Professional Documents
Culture Documents
Parabolic Equations
in Population
Dynamics
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
2016 by Taylor & Francis Group, LLC
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
1 Introduction: Preliminaries
1.1
1.2
1.3
1.4
1.5
1.6
1.7
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
33
. . . .
. . . .
. . . .
. . .
. . . .
.
.
.
.
.
.
.
.
.
.
3 A priori bounds in
3.1
3.2
3.3
3.4
3.5
34
37
41
42
47
49
. . .
. . .
. .
. . .
. . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8
11
15
17
20
27
30
50
57
59
61
64
67
69
76
80
vii
viii
Contents
4.4
4.5
4.6
4.7
4.8
4.9
5 Dynamics: Metasolutions
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
II
.
.
.
.
.
.
.
.
.
. .
. .
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
137
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7.3
7.4
7.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
139
148
159
164
165
169
.
.
.
.
.
.
.
.
.
.
.
.
.
.
106
108
111
117
125
128
129
131
135
7.1
7.2
96
98
101
105
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
. . . . . . . . . . . . . . . . . . . .
207
204
209
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
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
xv
1.1
1.2
1.3
1.4
1.5
.
.
.
.
.
5
10
13
14
30
2.1
2.2
40
44
3.1
59
4.1
4.2
4.3
4.4
4.5
.
.
.
.
.
68
70
72
73
89
5.1
5.2
5.3
5.4
109
112
113
5.5
5.6
min
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
max
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
114
115
116
6.1
139
7.1
and Lmin
The functions L
[,BR (x0 )] . . . . . . . . . . . . . .
173
8.1
8.2
195
199
9.1
9.2
211
218
xi
xii
List of Figures
9.3
9.4
9.5
9.11
9.12
10.1
10.2
10.3
10.4
10.5
10.6
10.7
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
u(, 0) = u0 > 0
in
(P.1)
(P.2)
,
a(x)
x ,
(P.3)
xiv
Preface
Preface
xv
1 [, B1 ]
R2
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,
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)
and
0,2
4
[
i=1
i0,1
Preface
xvii
0,2
4
[
i0,2
and
0,3
i=1
4
[
i0,3
i=1
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
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
Preface
xix
Part I
Chapter 1
Introduction: Preliminaries
1.1
1.2
1.3
1.4
1.5
1.6
1.7
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 ,
a C (),
[0, )),
f C ,1+ (
(a 0, a 6= 0),
for all u 0,
the general case when a(x) changes sign will be dealt with in Chapter 9.
3
2016 by Taylor & Francis Group, LLC
:= { x : a(x) < 0 },
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
j=1 i=1
1 i mj , 1 j q0 ,
1 j q0 1.
(1.3)
mj
[
i0,j ,
i=1
1 [, 0,j ] := j = 1 [, i0,j ], 1 i mj .
Introduction: Preliminaries
1 = 0,1 \ ,
2 := 0,2 ,
= 1 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 |)
Therefore,
lim 1 [, D] = .
|D|0
(1.5)
f C
( [0, )) satisfies f (x, 0) = 0 and u f (x, u) > 0 for all x
and u > 0.
,1+
[0, ),
(x, u)
(1.6)
u 0.
Introduction: Preliminaries
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)
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)
[ g(ut) 1] t dt
( Cup tp 1) t dt
1 2
Cup p+2
1
1
p+2
2
2
1
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)
1 2
Cup p+2
1
1 ,
p+2
2
1,
satisfies
R(1) = 0,
p+2
2
d =
R()
Cup
=
> 0,
p+2
p+2
lim
2
< .
p
1.1
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)
Introduction: Preliminaries
tL
g(s)s ds =
0
2
x A
2
g(s)s ds.
(1.15)
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,
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
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
Introduction: Preliminaries
11
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
it becomes apparent that
lim I(x) = .
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
12
u(, 0) = u0 > 0
in ,
whereas it provides us with the (linear)
u
t du = u
u=0
u(, 0) = u0 > 0
(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 ) =
Introduction: Preliminaries
13
and hence,
e(d0 )t e(t+1)(+d) u0 Ce(d0 )t .
Consequently,
lim u = 0
if < d0 ,
while
lim u = with exponential growth rate e(d0 )t
if > d0 .
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
Introduction: Preliminaries
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
16
D, then 1 [d + V, D] < 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]).
Introduction: Preliminaries
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,
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
18
(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)
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
Introduction: Preliminaries
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)
in Dk
for all n k.
k1 ) such that
for some function u
k1 C 2+ (D
d
uk1 =
uk1 + af (, u
k1 )
uk1
in Dk1 .
(1.25)
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)
20
provides us with the desired solution of (1.23). Note that (1.24) implies
uu
k u
in Dk
uuu
The proof is complete.
1.5
in D.
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)
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,
(1.28)
Introduction: Preliminaries
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)
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.
(resp.
[,D,b] u
).
u0 > 0,
(b) For every u0 C(D),
lim ku[,D,b] (, t; u0 ) [,D,b] kC(D)
= 0,
u(, 0) = u0
in D.
(1.30)
(1.31)
22
(1.32)
in D
for sufficiently small > 0, because 1 [d, D] < and, thanks to (Hf),
lim kaf (, )kC(D)
= 0.
0
Introduction: Preliminaries
23
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
in D,
on D,
(1.33)
(1.34)
f (, tu + (1 t)u ) dt > af (, u ).
24
and
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)
Introduction: Preliminaries
25
u > [,D,b]
(1.37)
(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 <
u and therefore, by Theorem 1.2,
(1.21) possesses a positive solution, u. Thanks to Lemma 1.6, u 0 and
= 1 [d af (, u), D].
Moreover, since u 0 and a < 0 in D, we find from (Hf) that
af (, 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
(1.40)
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.
Introduction: Preliminaries
27
(1.43)
(1.44)
1.6
(1.45)
28
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
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,
Introduction: Preliminaries
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.
(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
1.7
Comments on Chapter 1
for all x, y .
Introduction: Preliminaries
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.
|| u
.
1 x1 N xN
X
||=k
|D u(x) D u(y)|
,
|x y|
x,y
sup
u C k+ (),
x6=y
where
kukC k ()
=
kD ukC()
.
||k
a C(),
[0, )),
f C 1 (
(1.48)
32
1 i, j N,
with bj C (),
b = (bj )1jN
on 0 ,
on 1 ,
C 2+ (),
(1.49)
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
u(, 0) = u0 > 0
in D (0, ),
on D (0, ),
in D,
34
37
41
42
47
(2.1)
for all x D.
34
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,
2.1
(2.4)
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
1jn
(2.6)
36
min
1jn DD/2,j
,j > 0.
/2,j ,
,j
in D
1 j n,
n
:=
[
/2,j ,
in Dint := D \
D
(2.7)
(2.8)
j=1
(2.9)
Dint
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
2.2
in Dint .
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)
if
() = 0
in
C(D).
(2.12)
1 [d,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
38
for all R
and
Du F(, 0)u = u (d)1 u
(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
Du F(0 , u0 )u = u (d)
f
0 u + a (, u0 )u0 u + af (, u0 )u
u
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
As () 0 and
1 [L(, ()), D] > 0
40
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
41
2.3
in
C(D),
(2.17)
(resp.
[,D,M ] u
).
< 1 2 < ,
M2 M1 + 2 1 > 0,
imply
[1 ,D,M1 ] [2 ,D,M2 ] .
Moreover,
[,D] [,D,M ]
(2.18)
42
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.
1jn D
and
min
min
1jn DD/2,j
,j > 0.
(2.19)
on D
for all
0 ,
in D
(2.20)
2.4
The next result establishes the structural stability of (2.1) when M > 0 perturbs from M = 0.
43
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
if
0<M <M
(2.22)
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
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
(0, )
M
7
C(D)
S(M ) := [,D,M ]
is of class C 1 . Moreover,
lim S(M ) =
M 0
0,
[,D] ,
45
u = v + M,
in D,
on D,
(2.23)
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
(2.24)
[, ]
M
7
C0 (D)
v(M )
for all M [, ]
=
v = v(M ).
(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)
in D
in D,
on D.
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 )
for all M [, ],
G(M, v M ) = 0
|M | + kv [,D] kC(D)
47
=
v = v(M ).
2.5
Comments on Chapter 2
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
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
du = u + a(x)f (x, u)u
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
for all
x RN .
50
3.1
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
for all
M >0
(3.6)
and
M1 M2
if
0 < M1 < M2 .
(3.7)
r := |x x0 |,
x D = BR (x0 ),
0 < r < R,
d 00 (r) + N r1 0 (r) = h((r)),
0 (0) = 0,
(3.8)
(R) = M.
(3.9)
for all x D.
(3.10)
52
(3.11)
0 < r < R.
(3.13)
0 r 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
0 < r < R.
A priori bounds in
53
Consequently,
h(M (r)) d00M (r)
h(M (r))
,
N
0 < r < R.
(3.15)
0 s r,
(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)
r s R,
qR
qR
u
u
2 M (r)
2 M (r)
d
h
M (0)
M (0)
h
M (r)
h
M (0)
(3.17)
h
54
du
u
u
2 M (r)
2 M (r)
d
h
h
M (0)
M (r)
0 r < R,
(3.18)
z > g 1 (1/).
(3.19)
du
qR
u
z
h(s) ds
s = zt,
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
for all
x D = BR (x0 ),
(3.20)
A priori bounds in
55
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
rR
(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()
(3.22)
(3.23)
in BR (x0 )
in BR (x0 )
(3.24)
in BR (x0 ),
A priori bounds in
3.2
57
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 )
in BR (x0 )
in BR (x0 ).
in BR (x0 ).
in K
for all
M > 0.
(3.27)
(3.28)
By construction,
n Dn+1 D ,
D
max a < 0,
n
D
D=
[
nm
Dn ,
(3.29)
58
[,D,M ] C(n)
for all
M > 0.
for all n n0 .
in D \ Dn .
in Dn
and hence,
[,D,M ] L
in D
on D,
in D.
A priori bounds in
3.3
59
max
L[,D] L L[,D] .
Moreover,
min
(3.30)
and
max
min
(3.31)
where Dn , for sufficiently large n, are those defined by (3.28). See Figure 3.1.
min
max
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
in Dn ,
n n0 ,
in Dn ,
n n0 .
min
L[,Dn+1 ] L[,Dn ]
in Dn
for all n n0 .
(3.32)
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.
n := max L,
n
D
n n0 .
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 .
in D.
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
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
(3.35)
g(ut)
,
g(
ut)
uu
,
(3.36)
satisfies
lim Q(u) = .
(3.37)
62
Thus, since u
> u > g 1 (1/),
g(
ut)
g(
ut)
g(
ut) 1 > 0
g(ut)
for all t 1.
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)
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.
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)
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
Comments on Chapter 3
(3.42)
(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
> 0.
g 1 (/A)
(3.45)
u 0.
Indeed, we already know that (KO) holds for all p > 0, while a direct calculation shows that (3.45) fails if 0 < p < 2.
The generalized KellerOsserman condition (KO), which is optimal in the
context of this book, was originally introduced under the form
Z
dx
qR
< ,
lim I(u) = 0,
(3.46)
I(u) :=
x
u
u
h
u
for some u > 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)
Chapter 4
Generalized diffusive logistic
equation
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
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 ,
du = u + a(x)f (x, u)u
u=0
on j ,
(4.2)
u=
on j \ ,
where f satisfies (KO) and
0,1
0,j ,
j := \
1 j q0 .
(4.3)
mj
[
i0,j ,
1 j q0 ,
i=1
68
1 i mj , 1 j q0 ,
1 j q0 1.
(4.4)
By construction,
0 = .
q0 = \
(4.5)
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 = ,
1 {1, 2, 3}.
4.1
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
d0
in
C(),
(4.8)
and
lim [,] =
d1
0,1 .
(4.9)
(4.10)
71
1 j q0 ,
1 i mj .
1 j n.
for all 1 j n.
72
1jn
(4.12)
/2,j ,
,k,/2 , 1 k n,
:=
(4.13)
,k, in N
k=1
73
(4.14)
on ,
in K
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
[,] [,]
in
i0,1 ,
(4.16)
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)
in
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
i
=
d1 0,1
i0,1 ,
(4.21)
in i0,1 ,
on i0,1 .
d1 )
for all (,
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.
76
4.2
<d .
< d1
(4.23)
(resp.
u
[,D,M ] ).
0 < M1 M2 ,
+ M2 M1 > 0.
M 0
0,
[,] ,
if d0 ,
if d0 < < d1 ,
(4.24)
77
1 j q0 , 1 i mj ,
and
,k, := {x RN : dist (x, j ) < },
N
1 k n,
i,j
in
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)
(4.27)
78
denoted by w , satisfies
lim w = M
(4.28)
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),
M
x
1 [, BR ] + 1
in BR (x)
M
1 [, BR ] + 1
for all x K.
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 ,
du = u + a(x)f (x, u)u
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] ).
0 < M1 M2 ,
+ M2 M1 > 0.
M 0
0,
[,] ,
if d0 ,
if d0 < < d1 ,
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 ,
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
u=M
(4.33)
(resp.
u [,j ,M ] ).
0 < M1 M2 ,
+ M2 M1 > 0.
81
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 all 1 j n.
1jn
,j,/2 , 1 j n,
,j, in N
:=
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
n n0 .
in j,n ,
du = u + a(x)f (x, u)u
u=0
on j,n ,
(4.36)
u=M
on j,n \ ,
possesses a positive solution if, and only if,
< dj+1 ,
(4.37)
(resp.
u [,j,n ,M ] ).
0 < M1 M2 ,
+ M2 M1 > 0.
dj+1
4.4
83
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
(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
lim L[,j ] =
dj+1
(4.40)
in ,
on .
(4.41)
in .
(4.42)
(4.43)
84
min
L[,j ] L[, ]
in ,
and hence,
min
L[,j ] <
in ,
(4.44)
min
min
max L[,j ] .
j,n \
max [,j ,M ]
j,n \
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)
in j,n
for all n n0 .
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 \
in
and consequently,
[,j ,M ] L
in
j,n ,
85
L[,j ] L,
min
in
j,n
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
for all m n + 1.
max
max
n n0 .
in j,n ,
on j,n ,
in
j,n
n n0 .
86
L L[,j ]
in j .
max
in
4.5
j .
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
in 1 .
(4.49)
in
d1
(4.50)
in
0,1 .
L L[d1 ,1 ]
in
(4.51)
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 ).
d1
L = L[d1 ,1 ]
(4.52)
min [,] = .
d1 1 \
(4.53)
88
m := min [,] ,
1 \
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 \
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
(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.
90
0,1,R
Consequently, x
= x and
[,] (x) [,] (x )
for all
R (Y (x )).
xB
(4.56)
(x) := e|xY (x )| eR ,
R (Y (x )).
xB
min
R/2 (Y (x ))
d1 B
[,] = .
Thus, setting
c :=
91
we have that
lim c = ,
(4.58)
d1
because
[,] (x ) b for all
[d
1 , d1 ).
in AR .
According to (4.59),
v 0
on BR/2 (Y (x )).
on BR (Y (x )).
Consequently,
v 0
on AR
for all [d
1 , d1 ).
(4.60)
in AR
(4.61)
1 i m1 ,
owing to Theorem 1.1, we can infer from (4.60) and (4.61) that
v (x) > 0
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
d1
[,]
(x ) = .
n
(4.63)
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 ),
93
[,]
(x )
(x ).
n
n
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 \ .
4.6
0 1im1
= 0.
(4.65)
and
a(x) = 0
1 i m1 .
as 0
(4.66)
94
where
Z
i :=
i0,1 \
i0,1
ni
!2
dS < 0,
Ci,1 (x)
if x i,1 , 1 i m1 ,
m1
u (x) :=
(4.67)
[
0
if
x
,1
i=1
(4.68)
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
0,
(4.70)
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)
d2 i + O( 2 )
.
supi,1 \ i0,1 a
di
F 1 ( sup2 i
C = C() := max
+O( 2 )
a)
i
\
0,1
,1
C2
1im1
(4.72)
for all x
m1
[
Ki
i=1
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
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
min
L[,j ] L[,j ]
in
j .
min
lim L[,j ]
in
dj+1
(4.76)
(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)
dj+1
m :=
min
0,j+1 \
min
L[,j ] .
(4.80)
97
in
min
L[,j ] L[dj+1 , ]
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 .
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
u=M
on (j+1 \ ) \ 0,j+1 ,
(4.82)
u
=
m
on 0,j+1 \ ,
u=0
on j+1 .
98
Fix < dj+1 . Then, according to (4.38) and (4.80), there exists M () > m
such that, for every M M (),
min
in
j+1 .
(4.83)
L L[dj+1 ,j+1 ]
in
j+1 .
min
L = L[dj+1 ,j+1 ]
which ends the proof.
4.8
in
j+1 ,
(4.84)
(4.85)
R (x0 )
B
99
(4.88)
(4.89)
On the other hand, by Theorem 2.4, we have that, for every < ,
[,D,] [,D,]
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
0 < (1 [d, D] )
L dx d
L
dS d
L0
dS
n
n
D
D
D
for all < 0, because
100
As (4.89) implies
Z
L (x0 )
Z
dx
L dx,
(4.92)
in D.
du = u + Ag(u)u
u=
in Dx ,
on Dx ,
(4.93)
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.
uniformly in D.
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 \
4.9
Comments on Chapter 4
and
V (x) := lim ( + a(x)f (x, u)) =
u
if x ,
if x 0 .
(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
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
H H,
[d1 ,d1 )
d1
(4.98)
104
because
0,1 \ ,
lim [,] = uniformly in compact subsets of
d1
d1
Chapter 5
Dynamics: Metasolutions
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
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 \ ,
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
(5.1)
in j ,
du = u + a(x)f (x, u)u
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 ,
Dynamics: Metasolutions
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
C(),
in
in ,
on .
(5.2)
max
in
(5.3)
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
max
in
j .
0,i ,
108
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 .
5.2
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 = \ (
Dynamics: Metasolutions
109
(5.5)
R 7 M[,2 ]
(5.6)
d1
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
110
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 ,
< 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
Dynamics: Metasolutions
111
for all n n0 .
5.3
d = 1,
with
(r) :=
sin(5(r + 0.5)),
0,
and
f (x, u) = u3 ,
u 0,
x .
= A0.1,0.3 ,
q0 = 2,
0,1 = A0.3,0.5 ,
0,2 = B0.1 ,
112
(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)
2
[,]
12
,
0 < < 1 ,
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
Dynamics: Metasolutions
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}
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.
Dynamics: Metasolutions
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.
116
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.
Dynamics: Metasolutions
where u[,0,1 ] (x, t; u1 ) stands for the
u
t u = u
u=0
u(, 0) = u1
117
which is given by
u[,0,1 ] (x, t; u1 ) = et(+) u1 .
(5.12)
5.4
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
(5.14)
Actually, as
lim [d1 ,] =
0
(5.15)
t
u=M
u=0
u(, TM ) = u[,] (, TM ; u0 )
in
on
on
in
1 (TM , ),
(1 \ ) (TM , ),
(1 ) (TM , ),
1 .
(5.16)
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
in
1 .
Dynamics: Metasolutions
119
in
(5.17)
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
in
M Mt .
for all
in
in
1 .
in
(5.18)
(5.19)
(5.20)
120
in ,
on .
(5.21)
in
af (, ) d0
or, equivalently,
in
in
in
(5.22)
in
in
(5.23)
Consequently, u[,] (, t; u0 ) is uniformly bounded above, in any compact subset of , for all t > 0.
Dynamics: Metasolutions
121
on 1,n \
t
u=M
u=0
u(, 0) = u0
in
1,n
(5.24)
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 ,
in
1,n .
(5.25)
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
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)
(5.28)
Actually, as
lim Mmin
[dj+1 ,j ] =
0
j+1
[
0,i \ . (5.29)
i=1
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 .
(5.30)
t
u=M
on (j+1 \ ) (TM , ),
u
=
0
on (j+1 ) (TM , ),
u(, 0) = u
0
in j+1 ,
Dynamics: Metasolutions
123
for all u
0 > 0. So, letting t in (5.30) shows that
lim inf u[,] (x, t; u0 ) [,j+1 ,M ]
in
j+1 .
in
(5.31)
in
j+1
t.
for all M M
Hence,
min
in
j+1
in
j+1 .
in
(5.32)
124
on j+1,n \
t
u=M
u=0
u(, 0) = u0
in
j+1,n
(5.33)
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 ,
in
j+1,n .
(5.34)
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
in
j+1 ,
Dynamics: Metasolutions
5.5
125
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
if d0 ,
0,
lim u[,,] (, t; u0 ) =
in C(),
(5.37)
t
[,,] ,
if > d0 ,
where u[,,] stands for the unique solution of (5.36).
The main result of this section can be stated as follows.
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)
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 := [,] .
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)
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
Dynamics: Metasolutions
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)
for all t t .
(5.46)
(5.47)
(5.48)
for all t t
Consequently,
lim [,,] = Mmin
[,j ]
0
(5.49)
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 ( \
u[,,] (, tC ; u0 ) > C
[,,] 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
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,
mj
q [
[
j=1 i=1
ij .
Dynamics: Metasolutions
129
5.7
Biological discussion
C := 1 [, B1 ]|B1 |2/N ,
130
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
Dynamics: Metasolutions
131
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
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
Comments on Chapter 5
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 .
Dynamics: Metasolutions
133
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
Chapter 6
A canonical one-dimensional
problem
6.1
6.2
6.3
6.4
6.5
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
2016 by Taylor & Francis Group, LLC
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.
Z
z
r
21
Z
2(p + 1)
dx
< .
h(s) ds
dx
p+1 z p+1
H
x
z
lim u(x) = 0.
x0
and
h(u) := g(u)u.
Z
b(x) =
a
x
1
p+1
p+1
p1
dy,
x (0, R].
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, ).
6.1
139
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)
(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,
u1 (x2 ) = u2 (x2 ),
and
x (x1 , x2 ).
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)
0 < x < L2 ,
u(L2 ) 0.
(6.9)
141
(6.10)
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 ) =
x[x1 ,x2 ]
142
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 ),
, 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
< 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)
(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
is a supersolution of
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
(
).
uM uM
in [
, ).
`min `min
in
[
, ).
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
(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
in
[, )
for all
M > 0, > 0.
(6.21)
`()
= , (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 ).
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
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
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
x L,
147
satisfies
() = U (0) = ,
U
(L) = U (L ) U (L) = M,
U
v 0 = a(x)h(u),
u(L) = M, u0 (L) = v0 ,
(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
boundary value problem
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
b0 (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)
Z
A(x) =
1
p+1
p+1
p1
p+1
a p1 (R) x p1
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)
A0 (x)
b00 (x)
=
lim
= .
x0 A(x)
b0 (x)
A(x)
x
1
A
!1
x (0, R],
1
A
lim A(x)
x0
!
= 0.
(6.32)
A(x)
x
1
=
A
A(x)
ds + A(x)
A(s)
1
x + A(x)
A
1
.
A
1
A
!
,
b00 (x)
b00 (x)
b0 (x)
= lim
lim
= ,
x0 b0 (x) x0 b(x)
b(x)
150
b(x)b00 (x)
= I0 (0, ).
[b0 (x)]2
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)
p+1 1
b (x) =
p 1 A(x)
00
1
1
p+1
x (0, R).
Z
1
p+1
2
p1
x [0, R].
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].
b00 (x)b(x)
.
[b0 (x)]2
b(x)b00 (x)
.
[b0 (x)]2
(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
and
for all
x (0, ].
(6.38)
1 < A (x)
x
1
M
A
(6.39)
152
x (0, ],
(6.40)
(6.41)
for x 0,
or
a(x) x ln x for x 0,
or
a(x) e1/x
for x 0.
for all
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)
a(x)
(> 0),
a0 (x)
x (0, ],
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) =
x (0, ].
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
1
p+1
= ,
(6.47)
x (0, ].
(6.48)
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
(6.49)
Hence,
Qa (xn ) < p + 1,
Qa (yn ) > p + 1,
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
(6.50)
(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,
[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
1
1
,
Qa (0) p + 1
Thus, setting
:=
Qa (0)(p + 1)
> 0,
p + 1 Qa (0)
x (0, ].
157
a(x) = (x + C) ,
x (0, ].
1
p1
Z
p+1
p+1
p1
1
ds
= p1
p+1
+p+1
p+1
p1
+p+1
p1
lim A (x)
x0
1
A
+p+1
,
+2
(6.54)
a 0
a0
(a0 )2 aa00
aa00
=
1
(a0 )2
(a0 )2
then,
Qa (0) := lim Qa (x) =
x0
p+1
.
(p + 1)a1 + 1
aa00
(a0 )2
is non-oscillating.
x (0, ],
(6.55)
Q0a (x)
,
p+1
x (0, ].
158
(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 ),
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 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
159
p+1
=
lim
p + 2 x0
Rx
0
a p+1
!
+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) =
which ends the proof.
6.3
p+1
[(1 a1 )Qa (0) + 1],
p+2
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)
if u .
(6.60)
Subsequently, we set
C := max{M, } + 1 > 0
(6.61)
160
2
3HC
1
p1
>
0
k0
<
2
Hm
1
p1
<
0
(6.62)
(6.63)
b0 (x) < 0,
lim b(x) = ,
x0
(6.64)
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)
.
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
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)),
(6.66)
`()
< 0
b()
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)
162
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)
x (0, ].
(6.74)
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))
b (s) ds ` () ` (x)
b (s) ds.
(6.75)
x h(b(s))
x h(b(s))
Subsequently, we set
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
s0
h(b(s))/(b(s))p p p
=
> 0.
h(b(s))/(b(s))p
163
QL := inf
(6.76)
is well defined. As b00 (x) > 0 for all x (0, ], substituting (6.76) into (6.75),
it becomes apparent that
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)
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)
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)
b00 (x)
=
b0 (x)
implies
lim
x0
` 00 (x)
= ,
`0 (x)
164
6.4
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)
`(x)
,
b(x)
x (0, ].
(6.80)
x (0, ].
Hence,
0 < qL := lim inf q(x) qM := lim sup q(x) .
x0
x0
lim q(tn ) = qM ,
lim q(sn ) = qL ,
q00 (tn ) 0,
q00 (sn ) 0.
on (0, ].
0
b
ah(qb)
0 b
+
2q
+q=
.
00
00
b
b
b00
(6.81)
165
b(tn )
a(tn )h(q(tn )b(tn ))
+ q(tn ) =
,
00
b (tn )
b00 (tn )
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
q0 = I 0
p+1
p1
6.5
p+1
p1
H p1 ,
Comments on Chapter 6
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,
+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
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
Z
b0 (x) :=
a
x
1
q+1
q+1
q1
dy,
provided
q+1
q1
q+1
q1
A q1
b0 (x)b000 (x)
(0, ).
x [b00 (x)]2
J0 := lim
Chapter 7
Uniqueness of the large solution
under radial symmetry
7.1
7.2
7.3
7.4
7.5
170
172
172
174
177
183
184
for all
t > 0,
(7.1)
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 .
170
(7.5)
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
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)
and
u > 0.
(7.10)
171
2
> 0,
r(g)
= % ,
(7.11)
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.
(7.14)
u > 0,
1.
(7.15)
for all 1, u 0.
Therefore,
g(u) g(u)
H := lim
for some p > 1. For example, for any integer n 1 and real numbers
0 < r1 < r2 < . . . < rn ,
(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
R
>1
R
0 |x x0 | R .
173
in BR (x0 ).
Then,
=
L
on BR (x0 )
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
Lmin
[,BR (x0 )] L
(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,
7.2.2
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),
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
if r (r0 , r1 ),
(7.20)
> 0.
N 1 0
(r0 ) = H((r0 )) < 0,
r0
(7.21)
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.
Subsequently, we set
:=
R2 Rm
> 1,
R2 Rm
(0, R2 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 ,
N 1 0
N 1
0
00
(r) = 2 min
min
(%)
(%)
r
r
N 1%
00
0
= 2 min
(%)
min
(%).
% r
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
N 1 0
(r) (r) 2 a(R2 r)g( (r)) (r).
r
Rm < r < R2 .
Then,
0 (Rm ) = 0,
lim (r) = ,
rR2
N 1 0
(r) (r) 2 a(R2 r)g( (r)) (r).
r
N 1 0
(r) (r) a(R2 r)g( (r)) (r),
r
Rm < r < R2 .
Rm < r < R2 ,
in (Rm , R2 )
7.3
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
lim
t0
(7.23)
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)
b(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)
b(t)
1
=
,
0
b (t)
(ln b)0 (t)
t (0, R],
with
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) = 0,
179
(7.31)
`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)
r = |x x0 |,
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)
and
lim ,A (r) = .
rR
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))
180
(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
(7.37)
as A .
(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
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)
lim z(C) = 0.
(7.40)
C0
rR
g( ,C (r))
g(`(R r))
= lim
rR
,C (r)
!p1
= (1 )p1 .
`(R r)
(7.41)
lim
rR
,C (r)
`(R r)
and
,C ,A
in [0, R).
= 1 ,
(7.42)
182
|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)
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 |,
in (R1 , R2 ),
(7.45)
Subsequently, we set
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 ),
183
rR2
(r)
(r)
= lim
1 + .
`(R2 r) rR1 `(r R1 )
(7.46)
7.4
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
in
C().
in C(),
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
Comments on Chapter 7
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
(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
a(t)
>0
t
(7.50)
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,
(7.52)
(N 1)H() ( + 1)C1
+p+3
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)
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
and Q(0) = 0,
(7.56)
(7.57)
188
R t ! 12
p
a
1
0
t 2 = tQ(t),
a(t)
t 0,
a(s) ds,
0
a1 a a2 ,
if t 0.
Chapter 8
General uniqueness results
8.1
8.2
8.3
8.4
8.5
8.6
190
192
193
201
202
204
This chapter establishes the uniqueness of the positive solution of the singular
boundary value problem
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)
mj
[
i0,j ,
1 j q0 .
i=1
in D,
du = u + a(x)g(u)u
Bu = 0
on B ,
(8.3)
u=
on ,
189
2016 by Taylor & Francis Group, LLC
190
,
B :=
+ b(x),
on 0B ,
on 1B ,
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
RN
nx
191
BR
z (z)
(z) + R
|z (z)|
D.
x D,
BR (zx ) D,
R (zx ) D = {x}.
B
(8.5)
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],
192
8.2
ax (0) = 0,
t (0, ),
(8.8)
3yx
t0
ay (t)
=1
ax (t)
uniformly in x .
(8.9)
p+1
p1
p+1
p1
H p1
(8.10)
bx (t)b00x (t)
.
[b0x (t)]2
(8.11)
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,
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)
Subsequently, we set
: 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)
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
194
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
(8.17)
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 )
in BR (x(R+)nx )
in BR (x Rnx )
(8.18)
195
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
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)
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 ,
197
|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
(8.25)
198
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 ].
in D.
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)
u = u (1+)
ax (dist (z, A0 ))g(u)u
u=M
in A0 ,
on A0 .
200
(8.28)
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)
p+1
p1
p+1
p1
H p1
p+1
p1
p+1
p1
H p1
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)0
z D.
(z) (z)
L2 (z)
L(z) (z)
L
(8.32)
J(x) := I0,x
p+1
p1
p+1
p1
x ,
H p1 ,
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
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
R
C0
t
2
p+1
2
p
p1
p+1
p1
p+1
p1
H p1 .
L(x tnx )
t
2
p1
1
p1
=a
(x)
2 p+1
p1p1
1
p1
H p1 ,
8.5
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, ) }.
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 .
in D \ D .
Consequently,
(1 )L2 L1 (1 + )L2
in D
204
Since
a0x (t) = 5 cos(5(0.8 + t)),
it is apparent that
ax (t) > 0,
ax (t)
a0x (t)
00
8.6
Comments on Chapter 8
(8.35)
205
(L(x))
= 1,
d(x)
where
1
(t) :=
2
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)
(8.39)
L(x)
=1
(d(x))
(8.40)
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
u(, 0) = u0 > 0
in (0, ),
on (0, ),
in ,
212
224
230
233
237
240
240
245
252
254
256
257
265
(9.1)
209
2016 by Taylor & Francis Group, LLC
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 ,
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,
= u
if a(x) = 0,
u + a(x)|u|p u
> 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 )
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
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
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
(9.6)
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,
214
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
(9.8)
p+1
u
(u u) =
p+2
u
u2 ( d0 ) a(x)p+2 .
(9.9)
(u u) = (p + 1)
u2 > 0,
u
u
u
( 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
(9.11)
(9.12)
|0
|\
(a) = ,
216
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 ]
for all (0 , 0 ).
217
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 [, \
218
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=
220
221
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,
222
sin(5x),
x 25 , 35 ,
a (x) :=
sin(5x),
x 15 , 25 35 , 45 ,
(9.15)
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
9.2
225
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)
226
(, 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)
(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,
(9.20)
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
(9.21)
(9.22)
0
u0
229
(9.23)
0
u0
2
Z
(0 u0 u0 0 ) = 2
0 2
0 u0 > 0,
u0
(9.24)
for all s (, ).
(9.25)
Since
u0 (s) = 0 + 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,
9.3
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
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,
' .
(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
< .
232
and hence,
lim u()(x ) = ,
a(x ) > 0,
a(x)p+2 (x) dx 0.
D :=
(9.27)
as s 0.
9.4
233
C0 ()
d0
lim k u0 kC()
= 0.
(9.29)
234
(9.30)
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 ()
for all
(0 , 0 ).
Moreover,
lim k u0 kC()
= 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
and
lim k kC()
= 0.
d0
Subsequently, we set
d0 0,
0 u0 .
236
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
+
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 ()
for all
(b , ],
but b = 0.
such that
(A3) There is t <
> 0,
1 [L(, ), ] > 0
for all
(t , ],
237
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
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)
[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
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)
x ,
9.6
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
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
p+1
D
241
where
Z
:= 1 [d, D] ,
:= d
D
[,a ] d,
n
(9.38)
(9.39)
in .
(9.40)
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
(9.41)
d
u
d + ( 1 [d, D]) I(t) + A
up+1 dx,
n
D
D
where
0
:=
d
,
dt
A min a = min a+ ,
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
(9.42)
(9.43)
x > 0,
(9.44)
Thus,
f (x) = 0
if and only if x = xL :=
(p + 1)A
p1
(p + 1)A
p+1
p
(p + 1)A
p1
+ 0.
Equivalently,
A
p p p+1
p+1
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
t t0 > 0,
t t0 > 0,
(9.47)
i
Ah
1 ep(tt0 )
t > t0 ,
(9.48)
and
lim I(t) = ,
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
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 +
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)
(9.50)
for all
and
lim u(xk , tk ; , a, u0 ) = ,
246
where
Z
I(w) :=
a(x)
|w|2 w2
|w|p+2
2
2
p+2
dx,
w 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
if n = 1,
if n > 1,
(9.51)
Eu0 (t) = .
(9.52)
(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
+
\
if x + ,
if x \ + ,
+,
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
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 < p < (N + 2)/(N 2) 1 = 4/(N 2) if N 3.
(H2) There exists (R1 , R2 ) such that a0 (r) 0 for each r [0, ], and
Z
a(r)rn1 dr > 0 and 1 [d, B ] < < d1 .
0
+ (Tc , ),
for all (x, t)
(9.54)
while
u
(, t) Lmin
+]
[,a ,\
+
in \
(9.55)
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
R2 R
and
R ] = .
lim 1 [, BR \ B
2
R2 R
R1 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 .
for all
x B , t > 0.
(9.59)
250
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 ) = +
(9.60)
satisfies
w(0, t) = 0,
w(R, t) < 0,
w(r, 0) 0,
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
du = u + a up+1
u=
u=0
+,
x\
+,
in \
on + ,
on .
251
In the contrary case when there exists a constant C > 0 such that
u(R1 , t; , a, u0 ) C
(9.61)
the limit
g(R1 ) := lim u(R1 , t; , a, u0 )
t
dw = w + a wp+1
w=g
w=0
in
+
\
+,
in \
on + ,
on .
d1
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
(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.
\
t
u=0
u
=M
u(, 0) = u0
in
on
on
in
( \ + ) (0, )
(0, )
+ (0, )
\ + .
254
in ,
9.8
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 .
In particular,
lim u(, t; , a, u0 ) Lmin
0]
[,a ,\
in D + .
(9.69)
255
in D + ,
0,
in
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)
uniformly in compact subsets of
256
lim u
(, t; u0 ) =
for all
+
Lmin
[,a , ]
t > Tb ,
0 \ ,
in
in ,
9.9
Dynamics for d2
0,
in
9.10
257
Comments on Chapter 9
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
L(x),
,
M(x) :=
L(1 x),
x [0, ),
x [, 1 ],
x (1 , 1],
(9.77)
(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
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
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
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
265
9.10.1
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
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
in (0, ),
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)
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
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
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,
u = and 0 v <
on D ,
such that
(Mu , Mv ) = (Lu , Lv )
in
and
Mu =
in
(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
in ,
on ,
(10.8)
270
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
(10.9)
1 i mj , 1 j q0 ,
1 j q0 1.
(10.10)
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 ).
q0 = q0 (1, 0),
1 j q0 .
(10.12)
(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
0,j (d, V ) :=
[
i=1
i0,j (d, V ),
1 j q0 (d, V ),
(10.14)
272
(10.15)
(10.17)
10.1
Preliminaries
10.1.1
in C().
in C(),
max
in
(10.18)
,
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
in
(10.19)
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
> d2 0 .
(10.21)
and
lim v(, t) = 0
in C().
274
and
lim v(, t) = ,v
in C(),
and
lim v(, t) = 0
in C(),
max
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)
(10.22)
for all x and t > 0. According to Theorem 10.2, we already know that
lim U (, t) = 0
in C()
in C().
(10.23)
in C(),
(10.24)
in C().
275
(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)
in C().
for all x , t T.
Consequently,
lim inf v(, t) lim w(, 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
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
(10.28)
which entails
lim inf u(, t) =
in \ j
and
min
in j .
min
L[,j ] [,j ,M ] ,
it becomes apparent that, for sufficiently small > 0 and every M > 0,
lim inf u(, t) [,j ,M ]
in j .
in j ,
which implies
lim inf u(, t) =
t
and
min
in \ j1
min
L[d1 j ,j ]
in j1 .
0,j \ ,
in
0,j ,
in j = j1 \
(10.29)
10.2
277
(10.30)
(10.31)
278
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
f1
(x, u0 )u0 0
u
and d
f2
(x, v0 )v0 0
v
in .
(10.34)
279
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 ).
> 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)
d2 0
lim 0 () = .
(10.38)
(10.39)
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
for all x ,
f2 (x, )
for all x ,
(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).
> 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)
281
. Moreover, by construction,
1 j q0 ,
1 i mj ,
(10.43)
(10.44)
d2 0
(10.45)
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
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)
0,j
b
b
< min .
d 10,j1 d
(10.47)
282
(10.49)
> d2 0 .
(10.51)
(10.52)
,v
= d1
283
Sj ()
d1
,v 1
b
= lim 1 [ + b
, 0,j ] = min
1 d
0,j
0,j d
0,j2
1
(10.53)
> d2 0 ,
1 i mj ,
> d2 0 ,
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
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
with (j1 , i1 ) =
6 (j2 , i2 ) the following holds
Z
Z
b 2i1 6=
i1
0,j
1
0,j1
b 2i2 .
(10.54)
(10.55)
i2
0,j
2
0,j2
q0
X
mj
j=1
,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
285
(10.58)
as d2 0 .
(10.60)
and
1 D = 0.
(10.61)
b
D .
d3
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
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.
for all 1 j 5,
287
:= { (, ) Q
:= { (, ) Q
:= { (, ) Q
:= { (, ) Q
:
:
:
:
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
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
in
max
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
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)
289
(10.65)
n1
for all n 1,
lim n = d1 j ,
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
V u2n C
u2n ,
n 1.
n 1.
290
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 .
|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.
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 ) = (, )
Setting
u
n :=
un
,
kun k
n 1,
for all n 1.
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
292
un = n
u
n
b
un vn ,
n 1,
0
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)
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 )
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
10.4
293
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) :=
.
(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
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)
(10.71)
(10.72)
(10.73)
1
( , ) P C+
(,0,,v ) ( , )
(10.74)
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)
(10.77)
and v ,v .
(10.78)
in ,
on .
(10.79)
296
in ,
on .
(10.80)
and
+
( , 0, ,v ) 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 ]
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
(10.82)
(10.83)
n 1.
(10.84)
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
300
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
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
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
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
1
min
0,1 \ ,
M[d1 1 ,1 ] = in
necessarily
lim C = .
0
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
1 [d2 + cMmin
[d1 1 ;1 ] , 1 ] 1 [d2 + cM[d1 1 ,1 ] , 1 ]
1 [d2 , 1 ] + cM C ,
cM := max c.
Hence,
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)
(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 < %.
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 } ]
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 ].
305
(10.90)
306
1 2
,v ]
= 1 [d1 + b1d
1 2
1 2
,v )].
= 1 [d1 + b1d
1 2
1 2
,v )]
= d1 2
1 2
,v ).
307
308
10.5
for all x .
(10.91)
=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
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
(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
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)
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 .
(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,
n 1,
that
= 0 [d1 , af1 (, un ) + bvn ],
n 1.
313
() < n < ,
n 1.
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
j {1, 2}.
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
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
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
(10.100)
(10.101)
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)
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.
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)
(10.104)
(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 )]
319
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)
10.7
on j
and
lim
dist(x,j )0
(x) =
320
Obviously,
j ; R+ )
n C (
and
n n+1
for all n 1.
n 1,
n n0 .
and
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,
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.
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
Z
h(1 ), 2 i +
d2
Z
(V + c)1 2 = 2
1 2 .
j
1 2 .
(10.112)
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)
weakly in H01 (j )
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.
324
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
and
j]
[,j ]
v(, 0) = v0
in .
Then
{,v;cMmax
[,
10.8
j]
[,j ]
}.
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
(10.116)
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 ]
< 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
in j ,
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.
327
(10.118)
0 < < 0 ,
(10.120)
328
for all n 1.
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
Comments on Chapter 10
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, ),
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.
[7] D. Aleja and J. L
opez-Gomez, Concentration through large advection,
J. Diff. Eqns. 257 (2014), 31353164.
[8] D. Aleja and J. L
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
2016 by Taylor & Francis Group, LLC
332
Bibliography
J. L
opez-G
omez and F. R. Ruiz del Portal Eds.), pp. 160, Elsevier B.
V., Amsterdam, 2005.
Bibliography
333
334
Bibliography
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
Bibliography
337
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 probl`eme danalyse relatif `a 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
Bibliography
341
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
Bibliography
347
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-Amp`ere 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
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
356
Index
of characterization of the
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
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