .
d1
Consequently, the Harnack inequality cannot be applied uniformly in compact

intervals of R in a degenerate nonlinear problem like (4.1).
More recently, F. C. Cirstea and V. Radulescu [57] considered the following
singular problem

u = au b(x)f (u)
in ,
(4.99)
u=
on ,
satisfies
where is a regular domain of RN , N 1, a is a constant, b C()
1
b 0, b 6= 0, and f C is a positive function satisfying (KO) and such that
f (u)/u is increasing in R+ . Setting
1 := 1 [, int b1 (0)],
the main result of [57] can be stated as follows.
Theorem 4.11 Problem (4.99) has a solution if and only if a (, 1 ).
Moreover, in this case, the solution is unique.
F. C. Cirstea and V. Radulescu [57] acknowledged that
We point out that our framework in the above result includes the case when
b vanishes at some points of , or even if b = 0 on . In this sense, our result
responds to a question raised to one of us by Professor Haim Brezis in Paris, May
2001.
The existence result had been already found by R. Gomez-Re

nasco and
J. L
opez-G
omez [109]. The uniqueness was also known, but the uniqueness
discussion is postponed to Part II.
Theorem 4.10 refines a result going back to the proof of Part (b) of J.
L
opez-G
omez and M. Molina-Meyer [166, Th. 1.1].
This chapter grew from R. Gomez-Re
nasco and J. Lopez-Gomez [109], and
J. L
opez-G
omez [147, 151, 160].
All the results of this paper are valid for a slightly more general family of
differential equations of the form
(d + V (x))u = u + a(x)f (x, u)u
or simply V L () if we seek for solutions in W 2,p (),
with V C (),
p > N.
Chapter 5
Dynamics: Metasolutions
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
Concept of metasolution: The main theorem . . . . . . . . . . . . . . . . . . . .

Paradigmatic bifurcation diagram with q0 = 2 . . . . . . . . . . . . . . . . . .
Numerical example with q0 = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Proof of Theorem 5.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Approximating metasolutions by classical solutions . . . . . . . . . . . . .
Pattern formation in classical logistic problems . . . . . . . . . . . . . . . . .
Biological discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
106
108
111
117
125
128
129
131
The main goal of this chapter is ascertaining the asymptotic behavior of the
solutions of (1.1) as t according to the each of the different ranges of values
of the parameter R. From the point of view of population dynamics, these
behaviors can be briefly sketched as follows:
If d0 , then the inhabiting region cannot support the species u.
If d0 < < d1 , then the species u exhibits logistic growth in .
If q0 2 and dj < dj+1 for some j {1, ..., q0 1}, then u has
Malthusian growth in

0,1
0,j \ ,
and logistic growth in the complement

0,1
0,j .
j := \
If dq0 , then the species u exhibits Malthusian growth in \

and logistic growth in .
These findings are extremely relevant from the point of view of the applications
of the abstract mathematical theory developed in this book to population dynamics, since they provide simultaneous effects of the growth laws of Malthus
and Verhulst within the same territory , which seems extremely realistic in
applications. Rather naturally, the growth of the species should be severely
limited in the regions with a serious shortcoming of resources; however, growth
may be huge within patches where resources are abundant, which might help
show why agriculture facilitated the emergence of human groups whose size
105
106
gradually increased over the last ten thousands years; almost nothing on an
evolutionary scale, until humans developed the extremely populated areas
they inhabit today. Simultaneously, in unfavorable areas, where agriculture
was not facilitated by the intricate nature of the territory, as occurs in most
rain forest areas, the human population numbers did not vary substantially.
The distribution of this chapter is the following. Section 5.1 states the
main theorem of the chapter, Section 5.2 applies it to discuss the dynamics of
(1.1) in a special case where q0 = 2, Section 5.3 studies a numerical example
within the setting of Section 5.2, and, finally, Section 5.4 gives the detailed
proof of the main theorem.
5.1
Concept of metasolution: The main theorem
The concept of metasolution will substantially shorten the statement of the

main theorem of this chapter. It should be remembered that the j s, 1
j q0 , were already defined by (4.3). Roughly defined, a metasolution is a
generalized solution whose set of singularities might have positive measure. In
the context of this book, they are the extensions by of the large solutions
in the j s.
Definition 5.1 (Metasolution) For every 1 j q0 , a function
[0, ]
M:
is said to be a positive metasolution of
(5.1)
supported in j if there exists a positive solution, L, of
in j ,
u=0
on j ,
u=
on j \ ,
such that
M=
in j ,
in \ j ,
on .
Thanks to Theorem 4.7, if a(x) satisfies (Ha) and f satisfies (KO), then, for
every 1 j q0 , (5.1) possesses a positive metasolution supported in j if
and only if < dj+1 . Moreover, should it be the case, (5.1) admits minimal
and maximal positive metasolutions. Namely,
min
max
in j ,
L
in
L[,j ]
j
[,
]
j
min
max
M[,j ] :=
M
:=
in \ j ,
in \ j ,
[,j ]

0
on .
0
on ,
107
The next result, which is the main theorem of this chapter and of Part I,
provides us with the dynamics of (1.1) according to each of the values of the
parameter R.
Theorem 5.2 Suppose a(x) and f satisfy (Ha) and (KO), respectively, and
denote by
u(x, t) := u[,] (x, t; u0 )
the unique solution of (1.1). Then, the following properties hold:
(a) If d0 , then
lim u(, t) = 0
C().
in
(b) If d0 < < d1 , then

lim u(, t) = [,]
C(),
in
where [,] is the unique positive solution of

u=0
in ,
on .
(5.2)
(c) If dj < dj+1 , 1 j q0 , then

min
max
M[,j ] lim inf u(, t) lim sup u(, t) M[,j ]

t
in
(5.3)
If, in addition, u0 > 0 is a subsolution of (5.2), then

min
lim u(, t) = M[,j ]
in
(5.4)
According to (5.3) and the definition of the maximal and the minimal metasolutions, it becomes apparent that
lim u(, t) = uniformly in compact subsets of
\ j =
j
[
i=1
while in j the following estimate holds

min
max
L[,j ] lim inf u(, t) lim sup u(, t) L[,j ] .

t
If, in addition, u0 is a subsolution of (5.2) in , then

min
lim u(, t) = L[,j ]
in
j .
0,i ,
108
As a result of Theorem 5.2, the dynamic of (1.1) is governed by the maximal

non-negative classical solution of (5.2) if < d1 , while it is regulated by the
minimal and the maximal metasolutions of (5.1) in j if dj < dj+1 for
some 1 j q0 . It should be remembered that q0 +1 := . Consequently,
from the point of view of population dynamics, for every [dj , dj+1 ), the
species u exhibits Malthusian growth in

0,1
0,j \ ,
whereas it has logistic growth in

0,1
0,j ,
j = \
1 j q0 .
max
min
max
Naturally, unless Mmin
[,j ] = M[,j ] or, equivalently, L[,j ] = L[,j ] , (5.3)
does not provide us with the exact behavior of u(x, t) for large t. This is why
the second part of this book will focus attention on the uniqueness of the large
positive solutions of (5.1).
Remark 5.3 All the results of Part I and, in particular, Theorem 5.2 are also
valid if, instead of d, we consider the more general differential operator
L = d + V
Even the function V might be taken in L () if solutions
for some V C ().
are regarded in W 2,p () with p > N . Naturally, in such case, the eigenvalues
d0 = 1 [d, ],
dj = 1 [d, i0,j ],
1 j q0 ,
should be inter-exchanged by
0 := 1 [d + V, ],
j := 1 [d + V, i0,j ],
1 j q0 .
In this context, q0 , mj , and even the components of , i0,j , might depend on

V and d. Although this was the case dealt with in [153], in this book we have
refrained from providing the most general setting available. Nevertheless, in
Chapter 10 we will need to work with this slightly more general setting.
5.2
Paradigmatic bifurcation diagram with q0 = 2
In Figure 5.1 we have represented the dynamics of (1.1) under the assumptions
of Theorem 5.2 in the special case when q0 = 2. Then,
0,1 ,
1 := \
0,1
0,2 ) = .
2 = \ (
109
Figure 1.1 shows an admissible nodal configuration of a(x) respecting this

situation. Precisely, Figure 5.1 shows the dynamics of (1.1) when
max
M[,j ] := Mmin
[,j ] = M[,j ]
(5.5)
for all j = 1, 2 and < dj+1 , where 3 := . In such case, thanks to

Theorem 4.7, it is easy to see that the maps
(, d2 ) 7 M[,1 ] ,
R 7 M[,2 ]
(5.6)
are continuous and increasing. Moreover, by Theorems 4.8 and 4.9,

lim [,] = M[d1 ,1 ] ,
d1
lim M[,1 ] = M[d2 ,2 ] .
d2
This explains why in Figure 5.1 the curves of metasolutions (, M[,1 ] ), defined for < d1 , and (, M[,2 ] ), defined for R, are increasing and meet
at = d2 . Note that, due to Theorems 4.1 and 4.8, the curve of classical
positive solutions (, [,] ), d0 < < d1 , bifurcates from u = 0 at = d0
and meets M[d1 ,1 ] at = d1 .
In Figure 5.1 we are representing the value of the parameter versus
the values of the classical positive solutions and metasolutions, u(x), at some
distinguished point x = 2 , where all of them are bounded, because the
large solution of (5.1) in is defined for all R. The global bifurcation
diagram shows four different types of solutions. The -axis represents u = 0,
which, according to Theorem 5.2(a), is a global attractor if d0 . As the
FIGURE 5.1: The dynamics of (1.1).
110
linearization of (1.1) at u = 0 is the linear problem

u
in (0, ),
t du = u
u=0
on (0, ),
u(, 0) = u0 > 0
in ,
whose unique solution is given by
u(x, t) = et(d+) u0 ,
when > d0 the trivial solution u = 0 is linearly unstable, because
u(x, t) = et(d+) = et(d0 )
as t ,
where 0 stands for any principal eigenfunction of associated to 0 .

This is why in Figure 5.1 the u axis has been represented with a dashed line
for d0 . As usual, continuous lines are filled in by stable steady states, or
stable metasolutions, while dashed lines are filled in by unstable steady states,
or unstable metasolutions. Then, we have represented the positive solution
[,] , d0 < < d1 , which is a global attractor of (1.1) in that range, the
curve of metasolutions M[,1 ] , < d2 , supported in 1 and, finally, the
curve of metasolutions M[,2 ] supported in 2 , R.
According to Theorem 5.2(a), u = 0 is a global attractor for the positive
solutions of (1.1) if d0 , by Theorem 5.2(b), [,] is a global attractor for
the positive solutions if d0 < < d1 , and, due to Theorem 5.2(c), M[,1 ]
is a global attractor for d1 < d2 , and M[,2 ] is a global attractor if
d2 . This describes the dynamics of (1.1) for all values of R.
As for < d1 the dynamics of (1.1) are governed by the classical nonnegative steady states, the metasolutions M[,j ] must be unstable with respect to these solutions. Consequently, = d1 provides us with the critical
value of where M[,1 ] becomes an attractor for all classical positive solutions. Similarly, M[,2 ] is unstable if < d2 , while it becomes a global
attractor for all classical positive solutions if d2 .
The monotonicity of the maps (5.6) is an easy consequence from Theorem
4.7, taking into account that, owing to Theorem 4.5(b),
[,j ,M ] [,j ,M ]
for all < < dj+1 , M > 0 and j = 1, 2. Indeed, letting M implies
L[,j ] < L[,j ]
provided < < dj+1 . The continuity of the mapping
7 L[,j ] ,
< dj+1 ,
under the uniqueness condition (5.5) can be established with the next argument. Suppose n , n 1, is a sequence of s, with n < dj+1 , such that
:= lim n < dj+1 .
n
111
Then, for sufficiently small > 0, there exists n0 N such that

L[n ,j ] L[ +,j ]
for all n n0 .
By adapting the compactness arguments of Chapter 4, it is easy to infer that,

along some subsequence relabeled by n, L[n ,j ] approximates uniformly in
compact subsets of j to a positive solution of (4.2). By uniqueness, the limit
must be L[ ,j ] . As this is true along any subsequence, the continuity holds.
Naturally, the continuity might fail if (4.2) has multiple solutions.
Obviously, the general discussion carried out in this section can be adapted
almost mutatis mutandis to cover the general case when q0 1.
5.3
Numerical example with q0 = 2
In this section we consider (1.1) in the special case when

N = 2,
= B0.5 = {x R2 : |x| < 0.5},
d = 1,
with

(r) :=
sin(5(r + 0.5)),
0,
a(x) = (r), r = |x|,
0.1 < r < 0.3,

r [0, 0.1] [0.3, 0.5],
and
f (x, u) = u3 ,
u 0,
x .
Figure 5.2 shows a plot of a(x).

By Lemma 3.6, f satisfies the conditions (Hf), (Hg) and (KO). Moreover,
a(x) satisfies (Ha) with
0 = B0.1 A0.3,0.5 ,
= A0.1,0.3 ,

Aa,b := { x RN : a < |x| < b }
for all 0 < a < b. As, due to R. Gomez-Re
nasco and J. Lopez-Gomez [109,
Sect. 5],
1 [, B0.5 ] ' 23.13 < 1 [, A0.3,0.5 ] ' 245.14 < 1 [, B0.1 ] ' 578.31,
it becomes apparent that, actually, a(x) satisfies Hypothesis (Ha) with
q = 1,
q0 = 2,
0,1 = A0.3,0.5 ,
0,2 = B0.1 ,
0 ' 23.13 < 1 ' 245.14 < 2 ' 578.31.
112
According to Theorems 4.1 and 4.8, the problem

u = u + a(x)u4
in ,
u=0
on ,
(5.7)
has a (unique) positive solution, [,] , if and only if 0 < < 1 . Moreover, by uniqueness, [,] is radially symmetric, the mapping 7 [,] (r) is
increasing for all r = |x| [0, 0.5),
lim [,] = 0
uniformly in = B0.5 ,
and
0,1 \ .
lim [,] = uniformly in compact subsets of
(5.8)
In particular, the mapping

Z
7 k[,] k2 := k[,] kL2 () =
2
[,]
12
,
0 < < 1 ,
is increasing and it satisfies

lim k[,] k2 = 0,
lim k[,] k2 = .
Figure 5.3 shows the graph of this curve computed in [109] through a pathfollowing solver (see [109] for technical details). It plots the L2 -norm of the
FIGURE 5.2: A plot of a(x).
113
non-negative solutions of (5.7) versus the parameter in the interval (0, 250).
Note that the L2 -norm of [,] indeed blows up as 1 ' 245.14. Continuous lines are filled in by stable solutions and dashed lines by unstable ones.
Each point on the continuous line represents a positive solution of (5.7). According to Theorem 5.2, u = 0 is a global attractor for the parabolic problem
u
in (0, ),
t u = u + a(x)u4
(5.9)
u=0
on (0, ),
u(, 0) = u0 > 0
in ,
if 0 , whereas [,] is a global attractor of (5.9) for all (0 , 1 ).
Figure 5.4 shows the plots of [,] for = 101.229218, = 191.845246,
= 217.71643 and = 232.193199, respectively. In agreement with Theorems
4.1 and 4.8, these solutions are point-wise increasing with respect to and
blow-up in
0,1 \ = {x R2 : 0.3 |x| < 0.5}
as 1 , while in the region

0,1 = B0.3 ,
1 := \
FIGURE 5.3: The curve 7 k[,] k2 .
114
they stabilize to the unique positive solution, L[1 ,1 ] , of the singular problem

u = 1 u + a(x)u4
in 1 ,
(5.10)
u=
on 1 ,
as 1 . The uniqueness of L[1 ,1 ] and the uniqueness of L[,1 ] for each
< 2 are guaranteed by Theorem 8.4.
Note that in Figure 5.4 the scales on the vertical axis are different in each
of the four plots. This is why the plots of the positive steady states do not
seem to grow as increases, though they actually do. Furthermore, by (5.8),
they do it at a much faster rate in 0,1 than in \ 0,1 .
Figure 5.5 shows a bifurcation diagram computed in [109] of stable positive
steady states, [,] , and stable metasolutions supported in 1 , M[,1 ] , for the
range 0 < < 2 . For a given x = A0.1,0.3 with r = |x| = 0.2, it plots
FIGURE 5.4: The classical positive steady states [,] for = 101.229218
(top left), = 191.845246 (top right), = 217.71643 (bottom left) and =
232.193199 (bottom right). They are point-wise increasing and blow up to
infinity in A0.3,0.5 as approximates 1 ' 245.14.
115
the value [,] (x) versus the parameter for 0 < < 1 , where, according to
Theorem 5.2(b), [,] is a global attractor of (5.9), and the value M[,1 ] (x)
versus for 1 < < 2 , where, according to Theorem 5.2(c), M[,] is a
global attractor of (5.9). According to Theorem 4.7, M[,1 ] exists if and only
if < 2 and, thanks to Theorem 5.2, it is unstable for < 1 and stable for
1 < 2 . The continuous line plots (, [,] (x)) for 0 < < 1 , while the
dashed-dotted line plots (, M[,1 ] (x)) for 1 < 2 . Only in this isolated
situation, in order to differentiate classical solutions from metasolutions, a
curve of generalized stable steady states will be represented through a dashed
line.
Figure 5.6 shows the plots of L[,1 ] for = 300, = 450, = 500
and = 525, respectively. They equal infinity on B0.3 and are point-wise
increasing with respect to up to = 2 , where, according to Theorem 4.9,
0,2 = B
0.1 reaching M[ , ] . As in Figure 5.4,
they blow up to infinity in
2
one should take into account that the scales of the vertical axis in the two rows
of Figure 5.6 are different, though in this occasion the increasing character of
the metasolution in is well differentiated.
As a consequence of Theorem 5.2, when the intrinsic growth rate of the
species u, measured by , is below the threshold 0 , the inhabiting area
cannot support the species, which is driven to extinction. Contrarily, when
(0 , 1 ), the territory is able to maintain u at the critical level [,] ,
independently of the size of the initial population u0 . So, as in this range of
values of the habitat cannot maintain an arbitrarily large population,
(5.9) exhibits a genuine logistic behavior.
FIGURE 5.5: Classical solutions and stable metasolutions supported in 1 .
116
Rather astonishingly, when [1 , 2 ) the population must be limited

0,1 , while, as an effect of the random
by L[,1 ] in the region 1 := \
dispersion of the individuals of the species u to the biggest protection zone,
0,1 , the population can grow arbitrarily in 0,1 . So, (5.9) exhibits a genuine
0,1 while it exhibits a typical exponential growth in
logistic growth in \
0,1 as t . Indeed, thanks to the parabolic maximum principle,
u1 := u[,] (, 1; u0 ) 0
and hence, for every t 0,
u[,] (, t+1; u0 ) = u[,] (, t; u1 ) > u[,0,1 ] (, t; u1 )
in 0,1 ,
(5.11)
FIGURE 5.6: The large solutions L[,1 ] for = 300 (top left), = 450
(top right), = 500 (bottom left) and = 525 < 2 (bottom right). They
are point-wise increasing and blow up to infinity in B0.1 as approximates
2 ' 578.31.
where u[,0,1 ] (x, t; u1 ) stands for the
u
t u = u
u=0
u(, 0) = u1
117
unique solution of the linear problem

in 0,1 (0, ),
on 0,1 (0, ),
in 0,1 ,
which is given by
u[,0,1 ] (x, t; u1 ) = et(+) u1 .
(5.12)
Suppose > 1 and let 0,1 be a principal eigenfunction associated with 1 .

Then, since u1 0, there exists > 0 such that
u1 > 0,1
and hence, (5.11) and (5.12) imply that
u[,] (, t + 1; u0 ) > et(+) 0,1 = e(1 )t 0,1 ,
which shows the exponential growth of the population in 0,1 for all > 1 .
Similarly, when > 2 the population must be bounded in , as a result
of the limited resources therein, though the individual of the species can disperse towards the most favorable areas 0,1 and 0,2 , where the population
density grows arbitrarily at the rates et(1 ) and et(2 ) , respectively. Naturally, in the applications, the principal eigenvalues 1 and 2 measure the
quality of the protection zones 0,1 and 0,2 .
5.4
Proof of Theorem 5.2
Part (a) is a direct consequence of the last assertion of Theorem 1.7 and Part
(b) is a direct consequence of Theorem 1.7(b), because, thanks to Theorem
4.1, (1.1) possesses a positive steady state for each (0 , 1 ). It remains to
prove Part (c). The proof will proceed by induction. First, we will establish
Part (c) in the special case when
d1 < d2 .
Since > d1 for each > 0, by the parabolic maximum principle,
u[,] (, t; u0 ) u[d1 ,] (, t; u0 )
for all > 0 and t 0. Thus, by Part (b), letting t yields
lim inf u[,] (, t; u0 ) lim u[d1 ,] (, t; u0 ) = [d1 ,]
t
(5.13)
118
for sufficiently small > 0. Thus, letting 0, Theorem 4.8 implies that
min
lim inf u[,] (, t; u0 ) lim [d1 ,] = M[d1 ,1 ] .

t
(5.14)
Actually, as
lim [d1 ,] =
0
0,1 \ , (5.13) implies that

0,1 \ .
lim inf u[,] (, t; u0 ) = unif. in compact subsets of
t
(5.15)
In particular, for every M > 0 there exists TM > 0 such that

u[,] (x, t; u0 ) M
(x, t) (0,1 \ ) [TM , ).
So, u[,] (, t; u0 ) is a supersolution of

u
du = u + af (, u)u
t
u=M
u=0
u(, TM ) = u[,] (, TM ; u0 )
in
on
on
in
1 (TM , ),
(1 \ ) (TM , ),
(1 ) (TM , ),
1 .
Consequently, by the parabolic maximum principle,

u[,] (x, t; u0 ) u[,1 ,M ] x, t TM ; u[,] (, TM ; u0 )
(5.16)
for all (x, t) 1 (TM , ), where u[,1 ,M ] (x, t; u

0 ) stands for the unique
solution of
u
du = u + af (, u)u
in 1 (TM , ),
t
u=M
on (1 \ ) (TM , ),
on (1 ) (TM , ),
u=0
u(, 0) = u
0
in 1 ,
which is the parabolic counterpart of (4.33) with j = 1. By Theorem 4.5,
(4.33) possesses a unique positive solution [,1 ,M ] for each < d2 . Thus,
thanks to Theorem 1.7(b),
lim u[,1 ,M ] (x, t; u
0 ) = [,1 ,M ]
for all u
0 > 0. So, letting t in (5.16) shows that
lim inf u[,] (x, t; u0 ) [,1 ,M ]
t
for all M > 0.
Therefore, letting M , we find from Theorem 4.7 that

min
lim inf u[,] (x, t; u0 ) L[,1 ]

t
in
1 .
119
Bringing together this estimate with (5.15) it becomes apparent that

min
lim inf u[,] (x, t; u0 ) M[,1 ]
in
(5.17)
which is the lower estimate of (5.3) for j = 1.

Suppose, in addition, that u0 is a subsolution of (5.2) in . Then, according
to D. Sattinger [220], the function
x ,
x 7 u[,] (x, t; u0 ),
is a subsolution of (4.1) for all t > 0, because t 7 u[,] (, t; u0 ) is nondecreasing. Fix t > 0 and set
Mt := max u[,] (, t; u0 ).
1
As u[,] (, t; u0 ) is a subsolution of (5.2), it is also a subsolution of (4.33)

with j = 1 for all M Mt . Thus, due to Theorem 4.5(a),
u[,] (, t; u0 ) [,1 ,M ]
in
M Mt .
for all
Hence, owing to (4.51),

min
u[,] (, t; u0 ) lim [,1 ,M ] = L[,1 ]
in
and therefore, letting t yields

min
lim sup u[,] (, t; u0 ) L[,1 ]
in
1 .
Consequently, according to (5.17), we can conclude that

min
lim u[,] (, t; u0 ) = M[,1 ]
in
(5.18)
which ends the proof of (5.4) for j = 1.

To obtain the upper estimates of (5.3), we need to get some a priori bounds
in for the solutions of (1.1). Fix d1 , set
u1 := u[,] (, 1; u0 ) 0
and let 0 be a principal eigenfunction associated to 0 . Then, since
u1 0, there exists > 1 such that
u1 < .
(5.19)
> max{, dq0 }
(5.20)
We claim that there exists
120
such that is a subsolution of

du = u + af (, u)u
u=0
in ,
on .
(5.21)
Indeed, since = 0 on , is a subsolution of (5.21) if, and only if,

d() + af (, )
in
af (, ) d0
or, equivalently,
in
which is true for sufficiently large > max{, dq0 }.

By the semigroup property and the parabolic maximum principle, it follows
from (5.19) that
u[,] (x, t + 1; u0 ) = u[,] (x, t; u1 ) u[,] (, t; )
for all x and t > 0. Similarly, (5.20) implies
u[,] (, t; ) < u[,] (, t; )
in
for all t > 0. Hence,

u[,] (, t + 1; u0 ) u[;] (, t; )
in
(5.22)
As is a subsolution of (5.21), it follows from D. Sattinger [220] that

u[;] (, t; ) is a subsolution of (5.21) for all t > 0. Fix t > 0 and set
t := max u[,] (, t; ).
M
Note that u[,] (, t; ) is a subsolution of (5.21) and also a subsolution of

t , since q = . Thus, due to Theorem
(4.33) for j = q0 and any M M
0
4.5(a),
t
u[,] (, t; ) [, ,M ] in for all M M
and hence, by (4.38),
min
u[,] (, t; ) lim [, ,M ] = L[, ]

M
in
Therefore, by (5.22), letting t yields

min
lim sup u[,] (, t; u0 ) L[, ]
in
(5.23)
Consequently, u[,] (, t; u0 ) is uniformly bounded above, in any compact subset of , for all t > 0.
121
Suppose d1 < d2 and for sufficiently large n N, say n n0 ,

consider the open sets defined by (4.35) with j = 1, i.e.,
1,n := {x 1 : dist (x, 1 \ ) > 1/n}.
Pick n n0 . As 1,n , it follows from (5.23) that there exists
M0 > 0 such that
u[,] (, t; u0 ) M
on 1,n \
for all M M0 and t > 0. Thus, by the parabolic maximum principle,

u[,] (, t; u0 ) u[,1,n ,M ] (, t; u0 )
for all t > 0, where u[,1,n ,M ] stands for
u
du = u + af (, u)u
t
u=M
u=0
u(, 0) = u0
in
1,n
(5.24)
the unique solution of

in
on
on
in
1,n (0, ),
(1,n \ ) (0, ),
(1,n ) (0, ),
1,n .
By Theorem 1.7,
lim u[,1,n ,M ] (, t; u0 ) = [,1,n ,M ]
in
1,n ,
where [,1,n ,M ] is the unique positive solution of (4.36) with j = 1. Hence,

letting t in (5.24) shows that, for every n n0 ,
lim sup u[,] (, t; u0 ) [,1,n ,M ]
in
1,n .
(5.25)
Consequently, letting M in (5.25), we find that

min
lim sup u[,] (, t; u0 ) L[,1,n ]
in
1,n
(5.26)
for all n n0 . As (5.26) holds for all n n0 and, due to (4.39), we already
know that
max
min
L[,1 ] = lim L[,1,n ]
n
we can conclude from (5.26) that

max
lim sup u[,] (, t; u0 ) L[,1 ]
in
1 ,
which ends the proof of (5.3) for j = 1. Naturally, this ends the proof of the
theorem if q0 = 1, because in such case 2 = .
Subsequently, we suppose q0 2 and, fixing 1 j q0 1, assume that
122
Part (c) holds for all < dj+1 . It remains to prove that in such case Part
(c) also holds for all [dj+1 , dj+2 ). Suppose
dj+1 < dj+2 .
Since > dj+1 for all > 0, by the parabolic maximum principle,
u[,] (, t; u0 ) u[dj+1 ,] (, t; u0 ).
Thus, as soon as dj < dj+1 < dj+1 , by the induction hypothesis, we
can infer that
lim inf u[,] (, t; u0 ) lim inf u[dj+1 ,] (, t; u0 ) = Mmin
[dj+1 ,j ] .
t
(5.27)
Thus, letting 0, Theorem 4.9 implies that

min
lim inf u[,] (, t; u0 ) lim Mmin
[dj+1 ,j ] = M[dj+1 ,j+1 ] .
t
(5.28)
Actually, as
lim Mmin
[dj+1 ,j ] =
0
0,j+1 \ , (5.27) implies that

lim inf u[,] (, t; u0 ) = unif. in compact subsets of
t
j+1
[
0,i \ . (5.29)
i=1
In particular, for every M > 0 there exists TM > 0 such that

u[,] (x, t; u0 ) M
(x, t) (j+1 \ ) [TM , ).
So, u[,] (, t; u0 ) is a supersolution of

u
du = u + af (, u)u
t
u=M
u=0
u(, TM ) = u[,] (, TM ; u0 )
in
on
on
in
j+1 (TM , ),
(j+1 \ ) (TM , ),
(j+1 ) (TM , ),
j+1 .
Consequently, by the parabolic maximum principle,

u[,] (x, t; u0 ) u[,j+1 ,M ] x, t TM ; u[,] (, TM ; u0 )
(5.30)
for all (x, t) j+1 (TM , ), where u[,j+1 ,M ] (x, t; u

0 ) stands for the unique
solution of
u
du = u + af (, u)u
in j+1 (TM , ),
t
u=M
on (j+1 \ ) (TM , ),
u
=
0
on (j+1 ) (TM , ),
u(, 0) = u
0
in j+1 ,
123
which is the parabolic counterpart of (4.33) at j + 1. By Theorem 4.5,

this parabolic problem possesses a unique positive steady-state solution,
[,j+1 ,M ] , for each < dj+2 . Thus, thanks to Theorem 1.7(b),
lim u[,j+1 ,M ] (x, t; u
0 ) = [,j+1 ,M ]
for all u
0 > 0. So, letting t in (5.30) shows that
lim inf u[,] (x, t; u0 ) [,j+1 ,M ]
for all M > 0.
Therefore, letting M , we find from Theorem 4.7 that

min
lim inf u[,] (x, t; u0 ) L[,j+1 ]
in
j+1 .
Bringing together this estimate with (5.29) it becomes apparent that

min
lim inf u[,] (x, t; u0 ) M[,j+1 ]
in
(5.31)
which is the lower estimate of (5.3) at j + 1.

Suppose, in addition, that u0 is a subsolution of (5.2) in . Then, arguing
is a subsolution of (5.2) for
as above, the function x 7 u[,] (x, t; u0 ), x ,
all t > 0, because t 7 u[,] (, t; u0 ) is non-decreasing. Fix t > 0 and set
t := max u[,] (, t; u0 ).
M
j+1
As u[,] (, t; u0 ) is a subsolution of (5.2), it is also a subsolution of (4.33)

t . Thus, due to Theorem 4.5(a),
with j + 1, instead of j, for all M M
u[,] (, t; u0 ) [,j+1 ,M ]
in
j+1
t.
for all M M
Hence,
min
u[,] (, t; u0 ) lim [,j+1 ,M ] = L[,j+1 ]
in
j+1
and therefore, letting t yields

min
lim sup u[,] (, t; u0 ) L[,j+1 ]
in
j+1 .
Consequently, according to (5.31), we can conclude that

min
lim u[,] (, t; u0 ) = M[,j+1 ]
which ends the proof of (5.4) for j + 1.
in
(5.32)
124
Suppose dj+1 < dj+2 and for sufficiently large n N, say n n0 ,

consider the open sets defined by (4.35)
j+1,n := {x j+1 : dist (x, j+1 \ ) > 1/n}.
Pick n n0 . As j+1,n , it follows from (5.23) that there exists
M0 > 0 such that
u[,] (, t; u0 ) M
on j+1,n \
for all M M0 and t > 0. Thus, by the parabolic maximum principle,

u[,] (, t; u0 ) u[,j+1,n ,M ] (, t; u0 )
for all t > 0, where u[,j+1,n ,M ] stands
u
du = u + af (, u)u
t
u=M
u=0
u(, 0) = u0
in
j+1,n
(5.33)
for the unique solution of

in
on
on
in
j+1,n (0, ),
(j+1,n \ ) (0, ),
(j+1,n ) (0, ),
j+1,n .
By Theorem 1.7,
lim u[,j+1,n ,M ] (, t; u0 ) = [,j+1,n ,M ]
in
j+1,n ,
where [,j+1,n ,M ] is the unique positive solution of (4.36) at j + 1. Hence,

letting t in (5.33) shows that, for every n n0 ,
lim sup u[,] (, t; u0 ) [,j+1,n ,M ]
in
j+1,n .
(5.34)
Consequently, letting M in (5.34), we find that

min
lim sup u[,] (, t; u0 ) L[,j+1,n ]
in
j+1,n
(5.35)
for all n n0 . As (5.35) holds for all n n0 and, due to (4.39), we already
know that
max
min
L[,j+1 ] = lim L[,j+1,n ] ,
n
we can conclude from (5.35) that

max
lim sup u[,] (, t; u0 ) L[,j+1 ]

t
which ends the proof of the theorem.
in
j+1 ,
5.5
125
Approximating metasolutions by classical solutions
In this section we will compare the dynamics of the parabolic problem

u
in (0, ),
t du = u + (a(x) + b(x))f (x, u)u
(5.36)
u=0
on (0, ),
u(, 0) = u0 > 0
in ,
satisfies
with the dynamics of (1.1) as 0. Here, b C ()
b(x) < 0
for all
0 = a1 (0)
x
and > 0 is a constant. Naturally, a and f are assumed to satisfy (Ha) and
(KO), respectively. Thus,
a(x) + b(x) < 0
for all
and (5.36) is a classical generalized diffusive logistic equation, of the same

type as those analyzed in Chapter 2. Consequently, owing to Theorem 2.2,
the next result holds.
Theorem 5.4 Suppose a(x) and f satisfy (Ha) and (Hf)(Hg), respectively.
Then, for every > 0, (5.36) possesses a unique positive steady state, [,,] ,
if and only if, > d0 . Moreover, [,,] 0 and
if d0 ,
0,
lim u[,,] (, t; u0 ) =
in C(),
(5.37)
t
[,,] ,
if > d0 ,
where u[,,] stands for the unique solution of (5.36).
Theorem 5.5 Suppose a(x) and f satisfy (Ha) and (KO), respectively, and
set q0 +1 := , as usual. Then,
(a) For every (d0 , d1 ),
lim [,,] = [,]
0
in C(),
(5.38)
where [,] is the unique positive steady-state of (1.1).

(b) For every 1 j q0 and [dj , dj+1 ),
lim [,,] = Mmin
[,j ]
0
in .
(5.39)
126
Proof: Fix (d0 , d1 ). Then, for every > 0, [,,] is the unique positive
solution of

du = u + (a(x) + b(x))f (x, u)u
in ,
(5.40)
u=0
on ,
whereas, according to Theorem 4.1, [,] is the unique solution of (5.40) with
= 0. Subsequently, we will set
:= [,,] ,
> 0,
0 := [,] .
As in Chapter 2, the positive solutions of (5.40) are regarded as the positive

C0 ()
defined by
zeroes of the nonlinear operator F : R C0 ()
F(, u) := u (d)1 [u + (a + b)f (, u)u] ,
R, u C0 (),
where (d)1 stands for the resolvent operator of d in under homogeneous Dirichlet boundary conditions. The operator F is of class C 1 and, by
elliptic regularity, F(, ) is a nonlinear compact perturbation of the identity
map for all R. Since F(0, 0 ) = 0, by Lemma 2.1, 0 0 and
= 1 [d af (, 0 ), ].
(5.41)
Moreover, differentiating F with respect to u yields

f
1
Du F(0, 0 )u = u (d)
u + a (, 0 )0 u + af (, 0 )u
u
Hence, Du F(0, 0 ) is a Fredholm operator of index zero.
for all u C0 ().
Moreover, as it is injective, it is also a linear topological isomorphism. Indeed,
if

f
1
u (d)
u + a (, 0 )0 u + af (, 0 )u = 0
u
then, by elliptic regularity, u C 2+ ()
and
for some u C0 (D),
0
Lu = 0
in
(5.42)
where
f
(, 0 )0 af (, 0 ).
(5.43)
u
On the other hand, since a(x) < 0 for all x , by the monotonicity
properties of the principal eigenvalue, it follows from (5.41) and (Hf) that
L := d a
1 [L, ] > 1 [d af (, 0 ), ] = 0.
Hence, (5.42) implies u = 0. So, Du F(0, 0 ) is a linear topological isomorphism.
Therefore, combining the uniqueness of the positive solutions with the implicit
127
function theorem, it becomes apparent that (5.38) holds, which ends the proof
of Part (a).
Pick 1 j q0 and [dj , dj+1 ). Let 0 be any principal eigenfunction associated to 0 . Then, u0 := is a subsolution of (5.40) for sufficiently small 0 and > 0. Moreover, > 0 can be shortened, if necessary,
so that
u0 < Mmin
[,j ] .
According to D. Sattinger [220], u[,] (, t; u0 ) 0 also is a subsolution of
(5.40) for sufficiently small > 0 and all t > 0. Thus, by Theorem 5.2(c),
lim u[,,0] (, t; u0 ) = Mmin
[,j ] .
(5.44)
In particular, for any compact subset K j and > 0 there exists a time
t > 0 such that
ku[,,0] (, t; u0 ) Mmin
[,j ] kC(K) /2
for all t t .
(5.45)
On the other hand, by continuous dependence with respect to the parameter

(e.g., Chapter 8 of A. Lunardi [186]), there exists 0 := (t ) > 0 such that,
for every (0, 0 ],
ku[,,] (, t; u0 ) u[,,0] (, t; u0 )kC(K) /2
for all t t .
(5.46)
Hence, by (5.45) and (5.46), we find that

ku[,,] (, t ; u0 ) Mmin
[,j ] kC(K)
for all 0 < 0 .
(5.47)
Shortening 0 , if necessary, we can assume that u0 also is a subsolution of

(5.40) and hence,
u[,,] (, t ; u0 ) u[,,] (, t; u0 ) < [,,] < Mmin
[,j ]
(5.48)
for all t t . According to (5.47) and (5.48), we find that

ku[,,] (, t; u0 ) Mmin
[,j ] kC(K)
for all t t
and 0 < 0 . Therefore, letting t , it follows from Theorem 5.4 that

k[,,] Mmin
[,j ] kC(K)
for all 0 < 0 .
Consequently,
lim [,,] = Mmin
[,j ]
0
uniformly on compact subsets of j .
(5.49)
To complete the proof of (5.39) it remains to prove that

lim [,,] =
0
j ) (j ).
in ( \
(5.50)
128
Indeed, by (5.44), for any positive constant C > 0, there exists a time tC > 0
such that, for every t tC ,
u[,,0] (, t; u0 ) u[,,0] (, tC ; u0 ) > C
j ) (j ). (5.51)
in ( \
On the other hand, by continuous dependence with respect to , we find from

(5.51) that there exists 0 > 0 such that
j ) (j )
in ( \
u[,,] (, tC ; u0 ) > C
for all 0 0 . Consequently, for every t tC and 0 0 ,

j ) (j ),
in ( \
u[,,] (, t; u0 ) u[,,] (, tC ; u0 ) > C
since t 7 u[,,] (, t; u0 ) is increasing. Thus, letting t , it follows from

Theorem 5.4 that
j ) (j )
in ( \
[,,] C
for all 0 0 , which shows (5.39) and ends the proof of the theorem.
Remark 5.6 By the Schauder interior estimates, it is easily seen that (5.49)
2+
implies the convergence of [,,] to Mmin
(K) for every
[,j ] as 0 in C
compact subset K of j .
Theorem 5.5(b) provides us with a reasonable numerical scheme to approximate the minimal metasolutions supported in j in the interval of s where
they are attractors from below, i.e., dj < dj+1 . By obvious reasons,
this scheme cannot be used to compute them outside these ranges of .
5.6
Pattern formation in classical logistic problems
Let ij , 1 j q, 1 i mj , be an arbitrary family of smooth subdomains

of satisfying
i
i =
if (i, j) =
6 (i, j)
j
j
and
j := 1 [, ij ],
j < j+1 ,
1 i mj , 1 j q,
1 j q 1,
and consider the open subset of defined by

O :=
mj
q [
[
j=1 i=1
ij .
129
Then, (1.1) is a classical diffusive logistic

Suppose a(x) > 0 for all x .
parabolic problem, whose dynamics are governed by its maximal non-negative
steady state solution. Suppose
max a ' > 0
minK a is separated away from zero,

while, for any compact subset K \ O,
then, by continuous dependence, Theorem 5.5 provides us with the shape of
u(x, t; u0 ), for sufficiently large t, according to the different ranges of values of
the parameter considered in Theorem 5.5. So, Theorems 5.2 and 5.5 provide
us with all the admissible patterns exhibited by the classical diffusive logistic
equations through the dynamics of the associated degenerated situations.
5.7
Biological discussion
From the perspective of population dynamics, according to Theorem 5.2, a

species u becomes extinct if d0 . For a given species with an intrinsic
growth rate this occurs if d d0 := /0 , which measures the critical size
of the dispersion rate d, d0 , so that the habitat can support it. If d d0 , then
the individuals of the species approximate the habitat edges at a sufficiently
high rate as to drive the species to extinction as a consequence of the hostility
of the surroundings of the inhabiting area .
Adopting another perspective, for any given > 0 and d > 0, the extinction can be avoided by placing the species u in a sufficiently large habitat .
Indeed, according to the FaberKrahn inequality, we already know that there
is a universal constant C > 0 such that
0 = 1 [, ] C||2/N ,
C := 1 [, B1 ]|B1 |2/N ,
where N is the spatial dimension. Therefore, as soon as the measure of the

habitat satisfies
|| (Cd/)N/2 ,
cannot support a population u with an intrinsic growth rate dispersing at
the rate d. Conversely, if is sufficiently large to contain some ball of radius
R > 0, BR (x0 ), then, by the monotonicity of 1 [, D] with respect to D, it
is apparent that
1 [, ] 1 [, BR (x0 )] = 1 [; BR (0)] = 1 [; B1 ]R2
and hence,
lim 0 = lim 1 [, ] = 0.
Thus, > d0 for sufficiently large R. Consequently, in sufficiently large
130
habitats the species u will be always permanent. Naturally, once we have

fixed , the bigger the dispersion rate, measured by d, the larger should be
the inhabiting area to avoid extinction.
It should be also noted that even the simplest linear problem
u
in (0, ),
t du = u
(5.52)
u=0
on (0, ),
u(, 0) = u0 > 0
in ,
predicts extinction if < d0 , as we have already discussed in Chapter 1. So,
we should not expect to be able to support any species u with the same
intrinsic growth rate, , and dispersion rate, d, if, in addition, we are incorporating interspecific competition of the individuals for the available resources,
measured by af (, u)u 0.
Theorem 5.2 predicts that the behavior of the species u should be of classical logistic type if d0 < < d1 . The condition > d0 allows to support
the species u independently of the level of intraspecific competition, measured
by af (, u)u. In the context of population dynamics, the condition
< d1 = d1 [, i0,1 ],
1 i m1 ,
means that the largest protection zones, measured by the size of the associated principal eigenvalues, cannot support the species u in the absence of
intraspecific competition effects, because the solutions of the linear problem
u
in i0,1 (0, ),
t du = u
u=0
on i0,1 (0, ),
(5.53)
u(, 0) = u0 > 0
in i0,1 ,
approximate 0 as t , for all 1 i m1 . This might be attributable to the
lack of room in these protection zones to accumulate the necessary resources
to maintain the species u therein. Note that we are measuring the size of the
protection zones through the size of the principal eigenvalue of . Theorem
5.2 establishes that the previous situation changes dramatically when
d1 < < d2 = d1 [, i0,2 ],
1 i m2 ,
as, in such case, the largest protection zones i0,1 , 1 i m1 , can support the
species u, in the sense that the solutions of (5.53) exhibit a genuine Malthusian
growth therein if > d1 , though the smaller protection zones, i0,j , 2
j q0 , 1 i mj , cannot support the species u because < d2 . In
such circumstances, Theorem 5.2 predicts exponential growth in the largest
protection zones,
m1
[
0,1 =
i0,1
i=1
and logistic growth in their complement

0,1 .
1 := \
131
More generally, when q0 3 and dj < < dj+1 for some 2 j q0 1,

any protection zone of the form i0,k , 1 k j, 1 i mk , can support u,
in the sense that the solutions of the linear problems
u
in i0,k (0, ),
t du = u
u=0
on i0,k (0, ),
(5.54)
u(, 0) = u0 > 0
in i0,k ,
grow exponentially for all 1 k j and 1 i mk . However, the smaller
protection zones i0,k , j + 1 k q0 , 1 i mk , cannot do it, in the
sense that the solutions of (5.54) become extinct for all j + 1 k q0 ,
1 i mk . In these circumstances, Theorem 5.2 predicts exponential growth
of the species u in the largest protection zones
j m
[
[k
i0,k
k=1 i=1
and logistic growth in the complement

j :=
q0
m
[
[k
i0,k .
k=j+1 i=1
When > dq0 , there is exponential growth in all the protection zones and
logistic growth in , the region where the behavior of the species has been
assumed to be of logistic type for the non-spatial model.
5.8
The first two parts of Theorem 5.2 go back to J. M. Fraile et al. [90] and
are valid for general second order elliptic operators like (1.48) under mixed
boundary conditions (see Theorem 3.7(a)(b) of [90]). Actually, according to
the proof of Theorem 3.7(c) of [90], it is apparent that
lim u[,] (x, t; u0 ) =
(5.55)
for all x 0,1 and d1 . The problem of the asymptotic behavior of the
solutions of (1.1) was addressed in J. M. Fraile et al. [90].
The first occasion that Theorem 5.2 was stated in its greatest generality
was in the four summarizing items on pages 573 and 574 of R. Gomez-Re
nasco
and J. L
opez-G
omez [109], submitted in September 1998. The pictures included in Section 5.3 were taken from [109]. The detailed proofs of these results were given in J. L
opez-Gomez [147], which was submitted for publication
132
in August 1999, some months before the work of Y. Du and Q. Huang [80]
was published, where, in the very special case when 0 is connected, (5.55)
was complemented by establishing the validity of the estimate
max
Lmin
[, ] lim inf u[,] (, t; u0 ) lim inf u[,] (, t; u0 ) L[, ]
t
in .
Consequently, though Y. Du and Q. Huang [80] got the previous estimates in

the simplest case when 0 consists of a single component, the first two parts
of Theorem 5.2 are attributable to J. M. Fraile et al. [90] and Theorem 5.2(c)
is attributable to R. G
omez-Re
nasco and J. Lopez-Gomez [109]. Indeed, [109]
was part of the Ph.D. thesis of R. Gomez-Re
nasco [105], under the supervision
of J. L
opez-G
omez, in the University of La Laguna (Tenerife, The Canary
Islands, Spain). The Ph.D. thesis was presented in February 1999 and defended
early May 1999. Professors H. Amann, C. J. Bud and V. Hutson took part in
the judging committee and received the thesis on March 1999.
Theorem 5.5 goes back to J. Lopez-Gomez [150] and it was published in
2001. It is an extremely sharp version of Theorem 5.9 of Y. Du [79], which
goes back to Y. Du [78], published in 2004.
As [147] contained a gap detected by the author from a deep question
raised by Professor H. Amann during a seminar given by the author in the
Institute of Mathematics of the University of Z
urich on June 20, 2002, the
complete technical details of the proof of Theorem 5.2(c) were given in J.
L
opez-G
omez [155].
Theorem 5.2 can be extended to a wide class of sublinear elliptic systems of
cooperative type according to P. Alvarez-Caudevilla

and J. Lopez-Gomez [9].
It also works if, instead of Dirichlet boundary conditions on all the components
of , we consider a boundary condition of the type
u
+ i (x)u = 0
n
along some of the components of which are components of , 1 i
m, where n stands for the outward unit normal to along and i are
appropriate continuous functions on the corresponding components of .
The only necessary change in the statement of Theorem 5.2 in this case is
that 0 should stand for the principal eigenvalue of under these boundary
conditions.
Going further, one might even impose these boundary conditions not only
on the common components of and but also on the common components of with 0 . However, in this case the solutions and large solutions
might blow up on these components of the boundary and, as a result, some
changes are necessary for the validity of the results of Part I. To keep the
number of technicalities under control in this book, we have preferred to work
with Dirichlet boundary conditions to reduce the complexity of the notations
to a reasonable level. Although Y. Du and Q. Huang [80] dealt with a general
boundary operator on , it should not be forgotten that in [80] the open set
133
0 . Naturally, in this special

0 consisted of a single component with
case, technicalities are much easier to overcome and one can even impose
Bu = 0
on , where B is a mixed boundary operator of the type (1.49). Naturally,
when working with these general boundary operators on it seems a categorical imperative to use Theorem 3.7 of J. M. Fraile et al. [90].
More recently, D. Aleja and J. Lopez-Gomez [6, 7, 8] analyzed the dynamics
of the positive solutions of the parabolic problem
in , t > 0,
t u = [Du um] + u(m aup )
Dn u un m = 0
on , t > 0,
(5.56)
u(, 0) = u0 > 0
in ,
where is a smooth bounded domain (open and connected set) of RN , N 1,
satisfies a 0, a 6= 0, n stands for the
D > 0, > 0, R, p 1, a C()
is a function such that, for
outward unit normal along , and m C 2 ()
some x+ , x ,
m(x+ ) > 0,
m(x ) < 0.
(5.57)
A less general version of this model, with = 1, p = 2 and a(x) > 0 for
was introduced by F. Belgacem and C. Cosner [21] to analyze
all x ,
the effects of advection along environmental gradients to reaction diffusion
models for population dynamics with dispersal. Although in the special case
it is well known that the dynamics of (5.56)
when a(x) > 0 for all x
are regulated by its non-negative steady states; in the general case when the
weight function a(x) vanishes somewhere in , the dynamics of (5.56) might be
governed by the associated metasolutions, as it has been recently established
in [8], which is the first work where metasolutions have been documented to
regulate the dynamics of a parabolic problem equation with drift terms.
Part II
Uniqueness of the large

solution
Chapter 6
A canonical one-dimensional
problem
6.1
6.2
6.3
6.4
6.5
Existence and uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Auxiliary function b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Getting sharp estimates for `(x) at x = 0 through b . . . . . . . . . . . .
The exact blow-up rate of `(x) at x = 0 . . . . . . . . . . . . . . . . . . . . . . . . .
139
148
159
164
165
The main goal of the second part of this book is to obtain general uniqueness
results for the large solutions constructed in the first part. The whole program
adopted to accomplish this task will be divided into three steps delimited
by each of the chapters included in this part. This chapter establishes the
existence and the uniqueness of the positive solution, `(x), of the singular
boundary value problem
00
u (x) = a(x)h(u(x)),
x > 0,
(6.1)
u(0) = , u() = 0,
under the following conditions on a(x) and h(u):
(A1) a C[0, ) satisfies a(x) a(y) > 0 whenever x y > 0.
(A2) h C 1 [0, ) satisfies h(0) = h0 (0) = 0, h0 (u) > 0 if u > 0, as well as the
KellerOsserman condition
Z
dx
qR
I(u) :=
< , u > 0,
lim I(u) = 0.
(6.2)
x
u
u
h
u
Moreover, it will ascertain the blow-up rate of `(x) at x = 0 when, besides
(A1) and (A2), the next conditions holds.
(A3) The quotient g(u) := h(u)/u, u > 0, satisfies g 0 (u) 0 for all u > 0 and
the limit
h(u)
H := lim p > 0
(6.3)
u u
is well defined for some p > 1.
137
138
Note that (6.3) implies (6.2). Indeed, by (6.3), there exists z > 0 such that
h(u)
H p
u ,
2
u z.
Hence, since p > 1, we have that

Z
z
Z
z
r
21
Z
2(p + 1)
dx
< .
h(s) ds
dx
p+1 z p+1
H
x
z
Naturally, by u(0) = and u() = 0 it is meant that

lim u(x) = ,
lim u(x) = 0.
x0
It should be noted that, according to the notations introduced in Part I,

a(x) := a(x)/d
and
h(u) := g(u)u.
Moreover, as discussed in the final section of Chapter 3, condition (6.2) for

u > u (see (3.46)) entails (KO) when h(u) is defined as in the proof of
Theorem 3.2. Consequently, as (6.2) is stronger than (KO), all the theory
developed in Chapter 3 can be applied here.
This chapter is organized as follows. Section 6.2 shows the existence and
the uniqueness of the positive solution of (6.1), and of some closely related
problems whose significance will become apparent later. For any given R > 0,
Section 6.3 studies some of the most basic properties of the function
Z
Z
b(x) =
a
x
1
p+1
p+1
p1
dy,
x (0, R].
Section 6.4 shows that (b, b) provides us with a sub-supersolution pair of

(6.1) in (0, ] for sufficiently small > 0 and > 0 and sufficiently large , in
order to establish that
lim
x0
p
`(x)
= I0p1
b(x)
where
I0 := lim
x0
p+1
p1
p+1
p1
H p1 ,
b(x)b00 (x)
(0, ).
[b0 (x)]2
Some sufficient conditions for 0 < I0 < are collected in Section 6.2. Note
that if b(x) is logarithmically convex, then I0 1 (see the end of Section
3.4). Precisely, if there exists > 0 such that a C 3 (0, ], a0 (x) > 0 and
aa00
(ln a)00 (x) < 0 for all x (0, ), and the function (a
0 )2 is non-oscillating in
(0, ), in the sense discussed in Definition 6.6, then I0 [1, ).
A canonical one-dimensional problem
6.1
139
Existence and uniqueness
The following result shows the existence and the uniqueness of the positive
solution of (6.1), `(x), and establishes some of its basic properties.
Theorem 6.1 Suppose a(x) and h(u) satisfy (A1) and (A2). Then, the problem (6.1) possesses a unique positive solution, `(x), x > 0. Moreover,
` 0 (x) < 0,
`(x) > 0,
` 00 (x) > 0,
x > 0,
(6.4)
and
lim ` 0 (x) = ,
x0
lim ` 0 (x) = 0.
(6.5)
Consequently, the graph of `(x) looks like Figure 6.1.

Proof: Subsequently, for every M > 0 and L > 0, we will consider the problem
00
u (x) = a(x)h(u(x)),
0 < x < L,
(6.6)
u(0) = M, u(L) = 0.
By our assumptions on a(x) and h(u), it is apparent that (u, u) := (0, M )
is an ordered sub-supersolution pair of (6.6). Thus, by Theorem 1.2, (6.6)
possesses a solution u C 2 [0, L] with u(x) [0, M ] for all x [0, L]. Since
u(0) = M > 0, u 6= 0. Thus,
u(x) > 0
for all x [0, L).
FIGURE 6.1: The graph of `(x).
(6.7)
140
Indeed, if there is x0 (0, L) such that u(x0 ) = 0, then u0 (x0 ) = 0 and (u, u0 )
solves the Cauchy problem
u0 = v,
v 0 = a(x)h(u),
u(x0 ) = v(x0 ) = 0,
whose unique solution is (u, v) = (0, 0), by the CauchyLipschitz theorem.

This contradiction shows (6.7) and establishes that u0 (L) < 0. Moreover,
since
u00 (x) = a(x)h(u(x)) > 0
for all x (0, L), u0 (x) is increasing. So, for every x [0, L]
u0 (x) u0 (L) < 0.
Therefore, u is decreasing.
Now, we will show the uniqueness of the solution of (6.6). Our proof will
proceed by contradiction. If u1 6= u2 are two positive solutions of (6.6) such
that u1 (x0 ) > u2 (x0 ) for some x0 (0, L), then there are x1 [0, x0 ) and
x2 (x0 , L] such that
u1 (x1 ) = u2 (x1 ),
u1 (x2 ) = u2 (x2 ),
and
x (x1 , x2 ).
u1 (x) > u2 (x)

Pick xm (x1 , x2 ) such that
u1 (xm ) u2 (xm ) =
max (u1 (x) u2 (x)) > 0.
x[x1 ,x2 ]
Then,
0 (u1 u2 )00 (xm ) = a(xm ) [h(u1 (xm )) h(u2 (xm ))] > 0,
by (A1) and (A2). This contradiction shows the uniqueness. Subsequently, we
will denote by u[M,L] the unique positive solution of (6.6).
Next, we will show that M1 M2 > 0 and L1 L2 > 0 imply
M1 u[M1 ,L1 ] u[M2 ,L2 ]
in
[0, L2 ].
(6.8)
Indeed, the function

u := u[M1 ,L1 ] |[0,L2 ]
00
u (x) = a(x)h(u(x)),
u(0) = M2 , u(L2 ) = 0,
0 < x < L2 ,
in the interval [0, L2 ], because it solves the differential equation and

u(0) = M1 M2 ,
u(L2 ) 0.
(6.9)
141
is a subsolution. Therefore, (6.8) follows from Theorem

Clearly, u := 0 u
1.2, since u[M2 ,L2 ] is the unique positive solution of (6.9). The first estimate
of (6.8) follows from the fact that u[M,L] is decreasing.
According to (6.8), for any M > 0, the point-wise limit
uM (x) := lim u[M,L] (x)
L
(6.10)
is well defined for all x [0, ). Moreover, 0 uM M in [0, ). Actually,

as the set of prolonged functions

u[M,L] ,
in [0, L],
F := u
[M,L] , L > 0 ,
where u
[M,L] :=
0,
in (L, ),
is bounded in C[0, ), one can infer from the AscoliArzela theorem that
uM 0 provides us with a solution of
u00 (x) = a(x)h(u(x)),
x > 0,
such that u(0) = M . Arguing as above, we find that uM (x) > 0 for all x > 0.
Moreover, by (6.10), uM is non-increasing. Hence, u0M (x) 0 for all x > 0.
Also, u00M (x) > 0 for all x > 0. Thus, u0M is increasing in [0, ). Consequently,
u0M (x) < 0 for all x > 0 and the next limit is well defined:
:= lim uM (x) 0.
x
Suppose > 0, Then, for every x > 1, we find from (A1) and (A2) that
u00M (x) = a(x)h(uM (x)) a(1)h() > 0
and hence, for every x > 1, we find that
Z x
u0M (x) = u0M (1) +
u00M (s) ds u0M (1) + a(1)h()(x 1),
1
which is impossible because this estimate implies u0M (x) > 0 for sufficiently
large x and we already know that uM is decreasing. Therefore, = 0 and uM
is a positive solution of
00
u (x) = a(x)h(u(x)),
x > 0,
(6.11)
u(0) = M, u() = 0.
Now, we will show that uM is the unique positive solution of (6.11). Suppose
that u1 =
6 u2 are two solutions of (6.11) with u1 (x0 ) > u2 (x0 ) for some x0 > 0.
Then, arguing as in the proof of the uniqueness for (6.6), there are x1 [0, x0 )
and x2 (x0 , ] with u1 (x1 ) = u2 (x1 ), u1 (x2 ) = u2 (x2 ), and u1 (x) > u2 (x)
for all x (x1 , x2 ). Pick xm (x1 , x2 ) such that
u1 (xm ) u2 (xm ) =
max (u1 (x) u2 (x)) > 0.
x[x1 ,x2 ]
142
Then, by (A1) and (A2),

0 (u1 u2 )00 (xm ) = a(xm ) [h(u1 (xm )) h(u2 (xm ))] > 0,
which is impossible. Therefore, uM is the unique positive solution of (6.11).
and L > 0 we have that u[M,L]
Owing to (6.8), for every 0 < M < M
u[M ,L] . Hence, letting L implies uM uM . Actually,
uM (x) < uM (x)
for all x 0.
(6.12)
Indeed, if uM (x0 ) = uM (x0 ) for some x0 > 0, then u0M (x0 ) = u0M (x0 ) because
uM uM , and hence, by the uniqueness of solution for the Cauchy problem,
u0 = v,
v 0 = a(x)h(u),
u(x0 ) = uM (x0 ),
u0 (x0 ) = u0M (x0 ),
, a contradiction. Consequently,
necessarily uM = uM , which implies M = M
(6.12) holds.
Subsequently, for every L > 0 we will consider the singular problem
00
u (x) = a(x)h(u(x)),
0 < x < L,
(6.13)
u(0) = , u(L) = .
As h satisfies (A2), it follows from Theorem 3.4 that (6.13) possesses minimal
and maximal positive solutions. Let denote by `min the minimal one.
Fix M > 0 and L > 0. Then, there is > 0 for which
uM M `min
[0, ] [L , L].
in
Thus, `min |[,L] is a supersolution of

00
u (x) = a(x)h(u(x)),
u() = uM (), u(L ) = uM (L ).
< x < L ,
(6.14)
The proof of the uniqueness of the positive solution for (6.6) and (6.11) can be
easily adapted to show that uM is the unique positive solution of (6.14). As
u := 0 is a subsolution of (6.14) with u `min |[,L] , we find from Theorem
1.2 that (6.14) possesses a solution between 0 and `min , necessarily uM . Thus,
uM `min in [, L ]. Therefore,
uM `min
in
[0, L].
(6.15)
So, by the monotonicity of uM with respect to M , the point-wise limit

` := lim uM
M
(6.16)
is well defined in [0, ). Moreover, by (6.15), the AscoliArzela theorem guarantees that ` solves (6.1). By (6.16), `(x) > 0 for every x > 0 since uM (x) > 0
143
and the mapping M 7 uM (x) is increasing. Moreover, ` is non-increasing,

because uM is decreasing for all M > 0. Thus, `0 0. As for all x > 0
` 00 (x) = a(x)h(`(x)) > 0,
the derivative `0 is increasing. Therefore, `0 (x) < 0 for all x > 0 and (6.4)
holds. Actually, ` is the minimal positive solution of (6.1), because any other
solution, u(x), satisfies u uM for all M > 0 and so, u `. Consequently,
we will subsequently denote it by
`min := `.
To show the existence of a maximal solution for the problem (6.1) we consider
the problems
00
u (x) = a(x)h(u(x)),
x > ,
(6.17)
u() = , u() = 0,
where > 0. By making the change of variable y := x , (6.17) fits into the
abstract setting of (6.1). Thus, for every > 0, (6.17) possesses a minimal
positive solution, denoted by `min . Similarly, for every M > 0 and > 0,
00
u (x) = a(x)h(u(x)),
x > ,
(6.18)
u() = M, u() = 0,
has a unique positive solution, denoted by uM . Moreover,
lim uM = `min .
Pick 0 < < . In the interval [
, ), uM
00
u (x) = a(x)h(u(x)),
x > ,
), u() = 0,
u(
) = uM (
(6.19)
because
uM (
) < uM () = M = uM
(
).
Moreover, 0 is a subsolution. Thus, by uniqueness,
uM uM
in [
, ).
Thus, letting M yields
`min `min
in
[
, ).
From these estimates, it is easily seen that the point-wise limit

`max := lim `min
0
is well defined and it provides us with a positive solution of (6.1). As for any
solution u of (6.1) and > 0, we have that u `min in [, ), letting 0
144
yield u `max . Therefore, `max is the maximal solution of (6.1). To establish

the uniqueness of the solution of (6.1) it remains to prove that
`min = `max .
(6.20)
Note that all previous solutions are non-negative. So, by the uniqueness of the
associated Cauchy problems, they must be positive.
Let ` be an arbitrary positive solution of (6.1) and, for any fixed > 0,
consider
:= `(x ),
`(x)
x > .
According to (6.1), for every x > we have that
`00 (x) = ` 00 (x ) = a(x )h(`(x )) a(x)h(`(x)).

Hence, ` is a supersolution of (6.18) for all M > 0. Thus, for every M > 0
and > 0, we have that
` uM
in
[, )
for all
M > 0, > 0.
(6.21)
To prove (6.21) we will proceed by contradiction. Suppose (6.21) fails. As
`()
= , (6.21) holds for sufficiently small x > 0. Thus, there exist
x1 > and x2 (x1 , ] such that
1 ) = u (x1 ),
`(x
M
2 ) = u (x2 ),
`(x
M
and
= `(x ),
uM (x) > `(x)
x (x1 , x2 ).
Let xm (x1 , x2 ) be such that

m) =
uM (xm ) `(x
max
x[x1 ,x2 ]
uM (x) `(x)
.
Then,

00 (xm ) = a(xm ) h(u (xm )) h(`(x
m )) > 0,
0 (uM `)
M
which is impossible. Thus, (6.21) holds. Now, letting M in (6.21) yields
`min = lim uM ` = `( )
M
in
[, ),
where `min stands for the minimal positive solution of (6.17). Consequently,
letting 0, we find that, for every x > 0,
`(x) lim `min (x) = `max (x).
0
In particular, `min `max . Therefore, (6.20) holds, which concludes the proof
of the uniqueness.
145
To complete the proof of the theorem, it remains to show (6.5), where `

stands for the unique positive solution of (6.1). Since ` 00 (x) > 0 for all x > 0,
` 0 is increasing. Thus, since ` 0 < 0, the limit
` 00 := lim ` 0 (x) [, 0)
x0
is well defined. Suppose ` 00 (, 0). Then, for every x > y > 0,

Z x
`(x) `(y) =
` 0 ` 00 (x y).
y
Hence,
`(y) `(x) ` 00 (x y),
which is impossible, because
lim `(y) = .
y0
` 00
Therefore,
= , which is the first limit of (6.5).
Lastly, for each integer n 2, there exists xn [n 1, n] such that
Z n
`(n) = `(n 1) +
` 0 = `(n 1) + ` 0 (xn ).
n1
Thus, since limx `(x) = 0, we find that

lim ` 0 (xn ) = 0.
Therefore, since ` 0 0 is increasing, it becomes apparent that

lim ` 0 (x) = 0.
This establishes (6.5) and ends the proof of the theorem.
Similarly, the next result holds.

Theorem 6.2 Suppose a(x) and h(u) satisfy (A1) and (A2), and there are
L > 0 and M [0, ) for which
00
u (x) = a(x)h(u(x)),
x (0, L),
(6.22)
u(0) = , u(L) = M,
has a positive solution U (x), 0 < x L, with U 0 (L) < 0. Then, U (x) is the
unique positive solution of (6.22).
Remark 6.3 (a) When M = 0, any positive solution, u, of (6.22) must
satisfy u0 (L) < 0. Indeed, necessarily u0 (L) 0. So, if, in addition,
u0 (L) = 0, then u(L) = u0 (L) = 0, which implies u = 0, because, due to
the Cauchy-Lipschitz theorem, (u, v) = (0, 0) is the unique solution of
u0 = v,
v 0 = a(x)h(u),
u(L) = v(L) = 0.
146
(b) As a direct consequence of Theorem 6.2, for every L > 0, the function
`(x) is the unique positive solution of
00
u (x) = a(x)h(u(x)),
x (0, L),
u(0) = , u(L) = `(L),
because `0 (L) < 0, by (6.4).
(c) For sufficiently large M > 0, (6.22) might not admit a solution U with
U 0 (L) < 0, but, instead, U 0 (L) > 0. Indeed, as M , these solutions
approximate a solution of
00
u (x) = a(x)h(u(x)),
x (0, L),
u(0) = , u(L) = .
In such cases, whether or not uniqueness occurs might depend on some
hidden growth properties of h(u). Therefore, U 0 (L) < 0 might be really
necessary for the validity of Theorem 6.2.
Proof of Theorem 6.2: Fix L > 0 and M 0, and consider the problem
(6.22). The existence of minimal and maximal solutions follows the general
scheme of Theorem 6.1. Indeed, for every N > 0, the problem
00
u (x) = a(x)h(u(x)),
x (0, L),
u(0) = N, u(L) = M,
possesses a unique positive solution, uN , and
`min := lim uN
N
provides us with the minimal positive solution of (6.22). Similarly, for sufficiently small > 0, we denote by `min the minimal positive solution of
00
u (x) = a(x)h(u(x)),
x (, L),
(6.23)
u() = , u(L) = M.
Then, the point-wise limit
`max := lim `min
0
provides us with the maximal positive solution of (6.22).

It remains to prove the uniqueness. Let L > 0 and M 0 for which (6.22)
admits a positive solution U (x) with U 0 (L) < 0. Then, for sufficiently small
> 0,
U (L ) U (L).
Hence, the auxiliary function
(x) := U (x ),
U
x L,
147
satisfies
() = U (0) = ,
U
(L) = U (L ) U (L) = M,
U
and, for each x (, L),

00 (x) = U 00 (x ) = a(x )h(U (x ))
U
(x)) a(x)h(U
(x)).
= a(x )h(U
(x) is a supersolution of (6.23). Hence, arguing as in the proof of
Thus, U
Theorem 6.1 yields
(x) = U (x ) ` (x)
U
min
for all x (, L].
Consequently, letting 0, we find that

U (x) lim `min (x) = `max (x)
0
for all x (0, L]. Therefore,

U = `max .
In other words, U = `max is the unique positive solution of (6.22) with U 0 (L) <
0. To prove that `min = `max we will proceed by contradiction. Suppose
`min < `max = U.
Then, `0min (L) 0 and, for every
v0 (`0max (L), `0min (L)) ,
the unique solution of the Cauchy problem
u0 = v,
v 0 = a(x)h(u),
u(L) = M, u0 (L) = v0 ,
must be globally defined in (0, L] and satisfy

`min (x) < u(x) < `max (x)
for all x (0, L).
(6.24)
This is impossible, since u would provide us with a solution of (6.1) such that
u0 (L) = v0 < 0 but u =
6 U , which contradicts the previous uniqueness result.
The estimate (6.24) holds true because for every (0, L) and N > 0, the
00
u (x) = a(x)h(u(x)),
x (, L),
u() = N, u(L) = M,
possesses a unique positive solution. This ends the proof.
148
6.2
Auxiliary function b
In this section we suppose that a satisfies (A1), fix R > 0 and p > 1, and
consider the function b defined through
Z
b(x) :=
x
Z
1
,
A
A(x) :=
1
p+1
p+1
p1
x (0, R],
(6.25)
i.e.,
R
Z
b(x) =
p+1
p1
x (0, R].
dy,
a p+1
0
The following result demonstrates the main properties of this function.

Lemma 6.4 The function b defined by (6.25) satisfies b C 2 (0, R] and
b(x) > 0,
b0 (x) < 0,
b00 (x) > 0,
x (0, R).
(6.26)
Moreover,
lim b0 (x) = ,
lim b(x) = ,
x0
and
lim
x0
(6.27)
x0
b0 (x)
b00 (x)
b00 (x)
= lim
=
lim
= .
x0 b0 (x)
x0 b(x)
b(x)
(6.28)
Proof: By (A1), A C 1 [0, R], A(0) = 0 and A(x) > 0 for all x (0, R]. Thus,
1/A C 1 (0, R] and b C 2 (0, R]. Clearly, b(x) > 0 for each x (0, R) and
b(R) = 0. Moreover,
Z x
1
1
1
p+1
0
p+1
A (x) =
A(x)
a p+1 (x) > 0
(6.29)
a
p1
0
for all x (0, R]. Hence,
b0 (x) =
1
< 0,
A(x)
b00 (x) =
A0 (x)
> 0,
A2 (x)
which concludes the proof of (6.26).

Next, we will prove (6.27). As a is non-decreasing and positive,
x
Z
A(x) =
1
p+1
p+1
p1
p+1
a p1 (R) x p1
for all x (0, R]. Thus,

Z
1
b(x) a p1 (R)
p+1
y p1 dy =
2

1
2
p 1 p1
a
(R) x p1 R p1 .
2
(6.30)
149
Consequently, the first limit of (6.27) holds true. The validity of the second
limit can be shown from the first identity of (6.30) taking into account that
A(0) = 0. This ends the proof of (6.27).
Lastly, we will show (6.28). By (A1), (6.29) implies that
A0 (x)
p+1
A(x) x1 ,
p1
x (0, R].
(6.31)
Thus, by (6.30) and (6.31), it becomes apparent that

lim
x0
A0 (x)
b00 (x)
=
lim
= .
x0 A(x)
b0 (x)
This establishes the second limit of (6.28).

On the other hand, since
b0 (x)
=
b(x)
A(x)
x
1
A
!1
x (0, R],
the first limit of (6.28) is equivalent to

Z
1
A
lim A(x)
x0
!
= 0.
(6.32)
To prove (6.32), let (0, R) and x (0, ). Then, since A is increasing,

Z
A(x)
x
1
=
A
A(x)
ds + A(x)
A(s)
1
x + A(x)
A
1
.
A
Thus, for every > 0,

Z
lim sup A(x)

x0
1
A
!
,
because A(0) = 0. Consequently, (6.32) holds. Finally, since

lim
x0
b00 (x)
b00 (x)
b0 (x)
= lim
lim
= ,
x0 b0 (x) x0 b(x)
b(x)
the proof is complete.
According to (6.29) and (6.30), we also have

b0 (R) < 0,
b00 (R) > 0.
The next lemma reveals an extremely important property of b that will be

very useful later.
150
Lemma 6.5 Suppose

lim
x0
b(x)b00 (x)
= I0 (0, ).
[b0 (x)]2
Then, the auxiliary function
a(x)bp (x)
b00 (x)
c(x) :=

p+1
p1
I0p ,
p+1
(6.33)
0 < x R,
(6.34)
x = 0,
lies in C[0, R] and it is bounded away from zero in [0, R ] for all 0 < < R.
Note that c(R) = 0, because b(R) = 0, a(R) > 0 and b00 (R) > 0.
Proof: Due to Lemma 6.4, c C(0, R] and c(x) > 0 for all x (0, R). Thus,
it suffices to prove that

lim c(x) =
x0
p1
p+1
p+1
I0p .
(6.35)
By (6.29) and (6.30), we find that

Z
p+1 1
b (x) =
p 1 A(x)
00
1
1
p+1
a p+1 (x) > 0,
x (0, R).
Thus, for every x (0, R),

p+1
Z x
p1
+1
1
1
p1
p
p+1
p+1
c(x) =
a(x)b (x)a
(x)
a
p+1
0
2p
p1
Z x
p
1
p1 p
p+1
p+1
=
b (x)a
(x)
a
p+1
0
"
2 #p
Z x
p1
1
1
p1
=
b(x)a p+1 (x)
a p+1
.
p+1
0
On the other hand, by (6.29), we have that

1
p + 1 p+1
A (x) =
a
(x)
p1
Z
1
p+1
2
p1
x [0, R].
Therefore, taking into account (6.30), we find that

c(x) =
p1
p+1
p+1
[A (x)b(x)] =
p1
p+1
p+1
b(x)b00 (x)
[b0 (x)]2
p
(6.36)
151
for all x (0, R). By (6.33), letting x 0 in (6.36) shows (6.35), which ends
the proof.
Subsequently, in order to discuss the meaning of condition (6.33), we will
introduce the quotient function
qb (x) :=
b(x)
1
=
,
b0 (x)
(ln b)0 (x)
x (0, R].
By Lemma 6.4, qb C 1 (0, R] and

lim qb (x) = 0.
x0
Thus, if we extend qb to [0, R] by qb (0) := 0, then qb C[0, R]. Moreover, for

every x (0, R], we have that
q0b (x) = 1
b00 (x)b(x)
.
[b0 (x)]2
Consequently, qb C 1 [0, R] if and only if the following limit exists

I0 := lim
x0
b(x)b00 (x)
.
[b0 (x)]2
In such case, by (6.26), I0 0. Thus, (6.33) can be equivalently expressed as

qb C 1 [0, R]
with q0b (0) < 1.
(6.37)
Note that if qb C 1 [0, R] but q0b (0) = 1, then I0 = 0. In such case, the function
c defined by (6.34) satisfies c(0) = 0 and the conclusions of Lemma 6.5 fail.
The next result will provide us with some sufficient conditions for (6.33), or,
equivalently, (6.37). To state it, we need the next concept.
Definition 6.6 A function C 1 (0, ], > 0, is said to be non-oscillating
in (0, ] if, either 0 (x) > 0 for each x (0, ], or 0 (x) < 0 for each x (0, ],
or = 0 in (0, ] for some constant 0 . The function is said to be nonoscillating at = 0 if it is non-oscillating in (0, ] for some 0 < .
Proposition 6.7 Suppose p > 1, a satisfies (A1), a(0) = 0, and there exists
0 < < R such that a C 2 (0, ],
a0 (x) > 0
(ln a)00 (x) < 0
and
for all
x (0, ].
(6.38)
Then, there exists M > 1 such that

0
1 < A (x)
x
1
M
A
for all x (0, ].
(6.39)
152
Suppose, in addition, that the function

Rx 1
a0 (x) 0 a p+1
Qa (x) :=
1
,
a(x) a p+1
(x)
x (0, ],
is non-oscillating in (0, ], as discussed in Definition 6.6. Then,

Z R !
1
b(x)b00 (x)
[1, M ].
I0 := lim A0 (x)
= lim
x0
x0 [b0 (x)]2
x A
(6.40)
(6.41)
In particular, (6.33), or equivalently (6.37), holds.

The local logarithmic concavity (6.38) is rather natural, since a(0) = 0 implies
lim ln a(x) = .
x0
Obviously, it allows most of the decaying rates for a at x = 0. Indeed, it holds

if there are two constants, > 0 and > 0, such that
a(x) x
for x 0,
or
a(x) x ln x for x 0,
or
a(x) e1/x
for x 0.
More generally, since

(ln a)00 (x) =
a00 (x)a(x) [a0 (x)]2

a2 (x)
for all
x (0, ),
a simple sufficient condition for (6.38) is

a0 (x) > 0
and a00 (x) 0
for all
x (0, ].
After we complete the proof of Proposition 6.7, we will give a very simple
criterion based on the quotient a/a0 , to guarantee that Qa (x) is non-oscillating
for all p 1.
Proof of Proposition 6.7: Subsequently, we consider the function (x) defined through
Z
(x) := A(x)
x
1
b(x)
= 0
,
A
b (x)
x (0, R].
153
By (6.28), (0) = 0. Moreover, (x) > 0 for all x (0, R). Thus, differentiating and rearranging terms, we find that
0
(x) = A (x)
x
and
00 (x) =
A0 (x)
A(x)
1
1
A
(6.42)

A(x)A00 (x) 0
[
(x)
+
1]
1
.
[A0 (x)]2
(6.43)
On the other hand, after some straightforward manipulations, from (6.25) it

becomes apparent that

A(x)A00 (x)
1
p1
=
Q
(x)
2
+
,
0 < x < R,
(6.44)
a
[A0 (x)]2
p+1
p+1
where Qa (x) is the function introduced in (6.40). Note that setting
qa (x) :=
a(x)
(> 0),
a0 (x)
x (0, ],
the function Qa (x) satisfies

(qa Qa )0 = 1
1
Qa
p+1
(6.45)
and hence,
qa (x)Q0a (x)

= Qa (x) q0a (x) +

1
+ 1,
p+1
x (0, ].
(6.46)
Moreover, by (6.38),
q0a (x) =
[a0 (x)]2 a(x)a00 (x)

= q2a (x)(ln a)00 (x) > 0,
[a0 (x)]2
x (0, ].
Thus, for every x

Q1
a (p + 1) (0, ], we find from (6.46) that

1
0
0
qa (
x)Qa (
x) = (p + 1) qa (
x) +
+ 1 = (p + 1)q0a (
x) < 0,
p+1
because q0a (
x) > 0. Consequently, Q0a (
x) < 0. Hence, Qa (x) > p + 1 for all
x (0, x
), if there exists x
(0, ] with Qa (
x) = p + 1. Suppose such a x
exists. Then,
Qa
(qa Qa )0 = 1
< 0 in (0, x
).
p+1
So,
d 1
>0
dx qa Qa
in (0, x
).
154
Thus, since
1
1
a p+1 (x)
d
= Rx 1 =
qa (x)Qa (x)
dx
a p+1
0
Z
ln
a p+1
x (0, x
),
we find that
d2
dx2
Z
ln

for all x (0, x
),
>0
a p+1
which is impossible, because

Z
lim ln
x0
1
p+1

= ,
and consequently the second derivative should be somewhere negative in (0, x)

for every x > 0. Therefore,
Qa (x) < p + 1
for all x (0, ].
(6.47)
Subsequently we will restrict ourselves to consider x (0, ]. First, we will

show that there exists (0, ] such that
0 (x) 0 ,
x (0, ].
(6.48)
To prove (6.48) we will argue by contradiction. As (0) = 0 and (x) > 0

for all x (0, R], 0 must be somewhere positive in every interval (0, ] with
0 < . Thus, if (6.48) fails for all (0, ), then 0 (x) must change sign in
(0, ] for all > 0. In such case, there exist two sequences, xn 0 and yn 0,
as n , such that
00 (xn ) = 00 (yn ) = 0,
0 (xn ) > 0,
0 (yn ) < 0,
n 1.
Thus, since A(x) > 0 and A0 (x) > 0 for all x (0, ], (6.43) yields
1
A(xn )A00 (xn )
= 0
< 1,
[A0 (xn )]2
(xn ) + 1
A(yn )A00 (yn )

1
= 0
> 1.
[A0 (yn )]2
(yn ) + 1
On the other hand, based on (6.44), we have that

A(x)A00 (x)
p 1 Qa (x)
1=
1 .
[A0 (x)]2
p+1 p+1
(6.49)
Hence,
Qa (xn ) p + 1,
which contradicts (6.47). Consequently, > 0 may be shortened if necessary

to get (6.48) with = .
155
Suppose there is x
(0, ] such that 0 (
x) = 0. Then, owing to (6.48), we
00
also have that (
x) = 0, because x
is a local minimum of 0 . Thus, by (6.43),
A(
x)A00 (
x)
= 1.
[A0 (
x)]2
So, by (6.49), Qa (
x) = p + 1, which contradicts (6.47). Therefore, > 0 can
be shortened if necessary so that
0 (x) > 0
for all x (0, ].
(6.50)
According to (6.42), (6.50) establishes the lower estimate of (6.39).

Also, thanks to (6.44), (6.47) and (6.49), it is apparent that
A(x)A00 (x)
2
<
<1
p+1
[A0 (x)]2
for all x (0, ].
(6.51)
Obviously, can be shortened if necessary so that some of the following excluding options occur:
(a) 00 0 on (0, ].
(b) 00 0 on (0, ].
(c) 00 changes sign infinitely many times in (0, ] for all 0 < < .
Suppose (a) occurs. Then, 0 (x) is non-decreasing and hence, owing to (6.42),
A0 (x)
1
= 0 (x) + 1 0 () + 1
A
for all x (0, ], which concludes the proof of (6.39). Note that in this case
the next limit exists
L := lim 0 (x) 0.
(6.52)
x0
Therefore,
0
lim A (x)
x0
1
A
!
= L + 1 1,
which ends the proof of (6.41).

Suppose (b) occurs. Then, according to (6.42), (6.43) and (6.51), it becomes apparent that, for every x (0, ],
A0 (x)
[A0 (x)]2
p+1
1
= 0 (x) + 1 00
<
.
A
A (x)A(x)
2
(6.53)
Consequently, (6.39) also holds in this case. Similarly, in this case the limit
(6.52) is well defined, since 0 (x) is non-increasing. Therefore, (6.41) holds.
156
Suppose (c) occurs and pick xj (0, ], j {1, 2, 3}, such that x1 < x2 <
x3 and 00 0 on [x1 , x2 ], while 00 0 on [x2 , x3 ]. By (6.42), (6.43) and
(6.51), it is apparent that (6.53) holds on [x2 , x3 ]. Since 00 (x2 ) = 0, we also
find that
Z R
p+1
1
[A0 (x2 )]2
A0 (x2 )
= 0 (x2 ) + 1 = 00
<
.
A (x2 )A(x2 )
2
x2 A
Moreover, since 00 0 on [x1 , x2 ], the function 0 is non-decreasing on
[x1 , x2 ]. Thus, for every x [x1 , x2 ],
Z R
1
p+1
0
A (x)
= 0 (x) + 1 0 (x2 ) + 1 <
.
A
2
x
Therefore,
Z
A (x)
x
p+1
1
<
A
2
for all x [x1 , x3 ].
As this is valid for every x1 , x2 , x3 satisfying the previous requirements, (6.39)

holds with M = (p + 1)/2. This ends the proof of (6.39) in all possible cases.
To complete the proof of the proposition it remains to show that cannot
satisfy alternative (c) if Qa is non-oscillating in (0, ], as in the remaining cases
the limit (6.41) has been already established. Indeed, by differentiating (6.43)
and rearranging terms it is easily seen that

0
A0
AA00
A0 AA00
000
00
=
2 0 2 1 +
(0 + 1).
A
(A )
A (A0 )2
Thus, by (6.50), for every x
(0, ] with 00 (
x) = 0, differentiating (6.44) with
respect to x yields

0
AA00
sign 000 (
x) = sign
(
x) = sign Q0a (
x).
(A0 )2
Consequently, for sufficiently small > 0, 00 cannot change sign in (0, ] if
either Q0a (x) > 0 for all x (0, ] or Q0a (x) < 0 for all x (0, ]. Note that we
are assuming that Qa (x) is non-oscillating in (0, ].
Suppose Qa Qa (0) is constant on (0, ]. Then, it follows from (6.46) and
(6.47) that
q0a (x) =
0 < Qa (0) 0,
p + 1 Qa (0)
there exists a constant C 0 such that

qa (x) = 1 x + C
for all x (0, ].
x (0, ].

a(x)
a0 (x) ,
Consequently, since qa (x) :=
157
there exists > 0 such that
a(x) = (x + C) ,
x (0, ].
Necessarily C = 0, because a(0) = 0. Thus, a(x) = x , x (0, ]. So,

A(x) =
1
p1
Z
p+1
p+1
p1

1
ds
= p1
p+1
+p+1
p+1
p1
+p+1
p1
for all x (0, ]. Therefore,

Z
lim A (x)
x0
1
A
which shows (6.41) and ends the proof.
+p+1
,
+2
(6.54)
Remark 6.8 As [0, ), the limit (6.54) ranges in [1, (p + 1)/2).

The following result provides us with a sufficient condition so that Qa (x) is
non-oscillating for all p 1.
Proposition 6.9 Suppose p > 1, a satisfies (A1), a(0) = 0, and there exists
0 < < R such that a C 3 (0, ] and (6.38) holds. Suppose, in addition, that
q0a =
a 0
a0
(a0 )2 aa00
aa00
=
1
(a0 )2
(a0 )2
is non-oscillating in (0, ]. Then, for each p 1, the function Qa (x) defined

by (6.40) is non-oscillating. If, in addition,
a1 = lim q0a (x) [0, ),
x0
then,
Qa (0) := lim Qa (x) =
x0
p+1
.
(p + 1)a1 + 1
Note that q0a is non-oscillating if and only if
aa00
(a0 )2
is non-oscillating.
Proof: By the proof of Proposition 6.7, we already know that

q0a (x) = q2a (x)(ln a)00 (x) > 0,
x (0, ],
(6.55)
by (6.38). Moreover, differentiating (6.45) shows that, for every p 1,

q00a (x)Qa (x) + 2q0a (x)Q0a (x) + qa (x)Q00a (x) =
Q0a (x)
,
p+1
x (0, ].
158
Suppose there is a p 1 for which Qa (x) is not non-oscillating. Then, there

are two sequences, xn , yn , n 1, such that
lim xn = 0 = lim yn
and, for every n 1,

Q0a (xn )Q0a (yn ) 0,
Q00a (xn ) = Q00a (yn ) = 0.
Thus, for every n 1, we find that

1
Q0a (xn ) = q00a (xn )Qa (xn ),
2q0a (xn ) +
p+1

1
Q0a (yn ) = q00a (yn )Qa (yn ).
2q0a (yn ) +
p+1
(6.56)
(6.57)
Since q0a is non-oscillating in (0, ], some of the following options occur. Either
q00a (x) > 0 for all x (0, ], or q00a (x) < 0 for all x (0, ], or q0a a1 for some
constant a1 > 0. Suppose q00a > 0, or q00a < 0, on (0, ]. Then, by (6.55), we find
from (6.57) that
sign Q0a (xn ) = sign q00a (xn ),
sign Q0a (yn ) = sign q00a (yn ),
for all n 1. Hence,

Q0a (xn )Q0a (yn ) > 0,
n 1,
which contradicts (6.56). Thus, q0a a1 on (0, ] for some constant a1 > 0.
Therefore,
qa (x) = a1 x
for all x (0, ].
Consequently, integrating in a yields
a(x) = Cx1/a1
for all x (0, ]
for some constant C > 0. So, Qa must be constant in (0, ], which shows the
non-oscillatory behavior of Qa (x) in (0, ] for all p 1.
Note that, thanks to (6.47), Qa (0) [0, p + 1]. Moreover,
a1 := lim q0a (x) =
x0
cannot be a priori excluded, because q0a does not need to be bounded if it is

non-oscillating.
159
Suppose a1 [0, ). Then, by the LHopitals rule,

Rx 1
1
a00 (x) 0 a p+1 + a0 (x)a p+1 (x)
1
Qa (0) = lim
1
x0
p+1 (x)a0 (x)
p+1 + 1 a
a00 (x)
p+1
=
lim
p + 2 x0
Rx
0
a p+1
a0 (x)a p+1 (x)

!
+1
a00 (x)a(x)
Qa (x) + 1
[a0 (x)]2
p+1
lim
p + 2 x0
p+1
lim [(1 q0a (x))Qa (x) + 1] .
p + 2 x0
Therefore,
Qa (0) =
6.3
p+1
[(1 a1 )Qa (0) + 1],
p+2
Getting sharp estimates for `(x) at x = 0 through b
Proposition 6.10 Suppose (A1)(A3) and (6.33). Then, there are (0, R)
and 0 < 0 < 0 such that for every (0, 0 ] and [0 , ), the functions
u := b and u := b are a subsolution and a supersolution, respectively, of the
problem
00
u (x) = a(x)h(u(x)),
x (0, ),
(6.58)
u(0) = , u() = `(),
where ` is the unique positive solution of (6.1) given by Theorem 6.1.
Proof: By Lemma 6.5, there exist two constants, 0 < m M , such that
0 < m c(x) M,
x [0, R/2].
(6.59)
Moreover, by (6.3), there exists > 0 such that

3H p
H p
u < h(u) 0
(6.61)
160
and pick 0 < 0 <

0 and := (
0 ) (0, R/2) satisfying

0 < 0 <
and
b(x) >
2
3HC
1
p1
>
0
k0

<
2
Hm
1
p1
<
0
for all x (0, ].
(6.62)
(6.63)
Obviously, since 3C > m, (6.62) can be accomplished by choosing a sufficiently

small 0 > 0 and a sufficiently large
0 . As, due to Lemma 6.4,
b(x) > 0,
b0 (x) < 0,
lim b(x) = ,
x0
to get (6.63) it suffices to choose with b() > C/

0 .
Now, fix 0 < 0 < 0 satisfying

`()
`()
0 <
0 max
0 ,
< 0 .
0 < 0 < min 0 ,
b()
b()
(6.64)
Obviously, (6.64) can be accomplished by choosing a sufficiently small 0 > 0

and a sufficiently large 0 . These are the choices of 0 , 0 and for which all
the assertions of the proposition hold.
Fix x (0, ]. Then, since b(x) > 0 and
0 > 0 , (6.61) and (6.63) imply
0 b(x) > 0 b(x) > C > .

Thus, by (6.60), the following estimates hold
h(
0 b(x)) <
3H p p
b (x),
2 0
h(
0 b(x)) >
H p p
b (x).
2 0
(6.65)
Hence, according to (6.59), it follows from (6.62), (6.64), (6.65) and (A3) that,
for every x (0, ],
0 bp (x)
2 1
0 bp (x)
<
<
< m c(x)
h(0 b(x))
h(
0 b(x))
H
p1
0
2 1
0 bp (x)
0 bp (x)
M <C<
<
<
.
p1
3H 0
h(
0 b(x))
h(0 b(x))
Therefore, by (A3), for every 0 < 0 , 0 , and x (0, ], the next
estimates hold
bp (x)
0 bp (x)
0 bp (x)
bp (x)
< m c(x) M <
.
h(b(x))
h(0 b(x))
h(0 b(x))
h(b(x))
In particular, by the definition of c, it becomes apparent that
bp (x)
a(x)bp (x)
bp (x)
<
<
00
h(b(x))
b (x)
h(b(x))
161
for all x (0, ], 0 and (0, 0 ], or, equivalently,
a(x)
< 00
<
.
h(b(x))
b (x)
h(b(x))
Therefore, using (6.26), we find that
b00 (x) < a(x)h(b(x)),
b00 (x) > a(x)h(b(x)),
(6.66)
for all x (0, ]. Finally, according to (6.64),

0 < 0 <
`()
< 0
b()
and, so, for every (0, 0 ] and 0 , it is apparent that

b() 0 b() < `() < 0 b() b(),
Based on Proposition 6.10, the next result holds.

Theorem 6.11 Suppose (A1)(A3) and (6.33), and let > 0, 0 > 0, 0 > 0
be the constants given by Proposition 6.10. Then, for every 0 < 0 and
0 > 0,
b(x) `(x) b(x),
x (0, ],
(6.67)
where `(x) is the unique positive solution of (6.1). Moreover, for every
(0, 0 ] and 0 , the following estimate holds
h(b(x)) 00
h(b(x)) 00
b (x) ` 00 (x)
b (x)
h(b(x))
h(b(x))
(6.68)
for all x (0, ]. Consequently, `(x) inherits most of the local properties of
b(x) at x = 0 established by Lemma 6.4. In particular,
lim `(x) = ,
x0
and
lim
x0
lim ` 0 (x) = ,
x0
` 0 (x)
` 00 (x)
` 00 (x)
= lim
=
lim
= .
x0 ` 0 (x)
x0 `(x)
`(x)
(6.69)
(6.70)
Proof: According to Remark 6.3(b), `(x) is the unique positive solution of

00
u (x) = a(x)h(u(x)),
x (0, ),
(6.71)
u(0) = , u() = `().
Subsequently, given (0, 0 ] and 0 , for each natural number n > 1
we consider the boundary value problem
00
u (x) = a(x)h(u(x)),
x (n1 , ),
(6.72)
+
1
1
u(n ) = 2 b(n ), u() = `().
162
By Proposition 6.10, (b, b) provides us with an ordered sub-supersolution

pair of (6.72). Thus, (6.72) possesses a solution, un , such that
b(x) un b(x)
x [n1 , ].
Actually, un is unique, but this uniqueness is far from necessary in our argument here. By a standard compactness argument, we can extract a subsequence of un , say unm , m 1, approximating to a solution of (6.71) as
m , necessarily `, by the uniqueness of `. Therefore, letting m in
the estimates
b(x) unm b(x)
x [n1
m , ],
(6.67) holds.
On the other hand, by (A2), it follows from (6.67) that
a(x)h(b(x)) a(x)h(`(x)) = ` 00 (x) a(x)h(b(x)),
x (0, ].
(6.73)
Moreover, owing to (6.66),

a(x)h(b(x))
a(x)h(b(x))
< b00 (x) <
,
x (0, ].
(6.74)
Thus, combining (6.73) and (6.74) yields
h(b(x)) 00
b (x) a(x)h(b(x)) a(x)h(`(x))
h(b(x))
h(b(x)) 00
= ` 00 (x) a(x)h(b(x))
b (x),
h(b(x))
for all x (0, ], which provides us with (6.68).

To complete the proof it remains to establish (6.69) and (6.70). The first
limit of (6.69) holds by the definition of `(x). For any given x (0, ), integrating in [x, ] the estimate (6.68) yields
Z
Z
h(b(s)) 00
h(b(s)) 00
0
0
b (s) ds ` () ` (x)
b (s) ds.
(6.75)
x h(b(s))
x h(b(s))
Q(s) :=
h(b(s))
,
h(b(s))
s (0, ].
The function Q is continuous and Q(s) > 0 for all s (0, ]. Moreover, since
lim b(s) = ,
s0
it follows from (6.3) that

lim Q(s) = lim
s0
s0
h(b(s))/(b(s))p p p
=
> 0.
h(b(s))/(b(s))p
163
Thus, the quantity

h(b(s))
(0, 1]
s(0,] h(b(s))
QL := inf
(6.76)
is well defined. As b00 (x) > 0 for all x (0, ], substituting (6.76) into (6.75),
0
0
QL (b0 () b0 (x)) ` 0 () ` 0 (x) Q1
L (b () b (x)),
x (0, ].
Therefore,
0
0
0
0
0
0
` 0 () Q1
L (b () b (x)) ` (x) ` () QL (b () b (x))
(6.77)
for all x (0, ]. By Lemma 6.4,

lim b0 (x) = .
x0
Consequently, (6.77) implies

lim `0 (x) = ,
x0
which ends the proof of (6.69). Now, for sufficiently small x > 0, it follows
from (6.77) and (6.67) that
0
0
` 0 () + Q1
` 0 () + QL (b0 () b0 (x))
` 0 (x)
L (b () b (x))
.
b(x)
`(x)
b(x)
Therefore,
lim
x0
b0 (x)
= implies
b(x)
lim
x0
` 0 (x)
= .
`(x)
Similarly, (6.67) and (6.68) imply

` 00 (x)
h(b(x)) b00 (x)
h(b(x)) b00 (x)
h(b(x)) b(x)
`(x)
h(b(x)) b(x)
for all x (0, ]. Thus,
QL
b00 (x)
` 00 (x)
b00 (x)
Q1
.
L
b(x)
`(x)
b(x)
Consequently,
lim
x0
b00 (x)
= implies
b(x)
lim
x0
` 00 (x)
= .
`(x)
Analogously, from (6.68) and (6.77), it becomes apparent that

lim
x0
b00 (x)
=
b0 (x)
implies
lim
x0
` 00 (x)
= ,
`0 (x)
164
6.4
The exact blow-up rate of `(x) at x = 0
The following result ascertains the exact blow-up rate of `(x) at x = 0.

Theorem 6.12 Suppose (A1)-(A3) and (6.33). Then,
p
`(x)
lim
= I0p1
x0 b(x)
p+1
p1
p+1
p1
H p1 ,
(6.78)
where `(x) is the unique positive solution of (6.1). In particular, the limit
(6.78) is independent of R > 0.
Proof: According to Lemma 6.5,
a(x)bp (x)
lim
=
x0
b00 (x)
p1
p+1
p+1
I0p .
(6.79)
Subsequently, we will consider the quotient

q(x) :=
`(x)
,
b(x)
x (0, ].
(6.80)
Owing to (6.67), we already know that

q(x) ,
x (0, ].
Hence,
0 < qL := lim inf q(x) qM := lim sup q(x) .
x0
x0
To show the existence of limx0 q(x) we will argue by contradiction. Suppose

qL < qM .
Then, there are two sequences, tn , sn , n 1, such that
lim tn = lim sn = 0,
lim q(tn ) = qM ,
lim q(sn ) = qL ,
and, for every n 1,

q0 (tn ) = q0 (sn ) = 0,
q00 (tn ) 0,
q00 (sn ) 0.
Since `00 = ah(`), we find from (6.80) that

q00 b + 2q0 b0 + qb00 = `00 = ah(qb)
So,
q00
on (0, ].
0
b
ah(qb)
0 b
+
2q
+q=
.
00
00
b
b
b00
(6.81)
165
Thus, according to (6.81), we find that, for every n 1,

q(tn ) q00 (tn )
b(tn )
a(tn )h(q(tn )b(tn ))
+ q(tn ) =
,
00
b (tn )
b00 (tn )
q(sn ) q00 (sn )
b(sn )
a(sn )h(q(sn )b(sn ))
+ q(sn ) =
,
00
b (sn )
b00 (sn )
because, owing to Lemma 6.4, b(x) > 0 and b00 (x) > 0 for all x (0, ].
Therefore, letting n in these inequalities, (6.3) and (6.79) imply that

p+1
p+1

p1
p1
p p
and
qL
I0 qM H
I0p qpL H.
qM
p+1
p+1
Therefore,
p+1
1
p + 1 p1 p1
H
p1
which contradicts qL < qM . Consequently, the next limit exists:
p
qL = qM = I0p1
q0 := lim
x0
`(x)
[, ].
b(x)
(6.82)
Finally, combining Lemma 6.4 and Theorem 6.11 with the LHopital rule, it
follows from (6.82), (6.3), (6.79) and (6.80) that

`00 (x)
a(x)bp (x) h(`(x))
= lim
q0 = lim 00
x0 b (x)
x0
b00 (x)
bp (x)

p+1
p1
a(x)bp (x) h(`(x)) p
p
q (x) =
I0p Hq0p ,
= lim
x0
b00 (x)
` (x)
p+1

because `(x) as x 0. Consequently, since q0 > 0, it is apparent that

p
p1
q0 = I 0
p+1
p1
which shows (6.78) and ends the proof.
6.5
p+1
p1
H p1 ,
This chapter is a refinement of S. Cano-Casanova and J. Lopez-Gomez [39],

going back to J. L
opez-G
omez [161], where, for a given p > 1, the special case
h(u) = up ,
u 0,
(6.83)
166
was analyzed. In the simplest case when (6.83) holds and there are two constants H > 0 and 0 such that
a(x) = Hx ,
x 0,
a direct calculation shows that actually

`(x) =
+2
p1
+2
+1
p1
1
p1
+2
H p1 x p1 ,
x > 0.
(6.84)
The blow-up rate of this function `(x) at x = 0 provides us with the most
classical blow-up rates of C. Loewner and L. Nirenberg [138], V. A. Kondratiev
and V. A. Nikishin [125], C. Bandle and M. Marcus [16, 17, 18], M. Marcus and
L. Veron [188], L. Veron [230], and A. C. Lazer and P. J. McKenna [129, 130],
for the special case when = 0, and Y. Du and Q. Huang [80], as well as J.
Garca-Meli
an, R. Letelier and J. C. Sabina de Lis [101], for the general case
when 0.
Indeed, it turns out that in the general multidimensional case when N 1,
at least for these particular choices of a(x) and h(u), the blow-up rate of `(x)
given by (6.84), +2
p1 , actually provides us with the blow-up rates on of
all the positive solutions of

u(x) = a(d(x))up ,
x ,
(6.85)
u = ,
on ,
provided
a(d(x))
= 1,
d(x)0 Hd (x)
lim

d(x) := dist (x, ),
x .
But this particular issue will be discussed in the next two chapters. So, we
refrain from giving more details herein.
As a corollary of Propositions 6.7 and 6.9, if there exists > 0 such that
a C 3 (0, ], a0 (x) > 0 and (ln a)00 (x) < 0 for all x (0, ), and, in addition, the
aa00
function (a
0 )2 is non-oscillating in (0, ), then (6.33), or equivalently (6.37),
holds for all p 1. Although we do not hope that (6.33) will be satisfied in
general without imposing some non-oscillation property on a(x), we did not
try to look for a counterexample yet.
S. Cano-Casanova [34] has recently used the techniques of this chapter to
ascertain the decay rate to zero at x = of the positive solutions of
00
u (x) = a(x)h(u(x)),
x > 0,
(6.86)
u(0) = M, u() = 0,
167
where M is a positive constant, a satisfies (A1), h satisfies (A2) with h(u)/u,

u > 0, increasing and
h(u)
A := lim q > 0
(6.87)
u0 u
for some q > 1. According to the proof of Theorem 6.1 we already know that
(6.86) possesses a unique positive solution, uM . Setting
Z
Z
b0 (x) :=
a
x
1
q+1
q+1
q1
dy,
the main theorem of S. Cano-Casanova [34] establishes that

q
uM (x)
u0 (x)
u00 (x)
lim
= lim M
= lim M
= J0q1
0
00
x b0 (x)
x b (x)
x b (x)
0
0
provided
q+1
q1
q+1
q1
A q1
b0 (x)b000 (x)
(0, ).
x [b00 (x)]2
J0 := lim
This result is a substantial improvement of some available results for a class

of generalized Thomas-Fermi equations (see, e.g., V. Maric and M. Tomic
[190] and S. D. Taliaferro [225], [226]) and it may present a huge number of
applications in analyzing exterior problems.
Chapter 7
under radial symmetry
7.1
7.2
7.3
7.4
7.5
The main uniqueness result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Proof of Theorem 7.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.1
The case when = BR (x0 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.2
The case when = AR1 ,R2 (x0 ) . . . . . . . . . . . . . . . . . . . . . . . . .
Exact blow-up rates on the boundary . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simple application in population dynamics . . . . . . . . . . . . . . . . . . . . . .
170
172
172
174
177
183
184
Throughout this chapter, we consider x0 RN , N 1, R > 0, R2 > R1 > 0,

BR (x0 ) := {x RN : |x x0 | < R},
AR1 ,R2 (x0 ) := {x RN : R1 < |x x0 | < R2 },
a continuous function a C[0, ) such that
a(t) > 0
for all
t > 0,
(7.1)
a function g C[0, ) C 1 (0, ) such that

g(0) = 0,
g 0 (u) > 0
for all
u > 0,
lim g(u) = ,
(7.2)
and
{BR (x0 ), AR1 ,R2 (x0 )}.
(7.3)
Also, we set
d(x) := dist(x, ),
x .
Under these assumptions, by Theorem 2.4, the boundary value problem

u = u a(d(x))g(u)u
in ,
(7.4)
u=M
on ,
possesses a unique positive solution, [,,M ] , for every M > 0. Moreover, the
map M [,,M ] is increasing, and, since (7.4) is invariant by rotations,
[,,M ] must be radially symmetric for all M > 0, by uniqueness. However,
unless the function
f (u) := g(u)u,
u 0,
169
170
satisfies the generalized KellerOsserman condition (KO), the limit

[,,] := lim [,,M ]
(7.5)
might be infinity somewhere. According to Theorem 3.4, when f satisfies

(KO), (7.5) equals the minimal large positive solution, Lmin
[,] , of the singular
problem

u = u a(d(x))g(u)u
in ,
(7.6)
u=
on .
Naturally, in such case, Lmin
[,] must be radially symmetric, because it is a
limit of radially symmetric functions. By the construction of the maximal
large positive solution, Lmax
[,] , carried out in the proof of Theorem 3.4, it
becomes apparent that Lmax
[,] is radially symmetric. It should be remembered
that
min
[,,] = Lmin
Lmax
(7.7)
[,] ,
[,] = lim L[, ] ,
0
where
:= {x : dist(x, ) > }
for sufficiently small > 0. In general, it is a very appealing open problem to
ascertain whether
max
Lmin
(7.8)
[,] = L[,] ,
which entails the uniqueness of the large solution of (7.6). This chapter gives
some rather satisfactory answers to this problem in the radially symmetric
case.
7.1
The main uniqueness result
Theorem 7.1 Suppose (7.3), 0, a C[0, ) satisfies

0 < a(t) a(s)
if
0 < t s,
(7.9)
g satisfies (7.2) and (KO), and there exists r := r(g) > 0 such that
r g(u) g(u)
for all > 1
and
u > 0.
Then, (7.8) holds, i.e., (7.6) admits a unique positive solution,

max
L[,] := Lmin
[,] = L[,] .
Moreover, L[,] is radially symmetric.
(7.10)
Uniqueness of the large solution under radial symmetry
171
Note that (7.10) also holds if = 1, or u = 0, because g(0) = 0. Moreover,

by (7.2), g(u) > 0 for all u > 0. Hence, (7.10) holds for every r (0, r(g)].
Moreover, the change of variable
:=
2
> 0,
r(g)
= % ,
(7.11)
transforms (7.10) into

%2 g(% u) g(u)
(%, u) (1, ) [0, ).
for all
(7.12)
The next result provides us with an important sufficient condition for (7.10),
or, equivalently, (7.12).
Lemma 7.2 Suppose g C[0, ) C 2 (0, ), g(0) = 0, and
g 00 (u) 0
for all
u > 0.
(7.13)
Then, (7.10) holds with r = 1. If, in addition, g(u) > 0 for all u > 0, then,
(7.10) also holds for every r (0, 1].
Proof: For every 1, consider the function
G (u) := g(u) g(u),
u 0.
Obviously, G C[0, ) C 2 (0, ), and, for every 1,

G (0) = 0.
(7.14)
Moreover, by (7.13), g 0 is non-decreasing. Thus,

G0 (u) = (g 0 (u) g 0 (u)) 0,
u > 0,
1.
(7.15)
Hence, from (7.14) and (7.15), it becomes apparent that

G (u) 0
for all 1, u 0.
Therefore,
g(u) g(u)
for all (, u) [1, ) [0, ).
Consequently, (7.10) holds with r = 1. The last assertion is obvious, because

g(u) > 0 if u > 0, and > r for all > 1 and r (0, 1).
Although at first glance condition (7.10) seems restrictive, according to
Lemma 7.2, it holds for a wide class of gs satisfying (7.2) and
g(u)u
>0
u up
H := lim
for some p > 1. For example, for any integer n 1 and real numbers
0 < r1 < r2 < . . . < rn ,
{a1 , ..., an } (0, ),
(7.16)
172
the function
g(u) :=
n
X
ak urk
k=1
satisfies (7.2), (7.16) and (7.10). Indeed, by construction, (7.2) holds. Moreover,
g(u)u
lim rn +1 = an > 0.
u u
So, g satisfies (7.16). Finally, for every > 1 and u > 0,
g(u) =
n
X
ak rk urk = r1
k=1
r1
n
X
rk r1 ak urk
k=1
n
X
ak urk = r1 g(u).
k=1
Consequently, (7.10) holds for every r (0, r1 ]. It should be noted that condition (7.13) does not necessarily hold for this choice of g.
7.2
This section consists of the proof of Theorem 7.1. First, we will prove it in the
ball, then in the annulus.
7.2.1
The case when = BR (x0 )
For every (0, R), we set

% :=
R
>1
R
and consider the function

(x) := Lmin
L
[,BR (x0 )] (x0 + % (x x0 )),
0 |x x0 | R .
= on BR (x0 ). Moreover, for every x BR (x0 ), we

By definition, L
have that
(x) = %2 Lmin
L
[,BR (x0 )] (x0 + % (x x0 ))
2
(x))L
(x)
= % L (x) %2 a(R % |x x0 |)g(L
(x) %2 a(R |x x0 |)g(L

(x))L
(x)
L
(x) %2 a(d(x))g(L
(x))L
(x),
= L
173
because % > 1, 0, and, owing to (7.9),

a(R % |x x0 |) a(R |x x0 |).
Let > 0 be satisfying (7.12) and consider the function
:= % L
in BR (x0 ).
Then,
=
L
on BR (x0 )
and, for every x BR (x0 ),

(x) L
(x) %2 a(d(x))g(%
L
L (x))L (x).
Thus, according to (7.12), we find that
(x) L
(x) a(d(x))g(L
(x))L
(x)
L
provides us with a supersolution of the
for all x BR (x0 ). Therefore, L
singular problem

L = L a(d(x))g(L)L
in BR (x0 ),
L=
on BR (x0 ).
and Lmin
FIGURE 7.1: The functions L
[,BR (x0 )] .
174
By Theorem 3.4, we already know that

min
Lmax
[,BR (x0 )] = lim L[,BR (x0 )] .
0
Moreover, as in a neighborhood of BR (x0 ),
Lmin
[,BR (x0 )] L
for each < ,
(7.17)
by Theorem 1.7, the estimate (7.17) holds in the entire BR (x0 ), as illustrated
in Figure 7.1. Therefore, letting 0 in (7.17) yields
min
Lmax
[,BR (x0 )] (x) L (x) = % L[,BR (x0 )] (x0 + % (x x0 ))
for all (0, R) and x BR (x0 ). So, letting 0 we find that
min
Lmax
[,BR (x0 )] (x) L[,BR (x0 )] (x),
|x x0 | < R,
which ends the proof of the theorem in this case.
7.2.2
The case when = AR1 ,R2 (x0 )
Then, setting
Rm :=
R1 + R2
,
2
r := |x x0 |,
we have that

d(x) := dist(x, ) =
R2 r
r R1
if Rm r R2 ,
if R1 r Rm .
max
Moreover, since Lmin
[,] and L[,] are radially symmetric, necessarily
Lmin
[,] (x) = min (r),
Lmax
[,] (x) = max (r),
x = AR1 ,R2 (x0 ),
where min (r) and max (r) are the reflections about r = Rm of the minimal and the maximal positive solutions, respectively, of the singular onedimensional problem

Rm < r < R2 ,
00 N r1 0 = a(R2 r)g(),
(7.18)
0 (Rm ) = 0, (R2 ) = .
Next, we will show that any positive solution of (7.18) satisfies
0 (r) 0
for all
r [Rm , R2 ).
(7.19)
This is clear if 0, because, in such case, multiplying the differential equation by rN 1 and integrating in (Rm , r) shows that, for every r (Rm , R2 ),
Z r
rN 1 0 (r) =
sN 1 [a(R2 s)g((s)) ](s) ds > 0.
Rm
175
When > 0, our proof of (7.19) will proceed by contradiction. Suppose > 0
and (7.18) possesses a positive solution, , for which there exists r (Rm , R2 )
such that
0 (
r) < 0.
Then, since
0 (Rm ) = 0,
lim (r) = ,
rR2
there exist r0 and r1 such that

Rm r0 < r < r1 < R2
and
0 (r0 ) = 0 (r1 ) = 0, 0 (r) 0

00 (r0 ) 0, 00 (r1 ) 0.
if r (r0 , r1 ),
(7.20)
Subsequently, we will consider the function H defined by

H() := a(R2 r0 )g(),
> 0.
By (7.2), the value

0 := g 1 (/a(R2 r0 ))
provides us with the unique positive zero of H(). Actually, H() > 0 if
(0, 0 ), H(0 ) = 0, and H() < 0 if > 0 .
Suppose (r0 ) > 0 . Then, due to (7.18) and (7.20), we find that
0 00 (r0 ) = 00 (r0 )
N 1 0
(r0 ) = H((r0 )) < 0,
r0
which is impossible. Thus,

(r0 ) 0 ,
(7.21)
and hence, it follows from (7.18) and (7.20) that

N 1 0
(r1 ) = (r1 ) a(R2 r1 )g((r1 ))(r1 )
r1
= (r1 )a(R2 r0 )g((r1 ))(r1 )+[a(R2 r0 )a(R2 r1 )]g((r1 ))(r1 )
= H((r1 )) + [a(R2 r0 ) a(R2 r1 )]g((r1 ))(r1 ).
0 00 (r1 ) = 00 (r1 )
As 0 0, 0 =
6 0, in (r0 , r1 ), necessarily (r1 ) < (r0 ). So, by (7.21), we find
that (r1 ) < 0 , which implies
H((r1 )) > 0.
On the other hand, r0 < r1 implies R2 r0 > R2 r1 and hence, due to (7.9),
a(R2 r0 ) a(R2 r1 ).
176
Therefore,
0 H((r1 )) + [a(R2 r0 ) a(R2 r1 )]g((r1 ))(r1 ) > 0,
which is impossible. Consequently, (7.19) holds.
:=
R2 Rm
> 1,
R2 Rm
(0, R2 Rm ),
and consider the function defined by

(r) := min ( (r Rm ) + Rm ),
Rm r < R2 .
By definition, we have
0
0 (Rm ) = min
(Rm ) = 0
and
lim (r) = lim min (%) = .
rR2
%R2
Moreover, setting
% := (r Rm ) + Rm ,
Rm % < R2 ,
from (7.18) it becomes apparent that, for every, r (Rm , R2 ),

00 (r)
N 1 0
N 1
0
00
(r) = 2 min
min
(%)
(%)
r
r
N 1%
00
0
= 2 min
(%)
min
(%).
% r
Thus, taking into account that

0,
0
min
0,
> 1,
Rm (1 ) + r
%
=
,
r
r
we find that
00 (r)
N 1 0
N 1%
00
0
(r) = 2 min
(%)
min
(%)
r
% r

N 1 0
00
2 min
(%)
min (%)
%
= 2 [min (%) a(R2 %)g(min (%))min (%)]
min (%) 2 a(R2 %)g(min (%))min (%).
177
On the other hand,

(r) := min (%),
a(R2 r) a(R2 %),
because r %. Consequently, for every r (Rm , R2 ),

00 (r)
N 1 0
(r) (r) 2 a(R2 r)g( (r)) (r).
r
Let > 0 be satisfying condition (7.12) and consider the function

(r) := (r),
Rm < r < R2 .
Then,
0 (Rm ) = 0,
lim (r) = ,
rR2
and, for every r (Rm , R2 ),

00 (r)
N 1 0
(r) (r) 2 a(R2 r)g( (r)) (r).
r
Consequently, thanks to (7.12), it becomes apparent that

00 (r)
N 1 0
(r) (r) a(R2 r)g( (r)) (r),
r
Rm < r < R2 .
Therefore, provides us with a supersolution of the singular problem

00 N r1 0 = a(R2 r)g(), Rm < r < R2 ,
0 (Rm ) = 0, (R2 ) = .
According to Theorem 1.7,
(r) = (r) max (r),
Rm < r < R2 ,
for sufficiently small > 0. Consequently, letting 0 yields

min max
in (Rm , R2 )
which concludes the proof of the theorem.
7.3
Exact blow-up rates on the boundary
The main result of this section provides us with the blow-up rate of the positive
large solution of the singular problem

u = u a(d(x))g(u)u
in ,
(7.22)
u=
on ,
where 0 and satisfies (7.3). Precisely, it can be stated as follows.
178
Theorem 7.3 Suppose (7.3), 0, a C[0, ) satisfies (7.9), g C[0, )

C 1 (0, ) satisfies (7.2), (7.10) and (7.16), and
b(t)b00 (t)
= I0 > 0,
[b0 (t)]2
lim
t0
(7.23)
for some R > 0, where b stands for the function

Z
Z
b(t) :=
a
t
1
p+1
p+1
p1
t (0, R].
dy,
(7.24)
Then,
p
L[,] (x)
= I0p1
d(x)0 b(d(x))
lim
p+1
p1
p+1
p1
H p1 ,
(7.25)
where L[,] stands for the unique positive solution of (7.22).

Owing to Lemma 6.4, we already know that b C 2 (0, R],
b0 (t) < 0,
b(t) > 0,
b00 (t) > 0,
lim b(t) = ,
t0
and
lim
t0
t (0, R),
lim b0 (t) = ,
(7.26)
(7.27)
t0
b00 (t)
b00 (t)
b0 (t)
= lim
= lim
= .
0
t0 b (t)
t0 b(t)
b(t)
(7.28)
As in Section 6.2, in terms of the quotient function

qb (t) :=
b(t)
1
=
,
0
b (t)
(ln b)0 (t)
t (0, R],
it is easily seen that (7.23) can be equivalently expressed in the form

qb C 1 [0, R]
with
q0b (0) < 1.
According to Propositions 6.7 and 6.9, if there exists > 0 such that a
aa00
C 3 (0, ], a0 (x) > 0 and (ln a)00 (x) < 0 for all x (0, ), and, in addition, (a
0 )2
is non-oscillating in (0, ), then (7.23) holds for all p 1.
Proof of Theorem 7.3: According to Theorem 6.1, the singular problem
00
u (t) = a(t)g(u(t))u(t), t > 0,
(7.29)
u(0) = , u() = 0,
possesses a unique positive solution, `(t), t > 0. Moreover,
`(t) > 0,
` 0 (t) < 0,
` 00 (t) > 0,
t > 0,
(7.30)

lim ` 0 (t) = ,
t0
lim ` 0 (t) = 0,
179
(7.31)
and, due to Theorems 6.11 and 6.12,

lim
t0
`0 (t)
`00 (t)
`00 (t)
= lim 0
= lim
=
t0 ` (t)
t0 `(t)
`(t)
and
p
`(t)
lim
= I0p1
t0 b(t)
p+1
p1
p+1
p1
H p1 .
(7.32)
(7.33)
Subsequently, we will distinguish two different situations according to whether

= BR (x0 ), or = AR1 ,R2 (x0 ). Suppose
= BR (x0 ).
Then,
L[,] (x) = (r),
r = |x x0 |,
where is the unique positive solution of

00 N r1 0 = a(R r)g()
0 (0) = 0, (R) = .
in (0, R),
(7.34)
Firstly, we will show that for sufficiently small > 0 there exists a constant
A > 0 such that, for every A > A , the function
,A (r) := A + (1 + )
r 2
R
`(R r),
0 r < R,
(7.35)
provides us with a positive supersolution of (7.34). Indeed, by construction,

0
,A
(0) = 0
and
lim ,A (r) = .
rR
Thus, ,A is a supersolution of (7.34) if, and only if,

2N
r 2
r(1 + ) 0
1+
`(Rr)
+
(N
+
3)
`
(Rr)
(1
+
)
`00 (Rr)
2
R2
R
R

r 2
A
`(Rr)
+ (1 + )
`(Rr)
R

r 2 g( (r))
A
,A
+ (1 + )
.
a(Rr) `(Rr)g(`(Rr))
`(Rr)
R
g(`(Rr))
Dividing this inequality by `00 (R r) and taking into account that

`00 = ag(`)` > 0,
180
it becomes apparent that ,A is a supersolution of (7.34) if, and only if,

r 2
1 + `(R r)
r(1 + ) `0 (R r)
2N 2 00
+ (N + 3)
(1
+
)
R ` (R r)
R2 `00 (R r)
R

2
`(R r)
A
r
+ (1 + )
00
(7.36)
` (R r) `(R r)
R

r 2 g( (r))
A
,A
+ (1 + )
.
`(R r)
R
g(`(R r))
As
lim `(R r) = lim ,A (r) = ,
rR
rR
it follows from (7.16) and (7.35) that

p1

,A (r)
g(,A (r))
= lim
= (1 + )p1 .
lim
rR `(R r)
rR g(`(R r))
(7.37)
By (7.32) and (7.37) it becomes apparent that, at r = R, (7.36) reduces to

(1 + ) (1 + )p ,
which is satisfied because 1 + > 1 and p > 1. By continuity, there exists
= () > 0 such that (7.36) holds for all r [R , R). By enlarging the
constant A, if necessary, we can assume that (7.36) also holds in the entire
interval [0, R], because ` and `00 are positive and bounded away from zero on
compact intervals of (0, ), ,A A and we are assuming (7.2) and (7.16).
Indeed, since g is increasing, we have that
g(,A ) g(A) HAp1
as A .
Therefore, ,A is a supersolution of (7.34) for sufficiently large A > 0.

Next, we will show that, for sufficiently small 0 < < 1, there exists a
negative constant, C < 0, such that

r 2
,C (r) := max 0, C + (1 )
`(R r) ,
0 r < R,
R
is a non-negative subsolution of (7.34). Indeed, by reversing the inequality in
(7.36), it is apparent that ,C is a subsolution of (7.34) if in the region
r 2
C + (1 )
`(R r) 0
R
the following inequality is satisfied
r 2
1 `(R r)
r(1 ) `0 (R r)
2N 2 00
+ (N + 3)
(1
)
R ` (R r)
R2 `00 (R r)
R

2
`(R r)
C
r
00
+ (1 )
(7.38)
` (R r) `(R r)
R

r 2 g( (r))
C
,C
+ (1 )
.
`(R r)
R
g(`(R r))
181
Subsequently, we introduce the function defined by

(r) := (1 )
r 2
R
`(R r),
r [0, R).
By construction,
lim (r) = .
(0) = 0,
rR
Moreover, is increasing, because ` > 0 and `0 < 0. Thus, for every C < 0
there exists z = z(C) (0, R) such that
C + (r) = C + (1 )
r 2
R
`(R r) < 0
if r [0, z(C)),
whereas
C + (r) = C + (1 )
r 2
R
`(R r) 0
if r [z(C), R).
(7.39)
Note that the mapping C 7 z(C) is decreasing and

lim z(C) = R,
lim z(C) = 0.
(7.40)
C0
Owing to (7.16), (7.29) and (7.39), it follows that

lim
rR
g( ,C (r))
g(`(R r))
= lim
rR
,C (r)
!p1
= (1 )p1 .
`(R r)
(7.41)
Thus, thanks to (7.32) and (7.41), the inequality (7.38) at r = R reduces to

(1 ) (1 )p ,
which holds true because 1 (0, 1) and p > 1. By continuity, there exists
= () > 0 such that (7.38) also holds in [R , R). Moreover, according to
(7.40), there exists C < 0 such that
z(C) = R .
For this choice of C, it is apparent that ,C provides us with a non-negative
subsolution of (7.34).
Lastly, by construction, it becomes apparent that
,A (r)
= 1 + ,
rR `(R r)
lim
lim
rR
,C (r)
`(R r)
and
,C ,A
in [0, R).
= 1 ,
(7.42)
182
Thus, by the uniqueness of L[,] as a solution of the singular problem (7.6),

it follows from Theorem 1.4 that
,C (|x x0 |) L[,] (x) ,A (|x x0 |),
|x x0 | < R,
for all 0 < < 1. Consequently, by (7.42), one can infer that
1 lim inf
d(x)0
L[,] (x)
L[,] (x)
lim sup
1+
`(d(x))
d(x)0 `(d(x))
(7.43)
for all (0, 1). Therefore, we can conclude that

lim
d(x)0
L[,] (x)
= 1.
`(d(x))
(7.44)
Clearly, (7.25) follows readily by combining (7.44) with (7.33). This ends the
proof of the theorem when is a ball.
Subsequently, we assume
= AR1 ,R2 (x0 ).
Then, setting
(r) := min{R2 r, r R1 },
we have that
r = |x x0 |,
L[,] (x) := (r),
where is the unique positive solution of

00 N r1 0 = a((r))g()
(R1 ) = (R2 ) = .
in (R1 , R2 ),
(7.45)
Rm
R1 + R2
:=
,
2
(r) :=

2
rRm
R1 Rm `(r R1 )

rRm
R2 Rm
2
`(R2 r)
if R1 < r < Rm ,
if Rm r < R2 .
Adapting the previous argument, for sufficiently small > 0 there exists a
constant A > 0 such that, for every A > A , the function
,A (r) := A + (1 + )(r),
r (R1 , R2 ),
is a positive supersolution of (7.45). Similarly, there exists C < 0 such that

,C (r) := max {0, C + (1 )(r)}
is a non-negative subsolution of (7.45). Moreover, by construction,
,C ,A .
183
Thus, by Theorem 1.4, (7.45) possesses a positive solution, , such that

,C ,A
for sufficiently large A. Consequently, since L[,] is the unique positive solution of (7.6), it is apparent that
,C ,A .
Therefore,
1 lim
rR2
(r)
(r)
= lim
1 + .
`(R2 r) rR1 `(r R1 )
(7.46)
As (7.46) holds for sufficiently small > 0, (7.44) holds.

Finally, combining (7.44) with (7.33) ends the proof of the theorem.
Remark 7.4 Let m 1 and K Rm be a compact subset. If the weight

function a = a is assumed to vary in a continuous way with respect to the
a , is continuous,
parameter K, in the sense that the map K C(),
then the limit (7.25) holds uniformly in K.
7.4
Simple application in population dynamics
Let RN be an arbitrary domain with smooth boundary, , such that

R (x0 ) and consider the function
B

R (x0 ),
a(d(x)),
xB
a(x) :=
R (x0 ),
0,
x\B
as well as the associated parabolic model
u
(x, t) (0, ),
t du = u + a(x)g(u)u,
u = 0,
(x, t) (0, ),
u(x, 0) = u0 (x),
x ,
(7.47)
where a C[0, ) satisfies (7.9), g satisfies (7.2), (KO) and (7.10), d > 0 and
satisfies u0 > 0, i.e., u0 0 and u0 6= 0.
u0 C()
In the context of population dynamics, (7.47) models the evolution of a
single species u dispersing randomly in the territory , which consists of two
regions. In BR (x0 ), a < 0 and u grows according to a generalized logistic law,
while in
R (x0 ),
0 := \ B
184
a = 0 and u exhibits a genuine Malthusian growth. As usual, the parameter

is the intrinsic growth rate of the population u and u0 is the initial population
distribution. As we are working under homogeneous Dirichlet conditions, the
inhabiting area is assumed to be entirely surrounded by completely hostile
regions.
Setting
0 := 1 [, ],
1 := 1 [, BR (x0 )],
and combining Theorem 5.2 with Theorem 7.1, the next result holds. Note
that > 0 if d1 . Consequently, in such case we are within the range of
applicability of Theorem 7.1.
Theorem 7.5 Suppose a C[0, ) and g satisfy (7.9), and (7.2), (KO) and
(7.10), respectively. Let u(x, t) denote the unique solution of (7.47). The following properties hold:
(a) If d0 , then
lim u(, t) = 0
in
C().
(b) If d0 < < d1 , then

lim u(, t) = [,]
in C(),
where [,] is the unique positive steady-state solution of (7.47).

(c) If d1 , then
lim u(, t) =
0 \ ,
in
while
lim u(, t) = L[,BR (x0 )]
in BR (x0 ),
where L[,BR (x0 )] stands for the unique positive solution of the singular
problem

du = u a(d(x))g(u)u
in BR (x0 ),
u=
on BR (x0 ).
7.5
Theorem 7.1 goes back to J. Lopez-Gomez [159, 160, 161] for the special case
when
g(u) = up1 ,
u 0,
(7.48)
185
for some p > 1, and to J. Lopez-Gomez [162] for the general case when g
satisfies (7.2), (KO) and (7.10). Lemma 7.2 goes back to S. Cano-Casanova
and J. L
opez-G
omez [39], as well as Theorem 7.3, which is the main result of
[39]. It extends some previous results of the author for the special case (7.48)
[159, 160, 161].
Theorem 7.1 is an extremely sharp result because it establishes the uniqueness of the positive large solution from the generalized convexity of g(u), imposed through condition (7.10), and the monotonicity of the weight function a,
by means of a tricky use of the strong maximum principle. Consequently, the
knowledge of the exact blow-up rates of the large positive solutions on is
far from necessary for the uniqueness; however it is imperative in most of the
specialized literature. The proof of Theorem 7.1 can be adapted to get more
general uniqueness results valid for general cooperative systems based on J.
L
opez-G
omez and L. Maire [164]. The tremendous advantage of our approach
relies on the fact that it does not invoke the nature of the decay rate of a(x)
as dist (x, ) 0, nor the type of growth of g(u) as u , though those are
indeed imperative for ascertaining the exact blow-up rate of the positive large
solution on .
When S. Cano-Casanova and J. Lopez-Gomez [39] was published, some
other results of very different nature, with completely different techniques,
had been found by Y. Du and Q. Huang [80], J. Garca-Melian, R. Letelier
and J. C. Sabina de Lis [101], F. C. Cirstea and V. Radulescu [55, 56], T.
Ouyang and Z. Xie [205, 206], and J. Garca-Melian [96, 97]. Other uniqueness
results in balls, even for degenerate weights, had been given by M. Chuaqui et
al. [53], [52]. Some of these results were based on ideas and technical devices
taken from [154, 161]. But, adopting the spirit of Y. Du and Q. Huang [80]
and J. Garca-Meli
an, R. Letelier and J. C. Sabina de Lis [101], to find out
the exact blow-up rates of the positive large solutions on the boundary, most
of these papers required a(x) to decay like a fixed power of
d(x) := dist(x, ),
as d(x) 0, which is unnecessary to get uniqueness, at least in the radially
symmetric case, as established by Theorem 7.1.
By some very classical results of C. Loewner and L. Nirenberg [138], V.
A. Kondratiev and V. A. Nikishin [125], C. Bandle and M. Marcus [17], A. C.
Lazer and P. J. McKenna [129], [131], M. Marcus and L. Veron [188], and L.
Veron [230], when a(0) > 0, the monotonicity condition (7.9) is not necessary
for the validity of Theorem 7.1. In such case, under condition (7.48), a very
classical result is the existence of a constant C > 0 such that
2
lim CL(x)d p1 (x) = 1
(7.49)
d(x)0
for any positive solution L(x) of (7.6). Consequently, according to the results
of Chapter 8, (7.6) possesses a unique positive solution.
186
These classical results were substantially sharpened by Y. Du and Q.

Huang [80] to cover the more general case when
C0 := lim
t0
a(t)
>0
t
(7.50)
for some constant 0. By establishing that any positive solution L(x) of

(7.6) satisfies
L(x)
lim
d(x)0
[( + 1)/C0 ]
1
p1
d (x)
=1
with
:=
+2
,
p1
(7.51)
Y. Du and Q. Huang [80] found the uniqueness of the positive solution of (7.6)
for a general class of smooth domains , not necessarily radially symmetric.
Some time later, J. Garca-Melian, R. Letelier and J. C. Sabina de Lis [101]
substantially sharpened (7.51) by establishing that if
a(x) = C0 d (x)[1 + C1 d(x) + o(d(x))]
as d(x) 0,
then, for every ,

1
L(x) = [( + 1)/C0 ] p1 d (x)[1 + B()d(x) + o(d(x))], x ,

where
B() :=
(7.52)
(N 1)H() ( + 1)C1
+p+3
and H() is the mean curvature of at . These results were inspired by

some previous findings of G. Daz and R. Letelier [76] and M. A. del Pino
and R. Letelier [71], which extended the pioneering results of C. Bandle and
M. Marcus [18] for radially symmetric problems. Essentially, all these results
established that the more curved towards the exterior a domain is around a
given point of its boundary, the higher the blow-up rate of that point of the
solution is.
Incidentally, although [80] and [101] were written independently, [80] was
received by the SIAM Journal on Mathematical Analysis on February 22, 1999,
whereas [101] was received by the Proceedings of the American Mathematical
Society on April 17, 2000, after publication of [80]. It should be noted that
[71] was received by the editors on December 14, 1999, and not published until
2002.
Also in a general domain with smooth boundary and not necessarily
radially symmetric, M. Chuaqui et al. [53], [52] characterized the existence
and the uniqueness of the positive large solution of
u = a(x)up
when a(x) satisfies
C1 d (x) a(x) C2 d (x),
x ,
187
for some 0. Essentially, it was established that the equation cannot admit
a positive large solution in if 2, while it admits a unique large positive
solution if (2, 0). Some further refinements of these results were given
by J. Garca-Meli
an [97].
Almost simultaneously, F. C. Cirstea and V. Radulescu [55, 56, 57] generalized some of the results of [80] and [101] by covering the more general
situation when a C 1 [0, ) satisfies a(0) = 0 and
Rt
0
lim p
t0
a(t)
Rt
d 0 a
C1 := lim p
[0, 1].
t0 dt
a(t)
= 0,
(7.53)
2
Note that (7.53) holds, with C1 = +2
(0, 1], under condition (7.50). Moreover, setting
Rt
a
Q(t) := p0
,
t 0,
a(t)
(7.53) also holds provided

a0 0,
Q C 1 [0, R],
Q(0) = 0.
(7.54)
Indeed, Q(0) = 0 entails the validity of the first limit of (7.53), and the second
one can be inferred from the identity
Rt
Rt !
a0 (t) 0 a
d 0 a
p
,
t > 0,
(7.55)
0
=2 1 p
a(t) a(t)
dt a(t)
because, taking into account that Q C 1 [0, R], the next limit is well defined:
Rt
d 0 a
p
C1 := Q (0) = lim
.
t0 dt
a(t)
0
Moreover, by (7.55), C1 1. Finally, since Q(0) = 0 and Q(t) 0 for all

t > 0, necessarily
C1 = Q0 (0) 0,
More recently, T. Ouyang and Z. Xie [205] demonstrated the uniqueness
in the special case = BR (x0 ) by imposing
Q C 1 [0, R]
and Q(0) = 0,
where the quotient function Q is defined by

Rt
a
t 0.
Q(t) := 0 ,
a(t)
(7.56)
(7.57)
188
Condition (7.56) is rather reminiscent of (7.54), though it might be stronger

than (7.54), because, thanks to the Holder inequality,
R 21 1
t
Rt
a t2
a
0
0
Q(t) = p
p
=
a(t)
a(t)
R t ! 12
p
a
1
0
t 2 = tQ(t),
a(t)
and hence, Q(0) = 0 if Q(0) = 0. For every t > 0,

Rt
a0 (t) 0 a
I 00 (t)I(t)
0
,
Q (t) = 1
=1
2
a (t)
[I 0 (t)]2
where
Z
I(t) =
t 0,
a(s) ds,
0
and it becomes apparent how condition (7.56) is closely related to (7.23).

Nevertheless, except in [159, 160, 161], to prove the uniqueness, the usual
strategy adopted in most of the available references consists of establishing
that all large solutions have the same blow-up rate on the boundary to infer
the uniqueness from this property, as will be discussed in Chapter 8.
Theorem 7.1 does not impose any special requirement on a, like (7.50),
(7.53), (7.54), (7.56), or (7.23), beyond its monotonicity. Condition (7.9) provides us with a uniqueness theorem for which the knowledge of the exact
blow-up rates of the large solutions on is not needed. We conjecture that,
for every g(u) satisfying (7.2), (KO) and (7.10), and any a C[0, ) such
that a(0) = 0 and a(t) a(s) > 0 for sufficiently small t s > 0, the problem
(7.6) has a unique solution. Our conjecture relies on Theorem 7.1 and on the
fact that there always exist a1 , a2 C[0, ) non-decreasing with
a1 (0) = a2 (0) = 0,
a1 a a2 ,
a1 (t) = a(t) = a2 (t)
if t 0.
According to Theorem 7.1, for each i {1, 2}, the problem

u = u ai (d(x))g(u)u
in ,
u=
on ,
has a unique positive solution if 0. Let denote it by Li . By comparison,
L2 L L1
for every solution L of (7.6). As the blow-up rates of the large solutions should
only depend on the values of a around and a = a1 = a2 in a neighborhood
of , necessarily L1 and L2 , and hence L, have identical blow-up rates on
.
The discussion carried out in the final section of Chapter 9 reveals that
imposing 0 might be imperative for the validity of most of the results of
this chapter. Indeed, at least in the context of superlinear indefinite problems,
the nature of the problem changes dramatically when changes sign.
Chapter 8
General uniqueness results
8.1
8.2
8.3
8.4
8.5
8.6
Boundary normal section of a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Boundary blow-up rate of the large solutions . . . . . . . . . . . . . . . . . . .
Proof of Theorem 8.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Special case when a(x) > 0 for some x . . . . . . . . . . . . . . . . . . . .
Uniqueness of the large solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
190
192
193
201
202
204
This chapter establishes the uniqueness of the positive solution of the singular
in j ,
du = u + a(x)g(u)u
u=0
on j ,
(8.1)
u=
on j \ ,
under the following assumptions:
(i) d > 0, 0, and a(x) satisfies Hypothesis (Ha).
(ii) g C[0, ) C 1 (0, ) satisfies (7.2), (7.10) and (7.16).
(iii) For each 1 j q0 ,

0,1
0,j ,
j := \
(8.2)

0,j =
mj
[
i0,j ,
1 j q0 .
i=1
It should be remembered that according to Hypothesis (Ha), i0,j , 1 j q0 ,

1 i mj , are the components of
0 = int a1 (0).
More generally, given any subdomain D such that D consists of entire
components of and 0 , we will establish the existence of at most one
positive solution of
in D,
du = u + a(x)g(u)u
Bu = 0
on B ,
(8.3)
u=
on ,
189
190
where is an open and closed subset of D,

B := D \ ,
and B is a general mixed boundary operator on B like (1.49). More precisely,
for a given open and closed subset of B , 0B ,
1B := B \ 0B
we will take
and C 1 (D),
,
B :=
+ b(x),
on 0B ,
on 1B ,
where b C 1+ (1B ) and C 1 (1B ; RN ) is an outward pointing nowhere

tangent vector field. Consequently, B is the Dirichlet boundary operator on 0B
and the Neumann or a first order regular oblique derivative boundary operator
on 1B . It should be noted that either 0B , or 1B or both simultaneously may
be empty.
Dividing by d the differential equation and setting
a
a := ,
:= ,
d
d
(8.3) can be equivalently written in the form
in D,
u = u a(x)g(u)u
Bu = 0
on B ,
(8.4)
u=
on ,
which is the problem considered in this chapter.
Based on the uniqueness results of Chapter 7, the main result of this
chapter establishes the uniqueness of a positive solution of (8.4) through the
boundary normal sections of a(x) on D, which are going to be introduced in
Section 8.1.
The organization of the chapter is the following: Section 8.1 introduces the
concept of boundary normal section, Section 8.2 states the main theorem of
this chapter, which provides us with the exact blow-up rates of the positive
solutions of (8.4) for a wide class of non-oscillatory weight functions a(x),
Section 8.3 gives the proof of the main theorem and, finally, Section 8.4 uses
it to establish the uniqueness of the positive solution of (8.4) if it exists.
8.1
Boundary normal section of a
Throughout this chapter,

n : D
x
RN
nx
191
stands for the outward normal vector field to D. As D is of class C 2+ ,

n is well defined. Actually, thanks to Theorem 1.9 of Lopez-Gomez [163],
D satisfies the uniform interior sphere property in the strong sense on D.
Similarly, it also satisfies an exterior sphere property in the strong sense on D.
These properties are extremely useful in this chapter. According to Definition
1.2 of [163], D is said to satisfy the uniform interior sphere property in the
strong sense on D if there is R > 0 such that for every z D with
dist (z, D) R
there exists a point (z) D such that

dist (z, ) = |z (z)|,
BR
z (z)
(z) + R
|z (z)|

D.
Shortening R > 0, if necessary, and setting

zx := x Rnx ,
x D,
we have that, for every x D,

|x zx | = R,
BR (zx ) D,
R (zx ) D = {x}.
B
(8.5)
Moreover, one can also get

R (x + Rnx ) D
= {x}
B
for all x D.
(8.6)
Throughout this chapter, we will assume that R has been chosen to satisfy
(8.5) and (8.6).
The next concept plays an important role in the general theory developed
in this chapter.
Definition 8.1 (Boundary normal sections) For each x D, the function ax C[0, R] defined by
ax (t) := a (x tnx ) ,
t [0, R],
is said to be the boundary normal section of a at x D.

Roughly speaking, the boundary normal section of a at x D is the local
restriction of a along the normal line to D through x. Subsequently, for every
x and (0, R], we will denote by ax, C[0, ) the function

ax (t)
if t [0, ),
ax, (t) :=
(8.7)
ax ()
if t ,
which is one of the multiple (global) prolongations of the (local) boundary
normal section.
192
8.2
Boundary blow-up rate of the large solutions
The main result of this chapter can be stated as follows.

a 0, a(x) > 0 for all x D near
Theorem 8.2 Suppose 0, a C(D),
for some (0, R], and the boundary normal
, a C 1 (( + B ) D)
sections ax , x , satisfy ax C 3 (0, ],
a0x (t) > 0
ax (0) = 0,
(log ax )00 (t) < 0
t (0, ),
(8.8)
(ax /a0x )0 is non-oscillating, as discussed by Definition 6.6, and

lim
3yx
t0
ay (t)
=1
ax (t)
uniformly in x .
(8.9)
Suppose, in addition, that g C[0, ) C 1 (0, ) satisfies (7.2), (7.10) and

(7.16). Then, any positive solution, L, of (8.4) satisfies
p
L(x tnx )
lim
= I0,xp1
t0
bx (t)
p+1
p1
p+1
p1
H p1
(8.10)
bx (t)b00x (t)
.
[b0x (t)]2
(8.11)
for all x , where

Z
Z
bx (t) :=
1
p+1
p+1
p1
ds,
ax
t
I0,x := lim
t0
As we have already discussed after the statement of Proposition 6.7, the local logarithmic concavity imposed in (8.8) is rather natural, since ax (0) = 0
implies
lim ln ax (t) = .
t0
The additional regularity ax C 3 (0, ] together with (8.8) and the nonoscillating behavior of (ax /a0x )0 in (0, ] is required to apply Proposition 6.9
since, according to it, the functions
1
R t p+1
a0x (t) 0 ax
,
Qax (t) :=
1
ax (t) a p+1
(t)
x
t (0, ],
(8.12)
are non-oscillating for all p > 1. Thus, owing to Proposition 6.7, for every p > 1
and x the limit I0,x in (8.11) is well defined and satisfies I0,x [1, ).
Naturally, for every p > 1 the choice
g(u) := up1 ,
u 0,
satisfies all the requirements of Theorem 8.2.
8.3
193
Fix > 0. Due to (8.9), there exist % = %() (0, ) and = () > 0 such
that, for every x, y with dist (x, y) %,
1
ay (t)
a (y tny )
=
1 + for all t [0, ].
ax (t)
a (x tnx )
(8.13)
: dist (z, ) }.
Q := {y tny : (y, t) [0, ]} = {z D
Since is of class C 2 , % and can be shortened if necessary so that, for
every z Q, there exists a unique
((z), t(z)) [0, ]
such that
z = (z) t(z)n(z) ,
(8.14)
t(z) = |z (z)| = dist (z, D) = dist (z, ).
Throughout the rest of the proof, % and are chosen to satisfy these requirements.
According to the discussion at the beginning of Section 8.1, there exists
R > 0 such that

%
,
R < min
2 4
and
BR (x Rnx ) int Q,
R (x Rnx ) D = {x},
B
for all x . For such choice of R, there exists 0 > 0 such that, for every
x and a (0, 0 ],
R (x (R + )nx ) int Q.
B
Thus,
Bx :=
R (x (R + )nx ) Q.
B
00
By construction, it follows from (8.13) and (8.14) that for every x ,

R (x (R + )nx ),
[0, 0 ] and z B
a(z) = a((z) t(z)n(z) ) (1 )a(x t(z)nx )
= (1 )ax (t(z)) = (1 )ax (dist (z, D))
(1 )ax (dist (z, BR (x (R + )nx )) .
194
R (x (R + )nx ), we have that

Moreover, since z B
dist (z, BR (x (R + )nx )) = R |z x + (R + )nx |
and hence,
a(z) (1 )ax (R r ),
r |z x + (R + )nx |,
(8.15)
R (x (R + )nx ).
for all x , [0, 0 ] and z B
Let L be a positive solution of (8.4). By (8.15), L provides us with a positive (bounded) subsolution of the semilinear singular boundary value problem

u = u (1)ax (Rr )g(u)u
in BR (x(R+)nx ),
(8.16)
u=
on BR (x(R+)nx ),
for all x and [0, 0 ]. Thanks to (8.8), Theorem 7.1 guarantees that,
for each x and [0, 0 ], (8.16) possesses a unique positive solution,
,x, . Moreover, by uniqueness,
L
,x, = L
,x,0 ( + nx )
L
for all (x, ) [0, 0 ].
(8.17)
On the other hand, according to Theorem 3.4, we already know that

,x, = lim u
L
[,x,,M ] ,
M
where u
[,x,,M ] stands for the unique positive solution of

u = u (1)ax (Rr )g(u)u
in BR (x(R+)nx ),
u=M
on BR (x(R+)nx ).
Moreover, thanks to, e.g., Theorem 1.7, the map
M 7 u
[,x,,M ]
is increasing. Hence, setting
M :=
max
x , (0, 0 ],
L,
BR (x(R+)nx )
it follows from Theorem 1.7 that

,x,
Lu
[,x,,M ] L
in BR (x(R+)nx )
for all x and (0, 0 ]. Thus, letting 0 yields

,x,0
LL
in BR (x Rnx )
for all x and sufficiently small > 0.

On the other hand, if > we have that
(1 )ax (1 )ax
(8.18)
195
for all x and hence, u

[,x,0,M ] provides us with a subsolution of

u = u (1)ax (Rr0 )g(u)u
in BR (x Rnx ),
u=M
on BR (x Rnx ).
Consequently, thanks to Theorem 1.7, we find that
u
[,x,0,M ] u
[,x,0,M ]
for all M > 0.
So, letting M , from Theorem 3.4 it becomes apparent that

,x,0 L
,x,0
L
for all x and < . By a rather standard compactness argument, this
estimate shows that
,x,0
x := lim L
L
0
is well defined. Actually, it provides us with the unique positive solution of

the singular problem

u = u ax (Rr0 )g(u)u
in BR (x Rnx ),
(8.19)
u=
on BR (x Rnx ),
,x, are supported.

FIGURE 8.1: The balls where the supersolutions L
196
for all x . It should be noted that in the special case when, for some
p > 1,
g(u) = Hup1
then, by a simple re-scaling argument,
1
,x,0 = (1 ) p1
x.
L
L
Now, letting 0 in (8.18) shows that

x
LL
in BR (x Rnx )
for all x .
(8.20)
p+1

p
x (x tnx )
1
p + 1 p1 p1
L
p1
= I0,x
H
.
lim
t0
bx (t)
p1
(8.21)
By Theorem 7.3, we already know that
Therefore, according to (8.20) and (8.21) we find that

p
L(x tnx )
p1
lim
I0,x
t0
bx (t)
p+1
p1
p+1
p1
H p1
(8.22)
for all x .
Subsequently, we suppose that R > 0 and 0 > 0 have been shortened, if
necessary, so that
R (x + Rnx ) D
= {x} and B
R (x + (R + )nx ) RN \ D
B
for all x and (0, 0 ]. Moreover, we will set
R1 := R,
Rm := R1 + max |z1 z2 | + 1,
z1 ,z2 D
R2 := 2Rm R1 ,
and consider, for every x and [0, 0 ], the open annulus

A := AR1 ,R2 (x + (R + )nx ) .
By construction, there exists 1 (0, 0 ) such that, for every x ,
\
D
A .
01
Next, we pick a sufficiently small > 0 such that, for every x ,

Q,
AR1 ,R1 +2 (x + Rnx ) D
and consider the constant
C := ka kC(D)
+ 1.
197
By (8.13), for every
z AR1 ,R1 + (x + Rnx ) D

the next estimate holds
a(z) = a((z) t(z)n(z) ) (1 + )a(x t(z)nx ),
because, since R < %2 ,
|x (z)| R1 + < %.
Thus,
a(z) (1 + )ax (t(z)) = (1 + )ax (dist (z, ))
(1 + )ax (dist (z, x)) (1 + )ax (dist (z, BR (x + Rnx ))
(1 + )ax (dist (z, BR (x + (R + )nx ))
for all [0, 1 ]. Hence,
a(z) (1 + )ax (dist (z, A ))
(8.23)

[0, 1 ].
(z, ) AR1 ,R1 + (x + Rnx ) D
(8.24)
for all
Subsequently, for sufficiently large n 1 with

1
%
< 1+
< R < < ,
n
2
x defined by
we will consider the function a
ax (t)
if 0 t ,

x (t) :=
n t 1 [C ax ()] + ax () if < t 1 + n1 ,
a
C
if 1 + n1 < t Rm .
It is a non-decreasing continuous extension of ax from the interval [0, ] to the
interval [0, Rm ] such that, for sufficiently large n 1,
x (dist (z, A ))
a(z) (1 + ) a
for all (z, ) D [0, 1 ].
(8.25)
Indeed, due to (8.23), (8.25) holds in the range (8.24). In particular, if

dist (z, A ) for some [0, 1 ], then (8.24) holds and hence,
a(z) (1 + )ax (dist (z, A )) = (1 + )
ax (dist (z, A )).
Similarly, if

1
dist (z, A ) > 1 +
n
198
for some [0, 1 ], then

a(z) C < (1 + )C = (1 + )
ax (dist (z, A ))
and hence, (8.25) also holds. Lastly, suppose that, for some [0, 1 ],

1
< dist (z, A ) 1 +
.
n
The set of these points z lies in the compact set
o
n
: dist (z, A ) %
K := z D
2
where, owing to (8.8), a is a function of class C 1 , because %2 < . In particular,
a is bounded in K and, consequently, since

n
n
1
0
ax (t) = [C ax ()]
if < t 1 +
,
n
it becomes apparent that (8.25) holds for sufficiently large n 1.
Due to (8.25), L is a supersolution of the singular problem

u = u (1 + )
ax (dist (z, A ))g(u)u
in A ,
u=
on A ,
(8.26)
for each [0, 1 ]. Thanks to (8.8), it follows from Theorem 7.1 that (8.26)
possesses a unique positive solution, L,x, . By the uniqueness,
L,x, = L,x,0 ( nx ),
[0, 1 ].
Actually, for each (0, 1 ], the restriction L,x, |D is a bounded solution of

u = u (1 + )
in D and, therefore, thanks to Theorem 1.7,
L,x, = L,x,0 ( nx ) L
in D.
On the other hand, by Theorem 3.4, we already know that

L,x, = lim u[,x,,M ] ,
M
where u[,x,,M ] stands for the unique positive solution of

u = u (1+)
in A ,
u=M
on A .
Moreover, the map
M 7 u[,x,,M ]
199
is increasing. Hence,
u[,x,,M ] L,x, L in D
for all x and (0, 0 ]. Thus, letting 0 yields
u[,x,0,M ] L,x,0 L in D
(8.27)
for all x and sufficiently small > 0.

Lastly, note that if > , then we have that
(1 + )
ax (1 + )
ax
for all x . So, u[,x,0,M ] provides us with a supersolution of

u = u (1+)
ax (dist (z, A0 ))g(u)u
u=M
in A0 ,
on A0 .
FIGURE 8.2: The annuli where the subsolutions are supported.
200
Consequently, thanks to Theorem 1.7, we find that

u[,x,0,M ] u[,x,0,M ]
for all M > 0.
So, letting M shows that

L,x,0 L,x,0 L in D
(8.28)
for all x and < . Therefore, by a rather standard compactness

argument, it follows from (8.28) that
Lx := lim L,x,0
0
is well defined and actually provides us with the unique positive solution of
the singular problem

x (dist (z, A0 ))g(u)u
u = u a
in A0 ,
(8.29)
u=
on A0 ,
for all x . Moreover, according to (8.28), we find that
Lx L
in D
(8.30)
for all x .
It should be noted that in the special case when g(u) = Hup1 for some
p > 1, then
1
L,x,0 = (1 + ) p1 Lx
for all > 0.
By Theorem 7.3 we already know that
p
L (x tnx )
p1
lim x
= I0,x
t0
bx (t)
p+1
p1
p+1
p1
H p1 .
(8.31)
Thus, from (8.30) and (8.31) we can infer that

p
L(x tnx )
p1
I0,x
lim
t0
bx (t)
p+1
p1
p+1
p1
H p1
for all x . Therefore, by (8.22),

p
L(x tnx )
p1
lim
= I0,x
t0
bx (t)
p+1
p1
p+1
p1
for all x , which ends the proof of Theorem 8.2.
H p1
Actually, as a consequence from the proof of Theorem 8.2 and of Remark

7.4 the next result holds.
201
Corollary 8.3 Under the same assumptions of Theorem 8.2, for any pair of
positive solutions, L1 and L2 , of (8.4), the function
q(z) :=
L1 (z)
,
L2 (z)
z D,
satisfies
lim q(z) = 1,
d(z) := dist (z, ),
d(z)0
z D.
Proof: According to (8.20) and (8.30), we have that

(z) (z)
L(z) (z)
L
L1 (z)
(z) (z)
L2 (z)
L(z) (z)
L
(8.32)
for all z D with d(z) R. Moreover, setting

p
p1
J(x) := I0,x
p+1
p1
p+1
p1
x ,
H p1 ,
we find from (8.8) and Remark 7.4 that

(z) (z)
L(z) (z)
L
= lim
= 1.
d(z)0 b(z) (d(z))J((z))
d(z)0 b(z) (d(z))J((z))
lim
Thus,
(z) (z)
(z) (z)
L
L
b(z) (d(z))J((z))
= lim
lim
= 1.
L(z) (z)
d(z)0 b(z) (d(z))J((z)) d(z)0
d(z)0 L(z) (z)
lim
Therefore, also
L(z) (z)
lim
=1
d(z)0 L(z) (z)
and letting d(z) 0 in (8.32) ends the proof.
8.4
Special case when a(x) > 0 for some x
Although we imposed a(x) = 0 in the statement of Theorem 8.2, this condition

is not necessary for its validity. Indeed, if a(x) > 0 for some x , then, in
a neighborhood of x, ax lies between two positive constants, e.g.,
a(x) < ax (t) < a(x) +
for all
t (0, ],
202
and the localization method of the proof of Theorem 8.2 can be easily adapted
to show the validity of (8.10) with
R
Z
p+1
p1
C0p+1
bx (t) :=
ds,
I0,x := lim
t0
bx (t)b00x (t)
,
[b0x (t)]2
where
C0 = ax (0) = a(x).
Since a(x) > 0, the non-oscillation conditions on a and its derivatives are not
necessary for the validity of (8.10), of course. In such case, a direct calculation
shows that

2
2
1 p1
t p1 R p1
bx (t) = C0 p1
2
for sufficiently small t > 0. Hence,
I0,x := lim
t0
bx (t)b00x (t)
p+1
=
[b0x (t)]2
2
and consequently, according to (8.10),

L(x tnx )
lim 1
=

2
2
t0
p1
p1
p1 p1
R
C0
t
2
p+1
2
p
p1

p+1
p1
p+1
p1
H p1 .
Therefore, rearranging terms and simplifying yields

lim
t0
L(x tnx )
t
2
p1
1
p1
=a

(x)
2 p+1
p1p1
1
p1
H p1 ,
which provides us with (7.51) if H = 1.
8.5
As an immediate consequence of Corollary 8.3, the following uniqueness result

holds.
Theorem 8.4 Under the same assumptions of Theorem 8.2, (8.4) admits, at
most, a unique positive solution, L.
Proof: Suppose L1 and L2 are two positive solutions of (8.4). Then, according
to Corollary 8.3, the auxiliary function

L1 (z)/L2 (z)
z D,
q(z) :=
1
z ,
203
is continuous in
: dist (z, ) R}
KR := {z D
for sufficiently small R > 0. As KR is compact, q(z) is uniformly continuous
in KR . Thus, for every > 0 there exists > 0 such that
|q(z) q((z))| = |q(z) 1| <
if d(z) = |z (z)| .
Therefore,
(1 )L2 L1 (1 + )L2
in D := {z D : dist (z, ) }.
Next, we will consider the boundary value
u = u a(x)g(u)u
Bu = 0
u = L1
problem
in D \ D ,
on B ,
on (D \ D ) \ B .
(8.33)
By adapting Theorem 1.7, through Theorem 7.1 of [163], it is easily seen that
L1 is the unique solution of (8.33).
On the other hand, since g is increasing, we find that
((1 )L2 ) = (1 )L2 a(x)g(L2 )(1 )L2
(1 )L2 a(x)g((1 )L2 )(1 )L2
in D \ D . Moreover,
B((1 )L1 ) = (1 )BL1 = 0
on B ,
and
(1 )L2 L1
on (D \ D ) \ B .
Thus, (1 )L2 is a subsolution of (8.33). Similarly, (1 +)L2 provides us with

a supersolution of (8.33). Therefore, thanks again to Theorem 7.1 of [163] and
adapting the argument of the proof of Theorem 1.7, it becomes apparent that
(1 )L2 L1 (1 + )L2
in D \ D .
Consequently,
(1 )L2 L1 (1 + )L2
in D
and letting 0 yields L1 = L2 , which ends the proof.
The numerical example of Section 5.3 in the domain 1 = B0.3 satisfies

all the assumptions of Theorem 8.2 and hence, the large solution L[,B0.3 ] is
unique for all < 2 . Indeed, for the choice of the weight function in Section
5.3, we have that, for sufficiently small > 0 and every x B0.3 ,
ax (t) = sin(5(0.8 + t))
for all t [0, ).
204
Since
a0x (t) = 5 cos(5(0.8 + t)),
a00x (t) = (5)2 sin(5(0.8 + t)),
it is apparent that
ax (t) > 0,
a0x (t) > 0,
a00x (t) < 0,
for all t (0, ]. Hence, (8.8) holds. Naturally, as ax is independent of x, (8.9)

also holds. Finally, since

0
[a0x (t)]2 a00x (t)ax (t)
ax (t)
=
= sec2 (5(0.8 + t))
a0x (t)
[a0x (t)]2
we find that

ax (t)
a0x (t)
00
= 2 sec2 (5(0.8 + t)) tan(5(0.8 + t)) > 0
for sufficiently small t > 0. Consequently, (ax /a0x )0 is indeed non-oscillating

and all the assumptions of Theorem 8.2 are fulfilled.
8.6
The results of this chapter go back to J. Lopez-Gomez [161]. The localization

method used in the proof of Theorem 8.2 to determine the exact blow-up rate
on the boundary of the large solutions goes back to J. Lopez-Gomez [154].
Since then, it has been used very often in the specialized literature.
Three years after [161] was published, J. Garca-Melian [98], inspired by
S. Dumont et al. [81], found some sharp uniqueness results for the simplest
prototype model

u = a(x)f (u)
in ,
(8.34)
u=
on ,
for a rather general class of functions f (u), but, once again, he had to impose
C1 d (x) a(x) C2 d (x)
(8.35)
for x near , with 0 and C1 , C2 positive constants. To make the

general results of this chapter valid for wide classes of weight functions a(x)
not satisfying (8.35), one must pay the price of imposing (7.16), i.e., f (u) up ,
with p > 1, for sufficiently large u.
In the very special case when a = 1, a necessary and sufficient condition
205
for the existence of a positive solution of (8.34) when f is increasing is the

classical KellerOsserman condition
Z
dx
qR
< for some u > 0
(8.36)
x
u
f
0
(see J. B. Keller [123] and R. Osserman [202]). However, the uniqueness of the
large solution for general weight functions a(x) is a very trickly problem, not
completely understood yet, even for (8.34), which is a very simplified version
of the general model dealt with in this book.
Although the uniqueness results of Chapter 7 are based on the maximum
principle, at this stage of the book it is apparent that the blow-up rates of the
positive solutions of (8.34) are closely related to the problem of the uniqueness.
Indeed, according to Theorem 6.8 of Y. Du [79] and Theorem 1.6 of S. Dumont
at al. [81], it is well known that, in general, the boundary behavior of the
positive solutions L(x) of (8.34) is given by
lim
d(x)0
(L(x))
= 1,
d(x)
where
1
(t) :=
2
d(x) := dist (x, ),
dx
qR
.
x
f
0
(8.37)
(8.38)
But, in order to get the exact blow-up rate of the positive solutions L(x) of
(8.34), one needs to impose some additional restrictions on the nonlinearity
f (u), for instance,
lim inf
r
(ru)
>1
(u)
for all r (0, 1).
(8.39)
Indeed, in such case, it is well known that

lim
d(x)0
L(x)
=1
(d(x))
(8.40)
where := 1 (e.g., Corollary 6.9 of Y. Du [79]). It turns out that under

condition (8.40)
L1 (x)
=1
lim
d(x)0 L2 (x)
for any pair, L1 and L2 , of positive solutions of (8.34). Consequently, imposing
the additional monotonicity assumption that
f (u) = g(u)u,
u>0
with g(u) increasing, the uniqueness of the positive solution of (8.34) holds,
206
which is a classical result of C. Bandle and M. Marcus [17]. Note that (8.39)
holds under condition
f (u)
H := lim p > 0
u u
for some p > 1. The reader is sent to J. Garca-Melian [98] for a complete
discussion from a classical perspective. This book pays attention to a much
more general spatially heterogeneous prototype model, where the nonlinearity
is far from being monotonic and most of the previous results fail.
The localization method used in the proof of Theorem 8.2 has been widely
used in the literature to find a number of refinements of some of the results
covered in the last three chapters. For example, T. Ouyang and Z. Xie [205,
206] adapted it to get their own results.
The literature on the uniqueness of large solutions for the equations covered
in this book and other related equations with gradient terms, singular coefficients or even p(x)-Laplace operators, is really huge. The interested reader
might wish to have a look at Z. Zhang [246, 247, 248, 249, 250, 251, 252], Q.
Zhang, X. Liu and Z. Qiu [243], Z. Xie [238], P. Feng [86], M. Wang and L.
Wei [231], C. Mu, S. Huang, Q. Tian and L. Liu [196], H. Li, P. Y. H. Pang
and M. Wang [135, 136], M. M. Boureanu [30], Q. Zhang, Y. Wang and Z.
Qiu [244], Y. Wang and M. Wang [233], L. Wei [234], S. Huang and Q. Tian
[114], S. Huang, Q. Tian, S. Zhang, J. Xi and Z. Fan [117], Y. Chen and M.
Wang [45, 46, 47, 48], Q. Zhang [242], S. Huang, Q. Tian, S. Zhang and J.
Xi [116], M. Marras and G. Porru [191], S. Huang, W. T. Li, Q. Tian and
C. Mu [118], L. Wei and J. Zhu [236], S. Alarcon, J. Garca-Melian and A.
Quaas [5], S. Alarc
on, G. Daz, R. Letelier and J. M. Rey [3], S. Alarcon, G.
Daz and J. M. Rey [4], Z. Xie and C. Zhao [239], C. Liu and Z. Yang [137],
C. Aneda and G. Porru [15], S. Huang, Q. Tian and Y. Mi [115], S. Nakamori
and K. Takimoto [198], Y. Chen, P. Y. H. Pang and M. Wang [49], N. Belhaj
Rhouma, A. Drissi and W. Sayeb [22], Q. Zhang and C. Zhao [245], Y. Liang,
Q. Zhang and C. Zhao [133], Y. Chen, Y. Zhu and R. Hao [51], L. Wei and
M. Wang [235], T. Shibata [221], X. Ji and J. Bao [121], Z. Zhang, Y. Ma,
L. Mi and X. Li [253], Y. Chen and Y. Zhu [50], H. Feng and C. Zhong, [85],
W. Lei and M. Wang [132], Z. Zhang and L. Mi [254], H. Yang and Y. Chang
[241], L. Mi and B. Liu [194], D. Repovs [218], J. Bao, X. Ji and H. Li [19], L.
Chen, Y. Chen and D. Luo [44], and W. Wang, H. Gong and S. Zheng [232],
among many others.
Some related results by the author and his coworkers on the porous media
equation and variable exponents are found in M. Delgado, J. Lopez-Gomez
and A. Su
arez [73, 74, 75], J. Lopez-Gomez and A. Suarez [182], and J. LopezG
omez [156].
Part III
Metasolutions do arise
everywhere
Chapter 9
A paradigmatic superlinear
indefinite problem
9.1
9.2
9.3
9.4
9.5
9.6
9.7
9.8
9.9
9.10
Components of positive steady states . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Local structure at stable positive solutions . . . . . . . . . . . . . . . . . . . . . .
Existence of stable positive solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Uniqueness of the stable positive solution . . . . . . . . . . . . . . . . . . . . . . .
Curve of stable positive solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dynamics in the presence of a stable steady state . . . . . . . . . . . . . .
9.6.1
Global existence versus blow-up in finite time . . . . . . . . . .
9.6.2
Complete blow-up in + . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dynamics for [d1 , d2 ) and a+ small . . . . . . . . . . . . . . . . . . . . . . .
Dynamics for [d1 , d2 ) and a+ large . . . . . . . . . . . . . . . . . . . . . . .
Dynamics for d2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.10.1 Abiotic stress hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
This chapter studies the parabolic problem

u
t du = u + a(x)|u|p u
u=0
u(, 0) = u0 > 0
in (0, ),
on (0, ),
in ,
212
224
230
233
237
240
240
245
252
254
256
257
265
(9.1)
where d > 0, p 1, R, is a bounded domain of RN with N 1

is arbitrary. Since in this chapter we are not
of class C 2+ , and a C ()
requiring a(x) to satisfy a 0, as we have done in the previous ones, (9.1)
is said to be a parabolic problem of superlinear indefinite type. Naturally, we
are most interested in the general case when a(x) changes sign in . Precisely,
throughout this chapter we assume that
209
210
Hypothesis (HA)
There are x+ , x such that a(x+ ) > 0 and a(x ) < 0, the open sets
+ := {x : a(x) > 0}, := {x : a(x) < 0}, 0 := int a1 (0),
are of class C 2+ , and the negative part of a(x),
x ,
a (x) = min{0, a(x)},

satisfies Hypothesis (Ha) if 0 is non-empty.
As they are of class C 2+ , the open sets + , and 0 at most have finitely
many components. Figure 9.1 shows an admissible nodal configuration of a(x)
with 0 non-empty. The problem (9.1) is superlinear because a > 0 somewhere, though of indefinite type because simultaneously it is of sublinear type
in . Indeed, for every u > 0,
 u
if a(x) > 0.
Naturally, these problems might be also called sublinear of indefinite type.
The positive part of the weight function a, a+ , is defined by
a+ (x) := max{0, a(x)} = a(x) a (x),
x .
Obviously,
a+ 0,
a 0,
a = a+ + a ,
and the relative sizes of a+ and a measure the sublinear or superlinear character of the parabolic problem (9.1). Indeed, for sufficiently small a the model
(9.1) should be utterly superlinear, while for small a+ the model should exhibit
a sublinear behavior, in a sense to be discussed later.
Closely related to problem (9.1) arises the associated sublinear problem
u
in (0, ),
t du = u + a (x)|u|p u
(9.2)
u=0
on (0, ),
u(, 0) = u0 > 0
in ,
whose dynamics has been already analyzed in Part I of this book. As the
solutions of (9.2) are subsolutions of (9.1), if we denote by u(x, t; , a, u0 ) and
u(x, t; , a , u0 ) the solutions of (9.1) and (9.2), respectively, then, thanks to
the parabolic maximum principle, we find that
u(x, t; , a , u0 ) < u(x, t; , a, u0 )
for all (x, t) (0, T )
A paradigmatic superlinear indefinite problem
211
for all time time T for which u(x, t; , a, u0 ) is well defined. In particular,
the metasolutions of (9.2) push up the solutions of (9.1). Consequently, these
metasolutions should play a significant role not only for describing the dynamics of (9.2) but also for ascertaining the global dynamics of (9.1). By the
superlinear character of (9.1) in + , the local solutions of (9.1), those defined
for t [0, T ), might not be globally defined for all time t > 0, as they can
blow up in a finite time. Indeed, it is well known that (9.1) possesses a unique
classical solution, u := u(x, t; , a, u0 ), which is defined in a maximal existence
interval of the form [0, Tmax ) for some Tmax +. Moreover, u blows up in
L () at time Tb := Tmax if Tmax < +, i.e.,
lim ku(, t; , a, u0 )kC()
=
tTb
(see, e.g., D. Henry [112] or P. Quittner and P. Souplet [213]). Nevertheless, in

many circumstances, the dynamics of (9.1) are regulated by its non-negative
steady states, which are the non-negative solutions of

du = u + a(x)|u|p u
in ,
(9.3)
u=0
on .
Because in this chapter is viewed as a continuation parameter, the solutions
of (9.3) are regarded as solution couples (, u). According to [163, Ch. 5],
FIGURE 9.1: An admissible nodal configuration of a(x).
212
Obviously, (9.3)
any weak solution of (9.3) is a classical solution in C 2+ ().
admits two types of non-negative solutions: the trivial one, u = 0, and the
positive solutions, u 0, u =
6 0. Arguing as in Lemma 1.6, it is apparent
that any positive solution of (9.3), u, satisfies u 0. These solutions are
supersolutions of the underlying sublinear elliptic problem

du = u + a (x)|u|p u
in ,
(9.4)
u=0
on ,
whose non-negative solutions are the steady states of (9.2). By Theorem 4.1,
(9.4) admits a positive solution if, and only if, d0 < < d1 , where
].
1 := 1 [, \
For the special nodal configuration represented in Figure 9.1, \ consists
of two components, 0 and + . Hence,
] := min {1 [, 0 ], 1 [, + ]} .
1 [, \
(9.5)
From the point of view of population dynamics, (9.1) is a generalized diffusive logistic equation that incorporates some facilitation effects between the
individuals of the species u, measured by a+ , in the patch + . It seems rather
reasonable that, according to the amount of resources in each patch of , the
interaction between the individuals of the species might change from either
patch. Naturally, in patches with shortcomings of resources, the individuals
will be forced to compete for them, whereas in abundance the individuals
will facilitate the others. So, models a region with harsh environmental
conditions, while + models a region where cooperative synergy between the
individuals occurs. In 0 the individuals are allowed to grow exponentially.
Adopting this perspective, the main goal of this chapter is to ascertain the
combined effects of intra-specific competition and facilitation between the individuals of a single species.
9.1
Components of positive steady states
Arguing as in the proofs of Theorems 2.3 and 4.1, = d0 is the unique

bifurcation value from (, u) = (, 0) to positive solutions of (9.3). Actually,
by the theorem of M. G. Crandall and P. H. Rabinowitz [59], a curve of
positive solutions of (9.3) emanates from (, u) = (, 0) at = d0 . More
precisely, if 0 stands for the unique principal eigenfunction associated
with 0 := 1 [, ] such that
Z
2 = 1,
then the following result holds.
213
Proposition 9.1 There exist s0 > 0 and two (unique) maps of class C 1
v : (s0 , s0 ) C 2+ (),
: (s0 , s0 ) R,
such that
Z
(0) = 0,
v(0) = 0,
v(s) dx = 0
for all s (s0 , s0 ),
and the couple

((s), u(s)) := (d0 + (s), s( + v(s)))
(9.6)
is a solution of (9.3) for every s (s0 , s0 ). Moreover, for sufficiently small

s0 , these are the unique non-trivial solutions of (9.3) in a neighborhood of
and
(, u) = (d0 , 0) R C 2+ (),
Z
(s)
=
a(x)p+2 (x) dx.

(9.7)
lim
s0 |s|p
Proof: It suffices to show (9.7), as the remaining assertions can be easily

derived from the main theorem of [59], as in Part I. Substituting (9.6) into
(9.3), dividing by s, using the definition of and rearranging terms yields
(d d0 )v(s) = (s)( + v(s)) + |s|p a( + v(s))p+1
for all s (s0 , s0 ).
Thus, multiplying this identity by , integrating in and applying the formula

of integration by parts, we find that
Z
Z
(s)
( + v(s)) dx = |s|p
a( + v(s))p+1 dx,
s (s0 , s0 ).
Dividing this relation by |s| and letting s 0 provides (9.7), which ends the
proof.
As v(0) = 0 and 0, for sufficiently small s0 > 0 the couple
((s), u(s)) := (d0 + (s), s( + v(s)))
is a positive solution of (9.3) for all s (0, s0 ). Thus, by setting
Z
D :=
a(x)p+2 (x) dx,
the bifurcation to positive solutions is supercritical if D > 0, i.e., (s) > d0

for small s > 0, while it is subcritical if D < 0, i.e., (s) < d0 for small s > 0.
According to [163, Th. 7.1.4] the component of positive solutions of (9.3),
C+ , bifurcating from (, u) = (, 0) at = d0 is unbounded in R C 2+ ().

A component is a closed and connected subset of the set of positive solutions
that it is maximal for the inclusion. The next result establishes that C+ bends
backward globally if D 0.
214
Proposition 9.2 If (9.3) admits a positive solution, (, u), with d0 ,

then D > 0. Therefore, it cannot admit a positive solution if d0 and
D 0. In particular,
C+ (, d0 ) C 2+ ()
if
D 0.
The proof of this proposition is based on the next result, which is a very
classical property going back to M. Picone [209].
Lemma 9.3 Let u, v C 2 () be two arbitrary functions such that
u=v=0
on
and
C 1 () C().
u
Then, for every f C 1 [0, ) the next identity holds

Z
Z
v v 2
v

(vu uv) =
f0
u2 .
f
u
u
u
(9.8)
Proof: The following chain of identities holds

v
v
v
(vu uv) = f
(v div u u div v) = f
div (vu uv)
f
u
u
u
i
vD v
E
h v
(vu uv) f 0
, vu uv
= div f
u
h u
i
u
v
v 2D v
vE
= div f
(vu uv) + f 0
u ,
.
u
u
u
u
Consequently, by integrating in , applying the divergence theorem, and taking into account that u = v = 0 on , (9.8) holds.
Proof of Proposition 9.2: Let (, u) be a positive solution of (9.3). Then,
and
since 0 and u 0, we have that u C 1 ()
d
p+1
u
(u u) =
p+2
u
u2 ( d0 ) a(x)p+2 .
(9.9)
On the other hand, by (9.8),

Z
Z p+1
p 2

(u u) = (p + 1)
u2 > 0,
u
u
u
since u cannot be a multiple of . Therefore, integrating (9.9) in yields

Z
Z p+2
( d0 )
u2 >
a(x)p+2 (x) dx.
u
Suppose
(9.3) admits a positive solution, (, u), with d0 . Then,
R
p+1
a
dx
< 0 and therefore, D > 0. This ends the proof.
215
According to the linearized stability principle, the local stability of u = 0

as a solution of (9.1) is given by the sign of the eigenvalues of the problem

(d ) =
in ,
(9.10)
=0
on .
Actually, the sign of the principal eigenvalue of (9.10) provides us with the
local qualitative character of (, u) = (, 0). Indeed, if d0 > 0, i.e.,
< d0 , then the trivial solution is linearly asymptotically stable, while if
d0 < 0, i.e., > d0 , then it is linearly unstable. At = d0 , the value of
where the local character of 0 changes, it is said that zero is neutrally stable
and is attractive depending on the global nature of the nonlinearity. Thus, by
the exchange stability principle of M. G. Crandall and P. H. Rabinowitz [60],
the bifurcated solution ((s), u(s)), s > 0, is linearly stable for sufficiently
small s > 0 if D > 0, while it is linearly unstable with a one-dimensional
unstable manifold, if D < 0, i.e., if it bifurcates to the left of d0 .
The next result provides us with an optimal necessary condition for the
existence of a positive solution of (9.3).
Proposition 9.4 Suppose (9.3) possesses a positive solution, (, u). Then,
].
< := (a) 1 [d, \
(9.11)
Proof: Arguing as in the proof of Lemma 1.6, it follows that

= 1 [d a(x)up (x), ].
(9.12)
Thus, by the monotonicity properties of the principal eigenvalue,

] 1 [d, \
],
< 1 [d a(x)up (x), \
and hence, the larger

According to (9.11), the larger the smaller \
(a). Actually, thanks to FaberKrahn inequality, we already know that
lim
|0
|\
(a) = ,
in complete agreement with Theorem 2.2, because the associated sublinear

elliptic problem (9.4) can be regarded as a limiting problem from (9.3) as
approximates . It should be noted that, in such case, 0 = .
Conversely, we have that (a) > d0 and, actually, (a) can be taken
as close as we wish to d0 by choosing sufficiently small. Therefore, by
playing with the weight function a(x), the threshold (a) can indeed take
any value within the interval (d0 , ).
According to Theorem 4.1 and (9.5), the associated sublinear problem (9.4)
admits a positive solution, [,a ] , if and only if
d0 < < (a) = 1 [d, \ ].
216
Thus, since provides us with a positive subsolution of (9.4) for sufficiently

small and any > d0 , and the positive solutions of (9.3) are supersolutions of (9.4), it becomes apparent that (9.3) cannot admit a positive solution
for (a), as has been already established by Proposition 9.4 with another argument. The next theorem establishes that if the superlinear indefinite
problem (9.3) is a perturbation of the associated sublinear model (9.4), in the
sense that
(a+ )M max a+ = max a+
is sufficiently small, then (9.3) admits a positive solution for arbitrarily close
to (a). From this perspective, the universal upper bound (a) for the set
of values of for which (9.3) has a positive solution, J , is optimal.
Theorem 9.5 For every > 0 there exists > 0 such that (9.3) possesses a
positive solution for each (d0 , (a) ] provided (a+ )M < .
R C0 ()
defined by
Proof: Consider the operator G : R C0 ()
G(, u, ) = u (d)1 [u + (a + a+ ) |u|p u] ,
which is of class C 2 , since p 1. Note that the positive zeroes of
F(, u) := G(, u, 0)
are the positive solutions of the associated sublinear problem (9.4). By Theorem 4.1, (9.4) possesses a positive solution, [,a ] , if, and only if, d0 < <
(a). Moreover, going back to the proof of Theorem 4.1, the positive solution
(, [,a ] ) is non-degenerate in the sense that
C0 ()
Du F(, [,a ] ) : C0 ()
is a linear topological isomorphism. Given > 0, set
0 := (a) .
Then, applying the implicit function theorem to the operator G at
(, u, ) = (0 , [0 ,a ] , 0)
shows that there exist 0 > 0 and a (unique) function of class C 2 ,
u : (0 , 0 ) C0 ()
such that
u(0) = [0 ,a ]
and G(0 , u(), ) = 0
for all (0 , 0 ).
By elliptic regularity, (0 , u()) is a positive solution of (9.3) if < 0 . Finally,

let (d0 , 0 ) be. It is very easy to see that (, ) provides us with a
217
subsolution of (9.3) for sufficiently small > 0. Moreover, since < 0 ,

(0 , u()) is a supersolution of (9.3). As < u() for sufficiently small ,
it follows from Theorem 1.4 that (9.3) admits a positive solution for each
This ends
(d0 , 0 ]. Note that Theorem 1.4 is valid for general a C ().
the proof.
But, nevertheless, although the previous results suggest that (9.3) and
(9.4) can share very close solutions for large intervals of , there is nothing
to do between the global behavior of the components C+ for either model.
Even for arbitrarily small a+ , the components behave in a rather different
way. Actually, though C+ is the unique component of positive solutions of
the sublinear problem (9.4), the superlinear indefinite problem (9.3) might
exhibit an arbitrarily large number of components. As a consequence, the
global dynamics of (9.1) and (9.2) are qualitatively different, even for an
arbitrarily small a+ supported in an arbitrarily small compact subset of .
Indeed, suppose = (0, 1), d = 1, p = 4 and
1

2 sin(3x) if x 0, 13 23 , 1 ,
(9.13)
a(x) =
sin(3x),
if x 13 , 23 .
In this example,

0 = ,
1 2
,
3 3

,
+ =
2
0,
3

2
,1 ,
3
and
] = (3)2 .
0 = 1 [, ] = 2 , = 1 [, \
Moreover, (x) = 2 sin(x). Hence,

Z
D :=
a(x) sin6 (x) dx ' 1.3816 > 0.
Consequently, the component C+ bifurcates from u = 0 at = 2 towards

the right of 2 . The pictures on the first row of Figure 9.2 show C+ at two
different scales. They plot the value of the parameter versus the L2 -norm
of the non-negative solutions. The remaining four plots show the profiles of
a series of positive solutions along C+ . They were computed by coupling a
pure spectral method with collocation by R. Gomez-Re
nasco and appeared in
J. L
opez-G
omez [149, Fig. 7.3]. In all those bifurcation diagrams, continuous
lines are filled in by stable solutions and dashed lines by unstable ones. The
left plot on the first row shows the bifurcation diagram for 0 20, while
the second one shows it for 103 102 . Naturally, the details shown in
the first plot are lost in the second one as an effect of the different scales.
The component C+ emanates supercritically from u = 0 at = 2 and
has a subcritical turning point at 17.1615, where it turns backward. The
218
solutions on C+ are linearly asymptotically stable and increase with until

reaches the turning point. Then, they become unstable for all further val-
FIGURE 9.2: Bifurcation diagram and solution plots for the choice (9.13).
219
ues of the parameter. The unstable manifolds of the positive solutions along
the upper half-branch are one-dimensional from the turning point until another critical value, ' 17.1142, where they become two-dimensional. So,
a secondary bifurcation occurs at ' 17.1142. The secondary bifurcation is
subcritical and each of the solutions on the secondary branches possesses a
one-dimensional unstable manifold, though the bifurcation diagrams on the
first row of Figure 9.2 can only show one of these secondary branches based
on the symmetry of the problem. Indeed, as a(x) is symmetric about 0.5, the
solutions on the secondary branches are reflections about 0.5 of the solutions
on the other. So, they have identical L2 -norms and cannot be differentiated
in those plots.
The first figure of the second row of Figure 9.2 shows the plots of the
solutions on the lower half-branch of C+ for the values = 9.8697, 10.1620,
12.0291, 15.2227, 17.1614 and 17.1137, whereas the second one shows the plots
of the solutions on the upper half of the primary branch for = 17.1137,
0.0, 200.0, 500.0 and 1000.0. These solutions exhibit a two-peak layer
behavior as with the two peaks approximating the two local maxima
of a(x). Outside these local maxima, the solutions approximate zero as
.
The third row of Figure 9.2 shows a series of plots of the solutions on
each of the two secondary branches, for = 17.1101, 0.0, 200.0, 500.0 and
1000.0. Each of the plots in the left picture is a reflection about 0.5 of the
corresponding one on the right. These solutions exhibit a single peak layer
behavior as . According to the secondary branch where the solution
lies, the peak is localized around one maxima or the other.
Now, instead of (9.13), we will make the choice
1

x 0, 15 25 , 35 45 , 1 ,
2 sin(5x),
(9.14)
a(x) =

sin(5x),
x 15 , 25 35 , 45 .
In this case,

0 = ,
Moreover, (x) =
1 2
,
5 5
3 4
,
5 5

,
+ =
0,
1
5
2 3
,
5 5

4
,1 .
5
2 sin(x), 0 = 2 and
] = (5)2 ,
= 1 [, \
Z
D=
a(x) sin6 (x) dx ' 0.6136 > 0.
Thus, also in this case C+ bifurcates supercritically from u = 0.

Figure 9.3 shows the bifurcation diagram computed for this special choice.
The first row plots it for (20, 15), while the second row shows it for
(200, 50). A dramatic change with respect to the previous case occurs, as
for this choice of a(x) we have computed four components of positive solutions,
220
namely, the component C+ that bifurcates from u = 0 at = 2 , plus three

additional global subcritical folding type components, F1 , F2 and F3 . As the
solutions of F2 are reflections about 0.5 of the solutions of F3 , they share
the L2 -norm and hence, cannot be differentiated in the bifurcation diagrams.
This explains why we decided to plot the component F2 in the left diagrams
of Figure 9.3, together with F1 , while F3 has been plotted on the right.
Subsequently, we explain briefly the scheme adopted to compute these
components. The component C+ was constructed by bifurcation from u = 0
FIGURE 9.3: Bifurcation diagram for the choice (9.14).
221
at = 2 and global continuation. It exhibits a subcritical turning point at

t ' 9.9719 where it turns backward. As for the choice (9.13), the solutions
on C+ are linearly asymptotically stable and increase with until reaches
t where they become unstable, with a one-dimensional unstable manifold
for the entire interval where we computed them. Figure 9.4 plots a series of
solution plots along C+ for = 9.8697 and 9.9544 on the lower half-branch
and = 0.5414 102 , 100.0025 and 200.0033 on the upper one. It should
be noted that the positive solution = 9.8697 is very small, because is very
close to 2 . All these solutions have a single peak around the central maximum
of a(x), 0.5; the smaller , the more emphasized the peak.
Our numerical experiments for the choice (9.14) suggest that (4.3) should
have, at least,
3
X
3
= 23 1 = 7
j
j=1
positive solutions for sufficiently small < 0: three among them with a single
peak around each of the three local maxima of a(x), another three with two
peaks and one further solution with three peaks. The solutions on C+ provide
us with the solutions having a single peak around the central maximum. To
compute the remaining solutions we adopted the following methodology. First,
FIGURE 9.4: Plots of a series of solutions on C+ for the choice (9.14).
222
we solved (9.3) with = (0, 1) and p = 4 for the special choices

1

sin(5x),
x 0, 15 45 , 1 ,

sin(5x),
x 25 , 35 ,
a (x) :=

sin(5x),
x 15 , 25 35 , 45 ,
(9.15)
where [0, 0.5] is regarded as a secondary real parameter. When = 0.5,

(9.15) equals (9.14), while for = 0 the nodal behavior of a0 (x) is reminiscent
of the nodal behavior of a(x) for the choice (9.13). Naturally, it is expected the
bifurcation diagram for the choice a0 (x) will be similar to the one plotted in
Figure 9.2. Indeed, the component C+ for the choice a = a0 , like the one shown
in Figure 9.2, bifurcates supercritically from u = 0 at = 2 and exhibits
a subcritical turning point at t ' 45.4077 where it turns backward. Then,
along the upper half-branch, it exhibits a pitchfork subcritical bifurcation
at b 45.4077 where the unstable manifolds of the solutions become twodimensional. Thus, for every (, b ) the problem (4.3) possesses at least
three positive solutions: two of them with a single peak around each of the
local maxima 0.1 and 0.9 and the third one with two peaks at these these
maxima. The second step consisted in fixing a value of far away from the
bifurcation point b , e.g., = 110.0575, and using instead of as the main
continuation parameter to compute the solution from = 0 up to = 0.5.
We began computing the perturbation of the solution with two peaks. After
we computed it, we switched off the path-following parameter from to to
compute the entire component F1 , which exhibits a subcritical turning point
at t,1 = 6.5421.
Each of the solutions on the upper half-branch of F1 has a threedimensional unstable manifold and three peaks, each around 0.1, 0.5 and 0.9,
while the solutions along its lower half-branch have bi-dimensional unstable
manifolds and two peaks around 0.1 and 0.9. Figure 9.6(a) shows a series of
solution plots along the upper half-branch of F1 , for = 6.5461, 99.9985
and 200.0005, as well as the plots of a series of solutions along its lower
half-branch, for = 6.5425, 100.0580 and 200.0555.
To compute the component F2 we proceeded as in the previous case, but
now perturbing the solution of (9.3) with a single peak around 0.9 from = 0
up to = 0.5 for = 110.0006. Then, we used instead of as the main
continuation parameter to construct F2 , which exhibits its turning point at
t,2 = 2.1657. The solutions on the upper half-branch of this component have
two-dimensional unstable manifolds and exhibit two peaks around 0.5 and
0.9, while the solutions on its lower half-branch have one-dimensional unstable
manifolds and one single peak around 0.9. Figure 9.6(b) shows the plots of
some of the solutions along F2 for the values = 2.1656, 0.7686 101 ,
100.0068 and 200.0062, on its upper half-branch, and for = 2.1656,
0.2747 101 , 100.0007 and 200.0027 on the lower one.
223
FIGURE 9.5: Plots of a series of positive solutions of (9.3) for the choice
(9.14) on the components F1 , (a), F2 , (b), and F3 , (c).
224
Finally, to construct the fourth component, F3 , we adopted the same strategy

as before, but on this occasion starting at = 109.9927 from the positive
solution with a single peak around 0.1. The solutions on its upper half-branch
have two-dimensional unstable manifolds and exhibit two peaks around the
local minima 0.1 and 0.5, while the solutions on its lower half-branch have
one-dimensional unstable manifolds and possess a single peak around 0.1, the
turning point of this component placed at the same value of , t,2 = 2.1657
as the turning point of F2 . Figure 9.6(c) shows the plots of a series of solutions
along F3 .
All those numerical experiments strongly suggest that we can have an
arbitrarily large number of solutions by simply choosing a(x) sufficiently wavy.
As a matter of fact, varying appropriately a(x) in (9.3) might provide us with
very intricate bifurcation diagrams even in one spatial dimension.
Our numerical experiments support the idea that varying coefficients in
a one-dimensional model has a similar effect as varying support domains in
higher dimensional problems. In the context of higher dimensional superlinear problems, it is well documented that breaking down the convexity of the
domain can provoke multiplicity of positive solutions even when a(x) is a positive constant. More precisely, joining two balls with the same radius by a thin
narrow strip provokes a similar effect as varying the coefficient a(x) in model
(9.3), E. N. Dancer [64]. Naturally, we expect that varying the coefficients in
higher dimensional problems will increase the complexity of the bifurcation
diagrams as much as we wish, since in these problems one can play not only
with the shape of the support domain but also with the nodal behavior of the
weight function a(x).
9.2
Local structure at stable positive solutions
Given a positive solution (0 , u0 ) of (9.3), its stability as a steady state of

(9.1) is determined by the spectrum of the linearized operator (2.16), which
in our present situation becomes
L(0 , u0 ) := d 0 (p + 1)a(x)up0
subject to homogeneous Dirichlet boundary conditions on . Indeed, when
(0 , u0 ) is hyperbolic, the dimension of its unstable manifold equals the sum
of the algebraic multiplicities of the negative eigenvalues of L(0 , u0 ). As
L(0 , u0 ) is self-adjoint, their eigenvalues are real and semi-simple. So, their
algebraic and geometric multiplicities coincide, equaling 1 if N = 1.
More generally, for any (, u) R C 2 () with u 0 in , we will denote
L(, u) := d (p + 1)a(x)up .
Subsequently, (0 , u0 ) is said to be
225
Linearly asymptotically stable if 1 [L(0 , u0 ), ] > 0

Linearly unstable if 1 [L(0 , u0 ), ] < 0
Neutrally stable if 1 [L(0 , u0 ), ] = 0
By the linearized stability principle, (0 , u0 ) is exponentially asymptotically
stable if it is linearly asymptotically stable and unstable if it is linearly unstable.
The next two results provide us with the structure of the solution set of
(9.3) around any linearly stable positive solution.
Proposition 9.6 Let (0 , u0 ) be a linearly asymptotically stable positive solution of (9.3). Then, there exist > 0 and a map of class C 2
u : (0 , 0 + ) C0 ()
with u(0 ) = u0 such that:
(, u()) is a positive solution of (9.3) for all (0 , 0 + ).
The map
(0 , 0 + )
C0 ()
u()
is increasing.
such that (, u) =
There exists a neighborhood N0 of (0 , u0 ) in RC0 ()
(, u()) for some (0 , 0 + ) whenever (, u) N0 is a positive
solution of (9.3).
Proof: As in the proof of Theorem 2.3, the solutions of (9.3) are the zeros of
C0 ()
defined by
the operator F : R C0 ()
F(, u) = u (d)1 (u + a|u|p u) .
(9.16)
Since p 1, F is of class C 2 . We are assuming that F(0 , u0 ) = 0 and, since

1 [L(0 , u0 ), ] > 0, the linearized operator Du F(0 , u0 ) is an isomorphism.
Thus, the local existence and the uniqueness of a solution curve through
(0 , u0 ) are guaranteed by the implicit function theorem. Moreover, by implicit differentiation, it follows that
L(, u())u0 () = u() > 0.
Hence, owing to Theorem 1.1, u0 () 0. It should be pointed out that the
linear asymptotic stability of (0 , u0 ) also entails
1 [L(, u()), ] > 0
for sufficiently close to 0 , because the principal eigenvalue varies continuously with the potential. The proof is complete.
226
Proposition 9.7 Let (0 , u0 ) be a neutrally stable positive solution of (9.3)

and let 0 > 0 be a principal eigenfunction associated with
1 [L(0 , u0 ), ] = 0.
Then, there exist > 0 and a map of class C 2
(, u) : (, ) R C0 ()
such that
((0), u(0)) = (0 , u0 ) and ((s), u(s)) is a positive solution of (9.3) for
all s (, ). Moreover,
v(s) := u(s) u0 s0 = O(s2 ),
(s) = 0 + s2 2 + o(s2 ),
as s 0, with
Z
v(s)0 dx = 0
for all
p(p + 1)
2 =
2
s (, )
and
aup1
03
0
< 0.
u
0 0
(9.17)
such that
There exists a neighborhood, N0 , of (0 , u0 ) in R C0 ()
(, u) = ((s), u(s)) for some s (, ) if (, u) N0 solves (9.3).
For every s (0 , 0 ),
sign 0 (s) = sign 1 [L((s), u(s)), ].
(9.18)
Summarizing, the solution set of (9.3) in a neighborhood of (0 , u0 ) has the

structure of a quadratic subcritical turning point. Moreover, the solutions on
the upper half branch of this turning point are linearly unstable, while those
on the lower one are linearly asymptotically stable.
Proof: We already know that the solutions of (9.3) are the zeros of the non C0 ()
defined in (9.16). Since
linear operator F : R C0 ()
Du F(0 , u0 )u = u (d)1 [0 u + (p + 1)a(x)up0 u] ,
L(0 , u0 )0 := (d 0 (p + 1)a(x)up0 ) 0 = 0,
and zero is a simple eigenvalue of L(0 , u0 ), it becomes apparent that
N [Du F(0 , u0 )] = span [0 ].
We claim that
D F(0 , u0 ) = (d)1 u0
/ R[Du F(0 , u0 )].
(9.19)
227
such that
Indeed, if there is u C0 ()
Du F(0 , u0 )u = u (d)1 [0 u + (p + 1)a(x)up0 u] = (d)1 u0 ,
and
by elliptic regularity we find that u C 2+ ()
du = 0 u + (p + 1)a(x)up0 u u0 .
Thus, multiplying this equation by 0 and integrating in yields
Z
0 u0 dx = 0,
which is impossible. Therefore, (9.19) holds and hence,

N [D(u,) F(0 , u0 )] = span [(0 , 0)].
(9.20)
and set Z := X R. According

Let X denote the L2 -orthogonal of 0 in C0 ()
to (9.20),
R = N [D(u,) F(0 , u0 )] Z.
C0 ()
R possesses a unique decomposition as
So, each element (u, ) C0 ()
(u, ) = s(0 , 0) + z
for some s R and z Z. Indeed, one can make the choice
R
u
R 0 20 .
z = (u s0 , ),
s :=
FIGURE 9.6: Local structure of the set of positive solutions of (9.3) around
a linearly stable solution and a neutrally stable solution.
228
Consequently, F(u, ) = 0 can be expressed, equivalently, as H(z, s) = 0,

where
H(z, s) := F((u0 , 0 ) + s(0 , 0) + z).
In particular, the solutions (u, ) F1 (0) in a neighborhood of (u0 , 0 ) are
in correspondence with the solutions (z, s) H1 (0) in a neighborhood of
(0, 0). By construction,
H(0, 0) = F(u0 , 0 ) = 0.
Moreover, the operator
Dz H(0, 0) = D(u,) F(u0 , 0 )|Z
is a topological isomorphism, because, since it is a compact perturbation of
the identity map, it is Fredholm of index zero. Thus, by the implicit function
theorem, there exist > 0 and a map of class C 2 , z : (, ) Z, such that
z(0) = 0 and
H(z(s), s) = 0 for all s (, ).
Moreover, there exists a neighborhood, N , of (0, 0) in Z R such that z = z(s)
if (z, s) N with H(z, s) = 0. Consequently, in a neighborhood of (u0 , 0 )
R, N0 , the solution set F1 (0) consists of a C 2 -curve, (u(s), (s)),
in C0 ()
such that
v(s) := u(s) u0 s0 = O(s2 ),
(s) = 0 + s0 (0) + s2 2 + o(s2 ),
for all s (, ). As for every s (, ) we have that
p+1

v(s)
0
p+1
+
;
du(s) = (s)u(s) + a(x)u0
1+s
u0
u0
(9.21)
differentiating with respect to s at s = 0 and rearranging terms yields

0 = L(0 , u0 )0 = 0 (0)u0
and hence, 0 (0) = 0. Now, differentiating (9.21) twice with respect to s,
particularizing at s = 0 and rearranging terms yields
L(0 , u0 )v 00 (0) = 00 (0)u0 + a(x)p(p + 1)up1
02 .
0
(9.22)
Note that, for every p > 0, the function

up1
02 = up0 0
0
0
u0
is well defined, since 0 0 and u0 0. Multiplying (9.22) by 0 , integrating

in and applying the formula of integration by parts we find that
R
aup1 03
00
(0) = p(p + 1) R 0
.
u
0 0

Thus, to complete the proof of (9.17) it suffices to show that
Z
03 > 0.
aup1
0
229
(9.23)
Thanks to Lemma 9.3,

Z
d
0
u0
2
Z
(0 u0 u0 0 ) = 2

0 2

0 u0 > 0,
u0
(9.24)
because 0 cannot be a multiple of u0 . On the other hand,

d(0 u0 u0 0 ) = 0 (du0 ) + u0 (d0 )
= 0 (0 u0 + aup+1
) + u0 [0 + (p + 1)aup0 ]0
0
= paup+1
0 .
0
Therefore, (9.23) can be easily inferred from (9.24).
To establish (9.18) we argue as follows. Since

u(s) = (d)1 (s)u(s) + aup+1 (s)
for all s (, ),
differentiating with respect to s yields
u0 (s) = (d)1 (0 (s)u(s) + (s)u0 (s) + a(p + 1)up (s)u0 (s))
d
for all s (, ), where 0 = ds
. Thus, by elliptic regularity, inverting d
and rearranging terms, shows that
L((s), u(s))u0 (s) = 0 (s)u(s)
for all s (, ).
(9.25)
Since
u0 (s) = 0 + O(s) 0
and u(s) = u0 + O(s) 0
for sufficiently small s, it follows from Theorem 1.1 that 0 (s) > 0 if, and only
if, 1 [L((s), u(s)), ] > 0. Indeed, by (9.25), 0 (s) > 0 implies
L((s), u(s))u0 (s) > 0
and hence, u0 (s) 0 provides with a strict positive supersolution of
L((s), u(s)) in . Thus,
1 [L((s), u(s)), ] > 0.
Conversely, if 1 [L((s), u(s)), ] > 0, then the resolvent of L((s), u(s)) in
is strongly positive and hence, (9.25) implies that 0 (s) > 0. Similarly, if
0 (s) = 0 then,
1 [L((s), u(s)), ] = 0
230
and, conversely, if 1 [L((s), u(s)), ] = 0 and we denote by s 0 any principal eigenfunction associated with 1 [L((s), u(s)), ] = 0, then multiplying
(9.25) by s and integrating in yields
Z
0 (s)
s u(s) dx = 0,
which implies 0 (s) = 0 and ends the proof of (9.18).

Finally, since 0 (s) = 2s2 + o(s), we find from (9.17) that 0 (s) > 0 if
s < 0, while 0 (s) < 0 if s > 0. Therefore,

>0
if s < 0,
1 [L((s), u(s)), ]
<0
if s > 0,
Corollary 9.8 Let (0 , u0 ) be a neutrally stable positive solution of (9.3).

Then, there exists > 0 such that (9.3) possesses two positive solutions for
each (0 , 0 ). Moreover, one of them is linearly asymptotically stable
and some other linearly unstable. Furthermore, (9.3) cannot admit a positive
solution if > 0 and
| 0 | + ku u0 kC()
< .
Figure 9.6 summarizes at a single glance the main findings of Propositions
9.6, 9.7 and Corollary 9.8.
9.3
Existence of stable positive solutions
The following result establishes that u = 0 is the unique linearly stable nonnegative solution of (9.3) for all d0 .
Theorem 9.9 Let (0 , u0 ) be a positive solution of (9.3) with 0 d0 .
Then, 1 [L(0 , u0 ), ] < 0. Consequently, it is unstable.
Proof: If a > 0 the proof is very easy. Indeed, by (9.12), the monotonicity of
the principal eigenvalue with respect to the potential yields
1 [L(0 , u0 ), ] = 1 [d 0 (p + 1)aup0 , ]
< 1 [d 0 aup0 , ] = 0,
since a > 0. Actually, in this case it also follows from (9.12) that
0 = 1 [d aup0 , ] < 1 [d, ] = d0 .
231
Hence, (9.3) cannot admit a positive solution if d0 .

The proof in the general case when a changes sign is less obvious. It proceeds by contradiction. Suppose (9.3) admits a positive solution (0 , u0 ) with
0 d0 ,
1 [L(0 , u0 ), ] 0.
u
Then, thanks to Propositions 9.6 and 9.7, there exists a positive solution (,
)
of (9.3) such that
d0 ,
u
1 [L(,
)] > 0.
u
of positive
By Proposition 9.6, (,
) lies on a smooth curve (, u()), ' ,
solutions of (9.3) such that
1 [L(, u()), ] > 0,
' .
some of the following

By global continuation of the curve (, u()) for < ,
options occur:
(O1) u() > 0 and 1 [L(, u()), ] > 0 for all .

such that
(O2) There is b <
u() > 0
and 1 [L(, u()), ] > 0
(9.26)
but u(b ) = 0.
for all (b , ],
such that (9.26) holds for all (t , ],
but
(O3) There is t <
u(t ) > 0
and 1 [L(t , u(t )), ] = 0.
The option (O2) cannot occur because

d0
b <
and = d0 is the unique bifurcation value to positive solutions from u = 0.
By Corollary 9.8, the option (O3) cannot occur either. Therefore, the first
option occurs. Note that, according to Proposition 9.6,
=u
u() < u()
for all
< .
Now, for every < 0, let x be such that

u()(x ) = ku()kC()
.
Since
0 du()(x ) = u(x ) + a(x )[u()]p+1 (x ),
+ a(x )[u()]p (x ) 0
232
and hence,
lim u()(x ) = ,
a(x ) > 0,
which contradicts the estimate

u()(x ) < u
(x ) k
ukC()
and ends the proof.
Corollary 9.10 Suppose

Z
a(x)p+2 (x) dx 0.
D :=
Then, all the positive solutions of (9.3) are linearly unstable.

Proof: Thanks to Proposition 9.2, (9.3) cannot admit a positive solution if
d0 . Theorem 9.9 ends the proof.
Theorem 9.11 The problem (9.3) admits a linearly stable positive solution,
(, u), if, and only if, D > 0. Moreover, > d0 .
Proof: Thanks to Corollary 9.10, D > 0 if (9.3) has a linearly stable positive
solution. Moreover, in such case, > d0 , by Theorem 9.9. Conversely, assume
D > 0.
Then, thanks to Proposition 9.1, the component C+ bifurcates supercritically
from (, 0) and in a neighborhood of (, u) = (d0 , 0) consists of the curve
((s), u(s)), s > 0, defined in (9.6). Moreover, since D > 0, it follows from
(9.7) that 0 (s) > 0 for sufficiently small s > 0. Hence, differentiating (9.3)
with respect to s yields
L((s), u(s))u0 (s) = 0 (s)u(s) > 0.
(9.27)
Let s 0 be a principal eigenfunction associated with 1 [L((s), u(s)), ].

Multiplying (9.27) by s , integrating in and applying the formula of integration by parts gives
Z
Z
0
0
1 [L((s), u(s)), ]
s u (s) = (s)
s u(s) > 0.
(9.28)
On the other hand,

u0 (s) = + O(s) > 0
as s 0.
Therefore, (9.28) implies

1 [L((s), u(s)), ] > 0.
This shows that in a neighborhood of the bifurcation point, (, u) = (d0 , 0),
the component C+ consists of linearly asymptotically stable positive solutions.
9.4
233
Uniqueness of the stable positive solution
The main result of this section is the following.

Theorem 9.12 Suppose D > 0 and let 0 > d0 be such that (9.3) possesses
a linearly stable positive solution (0 , u0 ), i.e.,
1 [L(0 , u0 ), ] 0.
Then, for every (d0 , 0 ], the problem (9.3) has a unique linearly stable
positive solution, (, ). Moreover:
(a) For every (d0 , 0 ), the positive solution (, ) is linearly asymptotically stable, i.e. 1 [L(, ), ] > 0.
(b) The map
(d0 , 0 )
7
C0 ()
is of class C 2 and, for every x , the map 7 (x) is increasing.

(c) The following relations hold
lim k kC()
= 0,
d0
lim k u0 kC()
= 0.
(9.29)
(d) For every (d0 , 0 ], (, ) is the minimal positive solution of (9.3).

Therefore, the minimal positive solution of (9.3) is the unique linearly stable
positive steady state solution of (9.1) and, actually, it is linearly asymptotically
stable if < 0 .
Theorem 9.12 establishes that, for every (d0 , 0 ) the minimal positive
solution is linearly asymptotically stable and that any other positive steady
state of (9.1) must be linearly unstable. Moreover, u0 must be the minimal
positive solution of (9.3) at = 0 .
Remark 9.13 Since D > 0, the existence of a linearly stable positive solution, (0 , u0 ), with 0 > d0 is guaranteed by Theorem 9.11. By Corollary
9.10, D > 0 is necessary for the existence of (0 , u0 ).
In the proof of Theorem 9.12 we will use the following result.
Proposition 9.14 Suppose > d0 and (9.3) possesses a positive solution,
, then admits a minimal positive solution, min . Moreover, the minimal positive solution is linearly stable, i.e., 1 [L(, min ), ] 0.
234
Proof: Assume > d0 and represents a positive solution of (9.3). Then,

u := is a subsolution of (9.3) for sufficiently small > 0. Moreover, by
shortening > 0, we can assume that u = < . Then, the unique solution of
(9.1) with u0 = , u(x, t; , a, ), is bounded above by and it is increasing
in time. In particular, it is globally defined in time and, according to M.
Langlais and D. Phillips [128], the limit
= lim u(, t; , a, )
t
is well defined and it provides us with a positive solution of (9.3). Moreover,

and if < , by construction. Naturally, the minimal positive
solution of (9.3) can be defined through
min := min 0.
>0
Indeed, necessarily min is a non-negative solution of (9.3). Moreover, min 0

because we are assuming > d0 and the unique bifurcation value to positive
solutions from u = 0 is = d0 . Let be an arbitrary positive solution of
(9.3). Then, the argument above establishes that for sufficiently small
> 0. Consequently, min . So, min is indeed the minimal positive solution
of (9.3). It remains to show that
1 [L(, min ), ] 0.
(9.30)
On the contrary, suppose that

1 := 1 [L(, min ), ] < 0.
Let 0 be any positive eigenfunction associated with 1 . Then,
u := min
is a supersolution of (9.3) for sufficiently small > 0. Indeed,
du = dmin + d
p+1
p
(p + 1)amin
1
= min + amin

p
p+1
= u + a min
(p + 1)min
1 .
On the other hand, setting

f () := (min )p+1 ,
it is apparent that
p+1
p
up+1 = f () = f (0) + f 0 (0) + o() = min
(p + 1)min
+ o()
235
as 0. Consequently,
du = u + aup+1 + o() 1 > u + aup+1
for sufficiently small > 0, because we are assuming that 1 < 0. Therefore,
u is a positive supersolution of (9.1) for sufficiently small > 0. As (9.3)
possesses arbitrarily small positive subsolutions, , the problem (9.3) has a
positive solution between and u < min , by Theorem 1.2. This contradicts
the minimality of min and ends the proof.
Proof of Theorem 9.12: Suppose D > 0 and let (0 , u0 ) be a positive
solution of (9.3) with 0 > d0 . Thanks to Propositions 9.6 and 9.7, there
exist > 0 and a map of class C 2 ,
(0 , 0 )
7
C0 ()
such that (, ) is a positive solution of (9.3) with

1 [L(, ), ] > 0
for all
(0 , 0 ).
Moreover,
lim k u0 kC()
= 0.
By global continuation of the curve (, ) towards the left of 0 , one of the

following options occurs:
(O1) > 0 and 1 [L(, ), ] > 0 for all < 0 .
(O2) There exists b < 0 such that
> 0,
1 [L(, ), ] > 0
for all
(b , 0 ]
for all
(t , 0 ],
and b = 0.
(O3) There exists t < 0 such that
> 0,
1 [L(, ), ] > 0
t > 0 and 1 [L(t , t ), ] = 0.

By Theorem 9.9, the option (O1) is excluded. According to Proposition 9.7,
the option (O3) cannot occur either. Therefore, (O2) occurs. Moreover, since
d0 is the unique value of for which bifurcation to positive solutions from
u = 0 occurs, necessarily
b = d0
and
lim k kC()
= 0.
d0
d0 0,
0 u0 .
236
To complete the proof of the theorem it remains to show that (, ) is the

unique linearly stable positive solution of (9.3) for every (d0 , 0 ] and
that is the minimal positive solution of (9.3).
Consider the arc of curve
C+
s := {(, ) : d0 0 }.
As the solutions of (9.3) can be viewed as fixed points of a compact operator,
C+
s is a compact arc of continuous curve in R C0 (). Thus, combining the
local uniqueness given by the main theorem of M. G. Crandall and P. H.
Rabinowitz [59] with the uniqueness results established by Propositions 9.6
and 9.7, it becomes apparent that there exists > 0 such that the unique
+
linearly stable positive solutions of (9.3) in C+

s B (d0 , 0) are those on Cs ;
B (d0 , 0) stands for the ball of radius centered at (d0 , 0).
To prove the uniqueness of the linearly stable positive solution we argue by
u
contradiction assuming that there is a positive solution, (,
), of (9.3) such
that
(d0 , 0 ],
u
1 [L(,
), ] 0,
u
=
6 .
Necessarily,
u
6 C+
s B (d0 , 0).
(9.31)
Thanks again to Propositions 9.6 and 9.7, there exist > 0 and a C 2 -map
, ]

(
7
C0 ()
such that (, ) is a positive solution of (9.3) with

1 [L(, ), ] > 0
, ).
Moreover, = u
for all (
. By global continuation of the curve
(, ) towards the left of , it is clear that some of the following alternatives

occur:
(A1) > 0 and 1 [L(, ), ] > 0 for all < .

such that
(A2) There is a b <
> 0
and 1 [L(, ), ] > 0
for all
(b , ],
but b = 0.
such that
(A3) There is t <
> 0,
1 [L(, ), ] > 0
t > 0 and 1 [L(t , t ), ] = 0.
for all
(t , ],
237
Arguing as above, it becomes apparent that (A2) holds. As d0 is the unique

value of for which bifurcation to positive solutions from u = 0 occurs,
b = d0 ,
lim k kC()
= 0.
d0
(9.32)
By (9.31),
= u
6 C+
s B (d0 , 0).
Consequently, since the unique linearly stable positive solutions of (9.3) in
+
C+
s B (d0 , 0) are those of Cs , the curve (, ), d0 , remains
outside C+
B
(d
,
0),
which
contradicts
(9.32)
and
concludes
the
proof of
0
s
the uniqueness.
Finally, by Proposition 9.14, (9.3) possesses a minimal positive solution
Moreover, it is linearly stable. Therefore, thanks to the
for each (d0 , ].
uniqueness of the linearly stable positive solution, must be the minimal
positive solution of (9.3) for all (d0 , 0 ]. This ends the proof.
Theorem 9.15 Suppose D > 0 and (9.3) admits a positive solution, (0 , u0 ),
such that
0 > d0 ,
1 [L(0 , u0 ), ] = 0.
(9.33)
Then, (9.3) does not admit a positive solution if > 0 .
Proof: On the contrary, suppose that (9.3) possesses a positive solution,
u
> 0 . By Proposition 9.14, (9.3) has a minimal positive so(,
), with
. Moreover,
lution at = ,
), ] 0.
1 [L(,
Thus, thanks to Theorem 9.12, (9.3) has a unique linearly stable positive
solution for = 0 , 0 . Moreover, 1 [L(0 , 0 ), ] > 0. Therefore, thanks to
(9.33), u0 =
6 0 , and hence, (9.3) possesses two linearly stable positive solutions
at = 0 , which is impossible. This ends the proof.
9.5
Curve of stable positive solutions
Subsequently, we will denote by J the set of values of the parameter for

which (9.3) possesses a linearly stable positive solution. The following result
provides us with the structure of J and the global behavior of the associated
curve of linearly stable positive solutions. By Theorem 9.9, J (d0 , ).
Also, J = if D 0. Moreover, thanks to Theorem 9.12, J must be an
interval. Consequently, the next result holds.
238
Theorem 9.16 Suppose a satisfies (HA) and D > 0. Then, some of the
following options occur:
(a) There exists (d0 , (a)] such that J = (d0 , ).
(b) There exists (d0 , (a)) such that J = (d0 , ].
Moreover, in both cases the set of linearly stable positive solutions consists of
a C 2 -curve
J 7 R C0 ()
7
(, )
such that 7 (x) is increasing for all x . Moreover,
lim k kC()
=0
d0
and
1 [L(, ), ] > 0
for all
int J .
Furthermore:
In case (a)
lim k kC()
=
(9.34)
and (9.3) cannot admit a positive solution for . Therefore, C+

consists of an increasing smooth curve filled in by linearly asymptotically
stable positive solutions blowing up to infinity as .
In case (b),
1 [L( , ), ] = 0
and hence, owing to Proposition 9.7, ( , ) is a quadratic subcritical
turning point of C+ . In particular, there exists > 0 such that for every
( , ) the problem (9.3) possesses at least two non-degenerate
positive solutions: one of them linearly asymptotically stable and another
linearly unstable. Moreover, according to Theorem 9.15, (9.3) cannot
admit a positive solution if > .
Remark 9.17 Thanks to Theorem 7.1 of H. Amann and J. Lopez-Gomez
[13], the case (b) occurs if any set of positive solutions of (9.3), S RC0 (),
where we are denoting by P the
with P (S) bounded, is bounded in RC0 (),
-projection operator. According to Theorem 4.3 of H. Amann and J. LopezG
omez [13], these a priori bounds are available under the next hypothesis.
239
Hypothesis (HB)
Either N {1, 2} or N 3 and
a+ (x) = + (x) [dist (x, + )]
for all x + near + ,
for some continuous function + positive and bounded away from zero in
a neighborhood of + and some positive constant > 0 such that

N +1+ N +2
p < min
,
.
N 1
N 2
Actually, in the presence of a priori bounds, (9.3) possesses at least two positive
solutions for each
(d0 , ) = J \ { },
as illustrated by Figures 9.2 and 9.3. This follows easily by using the fixed
point index in cones, as in Section 10.7. Indeed, (, 0) has local index zero,
while (, ) has local index one by the Schauder formula because
1 [L(, ), ] > 0.
As the global index is zero, because (9.3) cannot admit a positive solution for
> and we are imposing the existence of a priori bounds, (9.3) should
admit a further positive solution, with index 1, if it is non-degenerate. The
interested reader can review Theorem 7.4 of H. Amann and J. Lopez-Gomez
[13] for further technical details.
Remark 9.18 According to Remark 9.17, in all the numerical experiments
carried out in Section 9.1, (9.3) cannot admit any positive solution for >
t = , where t is the first (subcritical) turning point of C+ .
Remark 9.19 According to J. Lopez-Gomez [148], in the special case when
p = 1, it is easily seen that the option (b) occurs provided D > 0 is sufficiently
small, independently of the spatial dimension, N , and the decay rates of a+
along + . Moreover,
lim = d0 .
D0
Proof of Theorem 9.16: Thanks to Proposition 9.4 and Theorems 9.9 and
9.12 there exists (a) such that either J = (d0 , ), or J = (d0 , ].
In case J = (d0 , ], the alternative (b) occurs and the result is an easy
consequence of Theorem 9.12.
Suppose J = (d0 , ). Then, (a) occurs and, according to Theorem 9.12,
in order to complete the proof it suffices to show that condition (9.34) holds.
On the contrary, assume that there is a constant M > 0 and a sequence
n J , n 1, such that
lim n =
n
240
and
kn kC()
M,
n 1.
(9.35)
By Theorem 9.11, the map 7 k kC()

is increasing. Thus, it follows from
(9.35) that
k kC()
for all J .
M
Therefore, the point-wise limit
u (x) := lim (x),
x ,
is well defined. A standard bootstrapping argument combined with the fact

that the solutions of (9.3) are fixed points of a compact operator shows that
( , u ) provides us with a positive solution of (9.3), which is impossible,
because we are assuming 6 J . This ends the proof.
9.6
Dynamics in the presence of a stable steady state
This section studies the dynamics of (9.1) for (d0 , (a)). For the validity of the remaining results of this chapter it suffices to impose p > 0. The
stronger condition p 1 was only needed in order to apply the local bifurcation theorem of M. G. Crandall and P. H. Rabinowitz [59]. Actually, adopting
the methodology of Chapters 6 and 7 of J. Lopez-Gomez [163], all the results
of the chapter are valid for p > 0, even the regularities of the curves of positive solutions, (, u), because the strong positivity of u actually entails their
analyticities outside (d0 , 0).
9.6.1
Global existence versus blow-up in finite time
The next result shows how in the case when (d0 , (a)) the dynamics of
(9.1) depend on the relative size of a+ with respect to .
Theorem 9.20 Suppose (d0 , (a)) and u0 is a positive strict subsolution of (9.4). Then:
(a) There exists > 0 such that (9.3) has a positive solution if (a+ )M .
Moreover,
lim ku(, t; , a, u0 ) [,a] kC()
(9.36)
= 0,
t
where [,a] stands for the minimal positive solution of (9.3).

+ ,
(b) Assume that there exists a smooth subdomain D + , with D
such that
p p p+1
(9.37)
A min a+ >
p+1
D
241
where
Z
:= 1 [d, D] ,
:= d
D
[,a ] d,
n
(9.38)
[,a ] stands for the unique positive solution of (9.4),

R 0 is the
principal eigenfunction of 1 [, D] normalized so that D = 1,
and n stands for the outward unit normal vector field to D. Then,
u(, t; , a, u0 ) blows up in L () at some time Tb = Tb (u0 ) > 0. In
other words,
Tmax (u0 ) = Tb (u0 ) < +.
Proof: Fix (d0 , (a)) and let u0 > 0 be a strict subsolution of (9.4).
Such subsolution exists because > d0 . Indeed, it suffices to take for
sufficiently small > 0, where stands for a positive eigenfunction associated
to 0 in . On the other hand, by Theorem 4.1, (9.4) admits a (unique)
positive solution, [,a ] , if, and only if, d0 < < . By Lemma 1.8,
u0 [,a ] .
(9.39)
Naturally, as u0 also is a subsolution of (9.3), t 7 u(, t; , a, u0 ) is increasing.

Moreover, for every t > 0,
u(, t; , a, u0 ) u(, t; , a , u0 )
in .
Hence, thanks to Theorem 5.2(b), we find that

lim u(, t; , a, u0 ) lim u(, t; , a , u0 ) = [,a ]
(9.40)
provided Tmax (u0 ) = +!

Now, we will prove Part (a). The first assertion is a direct consequence of
Theorem 9.5. So, suppose (9.3) admits a positive solution and let [,a] denote
its minimal positive solution. By Theorem 9.12, [,a] is the unique linearly
stable non-negative solution of (9.3). Moreover, by Theorem 1.7,
[,a ] [,a] ,
because [,a] provides a positive strict supersolution of the associated sublinear problem (9.4). Therefore, thanks to (9.39), we have that
u0 [,a] .
Thus, (9.36) is a direct consequence from the fact that [,a] is the minimal
positive solution of (9.3), which concludes the proof of Part (a).
Subsequently, we will prove Part (b). Since
] 1 [d, + ] < 1 [d, D],
< = 1 [d, \
242
it is apparent that
= 1 [d, D] > 0.
Moreover,
Z
=
D
[,a ] d > 0,
n
because
n < 0 on D. Consequently, the constant on the right-hand side of
(9.37) is well defined and positive.
Suppose (9.37). To show that u(, t; , a, u0 ) blows up in a finite time we
will argue by contradiction. So, suppose Tmax (u0 ) = + and set
Z
u(x, t; , a, u0 )(x) dx (0, ) for all t > 0.
I(t) :=
D
As I(t) is non-decreasing because u0 is a subsolution of (9.3), the limit

L := lim I(t) (0, ]
t
(9.41)
is well defined. On the other hand, setting

u := u(x, t; , a, u0 ),
multiplying by the nonlinear parabolic equation of (9.1), integrating in D
and applying the formula of integration by parts yields
Z
Z
Z
0
I (t) = d
u dx +
u dx +
aup+1 dx
D
D
D
Z
Z
d
u
d + ( 1 [d, D]) I(t) + A
up+1 dx,
n
D
D
where
0
:=
d
,
dt
A min a = min a+ ,
+ . On the other hand,

because D
Z
Z
1
1
u dx =
1 p+1 p+1 u dx
D
Z
1
1
1 p+1
Z
p+1
p+1
u
dx
Z
=
D
1
p+1
p+1
u dx
since
R
D
= 1. Hence,
Z
D
up+1 dx
Z
D
p+1
u dx
243
and consequently,
I 0 (t) d
Z
u
D
d I(t) + AI p+1 (t)

n
(9.42)
for all t > 0. Suppose L < . Then,

lim I 0 (t) = 0
and letting t in (9.42), it follows from (9.40) that

0 L + ALp+1 ,
(9.43)
where is the constant defined in (9.38). Setting

f (x) := Axp+1 x + ,
x > 0,
(9.43) can be equivalently written as

f (L) 0.
(9.44)
The function f satisfies

f (0) = > 0,
lim f (x) = and f 0 (x) = (p + 1)Axp x > 0.
Thus,

f (x) = 0
if and only if x = xL :=
(p + 1)A
p1
and, due to (9.44), necessarily

f (xL ) = A
(p + 1)A
p+1
p
(p + 1)A
p1
+ 0.
Equivalently,
A
p p p+1
p+1
which contradicts (9.37). Therefore, L = .

Subsequently, we will use the estimate
I 0 (t) I(t) + AI p+1 (t),
t > 0,
(9.45)
which is a direct consequence from (9.42). Since limt I(t) = , there exists
t0 > 0 such that
p1
I(t0 ) >
.
(9.46)
A
244
On the other hand, the change of variable

I(t) = e(tt0 ) J(t),
transforms (9.45), (9.46) into
(
J 0 (t) Aep(tt0 ) J p+1 (t),
1
p
J(t0 ) = I(t0 ) > A
.
t t0 > 0,
t t0 > 0,
(9.47)
Thus, integrating the differential inequality of (9.47) gives

i
A h p(tt0 )
1 p
e
1
t > t0
J (t) J p (t0 )
p
p
and hence,
J p (t) J p (t0 )
i
Ah
1 ep(tt0 )
t > t0 ,
(9.48)
since p > 0. On the other hand, since

J(t) = e(tt0 ) I(t)
for all t > t0
and
lim I(t) = ,
it is apparent that limt J(t) = and therefore, letting t in (9.48)

yields
A
0 J p (t0 ) .
Equivalently,
p1
,
I(t0 ) = J(t0 )
A
which contradicts (9.46). This contradiction comes from the assumption that
u is globally defined in time. Consequently, the solution blows up in a finite
time, Tb (u0 ) = Tmax (u0 ).
Note that the mapping
7 1 [d, D]
is decreasing, while
Z
7 d
D
[,a ] d
n
is increasing, because 7 [,a ] |D is point-wise increasing. Therefore,

p p p+1
7
p+1
is decreasing. This is a rather natural feature establishing that the bigger is
the smaller can be taken minD a+ for Tmax (u0 ) < +.
9.6.2
245
Complete blow-up in +
In this section we assume that u(, t; , a, u0 ) blows up in a finite time, Tb =

Tmax (u0 ) < +, for example, in the context of Theorem 9.20(b), to give some
sufficient conditions so that it cannot admit a weak continuation for t > Tb ,
i.e., so that life after death cannot occur. To simplify the exposition as much as
possible, through the next discussion we will assume that the nodal behavior
of a(x) obeys the general patterns of Figure 9.1 with
1 := 1 [, 0 ] < 2 := 1 [, + ],
as in the numerical example of Section 5.3. Then,
] = min{1 [d, 0 ], 1 [d, + ]} = d1 .
= 1 [d, \
We also introduce the approximating functions
(
u + a(x) min{up+1 , k} if x + ,
fk (x, u) :=
u + a(x)up+1
if x 0 ,
as well as the associated approximating problems
x , t > 0,
ut du = fk (x, u),
u(x, t) = 0,
x , t > 0,
u(x, 0) = u0 (x),
x ,
k N,
k N.
(9.49)
For each k N, let

uk := uk (x, t; , a, u0 )
denote the solution of (9.49). As fk is bounded in + , uk is globally defined
in time. Moreover, since fk fk+1 , it satisfies uk uk+1 . Thus, the limit
u
(x, t) := lim uk (x, t; , a, u0 ) (0, ]
k
(9.50)
is well defined for all x and t > 0. Moreover, by construction,

u
(x, t) = u(x, t; , a, u0 )
for all
and t < Tb = Tmax (u0 ),
though the problem of finding u

(x, t) for t Tb may be a hard task. Nevertheless, u
provides us with a weak extension of u for times t Tb in a rather
natural way.
Also, we denote by B(u0 ) the set of points x for which there is a
sequence (xk , tk ) (0, Tb ), k 1, such that
lim (xk , tk ) = (x, Tb )
and
lim u(xk , tk ; , a, u0 ) = ,
often referred to as the blow-up set of u(, t; , a, u0 ). Finally, we will set

Eu0 (t) := I(u(, t; , a, u0 )),
246
where
Z
I(w) :=
a(x)
|w|2 w2
|w|p+2
2
2
p+2

dx,
w H01 (),
is the associated energy functional. According to P. H. Rabinowitz [215, Pr.

B 10], it is well known that, for every w 0,
Z

0
I (w) =
dw (w + a(x)wp+1 )(x) dx for all H01 ().
Thus,
Z

d Eu0
du ut (u + a(x)up+1 )ut dx
(t) = I 0 (u)ut =
dt
Z
Z

p+1
=
du u a(x)u
ut dx =
u2t dx 0
and hence, Eu0 is decreasing in [0, Tb ).

The following result shows that, under the general assumptions of Theorem
9.20(b),
u
(x, t) = for all x + and t > Tb
if B(u0 ) + and p is sufficiently close to zero. In these circumstances, it is
said that u(x, t; , a, u0 ) completely blows up in + at time Tb . Consequently,
it does not admit any weak extension in + after time Tb .
Theorem 9.21 Suppose (d0 , d1 ), u0 > 0 is a strict subsolution of (9.4)
+ . Then, Tmax (u0 ) <
and (9.37) holds in some smooth domain D with D
+ and
+ .
u
= u Lmin
in \
+]
[,a ,\
+ . Moreover, if either
In particular, B(u0 )
(
+,
p+1<
(3n + 8)/(3n 4),
if n = 1,
if n > 1,
(9.51)
or a+ (x) satisfies Hypothesis (HB) in Remark 9.17, then

lim
tTmax (u0 )
Eu0 (t) = .
(9.52)
If, in addition, B(u0 ) + , then

u
(x, t) = +
for all x + , t > Tmax (u0 ),
i.e., u blows up completely in + .
(9.53)
247
Proof: Tb := Tmax (u0 ) < + follows from Theorem 9.20(b). Since <
d1 , by Theorem 4.7 the large solution Lmin
+ ] is well defined. Moreover,
[,a ,\
since u0 is a subsolution of (9.3), the restriction u|\ + provides us with a
subsolution of
du = u + a up+1
+ for all t (0, Tb ). Hence,
in \
u(, t; , a, u0 ) Lmin
+]
[,a ,\
in
+
\
for all t < Tb . Consequently, by the definition of u

,
u
(x, t) = lim uk (x, t) Lmin
+]
[,a ,\
k
for all t > 0.
+ . It should be noted that this entails

+.
u(x, t; , a, u0 ) = u
(x, t) to be defined for all t > 0 if x \
Now, suppose (9.51) or Hypothesis (HB). Then, (9.52) is a consequence
from the results of J. L
opez-Gomez and P. Quittner [177, Section 5]. Therefore,
in case B(u0 ) + , (9.53) holds by applying [177, Th. 1.1] in
:= {x : dist(x, + ) < }
for a sufficiently small > 0. This ends the proof.
It should be noted that the point-wise limit

+
M[,a,] (x) :=
limt u
(x, t; u0 )
if x + ,
if x \ + ,
is well defined, and it provides us with the minimal positive solution of
+,
in \
dw = w + a wp+1
w = limt u
on + ,
w=0
on .
Indeed, as the initial data u0 which are subsolutions of (9.3) give rise to nondecreasing solutions of (9.1) and (9.49) because u0 also is a subsolution of
(9.49) for k > ku0 kC()
also is non-decreasing in t
, the extended solution u
and therefore, limt u
is well defined in .
In the radially symmetric case, Theorem 9.21 can be sharpened to obtain
the next result.
Theorem 9.22 Suppose 0 < R1 < R2 < R,
= BR := {x Rn : |x| < R},
R , 0 = BR \ B
R ,
+ = BR1 , = BR2 \ B
1
2
d0 := 1 [d, BR ] < < d1 := 1 [d, 0 ] < d2 := 1 [d, + ],
248
the weight function a C 1 [0, R] is radially symmetric, a(x) = a(|x|), and it

+ , and u0 (x) =
satisfies (9.37) for some smooth subdomain D with D
1
u0 (|x|), u0 C [0, R], is a positive strict subsolution of (9.4) in BR . Then,
the solution u(x, t) := u(x, t; , a, u0 ) of the problem
x BR , t > 0,
ut du = u + a(|x|)up+1 ,
u(x, t) = 0,
x BR , t > 0,
u(x, 0) = u0 (|x|),
x BR ,
blows up in a finite time Tb := Tb (u0 ) < in L (). Suppose, in addition,
the following:
(H1) 0 0 and 1 [d, B ] < < d1 .
0
(H3) u00 (r) 0 for all r [0, R].

Then, there exists Tc Tb such that
u
(x, t) = +
+ (Tc , ),
for all (x, t)
(9.54)
while
u
(, t) Lmin
+]
[,a ,\
+
in \
(9.55)
is a classical solution for each t > 0. If, in addition,

lim u(R1 , t) = +,
tTb
then,
lim u
(, t) = Lmin
+]
[,a ,\
+,
in \
+ and therefore,
uniformly in compact sets of \

+
lim u
(, t) = Mmin
:=
[,a ,]
Lmin
t
+]
[,a ,\
+,
in
+.
in \
Consequently, in such case, the asymptotic behavior of u(x, t; , a, u0 ) is governed by the metasolution Mmin
[,a ,] .
To construct examples satisfying the requirements of this theorem, we can
proceed as follows. First, fix p > 0 satisfying p < 4/(N 2) if N 3 and
choose R > 0. Then, pick R2 < R sufficiently close to R so that
R ] = 1 [, 0 ].
1 [, BR2 ] < 1 [, BR \ B
2
(9.56)
249
This is possible because we already know that

lim 1 [, BR2 ] = [, BR ]
R2 R
and
R ] = .
lim 1 [, BR \ B
2
R2 R
After reaching (9.56), pick R1 (0, R2 ) sufficiently small so that

1 := 1 [, 0 ] < 2 := 1 [, BR1 ] = [, + ],
which is possible because
lim 1 [, BR1 ] = .
R1 0
Then, pick a satisfying

R ]
d0 = 1 [d, BR2 ] < < d1 = 1 [d, BR \ B
2
and (R1 , R2 ) sufficiently close to R2 so that
R ].
1 [d, B ] < < 1 [d, BR \ B
2
Finally, one should choose appropriate u0 (r) and a(r), satisfying the remaining
requirements of the statement.
Proof of Theorem 9.22: By Theorem 9.21, u blows up in L () in a finite
time Tb and (9.55) holds.
Subsequently, we assume (H1)(H3). In particular,
Z
a(0) > 0 > a(),
a(r)rn1 dr > 0,
0
because 0 + , where a > 0, and any x with |x| = lies in , where a < 0.
Pick t0 (0, Tb ) and consider the auxiliary problem
x B , t > t0 ,
vt dv = v + a(|x|)v p+1 ,
v(x, t) = 0,
x B , t > t0 ,
(9.57)
v(x, t0 ) = ,
x B ,
where > 0 stands for a principal eigenfunction of 1 [, B ] and > 0 is
sufficiently small so that can be a strict subsolution of
dv = v + a v p+1
in B .
It may be accomplished because > 1 [d, B ]. Shortening , if necessary,

one can also get
.
< u(, t0 ; , a, u0 )
in B
(9.58)
Let v(x, t; ) denote the solution of (9.57) and consider v(x, t; ), its associated extended function through the approximating process (9.49), (9.50). By
(9.58), the parabolic maximum principle implies that
v(x, t; ) u
(x, t + t0 ; u0 )
for all
x B , t > 0.
(9.59)
250
Moreover, thanks to Theorem 9.20, v(x, t; ) blows up in L (B ) in a finite

time Tb Tb t0 . Thus, owing to [177, Th. 1.3], we find that
R (Tb , ).
for all (x, t) B
1
v(x, t; v) = +
Therefore, setting
Tc := Tb + t0 Tb ,
we find from (9.59) that
R (Tc , ),
for all (x, t) B
1
u
(x, t; u0 ) = +
which concludes the proof of (9.54).

Now, set u := u(r, t) for each r [0, R] and t [0, Tb ), and suppose
lim u(R1 , tk ) =
(9.60)
for some sequence tk Tb as k . Then, the auxiliary function

w(r, t) := rN 1 ur (r, t),
(r, t) [0, R] [0, Tb ),
satisfies
w(0, t) = 0,
w(R, t) < 0,
w(r, 0) 0,
and, differentiating with respect to r the differential equation, multiplying by

rN 1 and rearranging terms yields

N 1
wt = d wrr
wr + w + a(p + 1)up w + ar rN 1 up+1
r

N 1
wr + ( + a(p + 1)up ) w
d wrr
in (0, ) (0, Tb ),
r
because ar 0 in [0, ]. Thus, w 0 and hence, ur 0 in (0, ) (0, Tb ).
Therefore, it follows from (9.60) that
lim u(r, t) = +
tTb
uniformly in [0, R1 ],
since t 7 u(r, t) is increasing, because u0 is a subsolution of (9.4). The remaining assertions of the theorem follow straight from the fact that
L(x) := lim u
(x, t; u0 ),
t
provides us with a solution of
du = u + a up+1
u=
u=0
+,
x\
+,
in \
on + ,
on .

This ends the proof.
251
In the contrary case when there exists a constant C > 0 such that
u(R1 , t; , a, u0 ) C
for all t > 0,
(9.61)
the limit
g(R1 ) := lim u(R1 , t; , a, u0 )
t
is well defined and hence,

L := lim u(, t; , a, u0 )
t
provides us with a positive strong solution of
dw = w + a wp+1
w=g
w=0
in
+
\
+,
in \
on + ,
on .
Therefore, it is of great interest to ascertain whether (9.60) or (9.61) holds.

In Figure 9.7 we have represented the two possible limiting profiles of u
(x, t)
as t , according to either case (9.60), left, and (9.61), right. It should be
noted that u
= in + for sufficiently large t > 0.
As a consequence of Theorem 4.9 we already know that
min
lim Mmin
+ ] = M[d1 ,a , ] .
[,a ,\
d1
Consequently, the profile of Lmin

+ ] for close to d1 looks like Figure
[,a ,\
9.7. On the other hand, according to the proof of Theorem 9.22, ur 0 in
FIGURE 9.7: The asymptotic profiles of u

in cases (9.60) and (9.61).
252
(0, ) for all t < Tb , which might be incompatible with the graphs shown
in Figure 9.7. Therefore, it seems that the matasolution cannot be reached
in a finite time. Of course, such a possibility cannot be excluded when is
separated from d1 . By comparing the gradients from both sides, the interior
+ , it should be possible to prove that the
of + and the interior of \
situation illustrated by the right picture of Figure 9.7 cannot occur.
9.7
Dynamics for [d1 , d2 ) and a+ small
Throughout the rest of this chapter, to simplify the notations as much as

possible, we will assume that the nodal behavior of a(x) obeys the general
patterns of Figure 9.1 with
1 := 1 [, 0 ] < 2 := 1 [, + ].
Then, the next result holds.
Theorem 9.23 Suppose < d2 and the singular problem

0,
du = u + a up+1
in \
0 ),
u = +,
on ( \
(9.62)
possesses a unique positive solution, L. Then, there exists > 0 such that

0,
du = u + aup+1
in \
(9.63)
0 ),
u = +,
on ( \
possesses a minimal positive solution if (a+ )M .
Proof: Some sufficient conditions for the uniqueness of the large solution of
(9.62) had been already given in Part II. According to the Hardy inequality, it
follows from M. Bertsch and R. Rostamian [27] that the principal eigenvalue
of the linearization of (9.62) at L is well defined and satisfies
0 ] > 0,
1 [d (p + 1)a Lp , \
(9.64)
because
0 ] = 0.
1 [d a Lp , \
Actually, the existence of these eigenvalues for general singular potentials, as
well as some of their most pivotal properties, will be rigourously established
in Section 10.8, so we refrain from giving any further detail here.
The non-degeneracy condition (9.64) allows us to apply the implicit function theorem to construct a solution of (9.63) close to the unique solution of
253
(9.62). Once we establish the existence of a solution, the minimal one can be
constructed as in Chapter 4. This ends the proof.
The following result ascertains the dynamics of (9.1) when u0 is a subsolution of (9.4) and (a+ )M .
Theorem 9.24 Suppose [d1 , d2 ), u0 is a positive strict subsolution of
(9.4), and (9.63) possesses a minimal positive solution, Lmin
0 ] . Then, the
[,a,\
solution of (9.1), u(x, t; , a, u0 ), is globally defined in time, i.e., Tmax (u0 ) =
. Moreover,
lim u(x, t; , a, u0 ) = Lmin
0 ] (x)
[,a,\
0,
x\
for all
(9.65)
while
lim u(x, t; , a, u0 ) =
for all
0 \ .
x
(9.66)
Proof: The subsolution u0 exists because > d0 . Moreover, u0 is a subsolution of (9.3) because a a. Thus, t 7 u(, t; , a, u0 ) is point-wise increasing,
i.e., u(, t; , a, u0 ) is a subsolution of (9.3) for all t [0, Tmax ). Actually, as
u0 is a positive strict subsolution of (9.4) and Lmin
0 ] provides us with a
[,a,\
positive strict supersolution of (9.63), it follows from Lemma 1.8 that
u0 < [,a ,] < L[,a ,\ 0 ] < Lmin
0]
[,a,\
in
0.
\
Thus, by the the maximum principle,

u(x, t; , a, u0 ) Lmin
0 ] (x)
[,a,\
0 ) (0, Tmax ). (9.67)

for all (x, t) ( \
Subsequently, for sufficiently small > 0 with + we consider the

open set
+ := { x : dist (x, + ) < }.
Thanks to (9.67), there is a constant M > 0 such that
u(x, t; , a, u0 ) M
for all (x, t) + (0, Tmax ).
Let U denote the unique solution of

u
du = u + a up+1 u
t
u=0
u
=M
u(, 0) = u0
in
on
on
in
( \ + ) (0, )
(0, )
+ (0, )
\ + .
Note that U is globally defined in time. Moreover, by the parabolic maximum

principle,
u(x, t; , a, u0 ) U (x, t)
for all (x, t) ( \ + ) (0, Tmax ).
254
for all t (0, Tmax ) and

Therefore, u(, t; , a, u0 ) is bounded above in
consequently, Tmax = +.
On the other hand, for every t > 0,
u(, t; , a , u0 ) u(, t; , a, u0 )
in ,
and hence, (9.66) follows from Theorem 5.2(c). In particular,

0 ) (0, ).
lim u(x, t; , a, u0 ) = for all (x, t) ( \
Therefore, u(, t; , a, u0 ) must approach a positive solution of (9.63) bounded

min
above by Lmin
0 ] , (9.65) holds and the
0 ] . By the minimality of L[,a,\
[,a,\
proof is complete.
9.8
Dynamics for [d1 , d2 ) and a+ large
The next counterpart of Theorem 9.20(b) establishes that Tb = Tmax < +

if u0 > 0 is a strict subsolution of (9.4) and (a+ )M is sufficiently large.
Theorem 9.25 Suppose [d1 , d2 ), u0 > 0 is a strict subsolution of (9.4)
+ and
and there exists a smooth domain D such that D
p p p+1
A := min a+ >
(9.68)
p+1
D
where
Z
:= 1 [d, D] ,
:= d
D
min
L
d,
n [,a ,\0 ]
likewise in Theorem 9.20. Then, Tmax (u0 ) < +, i.e., u(, t; , a, u0 ) blows
up in a finite time Tb := Tmax in L (+ ).
Proof: Fix [d1 , d2 ) and let u0 > 0 be a strict subsolution of (9.4); it
exists because > d0 . Then, u0 provides us with a subsolution of (9.3) and
hence the mapping t 7 u(, t; , a, u0 ) is increasing. Moreover, for any t > 0,
u(, t; , a, u0 ) u(, t; , a , u0 )
in .
Thus, by Theorem 5.2(c),

lim u(, t; , a, u0 ) Mmin
0].
[,a ,\
In particular,
lim u(, t; , a, u0 ) Lmin
0]
[,a ,\
in D + .
(9.69)
255
Note that, since

< d2 = 1 [d, + ] < 1 [d, D],
> 0. Moreover, > 0, because
n 0 on D. So, the constant on the right
hand side of (9.68) is positive. The rest of the proof follows by adapting the
argument given in the proof of Theorem 9.20(b). Now, instead of (9.40), we
should use (9.69). The technical details are omitted here.
As for every [d0 , d1 ) and , % [d1 , d2 ) with < %, we have that
min
[,a ,] < Lmin
0 ] < L[%,a ,\
0]
[,a ,\
in D + ,
it becomes apparent that (9.37) implies (9.68).

As a consequence from Theorem 9.25, the next counterpart of Theorem
9.21 holds.
Theorem 9.26 Suppose [d1 , d2 ), u0 > 0 is a strict subsolution of
+ . Then,
(9.4), and (9.68) holds for some smooth domain D with D
u := u(x, t; , a, u0 ) blows up in L (+ ) in a finite time Tb = Tmax < +,

and the following assertions are true:
(a) u
= u Lmin
[,a , ] in is a classical solution for each t > 0, where
min
L[,a , ] stands for the minimal positive solution of

dw = w + a wp+1
in ,
(9.70)
w=
on .
(b) For sufficiently small > 0,
u(, t; , a , u0 ) u
(, t; u0 ) U (, t)
0,
in
where U stands for the unique solution of

u
du = u + a up+1
in (0, )
ut= 0
on (0, )
u = Lmin
[,a , ]
u(, 0) = u0
on ( ) (0, )
in
(9.71)
(9.72)
with
:= { x : dist (x, 0 ) < }.
Thus, u
(, t; u0 )| 0 is a classical solution for each t > 0 and
lim u
(, t; u0 ) =
0 \ . (9.73)
+ . If one further assumes that either (9.51) or a+ (x)

satisfies Hypothesis (HB) in Remark 9.17, then (9.52) holds. If, in addition,
B(u0 ) + , then (9.53) also holds and hence u(x, t; , a, u0 ) blows up completely in + .
256
Proof: The fact that u blows up in L (+ ) in a finite time Tb is guaranteed

by Theorem 9.25. Part (a) follows from the fact that u
(, t; u0 )| provides us
with a subsolution of the singular problem (9.70) for all t > 0. Now, choose
> 0 sufficiently small so that \ . Then, thanks to Part (a),
u
provides us with a subsolution of (9.72) and therefore the upper estimate
of (9.71) holds. The lower estimate follows from the fact that u[,a ,] is a
subsolution of (9.1). The property (9.73) follows from (9.71) and Theorem
5.2. The remaining assertions of the theorem are easy consequences from the
previous features by adapting the proof of Theorem 9.21.
Adapting Theorem 9.22 to the present situation, one can obtain some further sufficient conditions ensuring that
+
u
(, t; u0 ) = in
while

lim u
(, t; u0 ) =
for all
+
Lmin
[,a , ]
t > Tb ,
0 \ ,
in
in ,
though no additional details will be given here.
9.9
Dynamics for d2
For this range of values of no restriction on the size of (a+ )M is needed to

get the blow-up in + . Consequently, the following result holds.
Theorem 9.27 Suppose d2 and u0 > 0 is a strict subsolution of (9.4).
Then, u(x, t; , a, u0 ) blows up in L (+ ) in a finite time Tb and the following
assertions are true:
(a) u
(, t; u0 ) Lmin
[,a , ] in is a classical solution for all t > 0.
(b) For sufficiently small > 0,
u(, t; , a, u0 ) u
(, t; , a, u0 ) U (, t)
0,
in
where U (x, t) stands for the unique solution of (9.72). Thus,

u
(, t; u0 )| 0 is a classical solution for each t > 0 and (9.73) holds.
+ . If one further assumes that either (9.51) or a+ (x)
satisfies Hypothesis (HB) in Remark 9.17, then (9.52) holds. If, in addition,
B(u0 ) + , then (9.53) holds, i.e., u blows up completely in + .
9.10
257
This chapter has been elaborated from R. Gomez-Re

nasco and J. LopezG
omez [106], [107], [108], J. Lopez-Gomez [148], [157] and J. Lopez-Gomez
and P. Quittner [177].
Proposition 9.2 goes back to H. Berestycki, I. Capuzzo-Dolcetta and L.
Nirenberg [23], [24], as well as its proof through the Picone indentity, [209].
Theorem 9.5 goes back to J. Lopez-Gomez [148]. The numerical experiments
of Section 9.1 were carried out by R. Gomez-Re
nasco and J. Lopez-Gomez
[106]. Originally, they were intended for the Proceedings of the International
Conference on Operator Theory held in October 1998 in Winnipeg (Manitoba,
Canada), but [106] was rejected and never re-submitted. So, the results of the
numerical experiments with the example (9.14) have been published in this
book for the first time.
The conjecture on the number of solutions of (9.3) in the one-dimensional
setting goes back to R. G
omez-Re
nasco and J. Lopez-Gomez [106], [107], [108]
and it has been solved very recently in an extremely elegant paper by G.
Feltrin and F. Zanolin, [84]. Proposition 9.6 is folklore and Proposition 9.7
was inspired by H. Amann [11, Pr. 20.8].
The uniqueness of the stable positive solution established in Sections 9.3
and 9.4, as well as Theorem 9.16, goes back to R. Gomez-Re
nasco and J.
L
opez-G
omez [107], [108], and it was part of the Ph.D. thesis of R. GomezRe
nasco [105] under the supervision of the author. The nice global continuation argument used in the proof of Theorem 9.12 has been invoked in a number
of different contexts by other authors to obtain similar stability and multiplicity results. Among others, it was adapted by J. Lopez-Gomez, M. MolinaMeyer and A. Tellini [173] to establish the existence and the uniqueness of
the linearly stable positive solution in a general class of non-homogeneous superlinear indefinite boundary value problems, by J. Garca-Melian et al. [102]
to establish some stability and multiplicity results in an elliptic problem with
nonlinear absorption and a nonlinear incoming flux on the boundary, and by
J. Garca-Meli
an [99] to establish the stability of the minimal large solution
of (9.63) at = 0 as well as the existence of a further large solution.
Proposition 9.14 goes back, at least, to H. Amann [11, Th. 20.3], where it
was shown that (9.3) possesses a minimal positive solution if it has some, and
H. Amann [11, Pr. 20.4], where the linear stability of the minimal positive
solution was established. Our proof of Proposition 9.14 goes back to J. LopezG
omez, M. Molina-Meyer and A. Tellini [175].
The analysis of the dynamics of (9.1) in Sections 9.69.9 goes back to
J. L
opez-G
omez [157], [159] and J. Lopez-Gomez and P. Quittner [177]. The
concept of complete blow-up was coined by P. Baras and L. Cohen [20]. The
results of these sections show the crucial role played by the metasolutions
to describe the asymptotic behavior of the positive solutions not only in the
258
context of sublinear problems such as those studied in Parts I and II, but also
for superlinear indefinite problems such as those analyzed in this chapter.
Theorem 9.23 was the first available result establishing the existence of
large solutions in a class of semilinear elliptic equations of superlinear indefinite type. It goes back to J. Lopez-Gomez [157]. Seven years later it was
re-discovered by J. Garca-Melian [99]. Similarly, the multiplicity result of
[99] also is a direct consequence from the analysis carried out in this chapter.
In his masters thesis submitted to the University of Udine, A. Tellini [227]
introduced the one-dimensional prototype model

u00 = u + ab (x)up+1 ,
x (0, 1),
(9.74)
u(0) = u(1) = ,
for the choice

ab (x) :=
c if x [0, ) (1 , 1],
b
if x [, 1 ],
(9.75)
with (0, 0.5), b > 0 and c > 0, and found an extremely interesting
multiplicity result. Namely, using b as the main continuation parameter and
0 as the secondary one, A. Tellini [227] established that there is a value
of the parameter b, b , for which (9.74) has an arbitrarily large number of
positive solutions provided > 0 is sufficiently large. This multiplicity result
was substantially sharpened by J. Lopez-Gomez, A. Tellini and F. Zanolin in
[184]; by using some sophisticated phase portrait techniques all the possible
global bifurcation diagrams in b of (9.74) were ascertained according to the
different ranges of values of the secondary parameter . As a consequence
of the analysis carried out in [184], for any integer n 0 the problem (9.74)
possesses solutions with n strict critical points in the interval (, 1) if >
0 is sufficiently large. Moreover, these solutions are asymmetric if n = 2m 2,
in spite of the symmetry of the problem. The solution with 0 strict critical
points, referred to as the trivial solution, plays a pivotal role in understanding
the complexity of the global bifurcation diagrams of (9.74). Actually, it plays
a similar role as an organizing center in the context of singularity theory (see,
e.g., M. Golubitsky and D. G. Shaeffer [104]). It can be constructed as follows.
Let `(x) denote the unique solution of

u00 = u cup in (0, ),
u(0) = +, u0 () = 0,
and set m0 := `(). Then, the constant m0 solves u00 = u + bup+1 if, and
only if,
b = b := /mp0 .
For this value of b, the trivial solution ut
`(x),
m0 ,
ut (x) =
`(1 x),
is defined through
x [0, ),
x [, 1 ],
x (1 , 1].
259
Note that ut is symmetric around x = 0.5. Subsequently, given a solution u

of Problem (9.74) with n 0 strict critical points in (, 1 ), it is said that
u is of type (n, a) if u is asymmetric around x = 0.5, while it is said that it is
of type (n, s) if it is symmetric.
Basically, as illustrated by Figure 9.8, for 0 sufficiently close to 0,
the global structure of the bifurcation diagram of positive solutions of (9.74)
consists of a primary curve establishing a homotopy between the (unique)
solution of the sublinear problem

u00 = u + a0 (x)up+1 in (0, 1),
(9.76)
u(0) = u(1) = +,
and the metasolution
L(x),
,
M(x) :=
L(1 x),
x [0, ),
x [, 1 ],
x (1 , 1],
(9.77)
where L stands for the unique solution of the singular problem

u00 = u cup+1 in [0, ),
u(0) = u() = +.
(9.78)
The plots of Figure 9.8, as well as all subsequent plots of this section, were
computed by J. L
opez-G
omez, M. Molina-Meyer and A. Tellini [176] for the
special values of the parameters
= 1/3,
p = 1,
c = 1.
As in all the remaining bifurcation diagrams, in the left picture we are representing the value of b in abscissas versus the value of u() in ordinates.
The bifurcation diagram shows a single primary curve which emanates from
150
800
100
600
400
50
200
0.05
0.10
0.15
0.20
0.25
0.30
0.2
0.4
0.6
0.8
1.0
FIGURE 9.8: Global bifurcation diagram for = 5 and plots of some

solutions along it.
260
the unique positive solution of (9.76), whose existence and uniqueness can be
established as in Parts I and II, and can be path-followed toward the right to
reach the critical value b = 0.3401, where it turns backward exhibiting a subcritical turning point. Once switched to this turning point, the solutions on
the upper half-branch can be continued for all 0 < b < 0.3401. The numerical
experiments of [176] reveal that along the upper half-branch the solutions of
(9.74) blow up in [, 1 ] as b 0, while they approximate L in [0, ). Consequently, as b 0, these solutions approximate the metasolution M(x) defined
in (9.77). Therefore, the global bifurcation diagram establishes a homotopy
between the unique classical solution of the sublinear problem (9.76) and the
unique metasolution of
u00 = u + ab (x)up+1
supported in (0, )(1, 1). In particular, for every b (0, 0.3401), the problem (9.74) possesses at least two positive solutions. Moreover, the solutions
on the lower half-branch are linearly asymptotically stable, while those on the
upper half-branch are unstable with one-dimensional unstable manifolds.
Essentially, as > 0 increases, a piece of the primary curve rotates
counterclockwise around the trivial solution ut and, almost after every half
rotation, a closed loop of positive solutions bifurcates from it. The loop consists
of solutions of asymmetric type and it persists for all further values of .
The left plot of Figure 9.9 shows the global bifurcation diagram of the positive
solutions of (9.74) for = 300. It consists of two curves: the continuous line,
which is the primary branch, and the dashed line, which is the first solution
loop bifurcating from the primary curve at b = 12.8294 and b = 526.4099.
The subcritical turning point of the primary curve occurs at b = 527.4319.
The central plot of Figure 9.9 shows a magnification of this turning point
capturing the subcritical bifurcation point of the closed loop. All solutions on
the primary curve between the two squares marked in the left plot are of type
0.80
1.6
0.78
1.4
0.76
1.2
1.0
0.74
0.8
0.72
0.6
0
100
200
300
400
500
0.70
525.0
525.5
526.0
526.5
527.0
527.5
0.30
0.35
0.40
0.45
0.50
0.55
0.60
0.65
0.70
FIGURE 9.9: Global bifurcation diagram for = 300 (left picture), magnification of the turning point along the primary curve exhibiting the subcritical
bifurcation of the first closed loop (central picture), and plots of ut and two
solutions of type (3, s) (right picture).
261
(3, s), while the remaining ones are of type (1, s). The changes of type on the
primary branch occur at the level of the trivial solution, ut . The readers can
consult [176] for a detailed discussion about the types of the solutions along
each solution arc in the global bifurcation diagram. The lower half-branch
of the primary curve consists of linearly stable solutions, while the upper
one is filled by unstable solutions with one-dimensional unstable manifolds
if they are outside the loop and two-dimensional unstable manifolds if they
are encircled by the loop. All the solutions on the loop have one-dimensional
unstable manifolds. The solutions on the right picture of Figure 9.9 have
been plotted in the central interval (, 1 ), instead of in (0, 1), because
u(0) = u(1) = +.
As decreases from = 300 up to = 750, we get the global bifurcation diagram plotted at the top left in Figure 9.10. Now, the previous arc
of curve with the solutions of type (3, s) along the primary branch rotates
counterclockwise around ut generating two new turning points on it: one supercritical and another subcritical. As in the previous cases, in the first picture
of Figure 9.10 we are plotting with a dashed line the first loop bifurcated from
the solutions of type (1, s) on the primary curve, and the small squares along
it mark the values of b where the type of the solutions changed from (1, a)
0.12
0.12
0.10
0.10
0.08
0.08
0.06
0.06
0.04
0.04
0.02
5000
10 000
15 000
20 000
25 000
0.02
0.10
0.020
0.08
0.015
0.06
5000
10 000
15 000
20 000
25 000
0.010
0.04
0.005
0.02
0
5000
10 000
15 000
20 000
25 000
30 000
35 000
100 000
200 000
300 000
400 000
500 000
600 000
FIGURE 9.10: Global bifurcation diagram for = 750 (upper left), =

760.3 (upper right), = 800 (bottom left) and = 1300 (bottom right).
262
to (2, a), or vice versa. The maximal subcritical turning point on the primary branch arises at b = 2.6184588 104 and the two bifurcation points of
the first closed loop from the primary curve occur at b = 0.0011 104 and
b = 2.6184574 104 .
At the scale used to print the global bifurcation diagrams of Figure 9.10,
there was not enough room to plot the first closed loop of asymmetric solutions
emanated from the primary branch. Indeed, in the first plot, the values of u()
along the loop reached 2.15, which is substantially larger than 0.12. Changing
the scale on the vertical axis to plot the entire loop would not have allowed us
to appreciate the counterclockwise rotation of the branch of solutions of type
(3, s) on the primary curve, because the twist would have been compressed to
a 1/18 fraction of the vertical axis.
As decreased from = 750 and reached the value = 760.3, a second
closed loop emanates from the primary curve. As decreases, the loop grows,
approximating ut . The upper right picture of Figure 9.10 shows the global
bifurcation diagram of (9.74) for = 760.3. In Figure 9.10, the second
closed loops bifurcated from the primary curves have been represented with
a dot-dashed black line.
The second loop of the upper right diagram of Figure 9.10 bifurcates from
the primary branch at b = 1.3656 104 and b = 1.5716 104 . The trivial
solution arises at b = 1.5791 104 . The types of the solutions along the
second loop change at b = 1.4522 104 and b = 1.5695 104 . As the range of
values of b for which (9.74) admits a positive solution is very large, the last
value looks very close to b in the diagram. Hence, two of the small squares
marking the change of type along the second loop are almost superimposed
with the dot marking ut .
As decreases from = 760.3, the loop grows until it reaches ut and
encircles it for any smaller value of . The bottom left plot of Figure 9.10
shows the global bifurcation diagram of (9.74) for = 800. At this value,
the second loop bifurcates from the primary branch at b = 1.2110 104 and
b = 2.5941 104 and ut arises at b = 2.0530 104 .
Finally, in Figure 9.11 we have plotted the global bifurcation diagram
computed for = 2000, where one can clearly differentiate the third solution
loop bifurcating from the primary curve. It should be noted that, for every
b between the two bifurcation points of the third bifurcated loop from the
primary curve, the problem (9.74) possesses at least 12 solutions: 6 symmetric
and 6 asymmetric.
An extremely remarkable feature is that, in all the computed bifurcation
diagrams, the dimensions of the unstable manifolds of all solutions on the
primary curve increase by one at each bifurcation and turning point, until
reaching the interior of the last closed bifurcated loop, where they start to
decrease according to the same patterns to become one-dimensional again.
As far as the behavior of the n-th loop is concerned, it consists entirely of
solutions of type (2n 1, a) as occurs soon after it bifurcates from the primary
branch, or it consists of solutions of type (2n1, a) near the bifurcation points
263
from the primary curve together with two central arcs of solutions of type
(2n, a), for each n = 1, 2, 3, . . . . But interested readers are referred to [176] for
further technical or computational details.
Actually, according to J. Lopez-Gomez, A. Tellini and F. Zanolin [184],
there is a further sequence of values of ,
< n+1 < n < n1 < < 5 < 4 < 2000
0.0025
0.0020
0.0015
0.0010
0.0005
2.0
106
4.0
106
6.0
106
8.0
106
1.0
107
1.2
FIGURE 9.11: Global bifurcation diagram for = 2000.
107
264
such that whenever crosses k , a new closed loop of positive solutions bifurcates from the primary branch at some value of b close to b , and all these
closed loops persist for all smaller values of .
Naturally, to approximate the solutions of (9.74), J. Lopez-Gomez, M.
Molina-Meyer and A. Tellini computed the solutions of

u00 = u + ab (x)up+1 ,
x (0, 1),
(9.79)
u(0) = u(1) = M,
for sufficiently large M > 0, rather than the solutions of (9.74). Thanks to
the results of J. L
opez-G
omez, A. Tellini and F. Zanolin [184] the solutions of
(9.79) indeed approximate the solutions of (9.74) as M .
Rather astonishingly, when ab loses its symmetry about 0.5, all the previous global bifurcation diagrams spread out into an arbitrarily large number
of global components. Essentially, each closed loop generates an isola as illustrated by Figure 9.12. Most precisely, choosing
c0 if x [0, ),
b
if x [, 1 ],
ab (x) :=
(9.80)
c1 if x (1 , 1],
with c0 =
6 c1 , instead of (9.75), provokes a bifurcation diagram like the one
plotted on the left of Figure 9.9 to perturb into some of the bifurcation diagrams plotted in Figure 9.12 (A), if c0 > c1 , or in Figure 9.12 (B), if c0 < c1 ,
where the isolas provoked by the symmetry breaking have been plotted with
a dashed line, while the perturbed primary branches have been plotted by a
continuous line.
Actually, as a consequence of the work of J. Lopez-Gomez and A. Tellini
[183], there is a sequence of values of ,
n+1 <
n <
n1 < <
2 <
1 < 0
<
n the set of positive solutions of problem (9.74) with
such that for every <
(A)
10
8
20
40
60
(B)
10
80
100
120
20
40
60
80
100
120
FIGURE 9.12: A genuine symmetry breaking of a bifurcation diagram.
265
the choice (9.80) possesses at least n + 1 components: one unbounded, the

n n if c0 c1 = c.
primary one, and the remaining n bounded. Naturally,
Essentially, when the symmetry of the weight function ab (x) is broken,
by whatever mechanism, each of the closed loops generates an isola, like the
ones shown in Figure 9.12, changing the structure of the primary branch
accordingly. The plots of Figure 9.12 go back to J. Lopez-Gomez, M. MolinaMeyer and A. Tellini [174] where the interested reader is sent for any further
detail.
In general multi-dimensional problems, we should not expect to get a global
bifurcation diagram with the closed loops of asymmetric solutions bifurcating
from the primary curve, but rather a finite number of compact components
filled in by asymmetric solutions plus an additional (unbounded) primary
curve, as in Figure 9.12 unless the problem is radially symmetric. More precisely, we conjecture that in the absence of radial symmetry for any given
superlinear indefinite problem of the type

u = u + a(x)up+1 ,
x ,
(9.81)
u = M,
x ,
with sufficiently large 0 < M , there exists n , n 1, with n+1 < n < 0
such that for all (n+1 , n ) the solution set of positive solutions of (9.81)
consists of at least n + 1 components; among them, one unbounded and n
bounded.
9.10.1
Abiotic stress hypothesis
In population dynamics, when a(x) changes sign in and < 0 the parabolic
problem
u
in (0, ),
t du = u + a(x)|u|p u
(9.82)
u=M
on (0, ),
u(, 0) = u0 > 0
in ,
models the evolution of a single species in a harsh inhabiting region, , surrounded by territories where the population density equals M > 0. In these
models, u(x, t) stands for the density of the species, < 0 measures the
neat death rate of the species in , and u0 > 0 is the distribution of the
initial population. In nature, is negative when pesticides are used in high
concentrations or a certain patch of the natural environment is polluted by
introducing chemicals, waste products, or poisonous substances. Suppose
a = ab = a + ba+ .
Then, the parameter b 0 measures the intensity of the intra-specific facilitative effects of the species u in + , the interior of the support of a+ . In these
generalized logistic prototypes, the individuals of the species u compete for
266
the resources in , the interior of the support of a , but are facilitated by

the others in + .
Although there are extensive empirical studies on interspecific competition
(see, e.g., T. W. Shoener [222] and J. H. Connell [58]) and positive interactions
are well documented among organisms from different kingdoms as they can
make significant contributions to each others needs without sharing the same
resources (see, e.g., G. E. Hutchinson [119], J. L. Wulff [237], M. B. Saffo
[219]), finding positive interactions between similar organisms seems to be a
huge task in empirical studies, since they do not arise alone but in combination
with competition. However, according to the abiotic stress hypothesis of M. D.
Bertness and R. M. Callaway [26], the importance of positive interactions in
plant communities increases with abiotic stress or consumer pressure. Several
empirical studies support the validity of the abiotic stress hypothesis and,
actually, a substantial number of positive interactions in plant communities
have been isolated in harsh environmental conditions (see, e.g., R. M. Callaway
and L. R. Walker [33] and F. I. Pugnaire [211]). Consequently, (9.81) might
be a rather reasonable mathematical model for analyzing the combined effects
of facilitation and competition under abiotic stress.
According to the discussion in the previous section, under facilitative effects in competitive media, the harsher the environmental conditions measured
by the size of < 0, the richer the dynamics of the species measured by the
number of steady-state solutions of (9.81) in complete agreement with the
abiotic stress hypothesis of M. D. Bertness and R. M. Callaway [26].
Chapter 10
Spatially heterogeneous competitions
10.1
10.2
10.3
10.4
10.5
10.6
10.7
10.8
10.9
Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.1.1 Dynamics of the semi-trivial positive solutions . . . . . . . . .
10.1.2 Dynamics when d1 0 or d2 0 . . . . . . . . . . . . . . . . . .
Dynamics of the model when c = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A priori bounds for the coexistence states . . . . . . . . . . . . . . . . . . . . . . .
Global continua of coexistence states . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.4.1 Regarding as the main bifurcation parameter . . . . . . . .
10.4.2 Regarding as the main bifurcation parameter . . . . . . . .
Strong maximum principle for quasi-cooperative systems . . . . . . .
Multiplicity of coexistence states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dynamics of (1.1) when b = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Existence of meta-coexistence states . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Comments on Chapter 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
272
272
273
277
288
293
293
302
308
311
319
325
328
This chapter analyzes the dynamics of the parabolic problem
t u d1 u = u a(x)f1 (x, u)u b(x)uv
in (0, ),
t v d2 v = v d(x)f2 (x, v)v c(x)uv

u=v=0
on (0, ),
u(, 0) = u0 0 and v(, 0) = v0 0

in ,
(10.1)
where is a bounded domain of RN , N 1, with smooth boundary of
class C 2+ for some (0, 1], > 0, > 0, d1 > 0, d2 > 0, and a, b, c,
are non-negative functions satisfying the following assumptions:
d C ()
Hypothesis (HC)
The function a a satisfies Hypothesis (Ha), b(x) > 0 and d(x) > 0 for
and c(x) > 0 for all x 0 .
all x ,
As for the nonlinearities f1 and f2 we assume the following:
Hypothesis (HF)
[0, )), f1 satisfies Hypothesis (KO) and
For each j = 1, 2, fj C ,1+ (
f2 satisfies (Hf) and (Hg).
267
268
Under these assumptions, (10.1) provides us with a model for the evolution of
two competing species with densities u and v in the habitat . In the absence
of v, the species u grows according to the (non-classical) generalized logistic
parabolic problem
u
in (0, ),
t d1 u = u a(x)f1 (x, u)u
(10.2)
u=0
on (0, ),
u(, 0) = u0 0
in ,
whose dynamics were described in Chapter 5. Similarly, in the absence of u,
the species v grows according to the (classical) generalized logistic problem
v
in (0, ),
t d2 v = v d(x)f2 (x, v)v
(10.3)
v=0
on (0, ),
v(, 0) = v0 0
in .
By the abstract theory of Chapter 2, the dynamics of (10.3) are described by
its maximal non-negative steady state, because d(x) > 0 for all x .
Thanks to the abstract theory of D. Daners and P. Koch [68], for any
with u0 0 and v0 0, (10.1) has a unique global strong
u0 , v0 C0 ()
solution, (u(x, t; u0 , v0 ), v(x, t; u0 , v0 )), such that
[0, )) C 2+,1+ 2 (
(0, )).
u, v C(
(10.4)
Indeed, by the parabolic maximum principle, for every t > 0,

0 u(, t; u0 , v0 ) T1 (t)u0
and
0 v(, t; u0 , v0 ) T2 (t)v0 ,
where
T1 (t) = et(d1 +) ,
T2 (t) = et(d2 +) ,
t 0,
are the evolution operators associated with dj + , j {1, 2}, under homogeneous Dirichlet boundary conditions. In particular, the solutions of (10.1)
are globally defined in time, t > 0.
This chapter establishes that most of the limiting profiles of the positive solutions of (10.1) as t are given by a certain generalized class of
non-negative equilibria. The classical equilibria of (10.1) are the non-negative
solutions of
d1 u = u a(x)f1 (x, u)u b(x)uv

in ,
d2 v = v d(x)f2 (x, v)v c(x)uv
(10.5)
u=v=0
on .
The generalized class of equilibria consists of the non-negative metasolutions
of (10.5) supported in one or several components of 0 := int a1 (0). The next
definition describes the most appropriate concept of metasolution for (10.1).
269
Definition 10.1 Given 1 q q0 , a subset {i1 , ..., iq } {1, ..., q0 }, and

D := \
q
[
0,i ,
k=1
a couple
([0, ], [0, ))
(Mu , Mv ) :
is said to be a positive metasolution of (10.5) supported in D if there exists a
classical large solution (Lu , Lv ) of the singular elliptic problem
in D,

(10.6)
u=v=0
on D ,
u = and 0 v <
on D ,
such that
(Mu , Mv ) = (Lu , Lv )
in
and
Mu =
in
(D ).
( \ D)
It should be noted that, adopting the notation introduced in Chapter 1, we

are denoting
mj
[
0,j :=
i0,j ,
1 j q0 .
i=1
By a large solution of (10.6), we mean a solution of the system in D, (u, v),

such that
u C 2+ (D) C(D (D )),
v C 2+ (D) C(D),
and
lim u(x) = ,
d(x) := dist(x, D ).
d(x)0
Besides (0, 0), there are two types of classical non-negative solutions of
(10.5): those with one component vanishing and the other positive, (u, 0) or
(0, v), where u and v solve

d1 u = u a(x)f1 (x, u)u
in ,
(10.7)
u=0
on ,
and
d2 v = v d(x)f2 (x, v)v

v=0
in ,
on ,
(10.8)
respectively, usually referred to as the semi-trivial positive solutions, and those

with both components positive, called coexistence states, though those may
be linearly unstable.
270
Similarly, there are two types of non-negative metasolutions: those of the

form (u, 0), where u is a metasolution of (10.7), as discussed by Definition 5.1,
referred to as the semi-trivial positive metasolutions, and those of the form
(u, v) with v > 0, which will be called meta-coexistence states.
Problem (10.5) might also have large solutions of the form (u, v) with u
and v large, of course, but those solutions are of no interest for describing
the dynamics of (10.1), because they cannot be limiting profiles of a positive
solution of (10.1). Indeed, if (u, v) is a classical solution of (10.1), then v is a
subsolution of (10.3) and hence, by the parabolic maximum principle,
v(, t; u0 , v0 ) v (, t; v0 )
for all t 0,
where v stands for the unique solution of (10.3). Thus, by Theorem 2.2, there
exists a constant C > 0 such that
v(, t; u0 , v0 ) C
for all t 0.
Consequently, we have decided to exclude all these solutions from the statement of Definition 10.1, though they are mathematically admissible, because
they do not play a role in describing the dynamics of (10.1).
From the point of view of population dynamics, and represent the
intrinsic birth rates of the species u and v, d1 and d2 are their respective
diffusivity rates in the habitat , which is assumed to be surrounded by inhospitable regions, as a consequence of the homogeneous Dirichlet boundary
conditions, and u0 and v0 are the initial population densities of the species.
As a = a satisfies Hypothesis (Ha), in the absence of v, in the non-spatial
model obtained by switching off to zero the diffusivity d1 of u, the species
u grows according to the Malthus low in 0 , while it has logistic growth
in each x where a(x) > 0. According to Theorem 5.2, as an effect
attributable to the diffusivity d1 > 0, in the spatial model the species u can
exhibit a logistic growth in some of the smaller components of a1 (0) provided
is sufficiently small. Nevertheless, the components of 0 = int a1 (0) are
protection zones where u is free from intra-specific competition. Contrarily,
according to Theorem 2.2
as we are assuming that d(x) > 0 for all x ,
the species v exhibits a genuine logistic type of behavior, as it must afford the
intra-specific competition effects throughout the entire habitat .
The coefficient functions b(x) and c(x) measure the level of the aggressions
of the species v on u and the response of the species v, respectively. As we are
assuming that |b1 (0)| = 0, u receives aggressions from v almost everywhere
in . Moreover, since c(x) > 0 for all x 0 = int a1 (0), the species v is aggressed by u in all the spatial locations where the non-spatial model predicts
an exponential growth of u as time passes. Naturally, the main problem addressed in this chapter is ascertaining the output of the competition according
to the possible values of the set of parameters arising in (10.1).
Next we are going to introduce some of the notations used throughout this
chapter. The set of components of 0 , or protection patches of the species u,
271
will be denoted by Ra . According to Hypothesis (Ha),

Ra := i0,j : 1 i mj , 1 j q0 ,
(10.9)
where the components of 0 are labeled so that (1.3) holds, i.e.,

j := 1 [, i0,j ],
j < j+1 ,
1 i mj , 1 j q0 ,
1 j q0 1.
(10.10)
It should be remembered that, for every 1 j q0 ,

0,j :=
mj
[
i0,j ,
1 [, 0,j ] := j = 1 [, i0,j ], 1 i mj .
i=1
Throughout this chapter, for any given d > 0 and V L (), the components
of Ra will be re-labeled, if necessary, so that

Ra = i0,j (d, V ) : 1 i mj (d, V ), 1 j q0 (d, V )
with
j [d, V ] := 1 [d + V, i0,j (d, V )],
j [d, V ] < j+1 [d, V ],
1im
j , 1 j q0 ,
1 j q0 1,
(10.11)
where
m
j := mj (d, V ),
q0 := q0 (d, V ),
1 j q0 (d, V ).
So, according to (1.3),

mj = mj (1, 0),
q0 = q0 (1, 0),
1 j q0 .
By (10.11), we are imposing

1 [d + V, i0,j (d, V )] < 1 [d + V, i0,j+1 (d, V )]
(10.12)
for all 1 j q0 1 and 1 i m

j , and
1 [d + V, i0,j (d, V )] = 1 [d + V, i+1
0,j (d, V )]
(10.13)
for all 1 i m
j 1 and 1 j q0 . As the total number of components of
0 cannot vary with the pair (d, V ), it is apparent that
card Ra =
q0
X
q0 (d,V )
mj =
j=1
mj (d, V )
for all
d>0
and V L ().
j=1
As in Chapter 1, when we use these notations we shall denote

mj (d,V )
0,j (d, V ) :=
[
i=1
i0,j (d, V ),
1 j q0 (d, V ),
(10.14)
272

1 [d + V, 0,j (d, V )] := 1 [d + V, i0,j (d, V )]
(10.15)
for all 1 j q0 (d, V ) and 1 i mj (d, V ), and

0,1 (d, V )
0,j (d, V ) , 1 j q0 (d, V ). (10.16)
j (d, V ) := \
Note that, for every d > 0 and V L (),
q0 (d,V ) = .
Thanks to (10.11) it is apparent that
0 [d, V ] < 1 [d, V ] < 2 [d, V ]
< < q0 (d,V ) [d, V ] < q0 (d,V )+1 [d, V ] := ,
(10.17)
where, as in Chapter 1, we have denoted

0 [d, V ] := 1 [d + V, ].
So, 0 = 0 [1, 0].
10.1
Preliminaries
10.1.1
Dynamics of the semi-trivial positive solutions
The dynamics of u in the absence of v, v0 = 0, are regulated by the parabolic

problem (10.2) and described by the next result, which is a direct consequence
of Theorem 5.2. Indeed, v0 = 0 implies v(, t; u0 , v0 ) = 0 for all t 0 and hence
u(, t; u0 , v0 ) solves (10.2).
with u0 > 0 and let u(x, t) := u (x, t; u0 )
Theorem 10.2 Suppose u0 C0 ()
denote the unique solution of (10.2). Then, the following assertions are true:
(a) If d1 0 , then
lim u(, t) = 0
in C().
(b) If d1 0 < < d1 1 , then

lim u(, t) = ,u
in C(),
where ,u stands for the unique positive solution of (10.7).

(c) If d1 j < d1 j+1 for some 1 j q0 , then
min
max

t
in
(10.18)
,

min
273
max
where M[,j ] (resp. M[,j ] ) stands for the minimal (resp. maximal)
metasolution of
d1 u = u a(x)f1 (x, u)u
supported in j . If, in addition, u0 is a subsolution of (10.7) in , then
min
lim u(, t) = M[,j ]
in
(10.19)
It should not be forgotten that

q0 +1 := ,
q0 := .
Similarly, according to Theorem 2.2, the next result provides us with the
dynamics of the species v in the absence of u, i.e., when u0 = 0, as in this case
u(, t; u0 , v0 ) = 0 for all t 0 and v(, t; u0 , v0 ) solves (10.3).
with v0 > 0 and let v(x, t) := v (x, t; v0 )
Theorem 10.3 Suppose v0 C0 ()
denote the unique solution of (10.3). Then,
if d2 0 ,
0,
lim v(, t) =
in C(D),
t
,v ,
if > d2 0 ,
where ,v stands for the unique positive solution of (10.8).
The existence and the uniqueness of ,u and of ,v in these results are
guaranteed by Theorems 2.2 and 4.1, respectively. If necessary, we might adopt
the following convention
,u 0
10.1.2
if d1 0
and
,v 0
if d2 0 .
(10.20)
Dynamics when d1 0 or d2 0
The next result provides us with the dynamics of (10.1) when d1 0 , or

d2 0 . So, throughout the rest of this chapter we will assume
> d1 0 ,
> d2 0 .
(10.21)
with u0 > 0 and v0 > 0 and let

Theorem 10.4 Suppose u0 , v0 C0 ()
(u(x, t), v(x, t)) denote the unique (classical) solution of (10.1). Then, the
following assertions are true:
(a) If d1 0 and d2 0 , then
lim u(, t) = 0
and
lim v(, t) = 0
in C().
274
(b) if d1 0 and > d2 0 , then,

lim u(, t) = 0
and
lim v(, t) = ,v
in C(),
where ,v is the unique (classical) positive solution of (10.8).

(c) if d2 0 and d1 0 < < d1 1 , then
lim u(, t) = ,u
and
lim v(, t) = 0
in C(),
where ,u is the unique (classical) positive solution of (10.7).

(d) if d2 0 and d1 j < d1 j+1 for some 1 j q0 , then
min
max

t
in
and
lim v(, t) = 0
in C().
Proof: Since (10.1) is of Kolmogorov type, u0 > 0 and v0 > 0 imply u(x, t) > 0
and v(x, t) > 0 for all x and t > 0. Thus, by the general assumptions on
the coefficients, u(x, t) and v(x, t) are positive subsolutions of the parabolic
problems (10.2) and (10.3), respectively, whose solutions are going to be denoted by U (x, t) and V (x, t). Hence, by the parabolic maximum principle,
u(x, t) U (x, t)
and v(x, t) V (x, t)
(10.22)
for all x and t > 0. According to Theorem 10.2, we already know that
lim U (, t) = 0
in C()
if d1 0 . So, by (10.22), we find that, for every d1 0 ,

lim u(, t) = 0
in C().
(10.23)
Similarly, according to Theorem 10.3, for every d2 0 ,

lim v(, t) = 0
in C(),
(10.24)
which completes the proof of Part (a).

Suppose d1 0 and > d2 0 . We already know that (10.23) holds.
Moreover, by Theorem 10.3,
lim V (, t) = ,v
in C().
275
Thus, by (10.22), we find that

lim sup v(, t) ,v .
(10.25)
On the other hand, due to (10.23), for any given > 0, there exists T = t() >
0 such that 0 c(x)u(x, t) for all x and t T . Thus,
v
d2 v = v d(x)f2 (x, v)v c(x)uv ( )v d(x)f2 (x, v)v
t
for all x and t T . Hence, v(x, t) provides us with a supersolution of the
parabolic problem
w
in (T, ),
t d2 w = ( )w d(x)f2 (x, w)w
w=0
on (T, ), (10.26)
w(, T ) = v(, T ) > 0

in ,
whose unique solution will be denoted by w(x, t). Suppose > 0 has been
chosen sufficiently small so that > d2 0 . Then, by Theorem 10.3,
lim w(, t) = ,v
in C().
Moreover, by the parabolic maximum principle, it is apparent that

v(x, t) w(x, t)
for all x , t T.
Consequently,
lim inf v(, t) lim w(, t) = ,v .
t
Therefore, for sufficiently small > 0, we find that

,v lim inf v(, t) lim sup v(, t) = ,v .
t
as 0 and
According to Theorem 2.3, ,v approximates ,v in C()
these estimates complete the proof of Part (b).
The proof of Part (c) follows similar patterns as the proof of Part (b), except that in this case one should invoke to Theorem 4.1, rather than Theorem
2.3, to establish the continuity of the map 7 ,u . So, the technical details
of the proof are omitted here.
Lastly, suppose d2 0 and d1 j < d1 j+1 for some 1 j q0 .
Then, (10.24) holds. Moreover, by (10.22) it follows from Theorem 10.2 that
max
lim sup u(, t) lim sup U (, t) M[,j ] .

t
On the other hand, by (10.24), for every > 0 there exists T = t() > 0 such
that 0 b(x)v(x, t) for all x and t T . Hence,
u
d1 u = u a(x)f1 (x, u)u b(x)uv ( )u a(x)f1 (x, u)u
t
276
for all x and t T . Consequently, u(x, t) is a supersolution of the problem

z
in (T, ),
t d1 z = ( )z a(x)f1 (x, z)z
(10.27)
z=0
on (T, ),
z(, T ) = u(, T ) > 0

in .
Let z(x, t) denote the unique solution of (10.27) and suppose, in addition,
that d1 j < < d1 j+1 . Then, for sufficiently small > 0,
d1 j < < d1 j+1 .
Thus, by the parabolic maximum principle, we find from Theorem 10.2 that
min
lim inf u(, t) lim inf z(, t) M[,j ] ,

t
(10.28)
which entails
lim inf u(, t) =
in \ j
and
min
lim inf u(, t) lim inf z(, t) L[,j ]

t
in j .
As, according to Theorem 4.7, we already know that

min
min
L[,j ] [,j ,M ] ,
L[,j ] := lim [,j ,M ] ,

M
it becomes apparent that, for sufficiently small > 0 and every M > 0,
lim inf u(, t) [,j ,M ]
in j .
Therefore, letting 0 and then M yields

min
lim inf u(, t) L[,j ]
in j ,
which ends the proof of Part (d) in this case.

In the limiting case when = d1 j , one has that
d1 j1 < < d1 j
for small > 0 and hence, instead of (10.28), we find that
min
lim inf u(, t) lim inf z(, t) M[d1 j ,j1 ] ,

t
which implies
lim inf u(, t) =
t
and
min
in \ j1
lim inf u(, t) L[d1 j ,j1 ]

t
As, according to Theorem 4.9,

(
min
lim L[d1 j ,j1 ] =

0
the proof is complete.
min
L[d1 j ,j ]
in j1 .
0,j \ ,
in
0,j ,
in j = j1 \
(10.29)
10.2
277
Dynamics of the model when c = 0
Throughout this section we impose c = 0. Then, the evolution of the species v

is unaltered by u, which simplifies substantially the mathematical analysis of
(10.1) and illuminates the analysis of the general case when c > 0. When c = 0,
the species u and v get uncoupled and hence studying the system reduces
to the problem of analyzing a semilinear parabolic equation with a varying
potential. The following result characterizes the existence of coexistence and
meta-coexistence states.
Theorem 10.5 Suppose c = 0. Then, the following assertions are true:
(a) The elliptic problem (10.5) possesses a coexistence state if, and only if,
> d2 0 = 0 [d2 , 0],
0 [d1 , b,v ] < < 1 [d1 , b,v ],
(10.30)
where ,v is the unique positive solution of (10.8). Moreover, it is

unique and linearly asymptotically stable if it exists.
(b) For every 1 j q0 (d1 , b,v ), (10.5) possesses a meta-coexistence state
of the form (M, ,v ) supported in j (d1 , b,v ) if, and only if,
> d 2 0 ,
< j+1 [d1 , b,v ].
(10.31)
Proof: It should be noted that

q0 (d1 ,b,v )+1 [d1 , b,v ] = ,
by definition. Thus, the second estimate of (10.31) is not imposing any real
restriction on the size of if j = q0 (d1 , b,v ).
According to Theorem 2.2, (10.5) admits a solution (u, v) with v > 0 if
and only if > d2 0 , which has been already denoted by ,v . Thus, (10.5)
possesses a coexistence state if, and only if, the nonlinear elliptic problem

(d1 + b,v ) u = u a(x)f1 (x, u)u
in ,
(10.32)
u=0
on ,
possesses a positive solution. According to Theorem 4.1 and Remark 5.3,
(10.32) admits a positive solution if, and only if, the second estimate of (10.30)
holds, which ends the proof of the first assertion of Part (a).
Subsequently, we will denote by ,u [b,v ] the unique positive solution of
(10.32) under condition (10.30). If it exists, the coexistence state of (10.5) is
given by the solution couple
(u0 , v0 ) := (,u [b,v ], ,v ) .
278
By construction, it is unique. Moreover, the associated spectral problem of the

linearized system of (10.5) at the coexistence state (u0 , v0 ) can be expressed
in the form
L1 u = bu0 v + u,
in ,
L2 v = v,
(10.33)
u=v=0
on ,
where
f1
(x, u0 )u0 + af1 (x, u0 ) + bv0 ,
u
f2
L2 := d2 + d
(x, v0 )v0 + df2 (x, v0 ) .
v
L1 := d1 + a
According to (HC) and (HF), we have that

a
f1
(x, u0 )u0 0
u
and d
f2
(x, v0 )v0 0
v
in .
Thus, by the monotonicity properties of the principal eigenvalues,

1 [L1 , ] > 1 [d1 + af1 (x, u0 ) + bv0 , ],
1 [L2 , ] > 1 [d2 + df2 (x, v0 ) , ].
(10.34)
As v0 = ,v is the unique positive solution of (10.8), Lemma 1.6 implies that

= 1 [d2 + df2 (x, v0 ), ]
and hence it follows from (10.34) that 1 [L2 , ] > 0. Similarly, since u0 satisfies

(d1 + af1 (x, u0 ) + bv0 ) u0 = u0
in ,
u0 = 0
on ,
necessarily
= 1 [d1 + af1 (x, u0 ) + bv0 , ],
which implies 1 [L1 , ] > 0, by (10.34). Consequently, owing to Theorem 1.1,
L1 and L2 satisfy the strong maximum principle.
Let (u, v) =
6 (0, 0) be a solution of (10.33) for some C. Suppose v 6= 0.
Then, R and, actually, since the principal eigenvalue is dominant, as
discussed by Theorem 7.9 of [163], we find that
1 [L2 , ] > 0.
Similarly, if v = 0, then L1 u = u and hence
1 [L1 , ] > 0.
Therefore, all the eigenvalues of (10.33) are real and positive and, consequently, by the linearized stability principle of Lyapunov (e.g., Lunardi [186]),
279
the coexistence state (u0 , v0 ) is linearly exponentially asymptotically stable.

This ends the proof of Part (a).
Part (b) is an immediate consequence of Theorem 4.7 and Remark 5.3
applied to the problem
in j ,
(d1 + b,v ) u = u + a(x)f1 (x, u)u
u=0
on j ,
(10.35)
u=
on j \ ,
where
j = j (d1 , ,v ),
The proof is complete
1 j q0 (d1 , ,v ).
Subsequently, we will look at the first quadrant of the (, )-plane, centered

at (d1 0 , d2 0 ), to ascertain the shapes of the regions described by (10.30) and
(10.31). According to Theorem 10.4, outside this quadrant the dynamics of
(10.1) are governed by the semitrivial solutions and metasolutions. Set
0 () := 0 [d1 , b,v ] = 1 [d1 + b,v , ],
> d2 0 .
(10.36)
By Theorem 2.3, 7 ,v is increasing. Thus, by the monotonicity of the principal eigenvalue with respect to the potential, 7 0 () also is increasing.
Consequently, the limit
lim 0 () (0, ]
(10.37)
is well defined. The next result establishes that it equals infinity.

Lemma 10.6 The following holds
lim 0 () = d1 0 ,
d2 0
lim 0 () = .
(10.38)
Proof: The first limit holds because ,v bifurcates from v = 0 at = d2 0 ,

by the continuity of the principal eigenvalue with respect to the potential. The
proof of the second limit is more delicate and it relies on the fact that
lim ,v = uniformly on compact subsets of
(10.39)
and on Theorem 9.5 of [163]. Let denote by the principal eigenfunction of

in under Dirichlet boundary conditions, normalized so that
max = 1.
The property (10.39) follows from the fact that, for every > d2 0 , there is
a positive constant () such that () ,v in and
lim () = .
(10.40)
280
Indeed, provides us with a subsolution of (10.8) if, and only if,

d(x)f2 (x, (x)) d2 0
for all x ,
which can be accomplished by taking any = () > 0 such that

d2 0
max d
f2 (x, )
for all x ,
because 1. As the right hand side of this inequality grows to infinity as

, it becomes apparent that () can be chosen so that (10.40) holds,
as required. By Theorem 1.7, we have that () ,v for all > d2 0 .
Therefore, (10.39) holds, since minK > 0 in any compact subset K .
Next, for sufficiently small > 0, we will consider the open subset of ,
, defined by
:= {x : dist (x, ) > }.
As is smooth, we have that
| = 0.
lim | \
0
Consequently, as in the beginning of Chapter 1,

] = .
lim 1 [, \
(10.41)
On the other hand, by the monotonicity of the principal eigenvalue with respect to the potential, we find that
0 () = 1 [d2 + b,v , ] 1 [d2 + b,v \ , ],
for every > d2 0 and sufficiently small > 0, where \ stands for the
. Thus, by (10.39), we find from [163, Th. 9.5]
characteristic function of \
that
].
lim 0 () lim 1 [d2 + b,v \ , ] = d2 1 [, \
As this estimate holds for all 0 < < 0 , (10.41) implies (10.38).
But the behavior of the curves

j () := j [d1 , b,v ] = 1 [d1 + b,v , i0,j (d1 , b,v )],
> d2 0 ,
for each 1 j q0 (d1 , b,v ), is substantially more involved than the behavior of 0 (), because even the number of these curves, q0 (d1 , b,v ), might
vary with the parameter . However, by the monotonicity of the principal
eigenvalue with respect to the domain,
0 () < j (),
> d2 0 ,
1 j q0 (d1 , b,v ),
(10.42)

because i0,j
281
. Moreover, by construction,
d1 j = j [d1 , 0] = 1 [d1 , i0,j ],
1 j q0 ,
1 i mj ,
1 = 1 [d1 , 0] < < q0 = q0 [d1 , 0].
(10.43)
(10.44)
Thus, since ,v bifurcates from v = 0 at = d2 0 , by the continuity of the

principal eigenvalue with respect to the potential, it is apparent that
lim 1 [d1 + b,v , i0,j ] = 1 [d1 , i0,j ] = j [d1 , 0] = d1 j
d2 0
(10.45)
for all 1 j q0 and 1 i mj . Therefore, owing to (10.44) and (10.45),

there exists > 0 such that
1
2
1 [d1 + b,v , i0,j
] < 1 [d1 + b,v , i0,j+1
]
for all (d2 0 , d2 0 + ] and

1 i1 mj ,
1 i2 mj+1 ,
1 j q0 1.
Hence, any of these eigenvalues may be separated from each other. Consequently, the next result holds.
Proposition 10.7 Suppose mj = 1 for all 1 j q0 , i.e., q0 equals the
total number of components of 0 . Then, there exists 0 > d2 0 such that, for
every (d2 0 , 0 ) and 1 j q0 ,
mj (d1 , b,v ) = 1
10,j (d1 , b,v ) = 10,j (d1 , 0) = 10,j .
and
Moreover,
j () = j [d1 , b,v ] < j+1 () = j+1 [d1 , b,v ]
for all 1 j q0 1
and
lim j () = d1 j
d2 0
for all 1 j q0 .
However, even in the simplest situation covered by Proposition 10.7, the curves
j () and j+1 () might meet as the parameter separates away from d2 0 .
Indeed, the next result holds.
Proposition 10.8 Suppose q0 > 1, mj = 1 for all 1 j q0 ,
f2 (x, v) = v
for all
[0, ),
(x, v)
(10.46)
and there are j1 , j2 {1, ..., q0 } with j1 < j2 such that

min
0,j
b
b
< min .
d 10,j1 d
(10.47)
282
Then, there exists 0 > d2 0 such that, for every (d2 0 , 0 ),

j1 () = 1 [d1 +b,v , 10,j1 ] < j2 () = 1 [d1 +b,v , 10,j2 ], (10.48)
whereas
j1 (0 ) = j2 (0 ).
(10.49)
Moreover, there exists 1 > 0 such that

1 [d1 + b,v , 10,j1 ] > 1 [d1 + b,v , 10,j2 ]
for all > 1 . (10.50)
Suppose j1 < j2 . Then, by definition,

1 [d1 + b,v , 10,j1 (d1 , b,v )] < 1 [d1 + b,v , 10,j2 (d1 , b,v )]
for all > d2 0 (see (10.12) if necessary). If we assume, in addition, that
q0 = 2 and m1 = m2 = 1, then j1 = 1, j2 = 2, and, for every < 0 ,
10,1 (d1 , b,v ) = 10,1 (d1 , 0) = 10,1 ,
10,2 (d1 , b,v ) = 10,2 (d1 , 0) = 10,2 ,
while, for every > 1 , we have the inversion of the protection patches
10,1 (d1 , b,v ) = 10,2 ,
10,2 (d1 , b,v ) = 10,1 .
Which one is the most beneficial for the species u depends on the size of birth
rate of the competitor v.
Proof of Proposition 10.8: By Proposition 10.7, (10.48) holds for sufficiently close > d2 0 . Thus, for these s we have that
10,j1 = 10,j1 (d1 , b,v ),
10,j2 = 10,j2 (d1 , b,v ).
Subsequently, for every j {j1 , j2 }, we consider the eigenvalue curves

Sj () := 1 [d1 + b,v , 10,j ],
> d2 0 .
(10.51)
The proof of the proposition will proceed by contradiction. Suppose

Sj1 () Sj2 ()
for all > d2 0 .
(10.52)
By (10.46), it follows from M. Delgado, J. Lopez-Gomez and A. Suarez [72,

Th. 3.4] that
lim
,v
= d1
283
Thus, according to J. E. Furter and J. Lopez-Gomez [92, Le. 3.1], it becomes

apparent that
lim
Sj ()
d1
,v 1
b
= lim 1 [ + b
, 0,j ] = min
1 d
0,j
for each j {j1 , j2 }. Thus, dividing (10.52)

from (10.53) that
b
min min
1
1
0,j d
0,j2
1
(10.53)
by and letting , we find

b
,
d
which contradicts (10.47) and ends the proof.
When mj 2 for some 1 j q0 = q0 (d1 , 0), the behavior of the curves

j () := j [d1 , b,v ] = 1 [d1 + b,v , i0,j (d1 , b,v )],
> d2 0 ,
where 1 j q0 (d1 , b,v ), can be more intricate. Indeed, according to

(10.46), the map F constructed in the proof of Theorem 2.3 is analytic and
hence 7 ,v also is analytic. Thus, by the simplicity of the principal eigenvalues, adapting the proof of [163, Th. 9.1], it follows from Theorem 2.6 on
page 377 and Remark 2.9 on page 379 of T. Kato [122] that each of the functions
Sji () := 1 [d1 + b,v , i0,j ],
1 i mj ,
> d2 0 ,
for 1 j q0 , is analytic. As generically the zeroes of analytic functions

should be simple, when the parameter perturbs from d2 0 , each of the
eigenvalues
d1 j = j [d1 , 0] = 1 [d1 , i0,j ],
1 i mj ,
will generically perturb into mj simple eigenvalues, Sji (), 1 i mj . Consequently, for each > d2 0 we find that
card Ra = q0 (d1 , b,v )
equals the total number of components of 0 , i.e.,
q0 (d1 ,0)
q0 (d1 , b,v ) =
mj ,
j=1
or, equivalently,
mj (d1 , b,v ) = 1
for all 1 j q0 (d1 , b,v ).
This should occur for all > d2 0 , except at most at a discrete set of values of where at least two of these curves meet. The next result provides
284
us with the simplest example exhibiting such phenomenology. Subsequently,

given any smooth subdomain D , we will denote by D > 0 the principal
eigenfunction of 1 [; D] with
Z
2D = 1.
D
Also, we will set := .

Proposition 10.9 Suppose (10.46) and for every
j1 , j2 {1, ..., q0 },
i1 {1, ..., mj1 },
with (j1 , i1 ) =
6 (j2 , i2 ) the following holds
Z
Z
b 2i1 6=
i1
0,j
1
0,j1
i2 {1, ..., mi2 },
b 2i2 .
(10.54)
for all (d2 0 , 0 ),
(10.55)
i2
0,j
2
0,j2
Then, there exists 0 > d2 0 such that

q0 (d1 , b,v ) =
q0
X
mj
j=1
i.e., q0 (d1 , b,v ) equals the total number of components of 0 .

Conditions (10.54) can be easily reached by choosing an adequate b(x). The
proof of Proposition 10.9 relies on the next perturbation result.
Lemma 10.10 Suppose (10.46). Then, as d2 0 , we have that
1
( d2 0 ) + O(( d2 0 )2 )
3
d
,v = R
(10.56)
and
R
1 [d1 + b,v , D] =
d1 0D
b2D
( d2 0 ) + O(( d2 0 )2 ) (10.57)
d3
R
+ D
for all smooth subdomains D .

Proof: The asymptotic expansion (10.56) is folklore (see, e.g., Lemma 4.3
of M. Delgado, J. L
opez-Gomez and A. Suarez [72]). To prove (10.57), we
adapt the proof of [72, Le 4.3]). As we have already discussed before stating
Proposition 10.9, by a classical result in perturbation theory, the map
7 SD () := 1 [d1 + b,v , D]
285
is analytic. Thus, there exists a constant K R such that

SD () = d1 0D + K( d2 0 ) + O(( d2 0 )2 )
(10.58)
as d2 0 . Let () denote the unique principal eigenfunction associated to

SD () for which
Z
()D = 1.
(10.59)
D
According to T. Kato [122, Section VII.2], 7 () is analytic and hence, it

possesses a unique expansion of the form
() = 0 + ( d2 0 )1 + O(( d2 0 )2 )
as d2 0 .
(10.60)
Using (10.59) yields

Z
0 = D
and
1 D = 0.
(10.61)
Now, substituting (10.56), (10.58) and (10.60) in the identity

(d1 + b,v )() = SD ()()
using (10.61) and identifying terms of order one in d2 0 we find that
(d1 d1 0D )1 = KD R
b
D .
d3
Finally, multiplying this identity by D and integrating by parts in D yields

R
b2D
R
,
K= D
d3
Proof of Proposition 10.9: Thanks to Lemma 10.10, for every 1 j q0

and 1 i mj , we have that
Sji () := 1 [d1 + b,v , i0,j ]
i
= d1 0 0,j + Ki,j ( d2 0 ) + O(( d2 0 )2 ),

where
R
Ki,j :=
i0,j
b2i
0,j
d3
Owing to (10.54),
Ki1 ,j1 =
6 Ki2 ,j2
if
(i1 , j1 ) 6= (i2 , j2 ).
(10.62)
d2 0 ,
286
Therefore, according to (10.62), for every 1 j q0 the eigenvalue

mj
0 0,j = = 0 0,j
perturbs into mj (different) eigenvalues. Therefore, (10.55) holds. This ends
the proof.
In Figure 10.1 we have represented the curves arising in the statement of
Theorem 10.5 in the special case when
q0 = 2,
m1 = 3,
m2 = 2.
As suggested by Proposition 10.9, for each > d2 0 we have that

q0 (d1 , b,v ) = 5
and mj (d1 , b,v ) = 1
for all 1 j 5,
except, at most, at a discrete set of values of the parameter , for which

q0 (d1 , b,v ) {1, ..., 4}.
Figure 10.1 shows one of these exceptional values with q0 (d1 , b,v ) = 4. It is
the coordinate of the crossing point between the second and third curves
FIGURE 10.1: The curves arising in Theorem 10.5.
287
emanating from = d1 1 at = d2 0 . The existence of these three (different)

curves is guaranteed by Proposition 10.9. More generally, the exceptional values of are the coordinates of the crossing points between the eigenvalue
curves represented in Figure 10.1, which have been listed on the right of their
graphs. These curves divide the first quadrant of the (, ) plane into several
regions. For the special configuration shown in Figure 10.1 the first quadrant,
Q := {(, ) R2 : > d1 0 , > d2 0 },
is divided into the regions Rj , 0 j 6, defined by
R0
R1
Rj
R6
:= { (, ) Q
:= { (, ) Q
:= { (, ) Q
:= { (, ) Q
:
:
:
:
d1 0 < < 0 [d1 , b,v ] = 1 [d1 + b,v , ] } ,

0 [d1 , b,v ] < < 1 [d1 , b,v ]},
j1 [d1 , b,v ] < j [d1 , b,v ] },
2 j 5,
5 [d1 , b,v ]}.
Thanks to Theorem 10.5(a), (10.5) possesses a coexistence state if, and only if,
(, ) R1 . Moreover, by Theorem 10.5(b), for every 1 k 5, it possesses
a metacoexistence state of the form (M, ,v ) supported in k (d1 , b,v ) if,
and only if,
k+1
[
0
(, ) R
Rj .
j=1
Thanks to Theorems 4.1, 4.8, 4.9 and Remark 5.3, fixing > d2 0 and varying
as the main bifurcation parameter, the u component of the curve of coexistence states grows to infinity in 0,1 [d1 , b,v ] as 1 [d1 , b,v ]. Moreover,
for each 1 k 4, the minimal metasolution of the form (M, ,v ) supported in k [d1 , b,v ] grows towards the minimal metasolution supported in
k+1 [d1 , b,v ] as k+1 [d1 , b,v ]. Owing to Theorem 5.2 and Remark 5.3,
the next result holds. It provides us with the dynamics of (10.1) when c = 0.
u0 > 0 and
Theorem 10.11 Assume (HC), (HF), c = 0, u0 , v0 C0 (),
v0 > 0. Then,
(a) If > d2 0 and d1 0 0 [d1 , b,v ], the semi-trivial state (0, ,v )
is a global attractor for the (classical) positive solutions of (10.1).
(b) If > d2 0 and
0 [d1 , b,v ] < < 1 [d1 , b,v ],
then, the (unique) coexistence state of (10.5) attracts to all (classical)
positive solutions of (10.1).
(c) If > d2 0 and, for some k {1, ..., q0 (d1 , b,v ) 1},
k [d1 , b,v ] < k+1 [d1 , b,v ],
288

then,
min
max
M[,k (d1 ,b,v )] lim inf u(, t) lim sup u(, t) M[,k (d1 ,b,v )]
t
and
in ,
lim v(, t) = ,v
uniformly in ,
(10.63)
where (u, v) stands for the unique solution of (10.1). If, in addition, u0
is a subsolution of (10.32) in , then
min
lim u(, t) = M[,k (d1 ,b,v )]
in
(d) If > d2 0 and

q0 (d1 ,b,v ) [d1 , b,v ],
then,
min
max
M[, ] lim inf u(, t) lim sup u(, t) M[, ]

t
and (10.63) holds. Further, if u0 is a subsolution of (10.32) in ,

in ,
then
min
lim u(, t) = M[, ]

in .
t
Most of the mathematical discussion in this section does not make sense when
q0 = 1 and m1 = 1, i.e., when 0 is connected.
10.3
A priori bounds for the coexistence states
In this section we establish the existence of uniform a priori bounds for the
coexistence states of (10.5) when (, ) varies in compact subsets of
B, := [(d1 0 , ) \ {d1 j , 1 j q0 }] (d2 0 , ).
(10.64)
Throughout the rest of this chapter, it is said that (, , u, v) is a coexistence

state of (10.5) if (u, v) is a coexistence state of (10.5). Moreover, we use the
notation
kwk := kwkL () = sup |w| < +
for any function w L (), and denote

H01 := W01,2 ().
The main result of this section follows.
289
Theorem 10.12 Suppose (HC) and (HF) and let (n , n , un , vn ), n 1, be

a sequence of coexistence states of (10.5) with
max {|n |, |n |}
(10.65)
n1
for some positive constant . Then,

vn ,v
for all n 1,
where ,v stands for the unique positive solution of (10.8) for = .

If, in addition,
lim sup kun k = ,
n
there exist 1 j q0 , 1 i mj , and a subsequence of (n , n , un , vn ),

n 1, relabeled by n, such that
lim kun kL (i0,j ) = .
lim n = d1 j ,
Therefore, the coexistence states are uniformly bounded for (, ) varying in

compact subsets of B, .
The proof of this theorem relies on the next result of technical nature.
Lemma 10.13 Suppose V L () and let un H01 () L (), n 1, be
a sequence of non-negative functions such that
(d + V )un C un ,
kun k C,
n 1,
(10.66)
for some constants C > 0 and d > 0. Then there exist a subsequence of un ,
n 1, labeled again by n, and a function u H01 () L () such that
weakly in H01 () and strongly in Lp () for all p > 1.
lim un = u
Necessarily, u 6= 0 if there exists > 0 such that kun k for sufficiently

large n 1.
Proof: Multiplying the differential inequality in (10.66) by un and integrating
in yields
Z
Z
Z
|un |2 +
V u2n C
u2n ,
n 1.
Thus, by the second estimate in (10.66),

Z
Z
2
d
|un | (C inf V )
u2n (C inf V )C 2 ||,
n 1.
Therefore, the sequence un , n 1, is bounded in H01 () L () and hence,

there exist u H01 () L () and a subsequence of un , n 1, labeled
290
again by n, satisfying all the requirements of the statement. Indeed, since

H01 is a Hilbert space, it is reflexive. Hence, by Theorem III.16 of H. Brezis
[32], any bounded subset of H01 () must be relatively compact with respect to
the weak topology of H01 (). Thus, there exists u H01 () such that, along
some subsequence, relabeled by n, we have that un * u in H01 (). Moreover,
by a celebrated compactness result of F. Rellich and V. I. Kondrachov [163,
Th. 4.5], the imbedding H01 () , L2 () is compact. So, u L2 (), and un
can be chosen so that, in addition, limn un = u in L2 (). On the other
hand, owing to Theorem IV.9 of [32], the subsequence can be refined so that
limn un (x) = u(x) almost everywhere in . Therefore, letting n in
|un (x)| C, n 1, also yields kuk C. In particular, u L (). As for
every p > 2,
Z
Z
Z
|un u|p =
|un u|2 |un u|p2 (2C)p2
|un u|2 ,
the L2 convergence of u entails the Lp convergence for all p > 1, which ends
the proof of the assertion above.
Lastly, suppose kun k , n 1, for some constant > 0, and u = 0.
Then, since
lim kun kLp () = 0
for all p > 1
n
we have that
lim un = 0
almost everywhere in .
This is impossible, because there should exist a subset D with positive

measure such that
|un (x)|
2
for almost all x D, which ends the proof.
Proof of Theorem 10.12: Let (n , n , un , vn ), n 1, be a sequence of
coexistence states of (10.5) satisfying (10.65) for some > 0. Then, it follows
from the v equation of (10.5) that
d2 vn n vn df2 (, vn )vn vn df2 (, vn )vn ,
n 1.
Hence, for every n 1, the function vn is a positive subsolution of

d2 v = v df2 (, v)v
in ,
v=0
on .
Thus, thanks to Theorem 1.7, vn ,v for all n 1. So, also
d2 vn vn df2 (, vn )vn ,v
for all n 1. So, the sequence vn satisfies (10.66). Consequently, according to Lemma 10.13, there exists v H01 () L () such that, along some
subsequence of vn , relabeled by n,
lim vn = v
291
Subsequently, we will focus our attention on this subsequence. Note that vn >
0 implies v 0. As n and n , n 1, are bounded, without loss of generality
we can also assume that
lim (n , n ) = (, )
for some (, ). Suppose

lim sup kun k = .
n
Then, we can consider a subsequence, labeled again by n, for which

lim kun k = .
Setting
u
n :=
un
,
kun k
n 1,
it follows from the u-equation of (10.5) that

d1
un
un
for all n 1.
Hence, owing to Lemma 10.13, there exists u

H01 () L (), with u
> 0,
such that
lim u
n = u
un k = 1 for all n 1. Moreover, since

Note that u
=
6 0, because k
d1 un un af1 (, un )un ,
n 1,
and f1 satisfies (KO), it follows from Theorem 1.7 and Proposition 3.3, that
for any compact subset K there exists a constant CK > 0 such that
kun kL (K) CK
for all n 1.
Consequently, u
= 0 in any compact subset of . Thus, setting
:= 0 { x : dist(x, 0 ) < }
for sufficiently small > 0, we find that
\
u
H01 ( ).
>0
As 0 satisfies the segment property, it is stable, as discussed at the end of

the proof of [144, Th. 4.2] and in [163, Th. 8.4]. Therefore, u
H01 (0 ).
292
Let C0 (0 ) be a test function supported in 0 . Then, multiplying the

un -equation by , dividing by kun k , and integrating by parts in 0 yields
Z
Z
Z
d1
un = n
u
n
b
un vn ,
n 1,
0
because a = 0 in 0 . Hence, letting n yields

Z
Z
Z
d1
u =
u

b
uv.
0
Therefore, u
provides us with a weak solution of

d1
u = ( bv)
u
in 0 ,
u
=0
on 0 .
(10.67)
i ) for all 1 j q0 and

As bv L (), by elliptic regularity, u
C01 (
0,j
= 0 in , there exist 1 j0 q0
1 i mj . Moreover, since u
=
6 0 and u
and 1 i0 mj0 such that
0
in D0 := i0,j
.
0
u
>0
Therefore, by (10.67), u
is a principal eigenfunction of
= 1 [d1 + bv, D0 ].
(10.68)
0 ).
In particular, u
lies in the interior of the cone of positive functions of C01 (D
By construction,
lim kun kL (D0 ) = .
n
C0 (D0 )
be a test function in D0 . Multiplying the vn equation by

Let
and integrating by parts in D0 shows that
Z
Z
Z
Z
d2
vn = n
vn
df2 (, vn )vn kun k
cvn u
n ,
D0
D0
D0
D0
for all n 1. So, dividing each of these identities by kun k and letting n
yields
Z
cv
u = 0.
D0
Consequently, v = 0 in D0 , because v(x) 0, u

(x) > 0 and c(x) > 0 for all
x D0 . Consequently, (10.68) becomes
= 1 [d1 , D0 ] = d1 j0 ,
10.4
293
Global continua of coexistence states
This section applies the abstract theory of J. Lopez-Gomez [149] to get some
necessary and sufficient conditions for coexistence states of (10.5). In order to
state the results, we need to introduce some notations. For each j {1, 2},
we will denote by ej the unique solution of

dj ej = 1
in ,
ej = 0
on .
SubseThanks to Theorem 1.1, e1 and e2 are strongly positive in C01 ().
quently, for every j {1, 2}, we will denote by
Xj := Cej ()
for which there
the ordered Banach space consisting of all functions u C()
is a positive constant > 0 such that
ej u ej ,
equipped with the Minkowski norm
kukej := inf{ > 0 : ej u ej }
and ordered by its cone of positive functions, Pj ; the interior of the cone Pj
will be denoted by int Pj , j = 1, 2 (see Section 7.2 of [163], if necessary). As
the Minkowski norm is monotone, according to [163, Th. 6.1], the cones Pj are
normal (see Section 6.1 of [163] to get familiar with the most basic concepts
in ordered Banach spaces).
Using these notations, for any K 0, the solutions of (10.5) are the fixed
points of the compact operator
K, : X1 X2 X1 X2
defined by

K, (u, v) :=
(d1 + K)1 [( + K)u af1 (, u)u buv]

(d2 + K)1 [( + K)v df2 (, v)v cuv]

.
(10.69)
Therefore, the abstract theory of [142] and Chapter 7 of [163] can be applied
to establish the existence of a component of the set of coexistence states of
(10.5) emanating from each of the semi-trivial positive solutions.
10.4.1
Regarding as the main bifurcation parameter
In the next result, is fixed and is regarded as the main bifurcation parameter. Later, the roles of these parameters will be exchanged. Throughout
the rest of this book, a component is defined as a closed and connected subset
which is maximal for the inclusion; P stands for the projection of (, u, v).
294
Theorem 10.14 Suppose (HC) and (HF), and regard > d2 0 as the main
bifurcation parameter. Then, for every
d1 0 < < d1 1 ,
(10.70)
= 0 [d2 , c,u ] is the unique bifurcation value to coexistence states of (10.5)

from (,u , 0). Moreover, the component of the set of coexistence states emanating from (,u , 0) at
= 0 [d2 , c,u ] = 1 [d2 + c,u , ],
denoted by
C+
(,,u ,0) R int P1 int P2 ,
meets the other curve of semi-trivial solutions (, 0, ,v ) at the unique value
of , , for which
= 0 [d1 , b ,v ] = 1 [d1 + b ,v , ].
(10.71)
In particular, (10.5) possesses a coexistence state if

( 0 [d2 , c,u ])( 0 [d1 , b,v ]) > 0.
(10.72)
Now, instead of (10.70), assume that

d1 1 .
Then, (, u, v) = ( , 0, ,v ), where is the unique value of for which
(10.71) holds, is the unique bifurcation point to coexistence states of (10.5)
from the curve (, 0, ,v ). Moreover, the component of coexistence states emanating from (, 0, ,v ) at = , denoted by
C+
(,0,,v ) R int P1 int P2 ,
is unbounded in R X1 X2 . If, in addition,
(d1 1 , ) \ {d1 j : 2 j q0 },
then
(10.73)

1
( , ) P C+
(,0,,v ) ( , )
where 1 stands for the unique value of for which

= 1 [d1 , b1 ,v ] = 1 [d1 + b1 ,v , 0 ].
(10.74)
Actually, there exists (1 , ] such that (10.5) cannot admit a coexistence

state if < .
Theorem 10.14 will be inferred from the next proposition.
295
Proposition 10.15 Suppose (10.70) is satisfied and (10.5) possesses a coexistence state. Then,
> 0 [d1 , b{,v;c,u } ],
(10.75)
where ,u is the unique positive solution of (10.7) and
:= {,v;c,u }
stands for the unique positive solution of

(d2 + c,u ) = df2 (, )
=0
in ,
on .
(10.76)
In particular, (10.5) cannot admit a coexistence state for sufficiently large .

Moreover, if it admits a coexistence state with d1 1 , then
< 1 [d1 , b,v ].
(10.77)
Proof of Proposition 10.15: Suppose (10.70) holds and (10.5) admits a

coexistence state, (u, v). Then, u and v are positive subsolutions of (10.7) and
(10.8), respectively. Thus, according to Lemma 1.9, we find that > d1 0 and
> d2 0 . Moreover, by Lemma 1.8,
u ,u
and v ,v .
(10.78)
Then, substituting (10.78) into the v equation of (10.5) yields

d2 v v df2 (, v)v c,u v.
Hence, v provides us with a positive supersolution of (10.76). So, according
to Theorem 1.7,
v {,v;c,u } .
Substituting this estimate into the u equation of (10.5) shows that
d1 u u af1 (, u)u b{,v;c,u } u.
In other words, u is a positive subsolution of

(d1 + b{,v;c,u } )u = u af1 (, u)u
u=0
in ,
on .
Therefore, thanks to Lemma 1.9, (10.75) holds.

By adapting the proof of (10.39), it is easily seen that
lim {,v;c,u } = uniformly in compact subsets of .
Thus, owing to Lemma 10.6,

lim 0 [d1 , b{,v;c,u } ] = .
(10.79)
296
According to (10.79), (10.75) cannot be satisfied for large . Therefore, as we

already know that (10.75) is necessary for the existence of a coexistence state,
(10.5) cannot admit a coexistence state for large .
Subsequently, we assume that d1 1 . Then, substituting (10.78) into
the u equation of (10.5) yields
d1 u u af1 (, u)u b,v u
and hence, u is a positive supersolution of

(d1 + b,v )u = u af1 (, u)u
u=0
in ,
on .
(10.80)
In case 0 [d1 , b,v ], (10.77) holds from the estimate

0 [d1 , b,v ] < 1 [d1 , b,v ].
So, suppose > 0 [d1 , b,v ]. Then, by adapting Theorem 1.7 it is easily seen
that (10.80) possesses a positive solution. Therefore,
= 1 [d1 + b,v + af1 (, u), ] < 1 [d1 + b,v , i0,j ]
for all 1 j q0 and 1 i mj , which entails (10.77) and ends the proof of
the proposition.
Proof of Theorem 10.14: Assume (10.70). The fact that
(, u, v) = (0 [d2 , c,u ], ,u , 0)
is the unique bifurcation point to coexistence states from (u, v) = (,u , 0)
follows easily by linearizing (10.5) at (,u , 0) and using the local bifurcation
theorem of M. G. Crandall and P. H. Rabinowitz [59], as in the proof of
Theorem 7.2.2 of [149]. The technical details are omitted here. Actually, the
existence of the component C+
(,,u ,0) follows from [149, Th. 7.2.2] and it
satisfies some of the following alternatives:
(A1) C+
(,,u ,0) is unbounded in R X1 X2 .
(A2) There exists R such that
= 0 [d1 , b ,v ]
and
+
( , 0, ,v ) C
(,,u ,0) .
(A3) There exists a further positive solution ,u =

6 ,u of (10.7) with
+
(0 [d2 , c,u ], ,u , 0) C
(,,u ,0) .
+
(A4) = d1 0 and (d2 0 , 0, 0) C
(,,u ,0) .
297
By (10.70), Alternative (A4) cannot occur. Thanks to Theorem 2.2 and Remark 5.3, ,u is the unique positive solution of (10.7). So, Alternative (A3)
cannot occur either. Moreover, owing to Proposition 10.15, (10.5) cannot admit a coexistence state for large , and any coexistence state, (u, v), of (10.5)
satisfies u ,u and v ,v . Therefore, C+
(,,u ,0) is bounded in RX1 X2 .
Consequently, Alternative (A2) occurs. Necessarily = , because the map
7 0 [d1 , b,v ] is increasing. Moreover, since P is continuous and C+
(,,u ,0)
is connected, the projection

I := P C+
(,,u ,0)
must be an interval. As a by-product, for any (, ) in the region enclosed by
the two curves,
= 0 [d1 , b,v ]
and = 0 [d2 , c,u ],
d1 0 < < d1 1 ,
described by (10.72), (10.5) has a coexistence state. This ends the proof of
the first part of the theorem.
Subsequently, we assume that d1 1 . As before, the existence of
+
C+
(,0,,v ) can be derived from [149, Th. 7.2.2]. Similarly, C(,0,,v ) satisfies
some of the following alternatives:
(B1) C+
(,0,,v ) is unbounded in R X1 X2 .
(B2) There exists a semi-trivial positive solution of the form (u, 0) such that
+
(, u, 0) C
(,0,,v ) .
,v of (10.8) such that
(B3) There exists a positive solution ,v =
6
+
(, 0, ,v ) C(,0,,v ) .
+
(B4) = d1 0 and (d2 0 , 0, 0) C
(,0,,v ) .
As d1 1 > d1 0 , Alternative (B4) cannot occur. By Theorem 4.1 and
Remark 5.3, (10.7) admits a positive solution if, and only if, d1 0 < <
d1 1 . Thus, since we are assuming that d1 1 , (10.7) cannot admit a
positive solution and hence Alternative (B2) also fails. Furthermore, since
(10.8) possesses a unique positive solution, Alternative (B3) cannot occur
either. Therefore, Alternative (B1) holds.
By Proposition 10.15,
< 1 [d1 , b,v ].
(10.81)
Thus, > 1 , where 1 stands for the unique value of for which
= 1 [d1 , b1 ,v ].
Hence,

1
P C+
(,0,,v ) ( , ).
298
Suppose, in addition, condition (10.73) holds. Then, by Theorem 10.12, the

coexistence states of (10.5) are uniformly bounded in compact subsets of
d2 0 . Consequently, since C+
(,0,,v ) is unbounded in R X1 X2 ,

+
( , ) P C(,0,
.
)
,v
It remains to establish the existence of . Let (, u, v) be a coexistence state
of (10.5) with < . Then,
= 0 [d1 , b ,v ] > 0 [d1 , b,v ].
(10.82)
Moreover, v ,v and u is a positive supersolution of (10.80). Thus, by

Remark 5.3, we find from Theorem 1.7 that u [,b,v ] , where [,b,v ] is
the unique positive solution of (10.80); it exists because, due to (10.81) and
(10.82),
0 [d1 , b,v ] < < 1 [d1 , b,v ].
By Theorem 10.12, there exists a positive constant, C, such that
kuk C.
Thus, since [,b,v ] u, we can infer from the previous estimate that
[,b,v ] C.
(10.83)
Suppose (10.5) admits a sequence of coexistence states, (n , un , vn ), n 1,

with
lim n = 1 .
n
Then, according to (10.83), we find that

[,bn ,v ] C,
n 1.
(10.84)
On the other hand, by the continuous dependence of the principal eigenvalue

with respect to the potential and the continuous dependence of ,v with
respect to , it becomes apparent that
lim 1 [d1 , bn ,v ] = 1 [d1 , b1 ,v ] = .
Consequently, according to Theorem 4.1,

lim [,bn ,v ] =
uniformly on compact subsets of some of the components of 0 , which contradicts (10.84) and ends the proof.
Theorem 10.14 establishes that under condition (10.70) the problem (10.5)
behaves much like its classical counterpart with 0 = . Naturally, as d1 1
299
the u components of the coexistence states might grow to infinity in some or

several components of 0 , but within the range
d1 0 < < d1 1 ,
the bifurcation diagrams of coexistence states of (10.5) look like those found
in the pioneering plots of J. C. Eilbeck, J. E. Furter and J. Lopez-Gomez [82].
The behavior of the model for > d1 1 is completely different, because,
according to Theorem 10.14, (10.5) possesses a coexistence state for every
> if > d1 1 but 6= d1 j for all 2 j q0 . Actually, we imposed
c(x) > 0 for all x 0 from the very beginning in order to get this result.
Indeed, thanks to Theorem 10.11, in the special case when c = 0, the problem
(10.5) admits a coexistence state if, and only if,
0 [d1 , b,v ] < < 1 [d1 , b,v ],
or, equivalently,
1 < < ,
in strong contrast with the case when c(x) > 0 for all x 0 .
Figure 10.2 represents an admissible bifurcation diagram of coexistence
states in the special case c = 0 with d1 1 .
According to Theorem 2.2 and Remark 5.3, for this choice (10.5) cannot
admit a semi-trivial positive solution of the form (u, 0). Figure 10.2 plots the
FIGURE 10.2: The component C+

(,0,,v ) in case c = 0 with d1 1 .
300
norm of the solutions

k(u, v)k := kuk + kvk
versus the main bifurcation parameter . The horizontal axis consists of the
trivial solutions (, 0, 0). The curve of semi-trivial positive solutions (, 0, ,v )
bifurcates from (, 0, 0) at the critical value of the parameter = d2 0 and it
is defined for all > d2 0 . By linearizing (10.5) around it, it is easily seen that
it is linearly unstable if < and linearly stable if > . At = there
is a secondary bifurcation from (, 0, ,v ) to a subcritical curve of coexistence
states of (10.5). Namely,

1
C+
(10.85)
(,0,,v ) = (, [d1 ,b,v ] , ,v ) : < < .
According to Theorem 10.11(b), each of these coexistence states is a global attractor with respect to the positive solutions of (10.1). According to Theorem
4.1 and Remark 5.3, since
= 1 [d1 , b1 ,v ],
the u component of these coexistence states blows up in 0,1 [d1 , b,v ] as
1 . As usual, solid lines in Figure 10.2 are filled in by linearly stable
solutions, while dashed lines indicate linearly unstable solutions.
Obviously, the situation sketched in Figure 10.2 for c = 0 is quite different
from the one described by Theorem 10.14 when c(x) > 0 for all x 0
and d1 1 . Although at first glance the huge differences between both
cases might seem paradoxical, since the solution diagrams for c = 0 and for
sufficiently small c > 0 should be reminiscent, the huge differences are rather
natural. Indeed, the next result explains whats going on when c > 0 perturbs
from zero. The notations introduced in Theorem 10.14 are maintained.
Theorem 10.16 Suppose (HC) and (HF), set
c := %
c
,
kck
% := kck ,
and regard to % as a parameter. Then, for each > 0 there exists % > 0 such
that (10.5) possesses a coexistence state for every (1 + , ) provided
% (0, % ). Therefore, if we denote by (c) the minimal value of for which
(10.5) has a coexistence state, whose existence is given by Theorem 10.14, then
lim (c) = 1 .
%0
Proof: In the special case c = 0, for each (1 , ) the unique coexistence

state of (10.5), (, [d1 ,b,v ] , ,v ), is non-degenerate. Therefore, applying the
implicit function theorem along the solution curve
(, [d1 ,b,v ] , ,v ),
1 < < ,
301
with % as the main continuation parameter, yields the result. Outside a neighborhood of = , one can complete very easily all technical details. In a
neighborhood of = , one should use the technical devices of [148].
It turns out that the curve of coexistence states of (10.5) for c = 0, given
by (10.85), perturbs into a compact arc of curve of coexistence states if c 0.
If one further assumes
[d1 1 , ) \ {d1 j : 1 j q0 },
then, according to Theorem 10.14, the perturbed component C+
(,0,,v ) for c
0 must turn backward at some turning point, (t (c), ut (c), vt (c)), as illustrated
in Figure 10.3, where it has been represented a genuine case where
(c) = t (c),
though, in general, (c) t (c), and (c) < t (c) might indeed occur.
Naturally, the global continuation of the component C+
(,0,,v ) for all further
> t (c) can be accomplished thanks to the existing a priori bounds guaranteed by Theorem 10.12. Arguing as in the proof of Theorem 10.16, it becomes
apparent that
lim t (c) = 1
c0
and that ut (c) must blow up in some component of 0 as c 0. Our perturbation analysis shows that C+
(,0,,v ) looks like Figure 10.3 for sufficiently
FIGURE 10.3: The component C+

(,0,,v ) for c 0 and satisfying (10.73).
302
small c. However, one cannot exclude the existence of a further global supercritical folding, or even an isola, separated away from C(,0,,v ) . Although
C+
(,0,,v ) consists of a smooth curve filled in by asymptotically stable coexistence states from the bifurcation point ( , 0, ,v ) to reach the turning point
(t (c), ut (c), vt (c)), one cannot guarantee anything about the fine structure of
C+
(,0,,v ) beyond the turning point. In Section 10.6, we will show that (10.5)
possesses at least two coexistence states for each ( (c), ), as illustrated
in Figure 10.3. In general, (c) should grow as c increases.
Summarizing, as c perturbs from zero, the component C+ (, 0, ,v ) shown
in Figure 10.2 bends backward, looking like Figure 10.3. Consequently, as in
Chapter 9, (10.5) or (10.1) exhibits a sort of superlinear indefinite character
as c perturbs from zero.
10.4.2
Regarding as the main bifurcation parameter
In this section, we will carry out the corresponding discussion fixing > d2 0
and using as the main bifurcation parameter, exchanging the roles played
by the parameters and in the previous section. The technical details of the
proofs are omitted here, since they can be be easily reconstructed from the
proof of Theorem 10.14. As the behavior of the model can be very intricate
for large c, we will focus our attention on the case when c is small. How small
should c be for the validity of our analysis will be discussed later.
By Remark 5.3, we find from Theorem 4.8 that
min
lim ,u = M[d1 1 ,1 ]
(10.86)
d1 1
(see (10.14)). Subsequently, for sufficiently small > 0, we will consider the
open neighborhoods
,1 := 0,1 {x : dist(x, 0,1 ) < } % 0,1 ,
,1 $ 1 .
1 := \
min
By construction, the metasolution M[d1 1 ,1 ] is uniformly bounded above in

by some positive constant C . Since
1
min
0,1 \ ,
M[d1 1 ,1 ] = in
necessarily
lim C = .
0
Thus, for every (d1 0 , d1 1 ) and sufficiently small > 0,
0 [d2 , c,u ] < 1 [d2 + c,u , 1 ] < 1 [d2 + cMmin

[d1 1 ,1 ] , 1 ], (10.87)
because 7 ,u is increasing and it approximates the minimal metasolution
303
Mmin
[d1 1 ,1 ] as d1 1 . Thanks to (10.86) and (10.87), the next limit is well
defined
1 [d2 + cMmin
[d1 1 ,1 ] , 1 ] := lim 0 [d2 , c,u ].
d1 1
Moreover, by (10.87), for sufficiently small > 0, we have

min
1 [d2 + cMmin
[d1 1 ;1 ] , 1 ] 1 [d2 + cM[d1 1 ,1 ] , 1 ]
1 [d2 , 1 ] + cM C ,
cM := max c.
Hence,
lim 1 [d2 + cMmin

[d1 1 ,1 ] , 1 ] 1 [d2 , 1 ].
cM 0
Therefore, by the continuous dependence of the principal eigenvalue with respect to the domain (see [163, Ch. 8]), letting 0 yields
lim 1 [d2 + cMmin
[d1 1 ,1 ] , 1 ] = 1 [d2 , 1 ],
cM 0
(10.88)
because, since cMmin

[d1 1 ,1 ] > 0, we also have that
1 [d2 + cMmin
[d1 1 ,1 ] , 1 ] 1 [d2 , 1 ].
Throughout the rest of this section we will assume that
1 [d2 , 1 ] < d1 1 ,
(10.89)
where, for every > d1 0 , is the unique value of > d2 0 for which
= 0 [d1 , b ,v ].
According to (10.88) and (10.89), there exists % > 0 such that
1 [d2 + cMmin
[d1 1 ,1 ] , 1 ] < d1 1
if cM < %.
Subsequently, we will assume cM < % and denote, to simplify the notation,

0 := 1 [d2 + cMmin
[d1 1 ,1 ] , 1 ].
Then,
0 lim 0 [d2 , c,u ] < d1 1 .
d1 1
Figure 10.4 shows four important curves in the (, ) plane from the point
of view of the existence of coexistence states for (10.5). Two of them enclose
the regions where, according to Proposition 10.15, (10.5) does not admit a
coexistence state,
0 [d1 , b{,v;c,u } ]
and 1 [d1 , b,v ].
304
The other two are the curves = 0 [d1 , b,v ], > d2 0 , and = 0 [d2 , c,u ],
d1 0 < < d1 1 , which provide us with all the bifurcation points to coexistence states from the semi-trivial solutions (0, ,v ) and (,u , 0), respectively.
Thanks to the assumption (10.89), the graph of the curve = 0 [d2 , c,u ]
lies below the graph of = 0 [d1 , b,v ] for cM sufficiently small, which is the
situation illustrated in Figure 10.4.
Subsequently, we consider d2 0 < < 0 and regard as the main bifurcation parameter. Let
C+
(,0,,u ) R int P1 int P2
denote the component of the set of coexistence states of (10.5) emanating
from (, 0, ,v ) at
= 0 [d1 , b,v ],
whose existence is guaranteed by [149, Th. 7.2.2]. For d2 0 , the local bifurcation analysis of J. C. Eilbeck, J. E. Furter and J. Lopez-Gomez [82] shows
that C+
(,0,,v ) meets the other branch of semi-trivial solutions, (, ,u , 0), at
= , the unique value of for which
= 0 [d2 , c ,u ].
FIGURE 10.4: Some significant curves in the (, ) plane.
305
Moreover, the component C+

(,0,,v ) is bounded. Consequently, the associated
global bifurcation diagram looks like those found by J. C. Eilbeck, J. E. Furter
and J. L
opez-G
omez [82] for the classical diffusive competing species model.
As separates from d2 0 and approximates 0 , the previous behavior might
change drastically, because the component of the set of coexistence states
of (10.5) emanating from (, ,u , 0) at := , which will be denoted by
+
C+
(,,u ,0) , might be unbounded, as may the component C(,0,,v ) .
Actually, both components must be unbounded if one is unbounded. Indeed, if one of them is bounded, according to [149, Th. 7.2.2], it must link the
two semi-trivial positive solutions. So,
+
C+
(,0,,v ) = C(,,u ,0) ,
as in the classical situations described by J. C. Eilbeck, J. E. Furter and J.

L
opez-G
omez [82]. Figure 10.5 shows an admissible global bifurcation diagram
of coexistence states when these components are unbounded.
Subsequently, we suppose that, in addition,
0 < 1d1 2
(10.90)
FIGURE 10.5: An admissible bifurcation diagram for ' 0 , < 0 .
306
as illustrated in Figure 10.4. It should be remembered that

d1 2 = 1 [d1 , b1d
1 2
,v ]
= 1 [d1 + b1d
1 2
,v , 0,1 (d1 , b1d
1 2
,v )].
By Proposition 10.15, (10.5) cannot admit a coexistence state for

1 [d1 , b,v ] = 1 [d1 + b,v , 0,1 (d1 , b,v )].
Pick 0 < < 1d1 2 (see Figure 10.4). For such range of values of the parameter , (10.5) cannot admit a semi-trivial positive solution of the form (,u , 0).
Thus, by [149, Th. 7.2.2], the component C+
(,0,,v ) is unbounded.
Moreover, by Proposition 10.15, whenever (10.5) has a coexistence state,
< 1 [d1 + b,v , 0,1 (d1 , b,v )]
< 1 [d1 + b1d ,v , 0,1 (d1 , b,v )]
1 2
= 1 [d1 + b1d
1 2
,v , 0,1 (d1 , b1d
1 2
,v )]
= d1 2
provided is sufficiently close to 1d1 2 so that

0,1 (d1 , b,v ) = 0,1 (d1 , b1d
1 2
,v ).
FIGURE 10.6: An admissible bifurcation diagram for 0 < < 1d1 2 .
307
Therefore, thanks to Theorem 10.12, the component C+

(,0,,v ) blows up at
= d1 1 . Consequently, it looks like Figure 10.6.
Thanks to Theorem 10.16, for sufficiently small c ' 0, the component
C+
(,0,,v ) is defined for values of as close as we wish to 1 [d1 , b,v ]. Thinking
of the special case c = 0 might help us understand whats going on. Therefore,
the smaller cM , the closer to 1 [d1 , b,v ] is the subcritical turning point of
the component C+
(,0,,v ) in Figure 10.6.
For larger values of , > 1d1 2 , the component of coexistence states
+
C(,0,,v ) might blow up at some d1 j with j 2. In such cases, the global
bifurcation diagram of coexistence states represented in Figure 10.7 is admissible.
If C+
(,0,,v ) blows up at d2 2 , then, by the multiplicity results of Section
10.6, the bifurcation diagram in Figure 10.7 occurs, because for each
(d1 1 , d1 2 ), (10.5) possesses at least one coexistence state. Consequently, a
further component of the set of coexistence states of (10.5), C+
isola , should lie
within that interval, as illustrated by Figure 10.7.
FIGURE 10.7: A possible bifurcation diagram for > 1d1 2 .
308
Naturally, some numerical experiments are imperative for ascertaining the

exact global bifurcation diagrams.
10.5
Strong maximum principle for quasi-cooperative

systems
Throughout this section we will assume that

b(x) > 0
and c(x) > 0
for all x .
(10.91)
If (u0 , v0 ) is a coexistence state of (10.5), its linearized stability is given

through the sign of the eigenvalues of the linearization of (10.5) at (u0 , v0 ),
i.e., by the signs of the s for which the linear boundary value problem

d1
0
u
u
u
=A
+
in ,
0
d2
v
v
v
(10.92)
(u, v) = (0, 0)
on .
has a solution (u, v) 6= (0, 0), where we have denoted

(x) (x)
A=
(x) %(x)
(10.93)
with
:= af1 (, u0 ) au0 u f1 (, u0 ) bv0 ,
% := df2 (, v0 ) dv0 v f2 (, v0 ) cu0 ,
:= bu0 ,
:= cv0 .
Thanks to (10.91), the off-diagonal entries of the coupling matrix A are negative. This is why (10.92) is said to be of quasi-cooperative type. More generally,
where
we will consider (10.92) with , , , % C()
(x) < 0
and (x) < 0
for almost all x .
Subsequently, to simplify notations, we will set

d1
0
d1 (x)
(x)
L :=
A=
(10.94)
0
d2
(x)
d2 %(x)
and suppose p > N . Given u, v Lp (), it is said that (u, v) q 0 if u 0
and v 0; the q emphasizes the quasi-cooperative character of this ordering.
309
If, in addition, u =
6 0, or v 6= 0, it is said that (u, v) >q 0. A couple (u, v)
W02,p () W02,p () is said to be strongly positive if u(x) > 0, v(x) < 0 for
all x and n u(x) < 0, n v(x) > 0 for all x , where n stands for
the outward unit normal to at x . In such case, we will simply write
(u, v) q 0. Subsequently, we shall always refer to this order. Note that we
are ordering X1 X2 with the cone P := P1 (P2 ) (see the beginning of
Section 9.5). As P is a normal cone, according to [163, Th. 6.1] there exists a
norm in X1 X2 , k kX1 X2 , which is monotonic. This will be the norm used
in Section 10.6 in the calculations of the topological indices.
Definition 10.17 The operator L defined by (10.94) is said to satisfy the
strong maximum principle in if x W02,p () W02,p () and Lx >q 0 imply
x q 0.
Definition 10.18 A function x W 2,p () W 2,p () is said to be a supersolution of (L, ) if Lx q 0 in and x q 0 on . If, in addition, Lx >q 0 in
, or x >q 0 on , then it is said that x is a strict supersolution of (L, ).
Adapting the proofs of Theorem 2.1 of J. Lopez-Gomez and M. MolinaMeyer [130] and Theorem 6.3 of J. Lopez-Gomez and J. C. Sabina de Lis
[180, Th. 6.3], the next results follow readily. Note that we are adapting to
a quasi-cooperative setting the main results of J. Lopez-Gomez [163, Sect. 7]
for the single equation.
Theorem 10.19 The operator L, under homogeneous Dirichlet boundary
conditions in , possesses a unique eigenvalue to a positive eigenfunction,
1 [L, ], called the principal eigenvalue of L in , or (L, ). The principal
eigenvalue is algebraically simple. Consequently, the principal eigenfunction,
>q 0, is unique up to a positive multiplicative constant. Moreover, q 0
and any other eigenvalue of (10.92) satisfy Re > 1 [L, ]. Furthermore,
for every > 1 [L, ], the operator
(L + )1 L(Lp () Lp ())
is compact and positive.
Actually, the last assertion is an immediate consequence of the next counterpart of Theorem 1.1.
Theorem 10.20 The following conditions are equivalent:
(a) 1 [L, ] > 0.
(b) (L, ) admits a positive strict supersolution in W 2,p () W 2,p ().
(c) (L, ) satisfies the strong maximum principle.
Actually, under any of these circumstances, the following result holds.
310
Theorem 10.21 Suppose 1 [L, ] > 0. Then, any strict supersolution

x := (
u, v) W 2,p () W 2,p ()
of (L, ) must be strongly positive in , in the sense that u(x) > 0 and v(x) <
0 for all x , n u(x) < 0 for all x (u)1 (0) and n v(x) > 0 for all
x (v)1 (0) .
From these results one can infer very easily the main monotonicity properties of 1 [L, ], as in J. Lopez-Gomez and M. Molina-Meyer [130] and H.
Amann [12]; technical details are omitted here.
Throughout the rest of this chapter, we will also consider the inhomogeneous problem

in ,
(10.95)
(u, v) = (u , v )
on ,
with u , v W 1,p ().
Definition 10.22 A function x = (u, v) W 2,p () W 2,p () is said to be
a subsolution of (10.95) if
d1 u u a(x)f1 (x, u)u b(x)u v
d2 v v d(x)f2 (x, v)v c(x)u v
in and x q (u , v ) on .
Similarly, a function x = (u, v) W 2,p () W 2,p () is said to be a
supersolution of (10.95) if
d1 u u a(x)f1 (x, u)u b(x)u v
d2 v v d(x)f2 (x, v)v c(x)u v
in and x q (u , v ) on .
The following result can be easily obtained combining the previous results
with H. Amann [11, Th. 9.4].
Theorem 10.23 Suppose (10.95) possesses a subsolution x = (u, v) and a
supersolution x = (u, v) such that x q x. Then (10.95) possesses a minimal
solution x = (u , v ) and a maximal solution x = (u , v ) in the order
interval [x, x]. If in addition u = v = 0, u > 0 and v > 0, then (10.5) has
a coexistence state.
Actually, most of the results of Chapter 1 remain valid for the system.
10.6
311
Multiplicity of coexistence states
Throughout this section, for every > d1 0 , we will denote by J the set of
R for which (10.5) possesses a coexistence state.
Theorem 10.24 Suppose b > 0 and c > 0 almost everywhere in and
d1 1 satisfies 6= d1 j , 1 j q0 . Then, some of the following alternatives
occur:
(a) J = ( , ), where = 0 [d1 , b ,v ], or
(b) J = [ (), ) for some () (1 , ], where = 1 [d1 , b1 ,v ].
Proof: Thanks to Theorem 10.14,
( , ) J (1 , ).
Consequently, option (a) occurs when (10.5) does not admit a coexistence
state for . It remains to show that if (10.5) has a coexistence state for
some
< , then it also admits a coexistence state for each [
, ], as
this establishes option (b) easily.
Suppose (10.5) has a coexistence state for some
< , (
u, v). Then,
u
= v = 0 on and for every (
, )
d1
u =
u af1 (, u
)
u b
u v
d2
v <
v df2 (, v)
v c
u v
in . Thus, (
u, v) provides us with a strict supersolution of (10.5). Moreover,
d2
v <
v df2 (, v)
v
and hence, by Lemma 1.8, v < ,v . So, substituting this estimate in the
u
-equation yields
d1
u =
u af1 (, u
)
u b
u v >
u af1 (, u
)
u b,v u
.
Thus, thanks again to Lemma 1.8, we find that
u
> [,b,v ] ,
(10.96)
where [,b,v ] stands for the unique positive solution of

(d1 + b,v )u = u af1 (, u)u in ,
u| = 0.
Note that [,b,v ] > 0 is well defined, because 1 < < implies
0 [d1 , b,v ] < < 1 [d1 , b,v ].
312
Set
u := [,b,v ] ,
v := ,v .
Then, since v < ,v , it follows from (10.96) that

u, v).
(u, v) <q (
(10.97)
Moreover,
d1 u = u af1 (, u)u b u v,
d2 v = v df2 (, v)v > v df2 (, v)v c u v.
Thus, (u, v) is a subsolution of (10.5). Therefore, by (10.97), we find from
Theorem 10.23 that (10.5) possesses a coexistence state for each [
, ).
To establish the existence of a coexistence state for = , let (n , un , vn ),
n 1, be a sequence of coexistence states such that n < , n 1, and
lim n = .
As =
6 d1 j for all 1 j q0 , according to Theorem 10.12, these coexistence
states are bounded in X1 X2 . Thus, by a rather standard compactness
argument, there is a solution of (10.5), (u , v ) X1 X2 , such that along
some subsequence, re-labeled by n,
lim kun u kX1 = 0,
lim kvn v kX2 = 0.
Necessarily, u 0 and v 0. If u > 0 and v > 0, we are done.

Suppose u > 0 and v = 0. Then, u = ,u and due to Theorem 4.1 we
find that d1 0 < < d1 1 , which contradicts > d1 1 . So, this cannot occur.
Suppose u = 0 and v = 0. Then, by the uniqueness of the principal
eigenvalue, we can infer from
d1 un = un af1 (, un )un bun vn ,
n 1,
that
= 0 [d1 , af1 (, un ) + bvn ],
n 1.
So, letting n yields = d1 0 , which is impossible.

Suppose u = 0 and v > 0. Then, v = ,v and, necessarily, C+
(,0,,v )
bifurcates towards the left of by the local uniqueness given by the main
theorem of M. G. Crandall and P. H. Rabinowitz [59]. Consequently, if (10.5)
does not admit a coexistence state for = , then

1
P C+
(,0,,v ) ( , ),
because the projection of the component C+
(,0,,v ) is an interval. But this
contradicts Theorem 10.14. Therefore, (10.5) also possesses a coexistence state
for = . So, [
, ) J .
313
Finally, set () := inf J . Thanks to the previous analysis, to complete

the proof we must prove that () J . Let (n , un , vn ), n 1, be a sequence
of coexistence states such that
lim n = (),
() < n < ,
n 1.
According to Theorem 10.12, these coexistence states are uniformly bounded

in X1 X2 . Thus, (10.5) admits a non-negative solution, ( (), u , v )
X1 X2 , such that along some subsequence, labeled again by n,
lim kun u kX1 = 0,
lim kvn v kX2 = 0.
Arguing as before, it becomes apparent that u > 0 and v > 0.
Adapting the arguments of the proof of Theorem 10.24, one can also show
that J is a bounded interval if < d1 1 , as in the most classical competing
species models dealt with in J. C. Eilbeck et al. [82] and J. E. Furter and
J. L
opez-G
omez [91]. Moreover, J might shrink to a single point when both
semi-trivial positive solutions are neutrally stable according to the shapes and
the sizes of the function coefficients b and c. Local bifurcation and singularity
theory were the approaches adopted in these pioneering papers to establish
some closely related features. Incidentally, re-elaborating on the results, techniques and ideas of J. C. Eilbeck, J. Lopez-Gomez and J. E. Furter [82], E. N.
Dancer [65] constructed some sharp examples establishing that some previous
existence results by L. Li and R. Logan [134] and R. Logan and A. Ghoreishi
[139] were incorrect as stated. Indeed, according to E. N. Dancer [65] (see
pages 239 and 240):
The main purpose of this paper is to show that for many smooth domains
(including some convex ones in all dimensions except two) there exists a, d > 1 such
that (1) has no positive solution for c = c, b = b. This result is particularly interesting
since there are two published proofs of the contrary result in the literature ([134],
[139]). The errors in those proofs are incorrect degree calculations as in Lemma 4.5
in [139]...
It is necessary to have a copy of [82] available in reading this paper.
Next, we will adapt the abstract theory of H. Amann in [11, Sect. 20],
further developed by H. Amann and J. Lopez-Gomez [13] and M. Delgado, J.
L
opez-G
omez and A. Su
arez [72], to prove the following multiplicity which is
the main result of this section; it deals with the situation described by Figure
10.3.
Theorem 10.25 Under the assumptions of Theorem 10.24, suppose, in addition, that
J = [ (), )
for some
() (1 , ).
(10.98)
Then, for every ( (), ) the problem (10.5) has at least two coexistence
states.
314
Condition (10.98) holds whenever the component C+

(,0,,v ) bifurcates towards
the left from the solution branch (, 0, ,v ) at the critical value of the parameter = . By Theorem 10.16, this occurs for sufficiently small cM .
Proof of Theorem 10.25: Proof is based on the properties of the fixed point
index in cones. First, we will establish that, for every ( (), ), (10.5)
has a minimal coexistence state. Indeed, according to the proof of Theorem
10.24, we already know that
(u, v) := ([,b,v ] , ,v )
is a positive subsolution of (10.5) such that (u, v) <q (u, v) for all coexistence
states (u, v). Suppose (10.5) has two coexistence states, (u1 , v1 ) and (u2 , v2 ).
Then, the couple
(u, v) := (max{u1 , u2 }, min{v1 , v2 })
provides us with a supersolution satisfying
(u, v) <q (u, v),
(uj , vj ) [(u, v), (

u, v)],
j {1, 2}.
Thus, by Theorem 10.23, there is a coexistence state

(u3 , v3 ) [(u, v), (u, v)]
such that (u3 , v3 ) q (uj , vj ), for each j {1, 2}. This gives the minimal
coexistence state when (10.5) has finitely many coexistence states, and it
shows that if (10.5) has infinitely many coexistence states but does not admit
a minimal coexistence state, there exists a sequence of coexistence states,
(
un , vn ), n 1, such that
lim vn = 0
in X2 .
As u
n > [,b,v ] for all n 1 and, due to Theorem 10.12, the coexistence
states of (10.5) possess uniform a priori bounds, by a rather standard compactness argument it becomes apparent that (10.5) admits a semi-trivial positive solution of the form (
u, 0), which is impossible because we are imposing
> d1 1 . This provides us with the minimal coexistence state, (u , v ).
Next, we will make sure that (10.5) fits into the abstract setting of H.
Amann [11]. Fix 1 < a < (), b > 0, and consider the interval I :=
[a, + b]. By the existence of uniform a priori bounds in I, the constant K
in (10.69) can be chosen sufficiently large so that, for every I and any
non-negative solution (u, v) of (10.5),
f1
(, u)u + af1 (, u) + bv < + K,
u
f2
df2 (, v) + cu d
(, v)v + df2 (, v) + cu < + K.
v
af1 (, u) + bv a
315
For this choice of K, the maps

u 7 [ + K af1 (, u) bv]u,
v 7 [ + K df2 (, v) cu]v,
are positive and increasing in u and v, respectively. Thus, for any pair of
coexistence states, (u1 , v1 ) and (u2 , v2 ), with (u1 , v1 ) q (u2 , v2 ), i.e., such
that u1 u2 and v1 v2 , one has that
[ + K af1 (, u1 ) bv1 ]u1 [ + K af1 (, u2 ) bv1 ]u2
[ + K af1 (, u2 ) bv2 ]u2 .
Similarly,
[ + K df2 (, v1 ) cu1 ]v1 [ + K df2 (, v2 ) cu1 ]v2
[ + K df1 (, v2 ) cu2 ]v2 .
Consequently,
K, (u1 , v1 ) q K, (u2 , v2 ).
Moreover,
(u1 , v1 ) <q (u2 , v2 )
K, (u1 , v1 ) q K, (u2 , v2 ),
because b(x) > 0 and c(x) > 0 for almost all x and the operators
(d1 + K)1 and (d2 + K)1 are strongly order preserving. Therefore,
K, is compact and strongly order preserving for all I.
Denote by B the unit ball of X1 X2 and, for every > 0, let P be the
positive part of B. Then, the fixed point index of K, in P is well defined
for sufficiently large > 0. Moreover, the next result holds.
Lemma 10.26 Assume (1 , ). Then, (0, 0) and (0, ,v ) are isolated
fixed points of K, in P1 P2 with
i(K, , (0, 0)) = i(K, , (0, ,v )) = 0.
Moreover, i(K, , P ) = 0 for sufficiently large .
Proof: Since < , it is easily seen that (0, ,v ) is linearly unstable. So,
from J. L
opez-G
omez [140, Le. 4.1], we have that i(K, , (0, ,v )) = 0. Moreover, by Lemma 13.1(ii) of H. Amann [11], i(K, , (0, 0)) = 0. The rest follows
from the homotopy invariance of the index, since (0, 0) and (0, ,v ) are the
unique non-negative solutions of (10.5) if (a, ()).
To ascertain the fixed point index of the minimal coexistence state,
(u , v ), which is the most delicate part of the proof, we need the following counterparts of Propositions 9.6, 9.7 and 9.14, whose proofs can be easily
adapted from the proofs of these propositions
316
Lemma 10.27 For every [ (), ) the minimal coexistence state

(u , v ) of (10.5) is weakly stable, in the sense that 1 [L , ] 0, where
L is the operator defined by (10.94) with (u0 , v0 ) = (u , v ).
Lemma 10.28 The following assertions are true:
(a) Let (, u, v) = (0 , u0 , v0 ) be a coexistence state with 1 [L0 , ] > 0,
where L0 is the operator defined by (10.94) with A(x) given by (10.93).
Then, there exist > 0 and a differentiable map
(u, v) : (0 , 0 + ) P1 P2
such that (u(0 ), v(0 )) = (u0 , v0 ) and (, u(), v()) is a coexistence state for each (0 , 0 + ). Moreover, the mapping
7 (u(), v()) is decreasing and there exists a neighborhood Q0 of
(0 , u0 , v0 ) in R X1 X2 such that the unique solutions of (10.5) in
Q0 are those of the form (u(), v()) with (0 , 0 + ).
(b) Suppose 1 [L0 , ] = 0 and let > 0 denote a principal eigenfunction
of 1 [L0 , ] = 0. Then, there exist > 0 and a differentiable map
(, u, v) : (, ) R P1 P2
such that ((0), u(0), v(0)) = (0 , u0 , v0 ) and ((s), u(s), v(s)) is a coexistence state for each s (, ). Moreover,
(s) = 0 +
(s),
(u(s), v(s)) = (u0 , v0 ) + s + (

u(s), v(s)),
where
(s) = 0(s), u
(s) = o(s) and v(s) = o(s) as s 0. In addition,
there exists a neighborhood, Q0 , of (0 , u0 , v0 ) in R X1 X2 such that
the unique solutions of (10.5) in Q0 are of the form ((s), u(s), v(s))
with s (, ). Furthermore,
sgn 0 (s) = sgn 1 [L(s) ],
(10.99)
where L(s) is the operator defined in (10.94) with A(x) given by (10.93)
and (u0 , v0 ) = (u(s), v(s)).
In the context of Lemma 10.28(a), differentiating with respect to yields
0

u ()
0
L
=
<q 0.
v 0 ()
v()
Thus, since 1 [L , ] > 0, it follows from Theorem 10.20 that
(u0 (), v 0 ()) q 0.
Therefore, 7 u() is decreasing, whereas 7 v() is increasing.
317
Similarly, in the context of Lemma 10.28(b), differentiating with respect

to the parameter s shows that
0

u (s)
0
0
L(s)
= (s)
.
v 0 (s)
v(s)
Thus, since
(u0 (s), v 0 (s)) q 0,
(10.99) holds from Theorem 10.20 taking into account that (0, v(s)) <q 0.
We are ready to ascertain the fixed point index of the minimal positive
solution, (, u , v ). According to Lemma 10.27,
1 [L , ] 0.
Suppose 1 [L , ] > 0. The Schauder formula implies that
i(K, , (u , v )) = 1.
Thus, according to Lemma 10.26, (10.5) admits a further coexistence state,
as stated in Theorem 10.25.
Suppose 1 [L , ] = 0 and let ((s), u(s), v(s)) be the curve of coexistence
states whose existences are guaranteed by Lemma 10.28(b), with
((0), u(0), v(0)) = (, u , v ).
Since >q 0, s 7 (u(s), v(s)) is strictly increasing. Hence, if (s) = for some
s 6= 0, then (10.5) possesses two coexistence states: (u(0), v(0)) = (u , v ) and
(u(s), v(s)). Thus, without loss of generality, to complete the proof of Theorem
10.25 we can assume that
(s) 6=
for all 0 < |s| < .
(10.100)
for all s (, 0).
(10.101)
In such case, we claim that

(s) >
Indeed, if 1 := (s1 ) < for some s1 < 0, since s (u(s), v(s)) is increasing
and 7 (u , v ) is non-increasing, we find that
(u(s1 ), v(s1 )) <q (u(0), v(0)) = (u , v ) q (u1 , v1 ),
(10.102)
where (u1 , v1 ) stands for the minimal coexistence state at = 1 . One

should take into account that (u1 , v1 ) is a supersolution of (10.5) if 1 < .
As (10.102) contradicts the minimality of (u1 , v1 ), (10.101) holds. Moreover,
by (10.100), either (s) < for all s (0, ) or (s) > for all s (0, ).
Subsequently, we will deal with each of these situations separately.
318
Case (i): Suppose (s) > for all s (0, ). Then, since [ (), ] J ,
there exists a sequence of coexistence states, (n , un , vn ), n 1, such that
lim n = ,
n < ,
n 1.
By the existence of a priori bounds, we can extract a subsequence, re-labeled

by n, such that
lim (un , vn ) = (u0 , v0 )
n
for some non-negative solution (u0 , v0 ). Adapting the arguments of the proof
of Theorem 10.24, it is easy to see that (, u0 , v0 ) is a coexistence state.
Moreover, by the uniqueness obtained as an application of Lemma 10.28(b),
we find that (n , un , vn ) 6 Q0 for all n 1, because n < < (s). Thus,
(, u0 , v0 ) 6 Q0 and hence (, u0 , v0 ) 6= (, u , v ). Therefore, (10.5) has two
coexistence states, as required.
Case (ii): Suppose
s (0, ).
(s) <
(10.103)
Then, owing to Lemma 10.28(b), (, u , v ) is an isolated solution of (10.5)

and the index i(K, , (u , v )) is well defined. According to Lemma 10.26, to
complete the proof of Theorem 10.25 it suffices to prove that
i(K, , (u , v )) = 1.
(10.104)
By (10.103), 0 (s1 ) < 0 for some s1 (0, ). Thus, by (10.99),

1 [L(s1 ) , ] > 0.
Consequently, owing to Theorem 10.19, ((s1 ), u(s1 ), v(s1 )) is exponentially
asymptotically stable and hence by the Schauder formula,
i(K,(s1 ) , (u(s1 ), v(s1 ))) = 1.
(10.105)
On the other hand, since ((s1 ), u(s1 ), v(s1 )) is non-degenerate and the mapping s (u(s), v(s)) is increasing, there exists > 0 with the property that
(10.5) cannot admit a coexistence state in
[(s1 ) , (s1 )] (P1 \ P 2 )
where
1 := k(u(s1 ), v(s1 ))kX1 X2 ,
2 := k(u , v )kX1 X2 .
Indeed, the fact that (10.5) cannot admit a coexistence state (, u, v) with
[(s1 ) , (s1 )]
and k(u, v)kX1 X2 = 1
319
follows easily from the uniqueness obtained as an application of Lemma

10.28(b), by the monotonicity of the norm k kX1 X2 , as in the proof of
[13, Th. 7.4]. That (10.5) cannot admit a coexistence state (, u, v) with
[(s1 ) , (s1 )]
and k(u, v)kX1 X2 = 2
follows easily from the fact that the minimal coexistence state, (, u , v ),
satisfies
k(u , v )kX1 X2 k(u , v )k > 2
for all [(s1 ) , (s1 )] because (s1 ) < . Moreover, thanks again to
the uniqueness obtained as an application of Lemma 10.28(a), > 0 can be
shortened, if necessary, so that (10.5) cannot admit a coexistence state in
P1 \ P 2 at = (s1 ) . For this choice, by the homotopy invariance of the
index, we find that
i(K(,(s1 )) , P1 \ P 2 ) = 0.
(10.106)
Finally, for sufficiently small > 0, we set
:= k(u(s1 ), v(s1 ))kX1 X2 + .
According to (10.105) and (10.106),
i(K(,(s1 )) , P \ P 2 ) = 1.
(10.107)
Moreover, by the monotonicity of (u(s), v(s)) and the uniqueness given by

Lemma 10.28(b), (10.5) cannot admit a coexistence state in
[(s1 ), ] (P \ P2 ).
Combining this property with (10.107) yields (10.104) and ends the proof.
10.7
Dynamics of (1.1) when b = 0
We begin by characterizing the existence of coexistence and metacoexistence

states in the special case b = 0. Then, we shall analyze their attractive character with respect to the classical positive solutions of (10.1). To state the
main result we need to introduce some notations. Throughout this section, for
every 1 j q0 , we denote by Xj the set of functions C (j ), 0,
such that
=0
on j
and
lim
dist(x,j )0
(x) =
where j is defined in (4.3), and for every 1 j q0 and Xj we denote

n := min {, n } for all integer n 1.
320
Obviously,
j ; R+ )
n C (
and
n n+1
for all n 1.
j ), the sequence of principal eigenvalues

Thus, for every V C (
1 [d2 + V + cn , j ],
n 1,
is non-decreasing. Moreover, it is bounded above. Indeed, pick x and

R (x) . Then, since C (B
R (x)), there exists n0 N
R > 0 with B
such that n = in BR (x) for all n n0 . Therefore, by monotonicity with

respect to the domain,
1 [d2 + V + cn , j ] 1 [d2 + V + c, BR (x)],
n n0 .
Consequently, the limit

1 [d2 + V + c, j ] := lim 1 [d2 + V + cn , j ]
n
is well defined in R. Using these notations the next result holds.

Theorem 10.29 Suppose b = 0 in . Then, the following properties hold:
(a) The problem (10.5) possesses a coexistence state if, and only if,
d1 0 < < d1 1
and
> 0 [d2 , c,u ],
where ,u stands for the unique positive solution of (10.7).

(b) Suppose 1 j q0 . Then, (10.5) has a metacoexistence state,
(Mu , Mv ), supported in j , as discussed by Definition 10.1, if, and only
if,
< d1 j+1 and > 1 [d2 + cMmin
(10.108)
[,j ] , j ],
where Mmin
[,j ] is the minimal metasolution of (10.7) supported in j .
Moreover, Mv can be chosen so that Mv = 0 on j .
As usual, when j = q0 we are taking q0 +1 and q0 = . In such case,
it should be noted that (10.108) is not imposing any restriction on the size of
, but only on . Our proof of Theorem 10.29 relies on the following result.
Proposition 10.30 Suppose 1 j q0 and Xj . Then, the problem

(d2 + c)v = v df2 (, v)v
in j ,
(10.109)
v=0
on j ,
admits a weak bounded positive solution if, and only if,
> 1 [d2 + c, j ].
Moreover, any weak bounded positive solution, v, of (10.109) satisfies v
j ), and is unique if it exists.
C 2+ (j ) C(
321
Note that 1 [d2 + c, j ] is the value defined just before the statement
of Theorem 10.29 with V = 0. To prove Proposition 10.30, we need the next
lemma.
Lemma 10.31 Consider the linear eigenvalue problem

(d2 + V + c)v = v
in j ,
v=0
on j ,
(10.110)
j ). Then,
where V C (
(a) (V ) := 1 [d2 + V + c, j ] is the unique value of for which
(10.110) admits a weak positive solution in H01 (j ). Moreover, for =
(V ) there exists a weak solution j H01 (j ) L (j ) such that
j (x) > 0 almost everywhere in j .
(b) If V < W , then (V ) < (W ).
Proof of Lemma 10.31: Consider the truncations
n := min{, n},
n 1,
and let j,n > 0, n 1, denote the (unique) principal eigenfunction of

1,n,j := 1 [d2 + V + cn , j ]
such that
kj,n kL (j ) = 1,
n 1.
Then,
(d2 + V )j,n (d2 + V + cn )j,n = 1,n,j j,n
for every n 1. Thus, there exists a constant C > 0 such that
(d2 + V )j,n Cj,n
for all n 1.
Consequently, by Lemma 10.13, there exist a subsequence of j,n , n 1,

re-labeled by n, and a function j H01 (j ) L (j ), j 6= 0, such that
lim j,n = j
weakly in H01 (j ) and in Lp (j ) for all p > 1. (10.111)
Necessarily, j > 0, because j,n > 0 for all n 1, and, owing to (10.111), it
is easily seen that j is a weak solution of (10.110) for = (V ). Moreover,
thanks to the weak Harnack inequality, j (x) > 0 almost everywhere (a.e.) in
j . To complete the proof of Part (a) we will argue by contradiction. Suppose
1 6= 2 are two values of for which (10.110) admits a weak positive solution,
1 and 2 , respectively. Then, thanks to the weak Harnack inequality, k (x) >
322
0 a.e. in j for k {1, 2}. Pick an arbitrary test function C0 (j ). Then,

k H01 (j ) for k {1, 2} and hence,
Z
Z
Z
d2
h(2 ), 1 i +
(V + c)1 2 = 1
1 2 ,
j
Z
h(1 ), 2 i +
d2
Z
(V + c)1 2 = 2
1 2 .
j
Thus, subtracting these two identities yields

Z
Z
d2
h, 2 1 1 2 i = (1 2 )
j
1 2 .
(10.112)
As (10.112) holds true for all C0 (j ), it becomes apparent that

Z
(1 2 )
1 2 = 0
j
and hence 1 = 2 , which is a contradiction. This ends the proof of Part (a).
To prove Part (b), let j and j denote two weak positive solutions of
(10.110) associated to = (V ) and = (W ), respectively. Then, for every
test function C0 (j ),
Z
Z
Z
d2
h, j j j j i +
(V W )j j = ((V ) (W ))
j j
j
and hence
Z
(V W )j j = ((V ) (W ))
j
j j .
j
As V < W , the left hand side of this identity is negative. Therefore, (V ) <
(W ), as claimed.
Proof of Proposition 10.30: Let 1 j q0 and Xj be, and suppose
(10.109) has a weak bounded positive solution, v. Then,
(d2 + df2 (, v) + c)v = v
and hence by Lemma 10.31,
= 1 [d2 + df2 (, v) + c, j ] > 1 [d2 + c, j ].
To show the converse, suppose
> 1 [d2 + c, j ].
Then, by definition, there exists n0 N such that
> 1 [d2 + cn , j ]
for all n n0 .
323
Consequently, by Remark 5.3 and Theorem 2.2, for every n n0 , the problem

(d2 + cn )v = v df2 (, v)v
in j ,
(10.113)
v=0
on j ,
has a unique positive (strong) solution, vn . As n+1 n ,
(d2 + cn+1 )vn (d2 + cn )vn = vn df2 (, vn )vn
and hence by Lemma 1.8,
vn vn+1
for all n n0 .
(10.114)
On the other hand, for each n n0 ,

d2 vn (d2 + cn )vn = vn df2 (, vn )vn vn .
By (10.114), it follows from Lemma 10.13 that there exists v H01 (j )
L (j ) such that
lim vn = v
weakly in H01 (j )
and strongly in Lp (j ) p > 1.
j ) and that v = 0
Also thanks to (10.114), it becomes apparent that v C(
on j . Naturally, it is easy to see that v 0 is a weak solution of (10.109).
By interior elliptic regularity, v C 2+ (j ). To complete the proof of the
existence, it suffices to show that v > 0. On the contrary, suppose v = 0.
Then, setting
vn
,
n n0 ,
vn :=
kvn kL (j )
we again have that
d2
vn
vn
in j
for all n n0 and hence, once again by Lemma 10.13, there exists
H01 (j ) L (j ) such that along some subsequence, re-labeled by n,
lim vn = weakly in H01 (j )
and strongly in
Lp (j ) p > 1.
Necessarily > 0, because k

vn kL (j ) = 1 for all n n0 . Moreover, dividing
the vn equation by kvn kL (j ) and letting n , it becomes apparent that
provides us with a weak bounded positive solution of
(d2 + c) = ,
which contradicts Lemma 10.31. Therefore, v > 0. The uniqueness of v can be
demonstrated from Lemma 10.31 by adapting the proof of the uniqueness in
Theorem 1.7.
Proof of Theorem 10.29: By Remark 5.3, Part (a) is an easy consequence
from Theorem 4.1, because b = 0. To prove Part (b) pick 1 j q0 and
324
suppose (10.5) possesses a meta-coexistence state, (Mu , Mv ), supported in j

such that Mv = 0 on j . Then, Mu is a metasolution of (10.7) supported in
j . Consequently, thanks to Theorem 4.5, < d1 j+1 . Moreover, Mu Xj
and Mv solves the problem

(d2 + cMu )v = v df2 (, v)v
in j ,
v=0
on j .
Thus, we find from Proposition 10.30 that
> 1 [d2 + cMu , j ] 1 [d2 + cMmin
[,j ] , j ],
because Mu Mmin
[,j ] . Therefore, the second estimate of (10.108) holds.
Conversely, suppose (10.108). Then, by Theorem 4.5, (10.7) possesses a
minimal metasolution Mmin

[,j ] Xj supported in j . Moreover, owing to
Proposition 10.30, the singular problem

(d2 + cMmin
in j ,
[,j ] )v = v df2 (, v)v
v=0
on j ,
has a positive solution. The proof is complete.
The relevance of the coexistence and meta-coexistence states constructed

in Theorem 10.29 relies on the fact that they provide us with the asymptotic
behavior of the positive solutions of (10.1) in the special case when b = 0.
Indeed, the next result is satisfied. The notations introduced in the previous
sections will be kept.
satisfy u0 > 0 and v0 > 0.
Theorem 10.32 Suppose b = 0 and u0 , v0 C0 ()
Let (u(x, t), v(x, t)) denote the unique classical solution of (10.1). Then, the
following properties are satisfied:
(a) If d1 0 < < d1 1 and 0 [d2 , c,u ], then
lim ku(, t) ,u kL () = 0,
lim kv(, t)kL () = 0.
(b) If d1 0 < < d1 1 and > 0 [d2 , c,u ], then

lim ku(, t) ,u kL () = 0,
lim kv(, t) {,v;c,u } kL () = 0.
(c) If there exists j {1, ..., q0 } such that

d1 j < d1 j+1
and
1 [d2 + cMmin
[,j ] ],
then
max
Mmin
[,j ] lim inf u(, t) lim sup u(, t) M[,j ]
t
and
lim kv(, t)kL () = 0.
(10.115)
325
(d) If there exists j {1, ..., q0 } such that

d1 j < d1 j+1
and
> 1 [d2 + cMmin

[,j ] ],
then (10.115) holds, and

{,v;cMmax
[,
j]
lim inf v(, t) lim sup v(, t) {,v;cMmin

t
[,j ]
where, given any Xj , {,v;c} stands for the extension by zero to

of the unique positive solution of

(d2 + c)v = v df2 (, v)v
in j ,
v=0
on j ,
whose existence is guaranteed by Proposition 10.30.
The proof of Theorem 10.32 follows readily by combining Theorem 5.2 and
Remark 5.3 with the next result whose proof is omitted. The technical details
can be reconstructed easily from the parabolic maximum principle. At this
stage of the book, the proofs should be routine.
Theorem 10.33 Under the same assumptions of Theorem 10.32, assume
that d1 j < d1 j+1 for some j {1, ..., q0 }, and let u(x, t) denote the
positive solution of the parabolic counterpart of (10.7). Lastly, let v := v(x, t)
be the (unique) positive solution of
v
in (0, ),
t d2 v + cu(x, t)v = v d(x)f2 (x, v)v
v=0
on (0, ),
v(, 0) = v0
in .
Then
{,v;cMmax
[,
10.8
j]
lim inf v(, t) lim sup v(, t) {,v;cMmin

t
[,j ]
}.
Existence of meta-coexistence states
This section reveals that the necessary and sufficient conditions given in Section 10.7 for a meta-coexistence state in the special case b = 0 are actually
sufficient when b(x) > 0 for all x .
Theorem 10.34 Suppose there exists j {1, ..., q0 } such that
< d1 j+1
and
> 1 [d2 + cMmin

[,j ] , j ].
(10.116)
Then, (10.5) possesses a meta-coexistence state (Mu , Mv ) supported in j

with Mv = 0 on j .
326
Condition (10.116) is optimal because, due to Theorem 10.29, it is necessary and sufficient if b = 0. When
min
Mmax
[,j ] = M[,j ] M[,j ]
is non-degenerate, the curve

= 1 [d2 + cM[,j ] , j ],
< d1 j+1 ,
provides us with the set of values of (, ) for which bifurcation to metacoexistence states, (Mu , Mv ), supported in j with Mv = 0 on j from
the semi-trivial metasolution (M[,j ] , 0) occurs. As this bifurcation can be
subcritical, (10.5) can indeed have a meta-coexistence state for
< 1 [d2 + bM[,j ] , j ].
Thus, in general (10.116) is not necessary for the existence of meta-coexistence
states supported in j with Mv = 0 on j .
Proof of Theorem 10.34: Suppose (10.116) holds for some j {1, ..., q0 }
and consider the nonlinear boundary value problems
d1 u = u af1 (, u)u buv
in j ,
d2 v = v df2 (, v)v cuv

u=n
on j ,
(10.117)
u
=
0
on
v=0
on j ,
where n N, n 1. For each integer n 1, let [,j ,n] denote the unique
positive solution of
in j ,
d1 u = u af1 (, u)u
u=n
on j ,
u=0
on j ,
whose existence and uniqueness are guaranteed by Theorem 4.5. The solutions
satisfy
[,j ,n] [,j ,n+1] Mmin
[,j ]
for all n 1 and, actually, thanks to Theorem 4.7, by elliptic regularity,
lim k[,j ,n] Mmin
[,j ] kC 2+ (j ) = 0.
Let > 0 be a principal eigenfunction of

j := 1 [d2 + cMmin
[,j ] , j ] < ,
whose existence is guaranteed by Lemma 10.31. We claim that there exists
0 > 0 such that for each 0 < < 0 and n N, n 1, the couple
(un , v) := ([,j ,n] , )
327
provides us with a supersolution of (10.117). Indeed, by definition,

d1 un = un af1 (, un )un un af1 (, un )un bun v,
because bun v 0. Moreover, un = n on j and un = 0 on j .
Thus, since = 0 on j , to prove the previous claim it suffices to show
that there exists 0 > 0 such that for each (0, 0 ) and n N, n 1, the
following estimate holds:
d2 () df2 (, ) c[,j ,n] .
(10.118)
Since (10.118) can be written in the form

min
(d2 + cMmin
[,j ] )() + c([,j ,n] M[,j ] ) df2 (, ),
and [,j ,n] Mmin

[,j ] for all n 1, it becomes apparent that (10.118) holds
provided
j df2 (, ).
(10.119)
As (10.119) is equivalent to
j df2 g(, ),
j < and f2 (, 0) = 0, there exists 0 > 0 such that (10.119) holds for all
(0, 0 ), which ends the proof of the claim above. Note that 0 can be
chosen sufficiently small so that
< ,v ,
0 < < 0 ,
(10.120)
where ,v is the unique positive solution of (10.8).

On the other hand, it is easily seen that
(u0 , v 0 ) := (0, ,v )
provides us with a subsolution of (10.117). Fix (0, 0 ). Thanks to (10.120),
(u0 , v 0 ) = (0, ,v ) <q (un , v) := ([,j ,n] , )
for all n 1. Thus, thanks to Theorem 10.23, the problem (10.117) with n = 1
has a minimal solution, (u1 , v1 ), in the interval [(u0 , v 0 ), (u1 , v)]. As 0 u1
and u1 = 1 on j , u1 > 0. Also v1 > 0, because v1 v = . Similarly,
(u1 , v 1 ) := (u1 , v1 ) provides us with a subsolution of (10.117) for n = 2 such
that
(u0 , v 0 ) q (u1 , v 1 ) q (u1 , v) q (u2 , v).
Consequently, by Theorem 10.23, we find that (10.117) with n = 2 has
a minimal solution, (u2 , v2 ), in the interval [(u1 , v1 ), (u2 , v)]. In particular,
(u1 , v1 ) (u2 , v2 ). By a simple induction argument, it is apparent that, for
328
every n 1, the problem (10.117) has a component-wise positive solution

(un , vn ) such that
(un , vn ) q (un+1 , vn+1 ) q (Mmin
[,j ] , )
for all n 1.
By monotonicity, the point-wise limits

u := lim un
n
and v := lim vn
n
are well defined. Moreover, u > 0 and v > 0. Finally, combining the
Lebesgue theorem on dominated convergence with some classical elliptic estimates, it is seen easily that indeed (u , v ) is a meta-coexistence state of
(10.5) supported in j with v = 0 on j . This ends the proof.
10.9
The non-spatial model

u0 = u Au2 Buv
v 0 = v Dv 2 Buv
(10.121)
with A > 0, B > 0, C > 0 and D > 0, was introduced by V. Volterra [229] and
independently by A. J. Lotka [185]. The first general results for its diffusive
counterpart,
t u d1 u = u Au2 Buv
in (0, ),
t v d2 v = v Dv 2 Buv
(10.122)
u=v=0
on (0, ),
u(, 0) = u0 0 and v(, 0) = v0 0

in ,
were given by J. Blat and K. J. Brown [29], where the global bifurcation theorem of P. H. Rabinowitz [214] was invoked to construct the first global continua
of coexistence states in the context of reaction-diffusion systems. Indeed, J.
Blat and K. J. Brown [29] established the existence of a component of coexistence states linking the two semi-trivial positive solutions. This methodology
had been previously used by J. Cushing [63] in the periodic LotkaVolterra
non-spatial model.
Some time later, these pioneering findings were substantially sharpened
and generalized to cover wide classes of spatially heterogeneous competing
species models by the author in [142], [149], where the global unilateral theory
of P. H. Rabinowitz [214] was revised and updated to get the most general
existence result for two species systems, Theorem 7.2.2 of [149], which is valid
for a very general class of nonlinear elliptic spatially heterogeneous systems:
L1 u = u + f (x, u)u + F (x, u, v)uv

in ,
L2 v = v + g(x, v)v + G(x, u, v)uv
(10.123)
B1 u = B2 v = 0
on ,
329
where Lk , k = 1, 2, are linear second order elliptic operators like (1.48) and
B1 and B2 are two general boundary operators of the mixed general type
(1.49). Theorem 7.2.2 of [149] has been the main bifurcation theorem used to
show the existence of coexistence states in this chapter.
Other closely related papers analyzing the effects of spatial heterogeneities
on the dynamics of competing species models are by J. Lopez-Gomez [143], V.
Hutson at al. [120] and J. Lopez-Gomez and J. C. Sabina [180]. The authors
analyzed the effects of vanishing the coefficient functions b(x) and c(x) on the
dynamics of (10.1). The patches where b(x) vanishes are protection zones for
the individuals of the species u, while the components of c1 (0) are protection zones for those of the species v. When the protection zones of each of the
competitors can maintain the species in isolation, there is permanence independent of the intensity of the competition. Consequently, as observed often in
empirical studies, the paradigm inherent to the principle of competitive exclusion is false in heterogeneous habitats. The reader is sent to J. Lopez-Gomez
[145], S. Cano-Casanova and J. Lopez-Gomez [36] and J. Lopez-Gomez and
M. Molina-Meyer [166], [168][172] for refinements and a detailed discussion
on the implications of these features in population dynamics. Undoubtedly,
[143], [120] and [180] constitute a paradigm of the discussion by R. S. Cantrell
and C. Cosner in Section 6.2.1 of [43], How Spatial Segregation May Facilitate
Coexistence.
The model (10.1) was introduced in J. Lopez-Gomez [153], whose materials
have been considerably polished in this chapter. It was the first time that the
carrying capacity of one of the competitors was allowed to be infinity on some
patches of the inhabiting area. Almost simultaneously, Y. Du submitted [77],
where the same problem in the very special case when 0 is connected was
dealt with.
From reading Chapter 10, it becomes apparent that the simplest way to
overcome most of the technical difficulties when 0 has an arbitrary number
of components, as has been always the case throughout this book, is assuming
that 0 is connected. Naturally, if 0 consists of a single component, then the
hierarchies of the protection zones, according to their relative sizes measured
by the underlying principal eigenvalues, cannot change as the parameters
and increase, as occurs when 0 has an arbitrary number of components.
Most precisely, when 0 is connected, the analysis carried out in Section 10.2
does not make any sense, while in the case of two or more components it is a
categorical imperative. It actually comes as a bolt out of the blue.
Bibliography
[1] S. Alama and G. Tarantello, On semilinear elliptic equations with indefinite nonlinearities, Calc. Var. Part. Diff. Eqns. 1 (1993), 439475.
[2] S. Alama and G. Tarantello, Elliptic problems with nonlinearities indefinite in sign, J. Funct. Anal. 141 (1996), 159215.
[3] S. Alarc
on, G. Daz, R. Letelier and J. M. Rey, Expanding the asymptotic explosive boundary behaviour of large solutions to a semilinear
elliptic equation, Nonl. Anal. 72 (2010), 24262443.
[4] S. Alarc
on, G. Daz and J. M. Rey, The influence of sources terms on the
boundary behaviour of the large solutions of quasilinear elliptic equations; the power-like case, Z. Angew. Math. Phys. DOI 10.1007/s00033012-0253-5, 2012.
[5] S. Alarc
on, J. Garca-Melian and A. Quaas, Keller-Osserman type conditions for some elliptic problems with gradient terms, J. Diff. Eqns.
252 (2012), 886914.
[6] D. Aleja and J. L
opez-Gomez, Some paradoxical effects of advection on
a class of diffusive equations in Ecology, Disc. Cont. Dyn. Systems B
19 (2014) 10, 30313056.
opez-Gomez, Concentration through large advection,
J. Diff. Eqns. 257 (2014), 31353164.
opez-Gomez, Dynamics of a class of advective-diffusive
equations in ecology, Adv. Nonl. Studies 15 (2015), 557585.
[9] P. Alvarez-Caudevilla
and J. Lopez-Gomez, Metasolutions in cooperative systems, Nonl. Anal. R.W.A. 9 (2008), 11191157.
[10] H. Amann, On the existence of positive solutions of nonlinear elliptic
boundary value problems, Ind. Univ. Math. J., 21 (1971), 125146.
[11] H. Amann, Fixed point equations and nonlinear eigenvalue problems in
ordered Banach spaces, SIAM Review 18 (1976), 620709.
[12] H. Amann, Maximum principles and principal eigenvalues, in Ten Mathematical Essays on Approximation in Analysis and Topology (J. Ferrera,
331
332
Bibliography
J. L
opez-G
omez and F. R. Ruiz del Portal Eds.), pp. 160, Elsevier B.
V., Amsterdam, 2005.
[13] H. Amann and J. L

opez-Gomez, A priori bounds and multiple solutions
for superlinear indefinite elliptic problems, J. Diff. Eqns. 146 (1998),
336374.
[14] A. Ambrosetti and J. L. Gamez, Branches of positive solutions for some
semilinear Schr
odinger equations, Math. Z. 224 (1997), 347362.
[15] C. Aneda and G. Porru, Boundary estimates for solutions of weighted
semilinear elliptic equations, Discrete Contin. Dyn. Syst. 32 (2012),
3801-3817.
[16] C. Bandle and M. Marcus, Sur les solutions maximales de problemes elliptiques nonlineaires: bornes isoperimetriques et comportement asymptotique, C. R. Acad. Sci. Paris 1 311 (1990), 9193.
[17] C. Bandle and M. Marcus, Large solutions of semilinear elliptic equations: Existence, uniqueness and asymptotic behavior, J. DAnalysis
Math. 58 (1992), 924.
[18] C. Bandle and M. Marcus, On second order effects in the boundary behavior of large solutions of semilinear elliptic problems, Diff. Int. Eqns.
11 (1998), 2334.
[19] J. Bao, X. Ji and H. Li, Existence and nonexistence theorem for entire
subsolutions of k-Yamabe type equations, J. Diff. Eqns. 253 (2012),
21402160.
[20] P. Baras and L. Cohen, Complete blow-up after Tmax for the solution of
a semilinear heat equation, J. Funct. Anal. 71 (1987), 142174.
[21] F. Belgacem and C. Cosner, The effects of dispersal along environmental gradients on the dynamics of populations in heterogeneous environments, Can. Appl. Math. Quart. 3 (1995), 379397.
[22] N. Belhaj Rhouma, A. Drissi and W. Sayeb, Existence and asymptotic
behavior of boundary blow-up weak solutions for problems involving the
p-Laplacian, J. Partial Differ. Equ. 26 (2013), 172192.
[23] H. Berestycki, I. Capuzzo-Dolcetta and L. Nirenberg, Superlinear indefinite elliptic problems and nonlinear Liouville theorems, Top. Meth.
Nonl. Anal. 4 (1994), 5978.
[24] H. Berestycki, I. Capuzzo-Dolcetta and L. Nirenberg, Variational methods for indefinite superlinear homogeneous elliptic problems, Nonl. Diff.
Eqns. Appl. 2 (1995), 553572.
Bibliography
333
[25] H. Berestycki and P. L. Lions, Some applications of the method of super

and subsolutions. Bifurcation and nonlinear eigenvalue problems (Proc.,
Session, Univ. Paris XIII, Villetaneuse, 1978), pp. 16-41, Lecture Notes
in Mathematics 782, Springer, Berlin, 1980.
[26] M. D. Bertness and R. M. Callaway, Positive interactions in communities, Trends in Ecology and Evolution 9 (1994), 191193.
[27] M. Bertsch and R. Rostamian, The principle of linearized stability for
a class of degenerate diffusion equations, J. Diff. Eqns. 57 (1985), 373
405.
[28] L. Bieberbach, u = eu und die automorphen Funktionen, Math. Annalen 77 (1916), 173212.
[29] J. Blat and K. J. Brown, Bifurcation of steady state solutions in
predator-prey and competition systems, Proc. Royal Soc. Edinburgh
97A (1984), 2134.
[30] M. M. Boureanu, Uniqueness of singular radial solutions for a class of
quasilinear problems, Bull. Belg. Math. Soc. Simon Stevin 16 (2009),
665-685.
[31] H. Brezis and L. Oswald, Remarks on sublinear elliptic equations, Nonl.
Anal. 10 (1986), 5564.
[32] H. Browder, Reflection on the future of mathematics, Notices of the
AMS 49 (2002), 658662.
[33] R. M. Callaway and L. R. Walker, Competition and facilitation: A synthetic approach to interactions in plant communities, Ecology 78 (1997),
19581965.
[34] S. Cano-Casanova, Decay rate at infinity of the positive solutions of
a generalized class of Thomas-Fermi equations, 8th AIMS Conference,
Discrete Contin. Dyn. Syst. Suppl. Vol. I (2011), 240249.
[35] S. Cano-Casanova and J. Lopez-Gomez, Properties of the principal
eigenvalues of a general class of non-classical mixed boundary value
problems, J. Diff. Eqns. 178 (2002), 123211.
[36] S. Cano-Casanova and J. Lopez-Gomez, Permanence under strong aggressions is possible, Ann. I. H. Poincare Anal. Nonlin. 20 (2003),
9991041.
[37] S. Cano-Casanova and J. Lopez-Gomez, Varying boundary conditions
in a general class of elliptic problems of mixed type, Nonl. Anal. 55
(2003), 4772.
334
Bibliography
[38] S. Cano-Casanova and J. Lopez-Gomez, Varying domains in a general

class of sublinear elliptic problems, El. J. Diff. Eqns. 74 (2004), 141.
[39] S. Cano-Casanova and J. Lopez-Gomez, Existence, uniqueness and blowup rate of large solutions for a canonical class of one-dimensional problems on the half line, J. Diff. Eqns. 244 (2008), 31803203.
[40] S. Cano-Casanova and J. Lopez-Gomez, Blow-up rates of radially symmetric large solutions, J. Math. Anal. Appl. 352 (2009), 166174.
[41] S. Cano-Casanova, J. Lopez-Gomez and M. Molina-Meyer, Permanence
through spatial segregation in heterogeneous competition, in Proceedings
of the 9th IEEE International Conference on Methods and Models in
Automation and Robotics, Miedzyzdroje, pages 123130. University of
Szczecin, Poland, 2003.
[42] S. Cano-Casanova, J. Lopez-Gomez and M. Molina-Meyer, Isolas: Compact solution components separated away from a given equilibrium
curve, Hiroshima Math. J. 34 (2004), 177-199.
[43] R. S. Cantrell and C. Cosner, Spatial Ecology via Reaction-Diffusion
Equations, Wiley, New York, 2003.
[44] L. Chen, Y. Chen and D. Luo, Boundary blow-up solutions for a cooperative system involving the p-Laplacian, Ann. Polon. Math. 109 (2013),
297310.
[45] Y. Chen and M. Wang, Large solutions for quasilinear elliptic equation
with nonlinear gradient term, Nonl. Anal. Real World Appl. 12 (2011),
455463.
[46] Y. Chen and M. Wang, Boundary blow-up solutions for elliptic equations
with gradient terms and singular weights: Existence, asymptotic behaviour and uniqueness, Proc. Roy. Soc. Edinburgh Sect. A 141 (2011),
717737.
[47] Y. Chen and M. Wang, Boundary blow-up solutions for p-Laplacian
elliptic equations of logistic type, Proc. Roy. Soc. Edinburgh Sect. A
142 (2012), 691714.
[48] Y. Chen and M. Wang, Boundary blow-up solutions of p-Laplacian elliptic equations with a weakly superlinear nonlinearity, Nonl. Anal. Real
World Appl. 14 (2013), 15271535.
[49] Y. Chen, P. Y. H. Pang and M. Wang, Blow-up rates and uniqueness of
large solutions for elliptic equations with nonlinear gradient term and
singular or degenerate weights, Manuscripta Math. 141 (2013), 171193.
[50] Y. Chen and Y. Zhu, Large solutions for a cooperative elliptic system
of p-Laplacian equations, Nonl. Anal. 73 (2010), 450457.
Bibliography
335
[51] Y. Chen, Y. Zhu and R. Hao, Large solutions with a power nonlinearity
given by a variable exponent for p-Laplacian equations, Nonl. Anal. 110
(2014), 130140.
[52] M. Chuaqui, C. Cortazar, M. Elgueta, C. Flores, J. Garca-Melian, On
an elliptic problem with boundary blow-up and a singular weight: The
radial case, Proc. Roy. Soc. Edinburgh 133A (2003), 12831297.
[53] M. Chuaqui, C. Cortazar, M. Elgueta, J. Garca-Melian, Uniqueness and
boundary behaviour of large solutions to elliptic problems with singular
weights, Comm. Pure Appl. Anal. 3 (2004), 653662.
[54] F. C. Cirstea and V. Radulescu, Existence and uniqueness of blow-up
solutions for a class of logistic equations, Comm. Contemp. Math. 4
(2002), 559586.
[55] F. C. Cirstea and V. Radulescu, Uniqueness of the blow-up boundary
solution of logistic equation with absorbtion, C. R. Acad. Sci. Paris 1
335 (2002), 447452.
[56] F. C. Cirstea and V. Radulescu, Asymptotics for the blow-up boundary
solution of the logistic equation with absorption, C. R. Acad. Sci. Paris
1 336 (2003), 231236.
[57] F. C. Cirstea and V. Radulescu, Solutions with boundary blow-up for
a class of nonlinear elliptic problems, Houston J. Math. 29 (2003),
821829.
[58] J. H. Connell, On the prevalence and relative importance of interspecific
competition: evidence from field experiments, Amer. Natur. 122 (1983),
661696.
[59] M. G. Crandall and P. H. Rabinowitz, Bifurcation from simple eigenvalues, J. Funct. Anal. 8 (1971), 321340.
[60] M. G. Crandall and P. H. Rabinowitz, Bifurcation, pertubation from
simple eigenvalues and linearized stability, Arch. Rat. Mech. Anal. 52
(1973), 161180.
[61] R. Courant and D. Hilbert, Methods of Mathematical Physics Vol. I,
Wiley, New York, 1962.
[62] R. Cui, J. Shi and B. Wu, Strong Allee effect in a diffusive predator-prey
system with a protection zone, J. Diff. Eqns. 256 (2014), 108129.
[63] J. Cushing, Two species competition in a periodic environment, J. Math.
Biol. 10 (1980), 385390.
[64] E. N. Dancer, The effects of domain shape on the number of positive
solutions of certain nonlinear equations II, J. Diff. Eqns. 87 (1990),
316339.
336
Bibliography
[65] E. N. Dancer, A counterexample on competing species equations, Diff.

Int. Eqns. 9 (1996), 239246.
[66] E. N. Dancer and Y. Du, Effect of certain degeneracies in the predatorprey model, SIAM J. Math. Anal. 34 (2002), 292314.
[67] E. N. Dancer and J. Lopez-Gomez, Semiclassical analysis of general
second order elliptic operators on bounded domains, Trans. Amer. Math.
Soc. 352 (2000), 37233742.
[68] D. Daners and P. Koch-Medina, Abstract Evolution Equations, Periodic
Problems and Applications, Longman Scientific, Harlow, Essex, 1992.
[69] Ch. Darwin, On the Origin of Species by Means of Natural Selection,
John Murray, London, 1859.
[70] M. A. del Pino, Positive solutions of a semilinear elliptic equation on a
compact manifold, Nonlinear Anal. 22 (1994), 14231430.
[71] M. A. del Pino and R. Letelier, The influence of domain geometry in
boundary blow-up elliptic problems, Nonl. Anal. 48 (2002), 897904.
[72] M. Delgado, J. L
opez-Gomez and A. Suarez, On the symbiotic Lotka
Volterra model with diffusion and transport effects, J. Diff. Eqns. 160
(2000), 175262.
opez-Gomez and A. Suarez, Nonlinear versus linear
diffusion: From classical solutions to metasolutions, Adv. Diff. Eqns. 7
(2002), 11011124.
opez-Gomez and A. Suarez, Characterizing the existence of large solutions for a class of sublinear problems with nonlinear
diffusion, Adv. Diff. Eqns. 7 (2002), 12351256.
opez-Gomez and A. Suarez, Singular boundary value
problems of a porous media logistic equation, Hiroshima J. Maths. 34
(2004), 5780.
[76] G. Daz and R. Letelier, Explosive solutions of quasilinear elliptic equations: Existence and uniqueness, Nonl. Anal. 20 (1993), 97125.
[77] Y. Du, Effects of a degeneracy in the competition model: Classical and
generalized steady-state solutions, J. Diff. Eqns. 181 (2002), 92132.
[78] Y. Du, Spatial patterns for population models in a heterogeneous environment, Taiwanese J. Math. 8 (2004), 155182.
[79] Y. Du, Order Structure and Topological Methods in Nonlinear Partial
Differential Equations, World Scientific Publishing, Singapore, 2006.
Bibliography
337
[80] Y. Du and Q. Huang, Blow-up solutions for a class of semilinear elliptic

and parabolic equations, SIAM J. Math. Anal. 31 (1999), 118.
[81] S. Dumont, L. Dupaigne, O. Goubet and V. Radulescu, Back to the
Keller-Osserman condition for boundary blow-up solutions, Adv. Nonl.
Studies 7 (2007), 271298.
[82] J. C. Eilbeck, J. E. Furter and J. Lopez-Gomez, Coexistence in the
competition model with diffusion, J. Diff. Eqns. 107 (1994), 96139.
[83] C. Faber, Beweis das unter allen homogenen Membranen von gleicher
Fl
ache und gleicher Spannung die kreisdormige den tiefsten Grundton
gibt, Sitzungsber. Bayer. Akad. der Wiss. Math. Phys. (1923), 169171.
[84] G. Feltrin and F. Zanolin, Multiple positive solutions for a superlinear
problem: A topological approach, J. Diff. Eqns., in press.
[85] H. Feng and C. Zhong, Boundary behavior of solutions for the degenerate logistic type elliptic problem with boundary blow-up, Nonl. Anal.
73 (2010), 34723478.
[86] P. Feng, Remarks on large solutions of a class of semilinear elliptic equations, J. Math. Anal. Appl. 356 (2009), 393404.
[87] A. Fick, Ueber Diffusion, Annalen der Physik XCII (1855), 5986.
[88] R. A. Fisher, The wave of advances of advantageous genes, Ann. Eugen.
7 (1937), 355369.
[89] J. B. J. Fourier, Theorie Analytique de la Chaleur, Firmin Didot, Paris,
1822.
[90] J. M. Fraile, P. Koch-Medina, J. Lopez-Gomez and S. Merino, Elliptic
eigenvalue problems and unbounded continua of positive solutions of a
semilinear equation, J. Diff. Eqns. 127 (1996), 295319.
[91] J. E. Furter and J. L
opez-Gomez, On the existence and uniqueness of coexistence states for the Lotka-Volterra competition model with diffusion
and spatially dependent coefficients, Nonl. Anal. 25 (1995), 363398.
[92] J. E. Furter and J. Lopez-Gomez, Diffusion mediated permanence problem for an heterogeneous LotkaVolterra competition model, Proc. Roy.
Soc. Edinburgh 127A (1997), 281336.
[93] M. Gaudenzi, P. Habets and F. Zanolin, Positive solutions of superlinear
boundary value problems with singular indefinite weight, Comm. Pure
Appl. Anal. 2 (2003), 403414.
[94] M. Gaudenzi, P. Habets and F. Zanolin, An example of superlinear
problem with multiple positive solutions, Atti Sem. Fis. Univ. Modena
LI (2003), 259272.
338
Bibliography
[95] J. Garca-Meli
an, Nondegeneracy and uniqueness for boundary blow-up
elliptic problems, J. Diff. Eqns. 223 (2006), 208227.
[96] J. Garca-Meli
an, Uniqueness for boundary blow-up problems with continuous weights, Proc. Amer. Math. Soc. 135 (2007) 27852793.
[97] J. Garca-Meli
an, Boundary behavior for large solutions of elliptic equations with singular weights, Nonl. Anal. 67 (2007) 818826.
[98] J. J. Garca-Meli
an, Uniqueness of positive solutions for a boundary
value problem, J. Math. Anal. Appns. 360 (2009), 530536.
[99] J. J. Garca-Meli
an, Multiplicity of positive solutions to boundary blowup elliptic problems with sign-changing weights, J. Funct. Anal. 261
(2011), 17751798.
[100] J. Garca-Meli
an, R. Gomez-Re
nasco, J. Lopez-Gomez and J. C. Sabina
de Lis, Point-wise growth and uniqueness of positive solutions for a class
of sublinear elliptic problems where bifurcation from infinity occurs,
Arch. Rat. Mech. Anal., 145 (1998), 261289.
[101] J. Garca-Meli
an, R. Letelier and J. C. Sabina de Lis, Uniqueness and
asymptotic behaviour for solutions of semilinear problems with boundary blow-up, Proc. Amer. Math. Soc. 129 (2001), 35933602.
[102] J. Garca-Meli
an, C. Morales-Rodrigo, J. D. Rossi and A. Suarez, Nonnegative solutions to an elliptic problem with nonlinear absorption and
a nonlinear incoming flux on the boundary, Ann. Mat. Pura Appl. 187
(2008), 459486.
[103] D. Gilbarg and N. S. Trudinger, Elliptic Partial Differential Equations
of Second Order, Springer, Berlin, 2001.
[104] M. Golubitsky and D. G. Shaeffer, Singularity and Groups in Bifurcation
Theory, Springer, Berlin, 1985.
[105] R. G
omez-Re
nasco, The Effect of Varying Coefficients in Semilinear
Elliptic Boundary Value Problems: From Classical Solutions to Metasolutions, Ph.D. dissertation, University of La Laguna, Tenerife, Spain,
March 1999.
[106] R. G
omez-Re
nasco and J. Lopez-Gomez, The effect of varying coefficients on the dynamics of a class of superlinear indefinite reaction diffusion arising in population dynamics, (unpublished) preprint 1999.
[107] R. G
omez-Re
nasco and J. Lopez-Gomez, The effect of varying coefficients on the dynamics of a class of superlinear indefinite reaction diffusion equations, J. Diff. Eqns. 167 (2000), 3672.
Bibliography
339
[108] R. G
omez-Re
nasco and J. Lopez-Gomez, The uniqueness of the stable
positive solution for a class of superlinear indefinite reaction diffusion
equations, Diff. Int. Eqns., 14 (2001), 751768.
[109] R. G
omez-Re
nasco and J. Lopez-Gomez, On the existence and numerical computation of classical and non-classical solutions for a family of
elliptic boundary value problems, Nonlinear Anal. 48 (2002), 567605.
[110] G. Green, An Essay of the Applications of Mathematical Analysis to
the Theory of Electricity and Magnetism, Bromley House, Nottingham,
1828.
[111] J. Hadamard, Memoires sur le problème danalyse relatif à lequilibre
des plaques elastiques encastrees, Mem. Acad. Sci. Paris 39 (1908),
128259.
[112] D. Henry, Geometric Theory of Semilinear Parabolic Equations, Lectures Notes in Mathematics 840, Springer, Berlin, 1981.
[113] E. Hopf, A remark on linear elliptic differential equations of the second
order, Proc. Amer. Math. Soc. 3 (1952), 791793.
[114] S. Huang and Q. Tian, Boundary blow-up rates of large solutions for elliptic equations with convection terms, J. Math. Anal. Appl. 373 (2011),
3043.
[115] S. Huang, Q. Tian and Y. Mi, General uniqueness results and blow-up
rates for large solutions of elliptic equations, Proc. Roy. Soc. Edinburgh
Sect. A 142 (2012), 825837.
[116] S. Huang, Q. Tian, S. Zhang and J. Xi, A second-order estimate for
blow-up solutions of elliptic equations, Nonl. Anal. 74 (2011), 2342
2350.
[117] S. Huang, Q. Tian, S. Zhang, J. Xi and Z. Fan, The exact blow-up rates
of large solutions for semilinear elliptic equations, Nonl. Anal. 73 (2010),
34893501.
[118] S. Huang, W. T. Li, Q. Tian and C. Mu, Large solution to nonlinear elliptic equation with nonlinear gradient terms, J. Diff. Eqns. 251 (2011),
32973328.
[119] G. E. Hutchinson, The Ecological Theater and the Evolutionary Play,
Yale University Press, New Haven, 1965.
[120] V. Hutson, J. L
opez-Gomez, K. Mischaikow and G. Vickers, Limit behaviour for a competing species problem with diffusion, World Scientific
Series Anal. Appl. 4 (1995), 343358.
340
Bibliography
[121] X. Ji and J. Bao, Necessary and sufficient conditions on solvability for

Hessian inequalities, Proc. Amer. Math. Soc. 138 (2010), 175188.
[122] T. Kato, Perturbation Theory for Linear Operators, Springer, Berlin,
1995.
[123] J. B. Keller, On solutions of u = f (u), Comm. Pure Appl. Math. X
(1957), 503510.
[124] A. N. Kolmogorov, I. G. Petrovsky and N. S. Piskunov, Etude de
lequation de la diffusion avec croissance de la quantite de matière et
son application `
a un problème biologique, Bull. Math. Univ. dEtat `
a
Moscow (Sèr. Int.) A 1 (1937), 1129.
[125] V. A. Kondratiev, and V. A. Nikishin, Asymptotics, near the boundary, of a solution of a singular boundary value problem for a semilinear
elliptic equation, Diff. Eqns. 26 (1990), 345348.
[126] E. Krahn, Uber

eine von Rayleigh formulierte Minimaleigenschaft des
Kreises, Math. Ann. 91 (1925), 97100.
[127] O. A. Ladyzenskaja, V. A. Solonnikov and N. N. Uraltzeva, Linear and
Quasilinear Equations of Parabolic Type, American Mathematical Society, Providence, 1968.
[128] M. Langlais, and D. Phillips, Stabilization of solutions of nonlinear and
degenerate evolution equations, Nonl. Anal. 9 (1985), 321333.
[129] A. C. Lazer and P. J. McKenna, On a problem of L. Bieberbach and H.
Rademacher, Nonl. Anal. 21 (1993), 327335.
[130] A. C. Lazer and P. J. McKenna, A singular elliptic boundary value
problem, Appl. Math. Comp. 65 (1994), 183194.
[131] A. C. Lazer and P. J. McKenna, Asymptotic behaviour of solutions of
boundary blow-up problems, Diff. Int. Eqns., 7 (1994), 10011019.
[132] W. Lei and M. Wang, Existence of large solutions of a class of quasilinear elliptic equations with singular boundary, Acta Math. Hungar. 129
(2010), 8195.
[133] Y. Liang, Q. Zhang and C. Zhao, On the boundary blow-up solutions of
p(x)-Laplacian equations with gradient terms, Taiwanese J. Math. 18
(2014), 599632.
[134] L. Li and R. Logan, Positive solutions to general elliptic LotkaVolterra
models, Diff. Int. Eqns. 4 (1991), 817834.
[135] H. Li, P. Y. H. Pang and M. Wang, Boundary blow-up solutions for
logistic-type porous media equations with nonregular source, J. Lond.
Math. Soc. 80 (2009), 273294.
Bibliography
341
[136] H. Li, P. Y. H. Pang and M. Wang, Boundary blow-up of a logistic-type

porous media equation in a multiply connected domain, Proc. Roy. Soc.
Edinburgh Sect. A 140 (2010), 101117.
[137] C. Liu and Z. Yang, Boundary blow-up quasilinear elliptic problems
with nonlinear gradient terms, Complex Var. Elliptic Eqns. 57 (2012),
687704.
[138] C. Loewner and L. Nirenberg, Partial differential equations invariant
under conformal or projective transformations, in: L. V. Ahlfors, I. Kra,
B. Maskit, L. Nirenberg (Eds.), Contributions to Analysis, pp. 245272,
Academic Press, New York, 1974.
[139] R. Logan and A. Ghoreishi, Positive solutions of a class of biological models in a heterogeneous environment, Bull. Aust. Math. Soc. 44
(1991), 7994.
[140] J. L
opez-G
omez, Positive periodic solutions of LotkaVolterra reactiondiffusion systems, Diff. Int. Eqns. 5 (1992), 5572.
[141] J. L
opez-G
omez, On linear weighted boundary value problems, in Partial Differential Equations; Models in Physics and Biology, pages 188
203, Akademie Verlag, Berlin, 1994.
[142] J. L
opez-G
omez, Nonlinear eigenvalues and global bifurcation: Application to the search of positive solutions for general LotkaVolterra reaction diffusion systems with two species, Diff. Int. Eqns. 7 (1994),
14271452.
[143] J. L
opez-G
omez, Permanence under strong competition, World Scientific Series Anal. Appl. 4 (1995), 473488.
[144] J. L
opez-G
omez, The maximum principle and the existence of principal
eigenvalues for some linear weighted boundary value problems, J. Diff.
Eqns. 127 (1996), 263294.
[145] J. L
opez-G
omez, Strong competition with refuges, Nonl. Anal. 30
(1997), 51675178.
[146] J. L
opez-G
omez, On the existence of positive solutions for some indefinite superlinear elliptic problems, Comm. Part. Diff. Eqns. 22 (1997),
17871804.
[147] J. L
opez-G
omez, Large solutions, metasolutions, and asymptotic behaviour of the regular positive solutions of sublinear parabolic problems,
El. J. Diff. Eqns. Conf. 05 (2000), 135171.
[148] J. L
opez-G
omez, Varying bifurcation diagrams of positive solutions for
a class of indefinite superlinear boundary value problems, Trans. Amer.
Math. Soc. 352 (2000), 18251858.
342
Bibliography
[149] J. L
opez-G
omez, Spectral Theory and Nonlinear Functional Analysis,
Research Notes in Mathematics 426, Chapman & Hall/CRC, Boca Raton, 2001.
[150] J. L
opez-G
omez, Approaching metasolutions by classical solutions, Diff.
Int. Eqns. 14 (2001), 739750.
[151] J. L
opez-G
omez, Metasolutions, Bol. Soc. Esp. Math. Appl. 24 (2003),
5990.
[152] J. L
opez-G
omez, Classifying smooth supersolutions in a general class of
elliptic boundary value problems, Adv. Diff. Eqns. 8 (2003), 10251042.
[153] J. L
opez-G
omez, Coexistence and metacoexistence states in competing
species models, Houston J. Math. 29 (2003), 485538.
[154] J. L
opez-G
omez, The boundary blow-up rate of large solutions, J. Diff.
Eqns. 195 (2003), 2545.
[155] J. L
opez-G
omez, Dynamics of parabolic equations: From classical solutions to metasolutions, Diff. Int. Eqns. 16 (2003), 813828.
[156] J. L
opez-G
omez, Varying stoichiometric exponents I: Classical steadystates and metasolutions, Adv. Nonl. Studies 3 (2003), 327354.
[157] J. L
opez-G
omez, Global existence versus blow-up in superlinear indefinite parabolic problems, Sci. Math. Jap. e-2004 (2004), 449472.
[158] J. L
opez-G
omez, Global existence versus blow-up in superlinear indefinite parabolic problems, Sci. Math. Jap. 61 (2005), 493517.
[159] J. L
opez-G
omez, Uniqueness of large solutions for a class of radially symmetric elliptic equations, Spectral Theory and Nonlinear Analysis with
Applications to Spatial Ecology (S. Cano-Casanova, J. Lopez-Gomez and
C. Mora-Corral, Eds.), 75110, World Scientific, Singapore, 2005.
[160] J. L
opez-G
omez, Metasolutions: Malthus versus Verhulst in population
dynamics: A dream of Volterra, in Handbook of Differential Equations,
Stationary Partial Differential Equations (M. Chipot and P. Quittner
eds.), pages 211309, Elsevier, Amsterdam, 2005.
[161] J. L
opez-G
omez, Optimal uniqueness theorems and exact blow-up rates
of large solutions, J. Diff. Eqns. 224 (2006), 385439.
[162] J. L
opez-G
omez, Uniqueness of radially symmetric large solutions, Disc.
Cont. Dynam. Systems, Proceedings of Sixth AIMS Conference, Supplement 2007, 677686.
[163] J. L
opez-G
omez, Linear Second Order Elliptic Operators, World Scientific, Singapore, 2013.
Bibliography
343
[164] J. L
opez-G
omez and L. Maire, Uniqueness of large positive solutions for
a class of radially symmetric cooperative systems, submitted.
[165] J. L
opez-G
omez and M. Molina-Meyer, The maximum principle for cooperative weakly coupled elliptic systems and some applications, Diff.
Int. Eqns. 7, (1994), 383398.
[166] J. L
opez-G
omez and M. Molina-Meyer, Singular perturbations in economy and ecology: The effect of strategic symbiosis in random competitive
environments, Adv. Math. Sci. Appl. 14 (2004), 87107.
[167] J. L
opez-G
omez and M. Molina-Meyer, Bounded components of positive
solutions of abstract fixed point equations: mushrooms, loops and isolas,
J. Diff. Eqns. 209 (2005), 416441.
[168] J. L
opez-G
omez and M. Molina-Meyer, The competitive exclusion principle versus biodiversity through segregation and further adaptation to
spatial heterogeneities, Th. Pop. Biol. 69 (2006), 94109.
[169] J. L
opez-G
omez and M. Molina-Meyer, In the blink of an eye, in Progress
in Nonlinear Partial Differential Equations and Their Applications 64,
pages 291327, Birkhauser, Basel, 2005.
[170] J. L
opez-G
omez and M. Molina-Meyer, Cooperation and competition,
strategic alliances, and the Cambrian explosion, in Spectral Theory and
Nonlinear Analysis with Applications to Spatial Ecology, pages 111126,
World Scientific, Singapore, 2005.
[171] J. L
opez-G
omez and M. Molina-Meyer, Superlinear indefinite systems:
beyond LotkaVolterra models, J. Diff. Eqns. 221 (2006), 343411.
[172] J. L
opez-G
omez and M. Molina-Meyer, Biodiversity through coopetition, Disc. Cont. Dyn. Syst. B 8(1) (2007), 187205.
[173] J. L
opez-G
omez, M. Molina-Meyer and A. Tellini, The uniqueness of
the linearly stable positive solution for a class of super-linear indefinite
problems with non-homogeneous boundary conditions, J. Diff. Eqns.,
255 (2013), 503523.
[174] J. L
opez-G
omez, M. Molina-Meyer and A. Tellini, Intricate bifurcation
diagrams for a class of one-dimensional superlinear indefinite problems
of interest in population dynamics, Disc. Cont. Dyn. Syst. Supplement
2013, 515524.
[175] J. L
opez-G
omez, M. Molina-Meyer and A. Tellini, Complex dynamics
caused by facilitation in competitive environments within polluted habitat patches, Eur. J. Appl. Math. 25 (2014), 213229.
344
Bibliography
[176] J. L
opez-G
omez, M. Molina-Meyer and A. Tellini, Spiraling bifurcation
diagrams in superlinear indefinite problems, Disc. Cont. Dyn. Syst. 35
(2015), 15611588.
[177] J. L
opez-G
omez and P. Quittner, Complete energy blow-up in indefinite
superlinear parabolic problems, Disc. Cont. Dyn. Syst. 14 (2006), 169
186.
[178] J. L
opez-G
omez and P. H. Rabinowitz, Nodal solutions for a class of
degenerate boundary value problems, Adv. Nonl. Studies, 15 (2015),
253-288.
[179] J. L
opez-G
omez and P. H. Rabinowitz, The effects of spatial heterogeneities on some multiplicity results, Disc. Cont. Dyn. Syst., 36 (2016),
in press.
[180] J. L
opez-G
omez and J. C. Sabina de Lis, Coexistence states and global
attractivity for some convective diffusive competing species models,
Trans. Amer. Math. Soc. 347 (1995), 37973833.
[181] J. L
opez-G
omez and J. C. Sabina de Lis, First variations of principal
eigenvalues with respect to the domain and point-wise growth of positive
solutions for problems where bifurcation from infinity occurs, J. Diff.
Eqns. 148 (1998), 4764.
[182] J. L
opez-G
omez and A. Suarez, Combining fast, linear and slow diffusion, Topol. Meth. Nonl. Anal. 23 (2004), 275300.
[183] J. L
opez-G
omez and A. Tellini, Generating an arbitrarily large number
of isolas in a superlinear indefinite problem, Nonl. Anal. 108 (2014)
223248.
[184] J. L
opez-G
omez, A. Tellini and F. Zanolin, High multiplicity and complexity of the bifurcation diagrams of large solutions for a class of superlinear indefinite problems, Comm. Pure Appl. Anal. 13 (2014), 173.
[185] A. J. Lotka, The growth of mixed populations: Two species competing
for a common food supply, J. Washington Acad. Sci. 22 (1932), 461469.
[186] A. Lunardi, Analytic semigroups and optimal regularity in parabolic
problems, in Progress in Nonlinear Partial Differential Equations and
Their Applications, Birkhauser, Berlin, 1995.
[187] Th. R. Malthus, An Essay on the Principle of Population, St. Pauls
Church Yard, London, 1798.
[188] M. Marcus and L. Veron, Uniqueness and asymptotic behavior of solutions with boundary blow-up for a class of nonlinear elliptic equations,
Ann. Inst. Henri Poincare 14 (1997), 237274.
Bibliography
345
[189] M. Marcus and L. Veron, The boundary trace and generalized boundary
value problem for semilinear elliptic equations with coercive adsorption,
Comm. Pure Appl. Math. 56 (2003), 689731.
[190] V. Maric and M. Tomic, Asymptotics of solutions of a generalized
Thomas-Fermi equation, J. Diff. Eqns. 35 (1980), 3644.
[191] M. Marras and G. Porru, Estimates and uniqueness for boundary blowup solutions of p-Laplace equations, Electron. J. Diff. Eqns. 119 (2011),
10.
[192] J. L. Mawhin, The legacy of P. F. Verhulst and V. Volterra in population
dynamics, in The First 60 Years of Nonlinear Analysis of Jean Mawhin,
pages 147160, World Scientific, Singapore, 2004.
[193] J. L. Mawhin, D. Papini and F. Zanolin, Boundary blow-up for differential equations with indefinite weight, J. Diff. Eqns. 188 (2003), 3351.
[194] L. Mi and B. Liu, Second order expansion for blow-up solutions of semilinear elliptic problems, Nonl. Anal. 75 (2012), 25912613.
[195] M. Molina-Meyer and F. R. Prieto-Medina, A collocation-spectral
method to solve the bi-dimensional degenerate diffusive logistic equation
with spatial heterogeneities in circular domains, preprint, May 2014.
[196] C. Mu, S. Huang, Q. Tian and L. Liu, Large solutions for an elliptic system of competitive type: existence, uniqueness and asymptotic behavior,
Nonl. Anal. 71 (2009), 45444552.
[197] J. D. Murray, Mathematical Biology, Springer, Berlin, 1989.
[198] S. Nakamori and K. Takimoto, Uniqueness of boundary blowup solutions
to k-curvature equation, J. Math. Anal. Appl. 399 (2013), 496504.
[199] L. Nirenberg, A strong maximum principle for parabolic equations,
Comm. Pure Appl. Math. 6 (1953), 167177.
[200] A. Okubo and S. A. Levin, Diffusion and Ecological Problems: Modern
Perspectives, Interdisciplinary Applied Mathematics 14, Springer, New
York, 2001.
[201] O. A. Oleinik, On properties of some boundary problem for equations
of elliptic type, Math. Sbornik, N.S. 30 (1952), 695702.
[202] R. Osserman, On the inequality u f (u), Pacific J. Math. 7 (1957),
16411647.
[203] T. Ouyang, On the positive solutions of semilinear equations u +
u hup = 0 on the compact manifolds, Trans. Amer. Math. Soc.,
331 (1992), 503527.
346
Bibliography
[204] T. Ouyang, On the positive solutions of semilinear equations u + u

hup = 0 on the compact manifolds, Part II, Indiana Univ. Math. J. 40
(1991), 10831141.
[205] T. Ouyang and Z. Xie, The uniqueness of blow-up for radially symmetric
semilinear elliptic equations, Nonl. Anal. 64 (2006), 21292142.
[206] T. Ouyang and Z. Xie, The exact boundary blow-up rate of large solutions for semilinear elliptic problems, Nonl. Anal. 68 (2008), 27912800.
[207] R. Pearl and L. L. Reed, On the rate of growth of the population of
the United States since 1790 and its mathematical representation, Proc.
Nat. Acad. Sci. 6 (1920), 275288.
[208] O. Perron, Eine neue Behandlung der Randwertaufgabe f
ur u = 0,
Mat. Z. 18 (1923), 4254.
[209] M. Picone, Sui valori eccenzionali di un pammetro da cui dipende
unequazione differenziale ordinaria del secondo ordine, Ann. Sc. Nor.
Sup. Pisa 11 (1910), 1141.
[210] M. W. Protter and H. F. Weinberger, Maximum Principles in Differential Equations, Prentice-Hall, Englewood Cliffs, 1967.
[211] F. I. Pugnaire (Editor), Positive Plant Interactions and Community Dynamics, CRC Press, Boca Raton, 2010.
[212] A. Quatelet, Sur lhomme et le developpement de ses facultes. Essai de
physique sociale, Bacheller, Paris, 1835.
[213] P. Quittner and P. Souplet, Superlinear Parabolic Problems. Blow-up,
Global Existence and Steady States, Birkhauser, Basel, 2007.
[214] P. H. Rabinowitz, Some global results for nonlinear eigenvaue problems,
J. Funct. Anal. 7 (1971), 487513.
[215] P. H. Rabinowitz, Minimax Methods in Critical Point Theory with Applications to Differential Equations, Regional Conference Series in Mathematics 65, American Mathematical Society, Providence, 1988.
[216] H. Rademacher, Einige besondere Probleme partieller Differentialgleichungen, in Die Differential und Integralgleichungen der Mechanik und
Physik I, (P. Frank and R. von Mises, eds.) pages 838845, Rosenberg,
New York, 1943.
[217] V. Radulescu, Singular phenomena in nonlinear elliptic problems: From
boundary blow-up solutions to equations with singular nonlinearities,
in Handbook of Differential Equations: Stationary Partial Differential
Equations, Vol. 4, pages 483591, Elsevier, Amsterdam, 2007.
Bibliography
347
[218] D. Repovs, Asymptotics for singular solutions of quasilinear elliptic

equations with an absorption term, J. Math. Anal. Appl. 395 (2012),
7885.
[219] M. B. Saffo, Invertebrates in endosymbiotic associations, Amer. Zool.
32 (1992), 557565.
[220] D. Sattinger, Topics in Stability and Bifurcation Theory, Lecture Notes
in Mathematics 309, Springer, Heildelberg, 1973.
[221] T. Shibata, Spectral asymptotics for inverse nonlinear Sturm-Liouville
problems, Electron. J. Qual. Theory Diff. Eqns. 58 (2009), 118.
[222] T. W. Shoener, Field experiments on interspecific competition, Amer.
Natur. 122 (1983), 240285.
[223] E. Shr
odinger, The Physical Aspect of the Living Cell, Cambridge University Press, Cambridge, 1944.
[224] B. Simon, Semiclassical analysis of low lying eigenvalues, I. Nondegenerate minima: Asymptotic expansions, Ann. Inst. Henri Poincare
A, XXXVIII (1983), 1237.
[225] S. D. Taliaferro, Asymptotic behavior of solutions of y 00 = (t)y . J.
Math. Anal. Appl. 86 (1978), 9534.
[226] S. D. Taliaferro, Asymptotic behavior of solutions of y 00 = (t)f (y),
SIAM J. Math. Anal. 12 (1981), 853-865.
[227] A. Tellini, Positive Blow-up Solutions for a Class of Nonlinear Differential Equations, Tesi di Laura, Udine, July 2010.
[228] P. F. Verhulst, Notice sur la loi que la population suit dans son accroissement, Corr. Math. Phys. 10 (1838), 113125.
[229] V. Volterra, Lecons sur la theorie mathematique de la lutte pour la vie,
Gauthier-Vilars, Paris, 1931.
[230] L. Veron, Semilinear elliptic equations with uniform blow up on the
boundary, J. DAnalyse Math. 59 (1992), 231250.
[231] M. Wang and L. Wei, Existence and boundary blow-up rates of solutions
for boundary blow-up elliptic systems, Nonl. Anal. 71 (2009), 2022
2032.
[232] W. Wang, H. Gong and S. Zheng, Asymptotic estimates of boundary
blow-up solutions to the infinity Laplace equations, J. Diff. Eqns. 256
(2014), 37213742.
348
Bibliography
[233] Y. Wang and M. Wang, The blow-up rate and uniqueness of large solution for a porous media logistic equation, Nonl. Anal. Real World Appl.
11 (2010), 15721580.
[234] L. Wei, The existence of large solutions of semilinear elliptic equations
with negative exponents, Nonl. Anal. 73 (2010), 17391746.
[235] L. Wei and M. Wang, Existence and estimate of large solutions for an
elliptic system, Nonl. Anal. 70 (2009), 10961104.
[236] L. Wei and J. Zhu, The existence and blow-up rate of large solutions of
one-dimensional pLaplacian equations, Nonl. Anal. Real World Appl.
13 (2012), 665676.
[237] J. L. Wulff, Clonal organisms and the evolution of mutualism. In Jackson, J.B.C., Buss, L.W., Cook, R.E. (Eds.), Population Biology and
Evolution of Clonal Organisms, pages 437466, Yale University Press,
New Haven, 1985.
[238] Z. Xie, Uniqueness and blow-up rate of large solutions for elliptic equation u = u b(x)h(u), J. Diff. Eqns. 247 (2009), 344363.
[239] Z. Xie and C. Zhao, Blow-up rate and uniqueness of singular radial
solutions for a class of quasi-linear elliptic equations, J. Diff. Eqns. 252
(2012), 17761788.
[240] H. Yamabe, On a deformation of Riemaniann structures on compact
manifolds, Osaka Math. J. 12 (1960), 2137.
[241] H. Yang and Y. Chang, On the blow-up boundary solutions of the
Monge-Ampère equation with singular weights, Comm. Pure Appl.
Anal. 11 (2012), 697708.
[242] Q. Zhang, Existence, nonexistence and asymptotic behavior of boundary
blow-up solutions to p(x)-Laplacian problems with singular coefficient,
Nonl. Anal. 74 (2011), 2045-2061.
[243] Q. Zhang, X. Liu and Z. Qiu [243], On the boundary blow-up solutions
of p(x)-Laplacian equations with singular coefficient, Nonl. Anal. 70
(2009), 40534070.
[244] Q. Zhang, Y. Wang and Z. Qiu, Existence of solutions and boundary
asymptotic behavior of p(r)-Laplacian equation multi-point boundary
value problems, Nonl. Anal. 72 (2010), 2950-2973.
[245] Q. Zhang and C. Zhao, Existence, uniqueness and blow-up rate of large
solutions of quasi-linear elliptic equations with higher order and large
perturbation, J. Partial Differ. Eqns. 26 (2013), 226-250.
Bibliography
349
[246] Z. Zhang, The asymptotic behaviour of solutions with boundary blowup for semilinear elliptic equations with nonlinear gradient terms, Nonl.
Anal. 62 (2005), 11371148.
[247] Z. Zhang, The asymptotic behaviour of solutions with blow-up at the
boundary for semilinear elliptic problems, J. Math. Anal. Appl. 308
(2005), 532540.
[248] Z. Zhang, Boundary blow-up elliptic problems with nonlinear gradient
terms, J. Diff. Eqns. 228 (2006), 661684.
[249] Z. Zhang, Boundary blow-up elliptic problems with nonlinear gradient terms and singular weights, Proc. Roy. Soc. Edinburgh Sect. A 138
(2008), 14031424.
[250] Z. Zhang, A boundary blow-up elliptic problem with an inhomogeneous
term, Nonl. Anal. 68 (2008), 34283438.
[251] Z. Zhang, Boundary behavior of large solutions to semilinear elliptic
equations with nonlinear gradient terms, Nonl. Anal. 73 (2010), 3348
3363.
[252] Z. Zhang, The second expansion of large solutions for semilinear elliptic
equations, Nonl. Anal. 74 (2011), 34453457.
[253] Z. Zhang, Y. Ma, L. Mi and X. Li, Blow-up rates of large solutions for
elliptic equations, J. Diff. Eqns. 249 (2010), 180199.
[254] Z. Zhang and L. Mi, Blow-up rates of large solutions for semilinear
elliptic equations, Comm. Pure Appl. Anal. 10 (2011), 17331745.
Index
abiotic stress, 265

Ahlfors L. V., 341
Alama S., 331
Alarcon S., 206, 331
Aleja D., 133, 331
Alvarez-Caudevilla P., 132, 331
Amann H., 17, 132, 238, 239, 257,
310, 313315, 331, 332
Ambrosetti A., 102, 332
Ampere A. M., 348
Aneda C., 206, 332
Arzela G., 141, 142
Ascoli C., 141, 142
Banach S., 293
Bandle C., 166, 185, 186, 206, 332
Bao J., 206, 332, 340
Baras P., 257, 332
Belgacem F., 133, 332
Belhaj Rhouma N., 206, 332
Berestycki H., 94, 257, 332, 333
Bertness M. D., 266, 333
Bertsch M., 252, 333
Bieberbach L., 64, 333, 340
Blat J., 328, 333
blow-up rate of large solutions, 164,
178, 192
boundary normal section, 191
Boureanu M. M., 206, 333
Brezis H., xiv, 48, 101, 102, 104, 290,
333
Browder F., 333
Brown K. J., 328, 333
Brown R., 30
Brownian motion, 30
Bud C. J., 132
Buss L. W., 348
Callaway R. M., 266, 333

Cano-Casanova S., 16, 47, 165167,
185, 329, 333, 334, 342
Cantrell R. S., 329, 334
Capuzzo-Dolcetta I., 257, 332
carrying capacity, xiii
Cauchy A. L., 55, 140, 144, 145, 147
Chang Y., 206, 348
Chen L., 206, 334
Chen Y., 206, 334, 335
Chipot M., 342
Chuaqui M., 185, 186, 335
Cirstea F. C., 104, 185, 187, 335
classical diffusive logistic problem, 33
coexistence states, 269
Cohen L., 257, 332
competitive exclusion, 47
complete blow-up, 246
component, 213
concept of metasolution, 106
Connell J. H., 266, 335
Cook R. E., 348
cooperative synergy, 212
Cortazar C., 335
Cosner C., 133, 329, 332, 334
Courant R., 15, 16, 335
Crandall M. G., 38, 41, 212, 215,
236, 240, 296, 312, 335
Cui R., 335
Cushing J., 328, 335
Dancer E. N., 16, 224, 313, 335, 336
Daners D., 35, 268, 336
Darwin Ch., 29, 336
De LHopital G., 159, 165
degenerate diffusive logistic equation,
xiii
351
352
Del Pino M. A., 102, 186, 336
Delgado M., 206, 282, 284, 313, 336
Diaz G., 186, 206, 331, 336
diffusive
logistic equation, xiii, 30
logistic problem, 12
Malthusian problem, 12
Dini U., 88
Dirichlet G. L., xiii, 4, 17, 20, 26, 32,
37, 38, 75, 78, 86, 126, 132,
184, 190, 224, 268, 270, 279,
309
Drissi A., 206, 332
Du Y., 103, 132, 166, 185, 186, 205,
329, 336, 337
Dumont S., 204, 205, 337
Dupaigne L., 337
Eilbeck J. C., 299, 304, 305, 313, 337
Elgueta M., 335
explosive
solution, 18
subsolution, 18
supersolution, 18
exponential growth, xiv
exterior problems, 167
Faber C., 5, 129, 215, 337
FaberKrahn inequality, 5
facilitation, 212
Fan Z., 206, 339
Feltrin G., 257, 337
Fen P., 206, 337
Feng H., 206, 337
Fermi E., 167, 333, 345
Ferrera J., 331
Fick A., 30, 337
Fisher R. A., xiv, 30, 337
Flores C., 335
Fourier J. B. J., 30, 337
Fraile J. M., xiv, 48, 102, 132, 133,
337
Frank P., 346
Fredholm I., 38, 39, 126, 228
function
Index
positive, 11
strongly positive, 11
Furter J. E., 283, 299, 304, 305, 313,
337
Gamez J. L., 102, 332
Garcia-Melian J., 103, 166, 185187,
206, 257, 331, 335, 338
Gaudenzi M., 337
Gauss C. F., 64
Ghoreishi A., 313, 341
Gilbarg D., 19, 43, 338
Golubitsky M., 258, 338
Gomez-Re
nasco R., xvii, 102104,
111, 131, 132, 217, 257, 338,
339
Gong H., 206, 347
Goubet O., 337
Green G., 37, 339
Habets P., 337
Hadamard J., 339
Hao R., 206, 335
Hardy G. H., 252
Harnack A., xvii, xviii, 103, 104, 321
heat semigroup, 12
Helmholtz H., xviii
Henry D., 339
Hess L. O., 340
Hilbert D., 15, 16, 290, 335
Holder continuous function, 30
Holder O., 30, 31, 188
Hopf E., 103, 339
Hsu S. B., xix
Huang Q., 103, 132, 166, 185, 186,
337
Huang S., 206, 339, 345
Hutchinson G. E., 266, 339
Hutson V., 132, 329, 339
Hypothesis
(HA), 210
(Ha), 4
(HB), 239
(HC), 267
(HF), 267
Index
(Hf), 6
(Hg), 6
(KO), 7
inhabiting area, xiii
initial distribution of the species, xiii
intra-specific competition, 212
intrinsic rate of natural increase, xiii
inversion of the protection patches,
282
Jackson J. B. C., 348
Ji X., 206, 332, 340
Kato T., 283, 285, 340
Keller J. B., 7, 49, 50, 55, 64, 65,
137, 170, 205, 331, 337, 340
KellerOsserman condition, 7, 8, 64,
137
classical, 205
Koch P., 268
Koch-Medina P., 35, 336, 337
Kolmogorov A. N., xiv, 30, 274, 340
Kondrachov V. I., 290
Kondratiev V. A., 166, 185, 340
Kra I., 341
Krahn E., 5, 129, 215, 340
Ladyzenskaja O. A., 340
Langlais M., 234, 340
Laplace operator, 3
Laplace P. S., 3, 206, 332, 334, 335,
340, 345, 347, 348
large
solution, 18
solution of order j, 68
subsolution, 18
supersolution, 18
large solution
blow-up rate, 164, 178, 192
Lazer A. C., 64, 103, 166, 185, 340
Lebesgue H., 5, 328
Lei W., 206, 340
Letelier R., 166, 185, 186, 206, 331,
336, 338
Levin S. A., xiii, 345
353
Li H., 206, 332, 340, 341
Li L., 313, 340
Li W. T., 206, 339
Li X., 206, 349
Liang Y., 206, 340
linearly asymptotically stable, 225
linearly unstable, 225
Lions P. L., 94, 333
Liouville J., 332, 347
Lipschitz R., 31, 140, 145
Liu B., 206, 345
Liu C., 206, 341
Liu L, 206, 345
Liu X., 206, 348
Loewner C., 166, 185, 341
Logan A., 313
Logan R., 313, 340, 341
logistic equation, 29
non-spatial, xiii
diffusive, xiii
logistic growth, xiv
Lopez-Gomez J., 16, 17, 19, 43, 47,
65, 93, 94, 101104, 111,
131133, 165, 184, 185, 191,
204, 206, 238240, 247, 257,
259, 263265, 282284, 293,
299, 304, 305, 309, 310, 313,
315, 329, 331334, 336339,
341344
Lotka A. J., 328, 336, 337, 340, 341,
344
lowest eigenvalue, xiv
Lunardi A., 127, 278, 344
Luo D., 206, 334
Lyapunov A. M., 278
Ma Y., 206, 349
Maire L., 185, 343
Malthus Th. R., xiv, xvi, 1215, 28,
29, 105, 130, 184, 270, 344
Marcus M., 166, 185, 186, 206, 332,
344, 345
Maric V., 167, 345
Marras M., 206, 345
Maskit B., 341
354
Mawhin J. L., 28, 345
maximal metasolution, 106
maximum principle
characterization, 15
parabolic, 12
strong, 17
theorem of characterization, 17
McKenna P. J., 64, 166, 185, 340
Merino S., 337
meta-coexistence states, 270
metasolution, xvii
concept, 106, 268, 269
maximal, 106
minimal, 106
Mi L., 206, 345, 349
Mi Y., 206, 339
minimal metasolution, 106
Minkowski H., 293
Mischaikow K., 339
Mises, R. von, 346
Molina-Meyer M., 17, 47, 104, 257,
259, 265, 309, 310, 329, 334,
343345
Monge G., 348
Mora-Corral C., 342
Morales-Rodrigo C., 338
Mu C., 206, 339, 345
Murray J. D., xiii, 345
Nakamori S., 206, 345
Neumann K. G., 102, 190
neutrally stable, 225
Ni W. M., xiv, 102
Nikishin V. A., 166, 185, 340
Nirenberg L., 12, 166, 185, 257, 332,
341, 345
non-oscillating function, 151
non-spatial logistic equation, xiii
Okubo A., xiii, 345
Oleinik O. A., 103, 345
Osserman R., 7, 49, 50, 55, 64, 65,
137, 170, 205, 331, 337, 345
Oswald L., xiv, 48, 101, 102, 333
Ouyang T., xiv, 102, 185, 187, 206,
345, 346
Index
Pang P. Y. H., 206, 334, 340, 341
Papini D., 345
parabolic logistic problem, xiii
parabolic problem, 3
Pearl R., 29, 30, 346
Perron O., 37, 346
Petrovsky I. G., xiv, 30, 340
Phillips D., 234, 340
Picone M., 214, 257, 346
Piskunov N. S., xiv, 30, 340
plant communities, 266
Poincare H., 333, 344, 347
polluted environment, 265
Porru G., 206, 332, 345
positive function, 11
positive strict supersolution, 17
Prieto-Medina F. R., 345
principal eigenfunction, 16
principal eigenvalue, 4, 16
continuity with respect to the
domain, 16
monotonicity with respect to the
domain, 16
monotonicity with respect to the
potential, 16
problem
sublinear, 210
superlinear, 210
superlinear indefinite, 210
protected areas, xiv, 15
protection zones, xiv, xvi, xvii, 15,
47, 116, 130, 270, 329
Protter M. W., 103, 346
Pugnaire F. I., 266, 346
Qiu Z., 206, 348
quasi-cooperative system
concept, 308
strong maximum principle, 309
subsolution, 309
supersolution, 309
Quass A., 206, 331
Quatelet A., 28, 29, 346
Quittner P., 211, 247, 257, 342, 344,
346
Index
Rabinowitz P. H., 38, 41, 102, 103,
212, 215, 236, 240, 246, 296,
312, 328, 335, 344, 346
Rademacher H., 64, 340, 346
Radulescu V., 104, 185, 187, 335,
337, 346
random motion, 30
Rayleigh L., 340
Reed L. L., 29, 30, 346
Rellich F., 290
Repovs D., 206, 347
Rey J. M., 206, 331
Riemann B., 64
Rossi J. D., 338
Rostamian R., 252, 333
Ruiz del Portal F. R., 332
355
strongly positive function, 11
structural stability, 42
Sturm J. C. F., 347
Suarez A., 206, 282, 284, 313, 336,
338, 344
sublinear problem, 3, 210
superlinear indefinite problem, 210
superlinear problem, 210
Takimoto K., 206, 345

Taliaferro S. D., 167, 347
Tarantello G., 331
Tellini A., 257259, 263265, 343,
344, 347
theorem
behavior of the large solution on
the boundary, 161
Sabina de Lis J. C., 47, 93, 94, 102,
exact blow-up rate of large
103, 166, 185, 186, 309, 329,
solutions, 164, 178, 192
338, 344
limiting behavior as of
Saffo M. B., 266, 347
the large solutions, 98
Sattinger D., 24, 119, 120, 127, 347
limiting behavior of the large
Sayed W., 206, 332
solutions of order
Schauder J., 19, 32, 43, 56, 58, 60,
1 j q0 1 as dj+1 ,
85, 87, 97, 128, 239, 317,
96
318
limiting behavior of the positive
semi-trivial positive metasolutions,
solution as d1 , 86
270
of approximation of
semi-trivial positive solutions, 269
metasolutions by classical
semilinear elliptic problem, 33
steady states, 125
Shaeffer D. G., 258, 338
of bifurcation of coexistence
Shi J., 335
states, 294
Shibata T., 206, 347
of characterization of the
Shoener T. W., 266, 347
dynamics, 287, 324
Shr
odinger E., 347
Simon B., 347
dynamics of the degenerate
singular problem, 8
diffusive logistic equation,
Solonnikov V. A., 340
107
solution
explosive, 9, 18, 49
existence of coexistence and
large, 9, 18, 49
meta-coexistence states, 277
Souplet P., 211, 346
spatially heterogeneous
existence of linearly stable
environments, xiv
positive solutions, 232
strong maximum principle, 17
356
Index
maximum principle, 17
of characterization of the strong
maximum principle for
quasi-cooperative systems,
309
of existence and uniqueness, 21
of existence and uniqueness for
the perturbed logistic
problem, 41
of existence and uniqueness for
the unperturbed classical
logistic problem, 35
of existence and uniqueness of
large solutions for a
one-dimensional problem in
(0, ), 139
of existence and uniqueness of
large solutions for a
one-dimensional problem in
(0, L), 145
of existence for hierarchic chains
of inhomogeneous problems,
80
of existence of classical
solutions, 69
of existence of coexistence states
and metasolutions, 320
of existence of large solutions for
superlinear indefinite
problems, 252
of existence of minimal and
maximal solutions, 59, 83
of existence of positive solutions
for the inhomogeneous
problem, 76
of existence of the minimal
solution, 50
of existence of the principal
eigenvalue for a
quasi-cooperative system,
309
of existence through
subsolutions and
supersolutions, 17, 18
of global existence and blow-up,

240
of global existence and complete
blow-up, 246
of instability of positive
solutions, 230
of multiplicity, 313
of non-existence of positive
solutions, 237
of structure for the unperturbed
problem, 37
of structure of the stable
positive solutions, 238
of uniqueness of the large
solution, 202
of uniqueness of the large
solution in the radially
symmetric case, 170
of uniqueness of the stable
positive solution, 233
Thomas L., 167, 333, 345
Tian Q., 206, 339, 345
Tomic M., 167, 345
Trudinger N. S., 19, 43, 338
Uraltzeva N. N., 340
Verhulst equation, 29
Verhulst P. F., xiii, xiv, 28, 29, 105,
345, 347
Verhulst-Pearl equation, 29
Veron L., 166, 185, 344, 345, 347
Vickers G., 339
Volterra V., 328, 336, 337, 340, 341,
345, 347
Walker L. R., 266, 333
Wang M., 206, 334, 340, 341, 347,
348
Wang W., 206, 347
Wang Y., 206, 348
Wei L., 206, 347, 348
Weinberger H. F., 103, 346
Wu B., 335
Wulff J. L., 266, 348
Index
Xi J., 206, 339
Xie Z., 185, 187, 206, 346, 348
Yamabe H., 102, 332, 348
Yang H., 206, 348
Yang Z., 206, 341
Z. Xie, 348
Zanolin F., 257, 263, 264, 337, 344,
345
Zhang Q., 206, 340, 348
Zhang S., 206, 339
Zhang Z., 206, 349
Zhao C., 206, 340, 348
Zheng S., 206, 347
Zhong C., 206, 337
Zhu J., 206, 348
Zhu Y., 206, 334, 335
357

1482238985

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1482238985

Uploaded by

Copyright:

Available Formats

Metasolutions of

2016 by Taylor & Francis Group, LLC

2016 by Taylor & Francis Group, LLC

2016 by Taylor & Francis Group, LLC

2016 by Taylor & Francis Group, LLC

Large solutions and metasolutions: Dynamics

The meaning of the KellerOsserman condition . .

2 Classical diffusive logistic equation

Unperturbed logistic problem . . . . . . . . . . . .

Singular problem in a ball BR (x0 ) b .

4 Generalized diffusive logistic equation

Classical positive solutions in . . . . . . . . . . . . . . . .

2016 by Taylor & Francis Group, LLC

Large solutions of arbitrary order j {1, ..., q0 } . . . . . . .

Concept of metasolution: The main theorem . . .

Existence and uniqueness . . . . . . . . . . . . . . .

7 Uniqueness of the large solution under radial symmetry

The main uniqueness result . . . . . . . . .

Boundary normal section of a . . . . . . . .

2016 by Taylor & Francis Group, LLC

8 General uniqueness results

6 A canonical one-dimensional problem

Uniqueness of the large solution

Metasolutions do arise everywhere

9 A paradigmatic superlinear indefinite problem

Components of positive steady states . . . . . . . .

10 Spatially heterogeneous competitions

2016 by Taylor & Francis Group, LLC

An admissible nodal configuration for a(x) . . . . . . . . . .

An admissible nodal configuration for a(x) . . . .

The dynamics of (2.1) for M = 0 . . . . . . . . . . . . . . .

Scheme of the construction of L[,D] and L[,D] . . . . . . .

An intricate nodal configuration for a(x) .

The dynamics of (1.1). . . . . . . . . . . . . . . . . . . . . .

The graph of `(x) . . . . . . . . . . . . . . . . . . . . . . . .

,x, are supported . .

An admissible nodal configuration of a(x) . . . . . . . . . .

2016 by Taylor & Francis Group, LLC

Bifurcation diagram for the choice (9.14) . . . . . . . . . . .

The curves arising in Theorem 10.5 . . . . . . . . . . . . .

2016 by Taylor & Francis Group, LLC

This book studies the dynamics of a generalized

prototype of the semilinear

where RN , with N 1, is a smooth bounded domain, t > 0 stands for

in . It is said to be degenerate because a(x) can vanish on some patches of

, in contrast to the classical case when a(x) < 0 for all x .

is the carrying capacity of at each location x . As we are imposing

Kolmogorov I. G. Petrovsky and N. S. Piskunov [124], and independently by

2016 by Taylor & Francis Group, LLC

As will be discussed in Chapter 1, from a biological perspective, d1 [, D]

FIGURE P.1: An admissible nodal configuration for a(x).

2016 by Taylor & Francis Group, LLC

it becomes apparent that

Naturally, for each j = 1, 2, 3, dj measures the critical size of so that the

whereas it has a limited logistic increase in

2016 by Taylor & Francis Group, LLC

If d3 , i.e., any refuge is able to maintain u at the increase rate ,

but according to a logistic law in the region

of the singular problem

2016 by Taylor & Francis Group, LLC

This might possibly explain the reaction of an anonymous reviewer of [195]

Although the strategy proposed by the reviewer in the previous paragraphs

2016 by Taylor & Francis Group, LLC

describe the dynamics of wide classes of parabolic equations and systems in

uniformly exponentially in compact subsets of , if > d0 . Indeed, let 0