Professional Documents
Culture Documents
Modelacion Matematica
Modelacion Matematica
FRACTIONAL REACTION-DIFFUSION
PROBLEMS
TESIS PARA OPTAR AL GRADO DE
MODELACION
MATEMATICA
EN COTUTELA CON LA UNIVERSIDAD PAUL SABATIER, TOULOUSE III
PROFESORES GUIA:
PATRICIO FELMER AICHELE
JEAN-MICHEL ROQUEJOFFRE
MIEMBROS DE LA COMISION:
MARTA GARCIA-HUIDOBRO CAMPOS
YANNICK SIRE
SANTIAGO DE CHILE
2014
Resume
Cette th`ese porte sur deux probl`emes differents : dans le premier, nous e tudions le comportement en temps long des solutions des e quations de reaction diffusion 1d-fractionnaire de
type Fisher-KPP lorsque la condition initiale est asymptotiquement de type front et decrot a`
linfini plus lentement que xb , o`u b < 2 et (0, 1) est lindice du laplacien fractionnaire
(Chapitre 2). Dans le second probl`eme, nous e tudions la propagation asymptotique en temps
des solutions de syst`emes cooperatifs de reaction-diffusion (Chapitre 3).
Dans le premier probl`eme, nous demontrons que les ensembles de niveau des solutions se
deplacent exponentiellement vite en temps quand t tend vers linfini. De plus, une estimation
ii
Resumen
Esta tesis trata sobre dos problemas diferentes: en el primero, se estudia el comportamiento
en tiempos grandes de las soluciones de la ecuacion de reaccion diffusion 1d-fraccionaria del
tipo Fisher-KPP cuando la condicion inicial es asintoticamente como un frente y decae al infinito
mas lento que xb , donde b < 2 y (0, 1) es el ndice del laplaciano fraccionario (Captulo
2). En el segundo problema, se estudia la propagacion asintotica en tiempo de las soluciones de
sistemas cooperativos de reaccion difusion (Captulo 3).
En el primer problema, se demuestra que los conjuntos de nivel de las soluciones se desplazan exponencialmente rapido cuando el tiempo t tiende a infinito. Mas aun, una estimacion
cuantitativa sobre el movimiento es obtenida en funcion del decrecimiento al infinito de la condicion inicial.
En el segundo problema, se demuestra que la velocidad de propagacion es exponencial
en tiempo y se encuentra un exponente preciso que depende del orden mas pequeno de los
laplacianos fraccionarios considerados y de la nolinearidad. Se hace notar que este ndice no
depende de la direccion espacial de la propagacion.
Palabras Clave: Problemas de reaccion-difusion, KPP, sistemas cooperativos, propagacion
rapida.
iii
iv
Acknowledgements
First and foremost I want to thank my advisors Patricio Felmer and Jean Michel Roquejoffre.
I appreciate all their contributions of time, ideas and support to make my Ph.D. experience
productive and stimulating. Also, I am grateful for his trust, his encouragement, and his good
disposition. Their joy and enthusiasm for the mathematics was contagious and motivational for
me.
The group of students of Patricio Felmer, Ying, Huyuan, Cesar and Erwin has contributed
immensely to my personal and professional time at Santiago. The group has been a source of
friendships as well as good advice and collaboration. In Toulouse, I am especially grateful with
Anne-Charline for the help that she gave me when I arrived for first time to France and for the
time and effort spent to complete our paper. My experience as Ph.D student was made enjoyable
in large part due to the many friends that became a part of my life. Clara, Natalia, Maximiliano,
Luis and Juan Carlos, I am grateful for time spent with you and for all good moments that we
lived together.
For this dissertation I would like to thank my reading and oral committee members: Marta
Garcia-Huidobro and Yannick Sire for their time, interest, helpful comments and insightful
questions.
I gratefully acknowledge the funding sources that made my Ph.D. work possible. My studies in Chile were funded by CONICYT (Comision Nacional de Investigacion Cientfica y
Tecnologica de Chile) and during my stay in France I was founded by SENESCYT (Secretaria
Nacional de Educacion Superior, Ciencia, Tecnologa e Innovacion del Ecuador). Also I am
grateful with Escuela Politecnica Nacional del Ecuador and Center of Mathematical ModelingChile for the support that gave me to study my Ph.D.
Lastly, I would like to thank my family for all their love and encouragement during these
four years that I was abroad. To my parents who are with me at each moment, for their advices
and support in all my pursuits. To my grandmother for all the love and happiness that gives me.
And most of all to my wife Carolina, for the love, support, encouragement, and patience during
all the stages of my Ph.D. Thank you.
Table of Contents
Abstract-Resume
ii
Acknowledgements
Introduction
14
17
21
2. Fast propagation for fractional KPP equations with slowly decaying initial conditions.
26
2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
27
31
33
40
43
43
Table of Contents
46
52
52
63
66
70
77
Bibliography
84
Table of Contents
Introduction
Reaction-diffusion models have found widespread applicability in a surprising number of
real-world models, including areas as, chemistry, biology, physics and engineering. But not
only physical phenomena can be the result of a diffusive models. Stochastic processes in mathematical finance are often modeled by a Wiener process or Brownian motion, which lead to
diffusive models. The simplest reaction-diffusion models are of the form
ut u = f (u)
(1)
where f is a nonlinear function representing the reaction kinetics. One of the most important
examples of particular interest for us include the Fisher-KPP equation for which f (u) = u(1
u). The nontrivial dynamics of these systems arises from the competition between the reaction
kinetics and diffusion.
At a microscopic level, diffusion is the result of the random motion of individual particles,
and the use of Laplacian operators in the model rests on the key assumption that this random
motion is an stochastic Gaussian process. However, a growing number of works have shown
the presence of anomalous diffusion processes, as for example Levy processes, thus, reactiondiffusion equations with fractional Laplacian instead of standard Laplacian appear in physical
models when the diffusive phenomena are better described by Levy processes allowing long
jumps, than by Brownian processes, see for example [39] for a description of some of these
models. The Levy processes occur widely in physics, chemistry and biology and recently these
models that give rise to equations with the fractional Laplacians have attracted much interest.
Our particular aim on this type of anomalous diffusion problems is focussed to the study
of large-time behavior of the solution of the Cauchy problem for fractional reaction-diffusion
equations
ut + () u = f (u)
u(0, x) = u0 (x)
(2)
(3)
with (0, 1) in one spatial dimension, where () denote the fractional Laplacian. The
nonlinearity f is assumed to be in the Fisher-KPP class. More precisely, the nonlinearity is
assumed to have two zeros, an unstable one at u = 0 and a stable one at u = 1.
The reaction diffusion equation (1) with Fisher-KPP nonlinearity has been the subject of
intense research since the seminal work by Kolmogorov, Petrovskii, and Piskunov [33]. Of
particular interest are the results of Aronson and Weinberger [2] which describe the evolution of
the compactly supported data. They showed that for a compactly supported initial valuep
u0 , the
movement of the fronts are linear in time. In addition, there exists a critical speed c = 2 f 0 (0)
for which the problem (2)-(3) admits planar traveling wave solutions connecting 0 and 1, that
is, solutions of the form u(t, x) = (x ct), which move with speeds c c .
Many papers have been concerned with the large-time behavior of solutions of equation
(1) or more general reaction diffusion equations with exponentially decaying initial conditions,
leading to finite propagation speeds, see for example [11], [32], [35], [41] and references in
[29].
In contrast with the results just mentioned, where finite speed of propagation is obtained
whenever the initial value decays faster that an exponential, it is shown by Hamel and Roques
[29] that when the initial condition is globally front-like, and it decays slower than any exponential then the asymptotic behavior of the front exhibits infinite speed and a very precise estimate
can be obtained for the propagation of the level sets of the front in terms of the initial value,
giving a precise superlinear behavior.
Moreover, Berestycki, Hamel and Roques [7] prove existence and uniqueness results for the
stationary solution associated to (2) and they then analyze the behavior of the solutions of the
evolution equation for large times. These results are expressed by a condition on the sign of the
first eigenvalue of the associated linearized problem with periodicity condition.
Regarding (2)-(3) with (0, 1) and Fisher-KPP nonlinearity, in connection with the discussion given above for the case = 1, in the recent papers [14] and [15], Cabre and Roquejoffre show that for compactly supported initial value, or more generally for initial values
decaying like |x|d2 , where d is the dimension of the spatial variable, the speed of propagation becomes exponential in time with a critical exponent c = f 0 (0)(d + 2)1 , they also
show that no traveling waves exists for this equation, all results in great contrast with the case
= 1. Moreover, if the initial data u0 [0, 1] is nonincreasing and for x > 0 decay faster than
x2 , then the mass at + makes the front travel faster to the left, in this case with a speed
c = f 0 (0)/2. All these results will be formally established in Chapter 1. Additionally we
recall the earlier work in the case (0, 1) by Berestycki, Roquejoffre and Rossi [10], where
it is proved that there is invasion of the unstable state by the stable one, also in [10], the authors
derive a class of integro-differential reaction-diffusion equations from simple principles. They
then prove an approximation result for the first eigenvalue of linear integro-differential operators of the fractional diffusion type, they also prove the convergence of solutions of fractional
evolution problem to the steady state solution when the time tends to infinity. For a large class of
nonlinearities, Engler [25] has proved that the invasion has unbounded speed. For another type
of integro-differential equations Garnier [27] also establishes that the position of the level sets
moves exponentially in time for algebraically decaying dispersal kernels. And in a recent paper
Stan and Vazquez [42] study the propagation properties of nonnegative and bounded solutions
of the class of reaction-diffusion equations with nonlinear fractional diffusion.
Chapter 2 is concerned with the study of the phenomena described by Hamel and Roques in
[29] in the case of nonlocal diffusion, that is, when (0, 1) in equation (2)-(3). In particular,
we study the asymptotic behavior of solutions with slowly decaying, globally front-like initial
value. We state that, for b < 2, the central part of the solution moves to the right at exponential
speed f 0 (0)/b, which is faster than c , the exponential speed for solutions with initial values
decaying faster than x2 . Thus we show that the exponent 2 is critical regarding the speed of
propagation of the solution. Furthermore, we prove that the initial condition u0 can be chosen
so that the location of the solution u be asymptotically larger than any prescribed real-valued
function.
The study of propagation fronts was also done in reaction diffusion systems, in this line,
Lewis, Li and Weinberger in [37], studied spreading speeds and planar traveling waves for a
particular class of cooperative reaction diffusion systems with standard diffusion by analyzing
traveling waves and the convergence of initial data to wave solutions. It is shown that, for a
large class of such cooperative systems, the spreading speed of the system is characterized as
the slowest speed for which the system admits traveling wave solutions. Moreover, the same
authors in [44] establish the existence of a explicit spreading speed for which the solution of
the cooperative system spread linearly in time, when the time tends to +.
Follow the line, it is our aim to study the spread speeds of solutions of reaction diffusion
cooperative system when the standard Laplacians are replaced for instance by the fractional
Laplacian with different indexes. Chapter 3 is devoted to study the time asymptotic propagation
of solutions to the fractional reaction diffusion cooperative systems. We prove that the propagation speed is exponential in time, and we find a precise exponent depending on the smallest
index of the fractional Laplacians and of the principal eigenvalue of the matrix DF (0) where
F is the nonlinearity associated to the fractional system. Also we establish the existence of
a constant steady state solution for the system and we prove the convergence of the solution
towards the steady solution for large times.
Chapter 1
Preliminaries and background
In this chapter, we provide some elementary properties of fractional Laplacian operators, also
we introduce the main results concerning to the large time behavior of solutions of reactiondiffusion problems with Fisher-KPP nonlinearities. Our principal goal in this chapter is to
note the differences between the speed propagation obtained when the problem has standard
diffusion or fractional diffusion. Moreover, we present some results related with the behavior
of solutions of weakly coupled reaction-diffusion equation systems.
1.1.
Fractional Laplacian
Let consider the Schwartz space S of rapidly decaying C 1 functions in Rn , hence, for any u S
and (0, 1), the fractional Laplacian denoted by (4) is defined as
Z
u(x) u(y)
C(d, ) = (2+ 2 )
( d2 + )
()
Let note that the principal value is not necessary in the definition if (0, 1/2). Indeed,
for any u S, we have
Z
Z
Z
u(x) u(y)
|x y|
1
dy C
dy + kukL (Rd )
dy
d+2
d+2
d+2
Rd | x y |
BR (x) |x y|
BR (x)c |x y|
Z
Z
1
1
=C
dy +
dy
d+21
d+2
BR (x) |x y|
BR (x)c |x y|
Z R
Z +
1
1
dr +
dr < +
=C
2
r2+1
R
0 r
where C is a positive constant.
Now, by two changes of variables, z = y x and z 0 = x y, we can rewrite (4) u as:
Z
Z
1
u(x) u(x + z)
u(x) u(x z 0 ) 0
(4) u = C P.V.
dz + P.V.
dz
2
|z|d+2
|z 0 |d+2
Rd
Rd
Z
u(x + y) + u(x y) 2u(x)
1
dy
= C P.V.
2
|y|d+2
Rd
Using a second order Taylor expansion, we have
kD2 ukL
u(x + y) + u(x y) 2u(x)
|y|d+2
|y|d+22
which is integrable near 0, thus we can write the singular integral in (1.1) as a weighted second
order differential quotient
Z
1
u(x + y) + u(x y) 2u(x)
(4) u = C
dy
2
|y|d+2
Rd
To end this part, we state a result which show that the fractional Laplacian (4) can be
viewed as a pseudo-differential operator of symbol ||2 .
Theorem 1.1 Let (0, 1) and let (4) : S L2 (Rd ) be the fractional Laplacian operator
defined by (1.1). Then, for any u S,
() u = CF1 (||2 (Fu)),
Rd
1.2.
f (0) = f (1) = 0
and
(1.1)
We are interested in the large time behavior of solutions u = u(t, x) to the Cauchy problem
t > 0, x Rd
xR
t u 4u = f (u),
u(0, x) = u0 (x)
(1.2)
(1.3)
the following result of Aronson and Weinberger [2] implies that there is a critical speed c such
that for fairly general initial data u0 that are non-negative and compactly supported, the invasion
rate in which the stable state 1 invades the unstable state 0 is linear in time.
p
Theorem 1.2 Let c = 2 f 0 (0) and let u the solution of (1.2)-(1.3) with u0 6 0 compactly
supported in Rd and satisfying 0 u0 (x) 1. Then:
- if c < c , then limt+ inf |x|ct u(t, x) = 1,
- if c > c , then limt+ sup|x|ct u(t, x) = 0.
Freidlin and Gartner [28] extended the spreading properties to space periodic media and to
some classes of random media using probabilistic tools, proofs using dynamical systems or
PDE arguments are given in [6], [24], [46]. Hence, starting from a positive compactly supported
initial data, in [28] is stated that there exists an asymptotic directional spreading speed w (e) >
0 in each direction e, in the sense that
- if 0 c < w (e), then lim inf t+ u(t, x + cte) > 0
- if c > w (e), then limt+ u(t, x + cte) = 0
locally uniformly in x Rd , where w (e) is characterized by:
w (e) =
c (e0 )
e0 Sd1 ,e0 e>0 e0 e
min
where c (e0 ) is the minimal speed of pulsating traveling fronts in the direction e0 .
Moreover, if we consider the case d = 1, it is indeed well-known that the equation ut =
uxx + f (u)p
admits a family of planar traveling waves u(t, x) = c (x ct) for all speeds
c c := 2 f 0 (0), where, for each speed c [c , +), the function c : R (0, 1) satisfies
00c + c0c = f (c ) in R,
c () = 1, c (+) = 0,
Furthermore, the function c is decreasing in R and unique up to shifts. Now, suppose that
u0 : R [0, 1] is uniformly continuous and satisfy
u0 > 0 in R,
(1.4)
p
If u0 (x) is equivalent as x + to a multiple of ex with 0 < < = f 0 (0), then
0
u(t, x) converges to a finite shift of the front c (x ct) as t +, where c = + f (0) > c .
(1.5)
and the function u0 is assumed to decay more slowly than any exponentially decaying function
as x +, in the sense that
> 0, x R, u0 (x) ex f or x [x , )
(1.6)
Hamel and Roques [29] prove that all level sets of the solutions move infinitely fast as time goes
to infinity and the locations of the level sets are expressed in terms of the decay of the initial
condition. Hence, [29] contains the first systematic study of the large-time behavior of solutions
of KPP equations with slowly decaying initial conditions. These results are in sharp contrast
with the well-studied case of exponentially bounded initial conditions, where the solution u
converge in some sense to some traveling fronts with finite speed as t +.
Before to state the results, we need to introduce a few notations. For any (0, 1) and
t 0, denote by
E (t) = {x R : u(t, x) = }
the level set of u of value at time t. For any subset A (0, 1], we set
u1
0 (A) = {x R : u0 (x) A},
the inverse image of A by u0 .
Theorem 1.3 Let u be the solution of (1.2)-(1.3), where f satisfies (1.1) and the initial condition
u0 : R [0, 1] satisfies (1.4) and (1.6).
a) Then
lim u(t, x) = 0 t 0 and
x+
10
b) For any given (0, 1), there is a real number t > 0 such that E (t) is compact and
non-empty for all t t , and
min E (t)
= +
(1.7)
lim
t+
t
Part b) of Theorem 1.3 simply says that the level sets E (t) of all level values (0, 1)
move infinitely fast as t +, in the average sense (1.7). As already announced above, this
property is in big contrast with the finiteness of the propagation speeds of solutions which are
exponentially bounded as x + at initial time.
Now, we state the main result which show a relation between the initial condition and the
large-time behavior of the level sets of the solution u.
Theorem 1.4 Let u be the solution of (1.2)-(1.3), where f satisfies (1.1), (1.5) and the initial
condition u0 : R [0, 1] satisfies (1.4) and (1.6). Assume that there exists 0 R such that u0
is of class C 2 and nonincreasing on [0 , ), and u000 (x) = o(u0 (x)) as x .
Then, for any (0, 1), (0, f 0 (0)), > 0 and > 0, there exists = ,,, t such
that
0
(f (0)+)t
t , E (t) u1
, e(f (0))t ]}
0 {[e
0
(1.8)
Observe first that the real numbers e(f (0)+)t and e(f (0))t belong to (0, supR u0 ) for
large t. Now, if there is a a R such that u0 is strictly decreasing on [a, +), then the inclusion
of Theorem 1.4 means that
(f 0 (0))t
1
(f 0 (0)+)t
u1
e
min
E
(t)
max
E
(t)
u
e
(1.9)
0
0
for large t, where, u1
0 : (0, u0 (a)] [a, +) denotes the reciprocal of the restriction of the
function u0 on the interval [a, +). Furthermore, if u0 is nonincreasing over the whole real line
R, then the derivative ux (t, x) is negative for all t > 0 and x R, from the parabolic maximum
principle. Therefore, E (t) is either empty or a singleton as soon as t > 0. In particular,
E (t) = {x (t)} for all t > t , where x (t) satisfies u(t, x (t)) = . The inequalities (1.9)
then provide lower and upper bounds of x (t) for large t. However, it is worth pointing out that
Theorem 1.4 is valid for general initial conditions which decay slowly to 0 at + but which
may not be globally nonincreasing.
Furthermore, for any (0, 1), formula (1.8) with = = can be rewritten as
0
11
for t large enough. Roughly speaking, this means that the real numbers x (t) are then asymptotically given, in the above approximate sense, by the solution of the family of decoupled ODEs
t U (t, x) = f 0 (0)U (t, x)
U (0, x) = U0 (x)
parameterized by x R, and then, say, by solving U (t, x (t)) = . In other words, the behavior
of u(t, x) at large time is dominated by the reaction term, that is to say that the diffusion term
plays in some sense a negligible role as compared to the growth by reaction.
Moreover, Theorem 1.4 actually yields the following corollary, which states that the level
sets E (t) can be located as far to the right as wanted, provided that the initial condition is well
chosen.
Corolary 1.5 Under the assumptions and notations of Theorem 1.4, the following holds: given
any function : [0, ) R which is locally bounded, there are initial conditions u0 such that,
for any given (0, 1)
min E (t) (t)
for all t large enough.
To end this section, we present the principal results stated by Berestycki, Hamel and Roques
in [7]. The authors prove existence and uniqueness results for the stationary equation and also
they analyze the behavior of the solutions of the evolution equation for large times. These results
are expressed by a condition on the sign of the first eigenvalue of the associated linearized
problem with periodicity condition.
Thus, we are concerned with the equation
t u (A(x)u) = f (x, u),
t > 0, x Rd
(1.10)
x Rd
(1.11)
Let L1 , ..., Ld > 0 be d given real numbers. In the following, saying that a function g : Rd R
is periodic means that g(x1 , ..., xk + Lk , ..., xd ) = g(x1 , ..., xd ) for all k = 1, ..., d. Let us
now describe the precise assumptions. Throughout the paper, the diffusion matrix field A(x) =
(aij (x))1i,jd is assumed to be periodic, of class C 1, (with > 0), and uniformly elliptic, in
the sense that
X
0 > 0, x Rd , Rd ,
aij (x)i j 0 ||2
1i,jd
12
(1.12)
M 0, s M, x Rd , f (x, s) 0
(1.13)
and
Examples of functions f satisfying (1.12 and (1.13) are functions of the type f (x, u) = u((x)
(x)u) or simply f (x, u) = u((x) u), where and are periodic.
Namely, we define 1 as the unique real number such that there exists a function > 0
which satisfies
(A(x)) fu (x, 0) = 1 in Rd
> 0 is periodic, > 0, kR0 k = 1
We are now ready to state the existence and uniqueness result on problem (1.11). Let us start
with the criterion for existence.
Theorem 1.6 1) Assume that f satisfies (1.13) and 1 < 0. Then, there exists a positive and
periodic solution p of (1.11).
2) Assume that f satisfies (1.12) and that 1 0. Then there is no positive bounded solution
of (1.11) (i.e. 0 is the only nonnegative and bounded solution of (1.11)).
Next we state our uniqueness result.
Theorem 1.7 Assume that f satisfies (1.12) and 1 < 0. Then, there exists at most one positive
and bounded solution of (1.11). Furthermore, such a solution, if any, is periodic with respect
to x. If 1 0 and f satisfies (1.12), then there is no nonnegative bounded solution of (1.11)
other than 0.
The core part in the proof of the above theorem consists in proving that any positive solution
of (1.11) is actually bounded from below by a positive constant, which does not seem to be a
straightforward property.
Let us now consider the parabolic equation (1.10), and let u(t, x) be a solution of (1.10),
with initial condition u(0, x) = u0 (x) in Rd . The asymptotic behavior of u(t, x) as t + is
described in the following theorem:
Theorem 1.8 Assume that f satisfies (1.11) and (1.12). Let u0 be an arbitrary bounded and
continuous function in Rd such that u0 0, u0 6 0. Let u(t, x) be the solution of (1.10) with
initial datum u(0, x) = u0 (x).
13
2
(Rd ) as t +, where p is the unique positive
1) If 1 < 0, then u(t, x) p(x) in Cloc
solution of (1.11).
1.3.
We are interested in the time asymptotic location of the level sets of solutions to the equation
t u + () u = f (u),
u(0, x) = u0 (x)
t > 0, x Rd
x Rd
(1.14)
(1.15)
with 0 < < 1 and f a concave function satisfying (1.1). Given (0, 1), we want to describe
how the level sets {x Rd : u(t, x) = } spread as time goes to +.
The results established by Cabre and Roquejoffre in [15] and announced in [14], show that
there exist drastic changes in the behavior of solutions as soon as the Laplacian is replaced for
instance by the fractional Laplacian (4) with (0, 1). They prove that the front position
will be exponential in time, in contrast with the classical case where it is linear in time by
Theorem 1.2.
For (0, 1), the fractional Laplacian is the generator for a stable Levy process. It is reasonable to expect that the existence of jumps (or flights) in the diffusion process will accelerate
the invasion of the unstable state u = 0 by the stable one, u = 1.
The first result concerns a class of initial data in Rd , possibly discontinuous, which includes
compactly supported functions. The following theorem shows that the position of all level sets
moves exponentially fast in time.
Theorem 1.9 Let d 1, (0, 1) and f satisfy (1.1). Let define c =
solution of (1.14)-(1.15), where u0 6 0, 0 u0 1 is measurable, and
u0 (x) C|x|d2 x Rd
for some constant C. Then:
-if c < c , then limt inf |x|ect u(t, x) = 1,
-if c > c , then limt sup|x|ect u(t, x) = 0.
A delicate step to prove the previous theorem is the following lemma.
14
f 0 (0)
.
d+2
Let u be a
Lemma 1.10 Under the assumptions of Theorem 1.9, for every c < c there exists (0, 1)
and t > 0 such that
u(t, x)
for all t t and |x| ect
Even if this lemma concerns initial data decaying at infinity, from it we can deduce the nonexistence of planar traveling waves, under no assumption of their behavior at infinity, as in the
following statement.
Proposition 1.11 Let d 1, (0, 1), f satisfy (1.1). Then, there exists no nonconstant
planar traveling wave solution of (1.14)-(1.15). That is, all solutions of (1.14)-(1.15) taking
values in [0, 1] and of the form u(t, x) = (x + te), for some vector e Rd , are identically 0 or
1. Equivalently, the only solutions : Rd [0, 1] of
(4) + e = f ()
in Rd
(1.16)
are = 0 and = 1.
The last statement
pon the elliptic equation (1.16) has an analogue for the Laplacian. As shown
in [2], if |e| < 2 f 0 (0) then equation (1.16) with = 1 admits the constants 0 and 1 as only
solutions taking values in [0, 1].
In one space dimension, it is of interest to understand the dynamics of nonincreasing initial
data. As mentioned before, for the standard Laplacian the level sets of u travel with the speed
c , provided that u0 decays sufficiently fast at +. In the fractional case, the mass at +
has an effect and what happens is not a mere copy of the result of Theorem 1.9 for compactly
supported data. The mass at + makes the front travel faster to the left, indeed with a larger
exponent than c .
0
(0)
. Let
Theorem 1.12 Let d = 1, (0, 1) and f satisfy (1.1). Let define the quantity c = f2
u be a solution of (1.14)-(1.15) and suppose that 0 u0 1 is measurable and nonincreasing,
u0 6 0 and
u0 (x) Cx2 x > 0
f 0 (0)
f 0 (0)
>
= c
2
1 + 2
where c is the exponent in Theorem 1.9 for d = 1 and compactly supported data.
c =
15
Notice also the slower power decay assumed in the initial condition with respect to Theorem 1.9.
One could wonder whether a model with such features is physically, or biologically relevant. In
fact, this behavior is consubstantial to fast diffusion, and the model may be relevant to explain
fast recolonization events in ecology.
To end this subsection, we consider the function f periodic in each xi -variable and satisfies
f (x, 0) = 0 for all x Rd , moreover
x Rd , s 7 f (x, s)/s is decreasing in s > 0
(1.17)
M 0, s M, x Rd , f (x, s) 0
(1.18)
and
Examples of functions f satisfying (1.17 and (1.18) are functions of the type f (x, u) = u((x)
u), where is periodic, note that if = 1 (homogeneous media) the function f satisfies (1.1).
The nonlinearity (x)u u2 is often referred to as a Fisher-KPP type nonlinearity.
Let 1 be the principal periodic eigenvalue of the operator (4) fu (x, 0). From [10]
we know that if 1 > 0, every solution to (1.14)-(1.15) starting with a bounded nonnegative
initial condition tends to 0 as t +. Thus we assume 1 < 0. Then, by [10], the solution to
(1.14)-(1.15) tends, as t +, to the unique bounded positive steady solution to (1.14)-(1.15),
denoted by u+ . By uniqueness, u+ is periodic and the convergence holds on every compact set.
Moreover, if we consider the problem (1.14)-(1.15) with the nonlinearity f (u) = (x)uu2
with periodic. Cabre, Coulon and Roquejoffre proved in [16] that the speed of propagation is
exponential in time, with a precise exponent depending on a periodic principal eigenvalue, and
that it does not depend on the space direction. This result is in contrast with the Freidlin-Gartner
formula for the standard Laplacian.
Theorem 1.14 Assume that 1 < 0. Let u be the solution to (1.14)-(1.15) with u0 piecewise
continuous, nonnegative, u0 6 0, and u0 (x) = O(|x|d+2 ) as |x| +. Then, for every
(0, min ), there exist c > 0 and a time t > 0 (all depending on and u0 ) such that, for
all t > t ,
|1 |
|1 |
d+2 }
{x Rd | u(t, x) = } {x Rd | c e d+2 |x| c1
e
Note that if = 1 then = 1 and f 0 (0) = 1, therefore, the estimate obtained is much sharper
than that in Theorem 1.9.
16
1.4.
In the first part of this subsection, we are interested in the large time behavior of the solution
u = (ui )m
i=1 with m N , to the one-dimensional weakly coupled reaction-diffusion system of
the form:
t ui (t, x) di 4ui (t, x) = fi (u(t, x)),
ui (0, x) = u0i (x),
(t, x) R+ R
x R
(1.19)
(1.20)
for all i {1, ..., m}, where each di is a nonnegative constant and f = (fi )m
i=1 is independent
m
of x and t. If a = (ai )i=1 is a constant positive vector, we define the set of functions
Ca = {u(x) : u(x) continuous and 0 u(x) a}
As general assumptions, we suppose that f C([0, a]) and the only zeros of f in Ca are 0 and a,
also, the system (1.19)-(1.20) is cooperative; i.e., each fi (u) is nondecreasing in all components
of u with the possible exception of the ith one.
In what follow, we present the results of Lewis, Li and Weinberger [37]. They showed the
existence of a single spreading speed for which the system (1.19)-(1.20) admit a traveling
wave solution W (x t) for all speeds .
Before to state this result, we need some additional hypothesis.
H1. f has uniformly bounded piecewise continuous first partial derivatives for 0 u a,
and it is differentiable at 0.
H2. The Jacobian matrix f 0 (0), whose off-diagonal entries are nonnegative, has a positive
eigenvalue whose eigenvector has positive components.
Theorem 1.15 Suppose that the function f in the system (1.19)-(1.20) satisfies the additional
hypothesis (H1), (H2) and the initial condition u0 = (u0i )m
i=1 is 0 for all sufficiently large x, and
that there are positive constants 0 < < 1 such that 0 u0 a for all x and u0 a
for all sufficiently negative x.
Then there exist , such that for every the system (1.19)-(1.20) has a nonincreasing
traveling wave solution W (x t) of speed with W () = a and W (+) = 0.
Moreover, if there is a traveling wave W (x t) with W () = a such that for at least
one component i
lim inf Wi (x) = 0
x+
then .
17
(1.21)
Let (0, +] the value of at which this infimum in (1.21) is attained, and let () be the
eigenvector of B which corresponds to the eigenvalue 1 ().
Theorem 1.16 Suppose that the function f in the system (1.19)-(1.20) satisfies the additional
hypothesis (H3), (H4). Assume that is finite,
1 () > ()
f or all > 1
and
f (()) f 0 (0)()
f or all > 0
Then there exist defined by (1.21) such that for any initial condition u0 (x) in Ca which
vanishes outside a bounded set, the solution of (1.19)-(1.20) has the properties that for each
positive
lim
t+
max
|x|t( +)
18
|u(x)| = 0
and for any strictly positive constant vector there is a positive R with the property that if
u0 on an interval of length 2R , then
lim
max |a u(x)| = 0
t+
|x|t( )
It is easy to see that if a function f satisfies (H1),(H3) and (H4) then in Theorem 1.14 is
equal to defined by (1.21).
Following the line, Barles, Evans and Sougandis [4] present another approach to the study
of the limiting behavior of the solution of certain scaled reaction-diffusion systems. Hence, they
considered the system
1
t ui = di 4ui + fi (u ),
ui (0, x) = gi (x),
in R+ Rd
(1.22)
on x Rd
(1.23)
for all i {1, ..., m}. Here the constants dk (1 k m), and the functions
g : Rd Rm
and
f : Rm Rm
m
m
are given, where we write g = (gi )m
i=1 , f = (fi )i=1 . The unknown is u = (ui )i=1 . We will
assume that
di > 0
i {1, ..., m}
and that the functions g, f are smooth, bounded and Lipschitz. In addition we suppose
gi > 0
i {1, ..., m}
and
G0 = {x : gi (x) > 0}
i {1, ..., m}
is a bounded, smooth subset of Rd . Our essential assumptions all concern the reaction term f .
First of all we suppose
(H1) f (0) = 0
and also
(H2) fi (u1 , ..., ui1 , 0, ui+1 , ..., um ) > 0 if u1 , ..., ui1 , ui+1 , ..., um 0 and ul > 0 for some
index l 6= i
19
Consequently, the vector field f points strictly inward along the boundary of the positivity set
= {u Rm : u1 > 0, ..., um > 0}
except at the point 0, which is an equilibrium point for the system (1.22)-(1.23). To ensure that
our solutions u do not become unbounded as 0, we further hypothesize
(H3) there exists a constant such that fi (u) 0 for all i {1, ..., m}, if u and ui .
Next we set forth additional hypotheses which imply that the rest point 0 is unstable. Let us
define the m m matrix
C := Df (0)
Df denoting the gradient of f . Thus
cik = fi,uk (0),
1 i, k m
cik > 0,
1 i, k m
We assume
(H4)
and also
(H4)
fi (u) cik uk ,
u , 1 i, k m
20
for all other eigenvalues of A(p). Let us define then the Hamiltonian
H(p) = 1 (A(p))
We additionally set
L(q) = sup (p q H(p))
pRd
L is the Lagrangian associated with H. Finally we define for each point (t, x) R+ Rd the
action function
Z
t
L(z(s))ds
| z(0) G0 , z(t) = x
J(t, x) = inf
0
the infimum taken over all absolutely continuous functions z : [0, t] Rd , satisfying the stated
initial and terminal conditions. As we will see, J turns out to be the unique solution of the
Hamilton-Jacobi equation
in R+ Rd
on G0 {0}
on int(Rd G0 ) {0}
Jt + H(DJ) = 0,
J = 0,
J = +
in the viscosity sense.
We interpret this theorem as describing how the Hamiltonian H, which depends upon both
C = Df (0) and the diffusion constants dl , ..., dm , controls the instability of the equilibrium
point 0.
1.5.
This section is devoted to present the principal results that we will present is the following
chapters. Hence, in Chapter 2, we consider the problem
ut + () u = f (u)
u(0, x) = u0 (x)
(1.24)
(1.25)
with (0, 1) in one spatial dimension. We assume that the nonlinearity in (1.24) is of FisherKPP type, that is, f : [0, 1] R is of class C 1 , concave and it satisfies
f (0) = f (1) = 0, f 0 (1) < 0 < f 0 (0).
21
(1.26)
Also, we assume that the initial condition u0 : R [0, 1] is continuous and it satisfies
u0 > 0 in R,
(1.27)
furthermore
There exists 0 R, such that u0 is non-increasing in [0 , ).
(1.28)
In view of the result stated in the Subsection 1.2 obtained by Hamel and Roques for slowly
decaying front-like initial values in the case = 1, a natural question is what kind of asymptotic
behavior does a solution of (1.24)-(1.25) with initial value decaying slower than a power |x|b
have?. The purpose of Chapter 2 is to answer this question for the case of fractional laplacian,
with (0, 1) in the one dimensional case. The main result states that, for b < 2, the
central part of the solution moves to the right at exponential speed f 0 (0)/b, which is faster than
c (Theorem 1.12), the exponential speed for solutions with initial values decaying faster than
x2 . Furthermore, it is proved that the initial condition u0 can be chosen so that the location
of the solution u be asymptotically larger than any prescribed real-valued function.
The first result provides basic properties of the solutions of (1.24)-(1.25) and says that the
level sets E (t) move at least exponentially fast as t .
0
(0)
and let u be the solution of (1.24)-(1.25), where
Theorem 1.18 Let (0, 1) and c = f2
f satisfies (1.26) and the initial condition u0 : R [0, 1] satisfies (1.27) and (1.28). Then u
satisfies:
x+
and
t+ xect
b) For any given (0, 1), there is a real number t > 1 such that E (t) is compact and
non-empty for all t t .
As in Subsection 1.2, we denote by E (t) the level sets of u of value (0, 1) at time t 0.
Hence, the following result is the main theorem which states a more accurate understanding of
the behavior of E (t), actually we express the motion of E (t) in terms of the behavior of the
initial value u0 . For doing this we need some additional hypothesis that expresses the slow
decay of the initial values:
(H1) There exists b < 2, such that u0 (x) xb for all x 0 .
22
for x 0 .
(f (0)+)t
E (t) u1
, e(f (0))t ]}, t ,
0 {[e
23
in R+ Rd
on Rd
(1.29)
(1.30)
where i (0, 1] for all i {1, ..., m} with at least one i 6= 1 and u = (ui )m
i=1 with m
N . For what follows and without loss of generality, we suppose that i+1 i for all i
{1, ..., m 1} and we set := m < 1.
As general assumptions, we impose the initial conditions u0i 6 0 to be nonnegative, continuous and bounded by constants ai > 0 and satisfy
u0i (x) = O(|x|(d+2i ) )
(1.31)
fi (u)
> 0 i 6= j
uj
(1.32)
i.e., the system (1.29)-(1.30) is cooperative. Moreover, we assume fi (0) = 0 and some additional hypothesis, which are compatible with strongly coupled systems.
(H1) The principal eigenvalue 1 of the matrix DF (0) is strictly positive, where F = (fi )m
i=1 .
There exist positive constants 1 and 2 such that
(H2) Dfi (0)u fi (u) cui1+1 .
(H3) Dfi (0)u fi (u) ckuk1+2 .
(H4) F is concave, DF (0) is a symmetric matrix and
where 1 , 2
2
d+2
fi (0)
ui
The purpose of this work is to understand the time asymptotic behavior of solutions to (1.29)(1.30). We show that the speed of propagation is exponential in time, with a precise exponent
depending on the smallest index and of the principal eigenvalue 1 of the matrix DF (0). Also
we note that it does not depend on the space direction. Moreover, we prove that the solution of
the system (1.29)-(1.30), tends to unique positive steady state solution as t +.
Moreover, we consider the Banach space
C0 (Rd ) := {w is continuous in Rd and w(x) 0 as |x| }
with the L (Rd ) norm and we set D0 (Ai ) the domain of the operator Ai = (4)i in C0 (Rd ).
In what follows we assume that u0i D0 (Ai ) for all i {1, ..., m}.
Now we are in a position to state our main theorems, which show that the functions move
exponentially fast for large times.
24
Theorem 1.21 Let d 1 and assume that F satisfies (1.32), (H1), (H2) and (H3). Let u be the
solution to (1.29)-(1.30) with u0 satisfying (1.31). Then:
a) for every i > 0, there exists a constant c > 0 such that,
1
(0, )
since F is continuous, we deduce that F (u+ ) = 0. Also, since the function F is positive in a
small ball near to zero, we have that u+ > 0.
Moreover we assume that the initial condition u0 satisfies
u0 u+
in Rd
(1.33)
Theorem 1.22 Let d 1 and assume that F satisfies (1.32), (H1), (H2), (H3) and (H4). Let u
be the solution to (1.29)-(1.30) with u0 satisfying (1.31) and (1.33). Then:
a) If c <
1
,
d+2
then
lim
t+ |x|ect
b) If c >
1
,
d+2
then
lim sup ui (t, x) = 0
t+ |x|ect
25
Chapter 2
Fast propagation for fractional KPP
equations with slowly decaying initial
conditions.
Work in colaboration with Patricio Felmer.
2.1.
Introduction
In this chapter we study the large-time behavior of the solution of the Cauchy problem for
fractional reaction-diffusion equations
ut + () u = f (u)
u(0, x) = u0 (x)
(2.1)
(2.2)
with (0, 1) in one spatial dimension. Let us now provide a precise description of our
assumptions and results. We assume that the nonlinearity in (2.1) is of Fisher-KPP type, that is,
f : [0, 1] R is of class C 1 , concave and it satisfies
f (0) = f (1) = 0, f 0 (1) < 0 < f 0 (0).
This properties mean that the growth rate
f (s)
s
(2.3)
is maximal at s = 0.
26
(2.4)
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
(2.5)
2.2.
Basic properties
We first recall the notion of mild solution that suited to our problem and we state the Comparison
Principle that will be a crucial tool in our analysis. Then we present the Theorem 2.2 that follows
the line of the corresponding result in [14] and [15].
In studying the existence of solution of equation (2.1)-(2.2) we first consider the linear heat
equation for the fractional Laplacian
ut + () u = f (t, x)
u(0, x) = u0 (x),
(2.6)
(2.7)
whose solution may be obtained by the formula of variation of parameters or Duhamel formula
Z t
p(t s, x) f (s, x)ds,
(2.8)
u(t, x) = p(t, x) u0 (x) +
0
where the convolution is taken in the variable x. Here the kernel p is given by p(t, x) =
1
1
t 2 p (t 2 x), where
Z
1
2
p (x) =
eix|| d
2 R
and it satisfies the following properties:
a) p C((0, +), R).
R
b) p(t, x) 0 and R p(t, x)dx = 1 for all t > 0.
c) p(t, ) p(s, ) = p(t + s, ) for all t, s R+
d) There exists B > 1 such that, for (t, x) R+ R:
B 1
1
2
t (1+ | xt
1
2
|1+2 )
p(t, x)
27
B
1
2
t (1+ | xt 2 |1+2 )
(2.9)
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
x+
equipped with the supremum norm. Given u0 Clim , equation (2.1)-(2.2), with Fisher-KPP
nonlinearities f and initial condition u0 , has a unique solution u that exists for all x R and
t 0, moreover, u(t, ) C([0, ), Clim ). This solution u can be obtained as the limit of the
iteration scheme
Z t
n+1
p(t s, x) f (s, un (s, x))ds,
(2.10)
u (t, x) = p(t, x) u0 (x) +
0
with u0 (t, x) = p(t, x) u0 (x). The limit is uniform in x and locally in time, see [2] and [15]
for details. The solution obtained in this way is called mild solution and in this chapter this will
be the notion of solution we consider in all our statements.
To continue we recall the Comparison Principle, which will be frequently used in the following computations.
Theorem 2.1 (Comparison Principle) Let u, v C([0, T ], Clim ) be mild solutions of the equations
ut + (4) u = g(u), vt + (4) v = h(v),
where g, h : R R are locally Lipschitz continuous. If
g() h(),
and
u(0, x) v(0, x),
x R,
then
u(t, x) v(t, x)
(t, x) [0, T ] R.
For the proof of this result we refer the reader to [15] or [23].
Before to stating our results, we introduce some notation. For any (0, 1) and t 0, we
denote by
E (t) = {x R : u(t, x) = },
the level set of u of value at time t. For any subset A (0, 1], we set
u1
0 (A) = {x R : u0 (x) A},
the inverse image of A by u0 . Our first result provides basic properties of the solutions of
(2.1)-(2.2) and says that the level sets E (t) move at least exponentially fast as t .
28
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
0
(0)
Theorem 2.2 Let (0, 1) and c = f2
and let u be the solution of (2.1)-(2.2), where f
satisfies (2.3) and the initial condition u0 : R [0, 1] satisfies (2.4) and (2.5). Then u satisfies:
x+
t+ xect
b) For any given (0, 1), there is a real number t > 1 such that E (t) is compact and
non-empty for all t t .
Proof: Part a) We start by using the Comparison Principle, recalling that 0 u0 (x) 1, to
obtain that the solution u of (2.1)-(2.2) satisfies
0 < u(t, x) 1
(t, x) [0, ) R.
Next we analyze the limit of u(t, x) as x +. To do this, let us first notice that the function
Z
f 0 (0)t
u(t, x) = e
p(t, x y)u0 (y)dy
R
29
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
Mt := 1 + t 2
and consider x Mt . Then we use the definition of u and (2.9) to find that
Z
Z
0
0
ef (0)t
u0 (y)
ef (0)t
u0 (x r)
u(t, x) B 1
dy
=
B
dr
1
1
1
t 2 1+ | (x y)t 2 |1+2
t 2 1+ | rt 2 |1+2
= Be
f 0 (0)t
= Be
f 0 (0)t
u0 (x st 2 )
ds
1+ | s |1+2
u0 (x st 2 )
ds +
1+ | s |1+2
!
1
u0 (x st 2 )
ds .
1+ | s |1+2
f 0 (0)t
ds +
ds
u(t, x) Be
1+2
1+ | s |1+2
1+ | s |
0
Bef (0)t (C + ) < + = .
2 2
Hence 0 u(t, x) < . Thus we have proved that limx u(t, x) = 0.
Now we study u(t, x) when t . From (2.4), we may find a continuous non-increasing
function v0 : R [0, 1] in Clim , such that R+ supp(v0 ) is compact and it satisfies 0 v0 (x)
u0 (x). Denote by v the solution of the Cauchy problem (2.1)-(2.2) with initial condition v0 , then
by the Comparison Principle we find v(t, x) u(t, x) 1 for all t 0 and x R. Then we
can use Theorem 1.5 of [15] to conclude that
lim inf v(t, x) = 1
t xect
for all t t . By continuity of x 7 u(t, x) we conclude that E (t) is a non-empty compact set
for all t t .
t
u
Remark 2.3 As a direct consequence of Theorem 2.2, we see that
E (t)
= ,
t ect
lim
30
(2.13)
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
2.3.
It is the purpose of our main theorem to obtain a more accurate understanding of the behavior
of E (t), actually we express the motion of E (t) in terms of the behavior of the initial value u0
and we improve the estimate for c in (2.13). For doing this we need some additional hypothesis
that expresses the slow decay of the initial values:
(H1) There exists b < 2, such that u0 (x) xb for all x 0 .
(H2) There exist > 1 and k > 0 such that
u0 (x)
k,
u0 (x)
for x 0 .
(f (0)+)t
E (t) u1
, e(f (0))t ]}, t ,
0 {[e
31
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
only (2.4) and (2.5), see Propositions 2.6 and 2.9. Finally, we observe that the non-existence of
traveling waves for the fractional problem, as proved in [15], implies to change various other
arguments given in [29].
Now we would like to make some comments on hypothesis (H2). This condition complements hypotheses (H1) and it also expresses the slow decay of u0 . Actually, we observe that
any power u0 (x) = xb satisfies also (H2). More generally, any function u0 C 1 ([0 , )),
decreasing and convex in [0 , ), such that
1
u00 (x)
|= O
as x ,
|
u0 (x)
x
satisfies the hypothesis (H2).
Theorem 2.4 and Corollary 2.5 complement the results by Cabre and Roquejoffre, where
they estimate the asymptotic behavior of solutions with front-like initial values which decays
faster than x2 as x . In our case we assume the initial value decays slower than a
power xb , with b < 2, the complementary exponents. In a sense we generalize to the case
(0, 1) results proved by Hamel and Roques in [29], replacing the Laplacian by the fractional
Laplacian.
Let us assume that the initial value is a pure power, that is, u0 (x) = xb , with b < 2, for x
large. In this case we see that Theorem 2.4 implies that for all c1 and c2 such that
c =
f 0 (0)
f 0 (0)
< c1 <
< c2 ,
2
b
for all t .
These observations are in contrast with the results of Cabre and Roquejofre [15], where they
showed that all solutions with front-like initial conditions decaying slower that x2 , spread at
an exponential speed c independent of further properties of u0 . In our case, using comparison
principle and the discussion given above, we see that solutions with front-like initial conditions
decaying slower than xb , with b < 2, spread at an exponential speed f 0 (0)/b, which is larger
than c and depends explicitly on the exponent b.
In this sense, our results show that the exponent 2 is a critical exponent. If the initial value
decays faster than x2 then the exponential speed is c and if the initial value decays slower
than xb , with b < 2, then the exponential speed is f 0 (0)/b or larger. Above the exponent 2,
the solutions speed of propagation starts getting influenced by the initial value, propagating
faster the slower the decay is.
32
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
2.4.
In order to prove Theorem 2.4 we need to obtain an upper and a lower estimate for the set E (t)
for t large. In this section we obtain the lower estimate. It is important to notice that in getting
the lower estimate we do not require the initial condition satisfies hypothesis (H1) and (H2), but
only (2.4) and (2.5).
Proposition 2.6 Let (0, 1) and let u be the solution of (2.1)-(2.2), where f satisfies (2.3)
and the initial condition u0 satisfies (2.4) and (2.5).
Then, for any > 0, (0, 1) and (0, f 0 (0)), there exists a time u = u (, , ) t ,
such that
(f 0 (0))t
E (t) u1
]},
t u .
(2.14)
0 {(0, e
For proving this proposition, we first prove a lemma where we construct an appropriate subsolution of (2.1)-(2.2) which will enable us to prove the lower bound for small values of . Then
we will show that such an estimate can also be done for the remaining values of (0, 1).
Let us start setting up some notation. Given (0, f 0 (0)) we let d and 0 be such that
d (1,
f 0 (0)
)
f 0 (0)
(2.15)
33
(2.16)
(2.17)
(2.18)
(2.19)
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
Proof: Let > 0 be such that < inf (,0 ) u0 and let R be so that u0 () = and
u0 (x) < for all x > . By making smaller if necessary, we may assume that > 0 and we
can choose T > + 1, such that = eT s0 and let us define u0 (x) = inf(u0 (x), ). We let u1
be the solution of the equation
(u1 )t + () u1 = u1
u1 (0, x) = u0 (x).
This solution is given by
u1 (t, x) = e
Z
p(t, x y)u0 (y)dy,
R
so that, by the election of and T , we have that u1 (t, x) s0 for all (t, x) [0, T ] R.
Moreover, we have
Z
T
p(t, z)u0 (x z)dz = eT = s0 .
lim u1 (T, x) = lim e
x
for all t [0, T ]. Thus u1 (t, x) is non-increasing in x for all t [0, T ]. Now we perform a
recursive process to define un given un1 , for all n 2. We let
u0,n1 (x) = inf(un1 ((n 1)T, x), ),
where un1 ((n 1)T, ) is non-increasing and un1 ((n 1)T, ) = s0 . Then we define un
as the solution of
(un )t + () un = un
un ((n 1)T, x) = u0,n1 (x),
for (t, x) [(n 1)T, nT ] R. This solution may be written as
Z
(t(n1)T )
p(t (n 1)T, x y)u0,n1 (y)dy,
un (t, x) = e
R
so that, by the election of and T , we have that un (t, x) s0 , for all (t, x) [(n1)T, nT ]R.
Moreover, by definition
un ((n 1)T, x) = u0,n1 (x) un1 ((n 1)T, x),
34
for all
x R.
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
We also have
lim u0,n1 (nT, x) = lim e
Remark 2.8 We may define the function u : [0, +) R [0, s0 ] in such a way that, for all
integer n 1,
u(t, x) = un (t, x) for (t, x) [(n 1)T, nT ) R.
Since for all integer n 1 the function un satisfies
(un )t + () un f (un )
in ((n 1)T, nT ) R.
(2.20)
We finally observe that, from the monotonicity property of the functions un , the function u is
non-increasing in x R for all t 0.
We are now in a position to prove Proposition 2.6.
Proof of Proposition 2.6: Using the notation as in the last proof, we let be defined as
ed
=
B
Z
0
u0 ( s 2 )
ds.
1+ | s |1+2
(2.21)
We observe that does not depend on nor and that 0 < < s0 . In order to see this last fact
we recall that < T , d > 1 and C < 2B, where C was defined in (2.11) and B is given in
(2.9). Then
Z
1
e d u0 ( s 2 )
e d C
C
ds
<
s0 < s0 .
(2.22)
=
1+2
B 0 1+ | s |
2B
2B
35
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
(2.23)
C e(t(n1)T )
u0,n1 (x).
2B
In case that u0,n1 (x) = , we conclude that
u(t, x)
C e(t(n1)T )
C
.
2B
2B
Otherwise, we have that u0,n1 (x) = un1 ((n 1)T, x) and then, as before we obtain that
C e(t(n1)T )
un1 ((n 1)T, x)
2B
Z
C e(t(n2)T )
u(t, x)
36
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
eT
u(t, x)
2B
2B
!
1
C e(1 d )
C
C
,
2B
2B
2B
(2.24)
where we have used (2.15). Otherwise, we have that u0,n2 (x) = un2 ((n 2)T, x) and then
as before we have
2
C
u(t, x)
e(t(n2)T ) un2 ((n 2)T, x).
2B
Repeating this procedure, we will either reach
u(t, x)
C
2B
as in (2.24), or we would have that x satisfies u0,m (x) 6= for all m {1, 2, 3, ..., n 1}. In
the later case we have that u0,1 (x) = u1 (T, x) and then
n1
C
u(t, x)
e(tT ) u1 (T, x)
2B
n1 t Z
1
u0 (x sT 2 )
C
e
ds
2B
B 0 1+ | s |1+2
n1
Z
1
t
d
C
u0 (y (t) s 2 )
(1 d1 )(n1) e
ds
e
2B
B 0
1+ | s |1+2
!n1
1
C e(1 d ) )
=
> .
2B
Summarizing, we have obtained that
u(t, y (t)) min(
C
, ) for all
2B
t .
1
ii) In the second case, that is when t = and x (, y ( )], we have that x s 2
1
1
y ( ) s 2 = s 2 , hence
Z
1
e u0 (x s 2 )
u(, x) = u1 (, x)
ds
B 1+ | s |1+2
Z
1
e d u0 ( s 2 )
ds = > 0.
B 0 1+ | s |1+2
37
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
This completes the analysis on the boundary of . To complete the proof we consider (t, x) ,
that is, t > and x < y (t). Since u(t, ) is non-increasing for each t , from i) we deduce
that
C
u(t, x) u(t, y (t)) min(
, ).
2B
Thus, we have found > 0 such that
u(t, x) u(t, x) ,
(t, x)
(2.25)
Now we can get the upper estimate for (0, ). Let x E (t) for t max(, t ), then we
have
x > y (t) .
(2.26)
d t
=
Z0
Z
1
1
u0 (x s 2 )
u0 (y (t) s 2 )
ds
ds
1+ | s |1+2
1+ | s |1+2
0
u0 (x)
C
C
ds =
u0 (x) =
u0 (x),
1+2
1+ | s |
2
2
(2.27)
where the last equality holds since x > . From (2.27) and since > 0 and > f 0 (0) 0 ,
there exists 1 (, , ) max(, t ) such that for all t 1 (, , ) and x E (t),
u0 (x)
f 0 (0) 0
2B t
e d s0 e d t e d t .
C
f 0 (0) 0
d
if z 0
(1 z) if 0 < z < 1
u,0 (z) =
0
if z 1
38
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
and denote by u the solution of the Cauchy problem (2.1)-(2.2) with initial condition u,0 . It
follows from (2.25) that
u(s, x) u,0 (x y (s) + 1)
f 0 (0)
2
for all
(s, x) [, ) R and t 0.
for all
ds
=
1+ | s |1+2
1+ | s |1+2
0
0
Z
u0 (x)
C
C
ds =
u0 (x) =
u0 (x).
1+2
1+ | s |
2
2
0
Here, the last equality is satisfied because x > . Now we conclude as in the other case,
since > 0 and > f 0 (0) 0 , there exist 2 (, , ) max( + T , t ), such that for all
t 2 (, , ) and x E (t)
u0 (x)
f 0 (0) 0
2B (tT )
0
e d
< e d t = e(f (0))t .
C
t
u
The proof of Corollary 2.5 follows from Proposition 2.6, finding a suitable initial condition
that satisfies (2.4) and (2.5). This proof follow closely the ideas by Hamel and Roques in [29].
Proof of Corollary 2.5: Let : [0, ) R be any locally bounded function, then there exists
a continuous, increasing function g : [0, ) R such that
g(z) (2z),
39
z 0.
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
u0 (x) = ef (0)g
1 (x)
x g(0)
and extended by one, to the left of g(0). We easily see that u0 Clim is decreasing and that it
satisfies (2.4) and (2.5).
Let u be the solution
of Cauchy problem (2.1)-(2.2) with initial condition u0 and let
f 0 (0)
(0, 1) and (0, 2 ). Moreover, let us consider 1 > 0 large so that
0
(2.28)
It follows from Proposition 2.6 with = 1, that there exists 2 max(1 , t ) such that
0
(f (0))t
y u1
),
0 (e
t 2 , y E (t)
2.5.
t
u
In this section we prove the upper bound for the set E (t) and we complete the proof of Theorem
1.2. The proof of the upper bound is obtained by constructing an appropriate super-solution of
(2.1)-(2.2). The construction of such super-solution strongly relies on the hypotheses (H1) and
(H2). Precisely we prove
Proposition 2.9 Let (0, 1) and let u be the solution of (2.1)-(2.2), where f satisfies (2.3)
and the initial condition u0 satisfies (2.4), (2.5), (H1) and (H2).
Then, for any > 0, (0, 1) and (0, f 0 (0)), there exists a time ` = ` (, , , b) t
such that
(f 0 (0)+)t
E (t) u1
, 1]}, t ` .
0 {[e
Proof. Let u be the solution of the problem
ut + () u = f 0 (0)u
u(0, x) = u0 (x),
that can be expressed as
f 0 (0)t
Z
p(t, x y)u0 (y)dy.
u(t, x) = e
40
(2.29)
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
By the assumptions on f we see that u is a super-solution for (2.1)-(2.2) and then, the Comparison Principle implies that
(t, x) [0, ) R.
Since b < 2, there exists a > 0 small enough, such that a + b 2. Let (0, 1) and
(0, f 0 (0)), then there exists 1 = 1 (, , b) t > 1, such that for all t 1 we have
8B 2
< e2t
f 0 (0)t
e
< inf {u0 (x), 0b },
x(,0 )
8B 2
and
8B 2
a
2b
af 0 (0)
t
2b
1
>
1
t
B
(2.30)
1
2
(2.31)
Here B is the constant appearing in (2.9) and is given in (H2). Let t 1 and let us denote by
> 0 the number such that
0
4B 2 ef (0)t = .
(2.32)
2
By (2.30) it is possible to find 1 [0 , ), such that u0 (1 ) = , we assume that 1 is the
largest number with that property. Then, let us choose > 0 big enough, such that
Z
B
1
ds
=
s1+2
2
and we define t := 1 + t 2 . For x t we estimate the values of u(t, x). From (2.29) we
have that
Z
0
u0 (y)
ef (0)t
f 0 (0)t
dy
Be
(I1 + I2)
u(t, x) B 1
1
t 2 1+ | (x y)t 2 |1+2
where
I1 =
1
t
and
I2 =
1
2
1
t
1
2
u0 (x)
1
1+ | (x y)t 2 |1+2
| u0 (y) u0 (x) |
1
1+ | (x y)t 2 |1+2
dy
dy.
41
Chapter 2. Fast propagation for fractional KPP equations with slowly decaying initial
conditions.
Z
Z
1
1
2
ds + 2
ds
1+2
1+ | s |1+2
1+ | s |
< 2B + B = 3B.
Therefore,
x t .
From here we obtain that y < t for all y E (t). In fact, if y t and y E (t), then
y t > 1 and
0
0
0
= u(t, y) < B 2 ef (0)t (u0 (y) + 3) B 2 ef (0)t (u0 (1 ) + 3) = B 2 ef (0)t 4 = ,
2
which is a contradiction. Since u0 (1 ) = , from (H1), (2.31) and (2.32) we see that
a
1
2 2b
2
a
af 0 (0)
a
8B
1
t
t
12 2b =
e 2b
.
( 1) B
From here, since 1b u0 (1 ) = and by the choice of 1 and , we conclude that
1
1
1
t 2
(u0 (1 )) 2
t = 1 + t 2 = 1 +
B
1
2
b
a
b
1
1
1 +
t 2 12 1 + ( 1)12 12 1 .
B
0
Now, if 2 is such that u0 (2 ) = e(f (0)+ 2 )t , then since e(f (0)+ 2 )t < by (2.30) and (2.32),
we have that 1 < 2 . Therefore, for each y E (t)
0
t
u
Thanks to Proposition 2.6 and 2.9, we can prove Theorem 2.4 on the behavior of level sets
for large times, expressed in terms of the decay of the initial condition.
Proof of Theorem 2.4: It follows directly from Proposition 2.6 and 2.9, taking
= (, , , , b) = max(u , ` ).
42
t
u
Chapter 3
Exponential propagation for fractional
reaction-diffusion cooperative systems
Work in colaboration with Anne-Charline Coulon.
3.1.
Introduction
In this chapter, we are interested in the large time behavior of solution u = (ui )m
i=1 with m N ,
to the fractional reaction diffusion system:
(t, x) R+ Rd
x Rd
(3.1)
(3.2)
where i (0, 1] for all i {1, ..., m} with at least one i 6= 1. As general assumptions, we
impose the initial conditions u0i 6 0 to be nonnegative, continuous and bounded by constants
ai > 0 and satisfy
u0i (x) = O(|x|(d+2i ) )
(3.3)
fi (u)
> 0 i 6= j
uj
43
(3.4)
The aim of this chapter is to understand the time asymptotic location of the level sets of
solutions to (3.1)-(3.2). We show that the speed of propagation is exponential in time, with a
precise exponent depending on the smallest index := mini (i ) and of the principal eigenvalue of the matrix DF (0) where F = (fi )m
i=1 . Also we note that it does not depend on the
space direction. Moreover, we prove that the solution u(t, x) of (3.1)-(3.2) tends to the smallest
positive constant steady state solution, as t +.
For what follows and without loss of generality, we suppose that i+1 i for all i
{1, ..., m 1} and we set := m < 1. Before to state the main results, we need some
additional hypothesis on the nonlinearities fi , which allow us to identify precisely the propagation speed of solutions, also note that this hypothesis are compatible with strongly coupled
systems.
(H1) The principal eigenvalue 1 of the matrix DF (0) is strictly positive.
There exist positive constants 1 and 2 such that
(H2) Dfi (0)u fi (u) cui1+1 .
(H3) Dfi (0)u fi (u) ckuk1+2 .
(H4) F is concave, DF (0) is a symmetric matrix and
where 1 , 2
2
d+2
fi (0)
ui
Before going further on, let us state at least one example of nonlinearity F satisfying (3.4),
(H1), (H2), (H3) and (H4). Let consider
1 u1
2
u1
u1
F (u) =
with u =
2
1 u2
u2
u2
Indeed, F satisfies (3.4), F (0) = 0, the principal eigenvalue of DF (0) is 1 = 3 and satisfies
(H2) and (H3) with 1 = 2 = , also F is concave and DF (0) is symmetric.
Moreover, we consider the Banach space
C0 (Rd ) := {w is continuous in Rd and w(x) 0 as |x| }
with the L (Rd ) norm and we set D0 (Ai ) the domain of the operator Ai = (4)i in C0 (Rd ).
In what follows we assume that u0i D0 (Ai ) for all i {1, ..., m}.
44
Now we are in a position to state our main theorems, which show that the functions move
exponentially fast for large times.
Theorem 3.1 Let d 1 and assume that F satisfies (3.4), (H1), (H2) and (H3). Let u be the
solution to (3.1)-(3.2) with u0 satisfying (3.3). Then:
a) for every i > 0, there exists a constant c > 0 such that,
1
+,
also
F
(u
)
=
0.
We
define
u+ = inf 0 u+
(0, )
since F is continuous, we deduce that F (u+ ) = 0. Also, since the function F is positive in a
small ball near to zero, we have that u+ > 0.
Moreover we assume that the initial condition u0 satisfies
u0 u+
in Rd
(3.5)
Theorem 3.2 Let d 1 and assume that F satisfies (3.4), (H1), (H2), (H3) and (H4). Let u be
the solution to (3.1)-(3.2) with u0 satisfying (3.3) and (3.5). Then:
45
a) If c <
1
,
d+2
then
lim
t+ |x|ect
b) If c >
1
,
d+2
then
lim sup ui (t, x) = 0
t+ |x|ect
3.2.
In this section, we prove the existence of unique solution of the system (3.1)-(3.2). In order to
prove this, we recall the notion of mild solution for the nonhomogeneous linear problem
t ui + (4)i ui = hi (t),
ui (0) = u0i
in (0, T )
(3.6)
where T > 0, u0i X, and hi C([0, T ], X) are given, where X is a Banach space. The mild
solution of (3.6) is given explicitly by Duhamel principle:
Z t
ui (t) = Tt,i u0i +
Tts,i hi (s)ds
(3.7)
0
Z
pi (t, x)w(x y)dy =
Tt,i w(x) =
Rn
pi (t, x y)w(y)dy
Rn
46
1
2i
1
2
i
(1+ | xt
|1+2i )
pi (t, x)
t
1
2i
1
2
(1+ | xt
(3.8)
i
|1+2i )
For more information about operators Tt,i see section 2 of [15]. So, from the equation (3.7), one
easily checks that ui C([0, T ]; X).
Now, we consider G : [0, ) X m X m , G = (Gi (t, u))m
i=1 be a function that satisfies
for all i {1, ..., m}
we note that X m
Gi C 1 ([0, ) X m ; X)
(3.9)
Gi (t, ) is globally Lipschitz in X m uniformly in t 0.
(3.10)
Pm
Q
= m
i=1 kui kX is a Banach space.
i=1 X with the product norm kukX m =
in (0, T )
(3.11)
(3.12)
where Tt = diag(Tt,1 , ..., Tt,m ). Now, we will use a fixed point principle to prove that (3.12)
has an unique solution. Define the map
Z t
Nu0 (u)(t) := Tt u0 +
Tts G(s, u(s))ds
(3.13)
0
47
ku(s) v(s)kX m ds
M Lipu (Gi )
0
= M Lipu (Gi )
Z tX
m
0 j=1
Pm
thus Nu0 is Lipschitz with constant M T [ i=1 Lipu (Gi )]. Recall that for any strongly continuous semigroup, we have that kTt,i k Ci ei t for some constants Ci and i . Now, it follows
by induction that (Nu0 )k is Lipschitz in C([0, T ]; X)m with Lipschitz constant
" m
#k
(M T )k X
Lipu (Gi )
k!
i=1
where k is any positive integer. This constant is less than 1 if we take k large enough. Then, we
conclude that Nu0 has a unique fixed point.
Moreover, if we consider the sequence of functions (Nu0 )i (u0 ) C([0, T ]; X)m , it is easy to
see that there exist u C([0, T ]; X)m such that
u = lim (Nu0 )i (u0 )
i+
(3.14)
where u0 (t) = Tt u0 , also, the limit u is the unique fixed point of Nu0 , and so, the unique mild
solution of (3.12) for all T > 0.
Given 0 < T < T 0 , the mild solution in (0, T 0 ) must coincide in (0, T ) with the mild solution
in this interval, by uniqueness. Thus, under assumption (3.9)-(3.10), the mild solution of (3.11)
extends uniquely to all t [0, ), i.e., it is global in time.
Moreover, let u = (ui )m
i=1 the unique solution of (3.12), if we define
Hi (t, w) = Gi (t, u1 , ..., ui1 , w, ui+1 , ..., um )
we have that
Hi C 1 ([0, ) X; X)
Hi (t, ) is globally Lipschitz in X uniformly in t 0.
48
(3.15)
(3.16)
in (0, T )
(3.17)
Following the computations in section 2.3 of [15], we conclude that this problem has a unique
mild solution in C([0, T ]; X) given by w = ui . Thus, if the initial datum belongs to the domain
of Ai = (4)i denoted by D(Ai ), we have further regularity in t of the mild solution ui =
ui (t). Under hypothesis (3.15)-(3.16) (here the continuous differentiability of Hi with values in
X is important), the mild solution ui of (3.17) satisfies
ui C 1 ([0, T ); X) and ui ([0, T )) D(Ai ) if u0i D(Ai )
(3.18)
and it is a classical solution, i.e., a solution satisfying (3.17) pointwise for all t (0, T ). Doing
the same procedure for all i {1, ..., m} and for all T > 0, we conclude that u = (ui )m
i=1 is a
classical solution of (3.11) global in time.
Now, we set a useful fact that we need in the followings computations. If u is the solution
of the system (3.11) with u0 X and G satisfying (3.9) and (3.10), then u(t) = eat u(t) is the
mild solution of the system
e u)
t u + L
u = G(t,
u(0) = u0
(3.19)
ei (t, eat u) and a R. This fact is proved in the same way as in [15].
ei (t, u) = a
with G
ui +eat G
We now apply all these facts to problem (3.1)-(3.2). Recall our standing assumptions for
the nonlinearities (fi )m
i=1 , from hypothesis (H2), we deduce the existence of a positive vector
M = np with n > 0 large enough and pi = 1 for all i {1, ..., m} such that F (M ) 0 and
u0 a M . Now, we extend fi outside of a compact set of Rm that contains [0, M ] to ensure
that:
fi C 1 (Rm ) is globally Lipschitz, fi (u) is nondecreasing in all components of u with the
possible exception of the ith one and Dfi is uniformly continuous in Rm .
We consider the Banach space X = C0 (Rd ) and taking Gi (t, u)(x) := fi (u(x)) we can verify (3.9)-(3.10). We use that Dfi is uniformly continuous and fi (0) = 0 to check that the
map u C0 (Rd )m 7 fi (u) C0 (Rd ) is continuously differentiable. Thus, by the previous
considerations, there is a unique mild solution u of
t u + Lu = F (u),
u(0, ) = u0
49
in (0, ) Rd
(3.20)
Q
for data u0 X. Moreover, if the initial datum u0 in (3.20) belongs to the domain m
i=1 D0 (Ai ),
d
where D0 (Ai ) is the domain of Ai in C0 (R ). Then the mild solution u of (3.20) satisfies (3.18)
for all i {1, ..., m} and for all T > 0, with D(Ai ) = D0 (Ai ) and it is a classical solution
global in time.
Before to state the upper bound for the solutions, we need to establish a comparison principle
for mild solutions.
1,2
C 1 (Rm ) functions that satisfy (3.4), globally
Theorem 3.3 Let F 1,2 = (fi1,2 )m
i=1 with fi
m
Lipschitz. Let u1,2 = (u1,2
i )i=1 mild solutions of
t u1 + Lu1 = F 1 (u1 ),
t u2 + Lu2 = F 2 (u2 )
belong to X
then
u1i (t, x) u2i (t, x)
(t, x) [0, ) Rd .
thus,
- if k = i, then
fij
(t, v)
vi
- if k 6= i, then
j = {1, 2}.
=a+
fij
(t, v)
vk
fij
(w)
wi
fij
(w)
wk
0, by the choice of a.
Moreover, since fi1 () fi2 (), we have fi1 (t, ) fi2 (t, ).
Now, let consider the system
t uj + L
uj = F j (
uj )
uj (0, ) = uj0
50
(3.21)
by the previous section, we know that uj (t, x) = eat uj (t, x) is the solution of the system (3.21)
for each j = 1, 2, where uj is the solution of the system (3.21) with F j replaced by F j . Therefore, it is enough to prove that u1 u2 .
Consider the mapping N j for j = {1, 2}, defined by
Z t
j
j
Tts Fej (s, w(s, ))ds
N (w)(t, ) := Tt u0 () +
(3.22)
Taking u0,j (t, ) = Tt uj0 (), we know that uj = limn+ (N j )n (u0,j ), thus, using a standard
induction argument, we only need to show that (N 1 )n (u0,1 ) (N 2 )n (u0,2 ) on [0, ) Rd for
all n.
Since u10 u20 , then u0,1 u0,2 on [0, ) Rd . Now, suppose that (N 1 )n (u0,1 )
(N 2 )n (u0,2 ); and by previous considerations, we have
fei1 (s, (N 1 )n (u0,1 )) fei2 (s, (N 1 )n (u0,1 ))
and
fei2 (s, (N 1 )n (u0,1 )) fei2 (s, (N 2 )n (u0,2 )).
Thus, for all i {1, ..., m}
(N 1 )n+1
(u0,1 )
i
Tt,i u10i ()
Tt,i u20i () +
Tt,i u20i () +
Z0 t
Z0 t
(t, x) [0, ) Rd
1
2
Remark 3.4 If we suppose fi1 fi2 in Rm
+ and 0 ui (0, ) ui (0, ) for all i {1, ..., m},
we obtain the same result as in Theorem 3.3.
Since F (0) = 0 by the previous theorem, we conclude that the solution of the system (3.1)-(3.2),
satisfies ui (t, x) 0 for all (t, x) [0, ) Rd .
51
3.3.
3.3.1.
Upper bound
Now, we are in position to establish an algebraic upper bound for the solutions of (3.1)-(3.2).
Since u = (ui )m
i=1 is the mild solution of system (3.1)-(3.2) with the extended function fi
1
m
C (R ) which is globally Lipschitz for all i {1, ..., m}, we have that
fi
Rm and i, j {1, ..., m}
uj () Lip(fi ),
Taking L = maxi{1,...,m} {Lip(fi )}, we have
m
Z 1
m
X Z 1 f
X
i
fi (w) =
Dfi (w)d w
wj
(w)d L
wj
w
j
0
0
j=1
j=1
(3.23)
(3.24)
(3.25)
In what follows, we prove that for each time t > 0, the solution of the system (3.1)-(3.2)
decay as |x|d2 when |x| is large enough, due to the inequality and convolution in (3.25) and
since u0 satisfies (3.3), we only need to prove that each term of the matrix F1 (e(A(||)+B)t ) has
the desired decay. To do this, we use the technique of rotating the interval of integration by a
52
small angle, in order to simplify the calculus at the moment to bound the integral that appear
in the previous Fourier transform. Hence, before to establish the upper bound, we state the
following lemma, which will help us to prove that the integral over Rd is equal to the integral
over ei Rd for some small enough, i.e., the bounds for the entries hi,j of e(A(r)+B)t which
appear in Lemma 3.5, decay fast enough, hence we can prove that the integral over the arc
obtained when we rotate the integration line, converges to zero.
Lemma 3.5 Let <
,
41
21
cos(21 ))t
if r < 1
(3.26)
if r 1
(3.27)
and
|hi,j (t, ei r)| e(cr
cos(21 ))t
(3.28)
where ej is the jth vector of the canonical basis of Rm , thus, we have that
m
i
w(t, r) = e(A(e r)+B)t .ej = hkj (t, ei r) k=1
Multiplying (3.28) by the conjugate transpose w, we have that
t w.w =
m
X
k=1
thus
m
X
1
t |w|2 +
cos(2k )r2k |wk |2 = Re(Bw.w) c|w|2
2
k=1
for some c > 0. By the choice of and using Gronwall Lemma, we get that for all j, k
{1, ..., m}
2
|hk,j (t, ei r)| |w(t, r)| e(cr 1 cos(21 ))t ,
if r < 1
and
|hk,j (t, ei r)| |w(t, r)| e(cr
cos(21 ))t
if r 1
Now, we divide the proof of the upper bound in two cases. First, for the sake of simplicity,
we consider the one space dimension case to underline the idea of the proof. The higher space
dimension case is treated after and requires the use of Bessel functions of first and third kind.
53
(3.29)
= Ct Dt
Also, we remember the Wilcox Formula [48] which states
Z t
e(A(r)+B)t
e(ts)(A(r)+B) A0 (r)es(A(r)+B) ds
=
r
0
Let t > 0, by the above formulas and integrating by parts, we have
Z
Z
1 (A(||)+B)t
ix (A(||)+B)t
F (e
) =
e e
d = 2
cos(|x|r)e(A(r)+B)t dr
R
0
Z
(A(r)+B)t
sin(|x|r) e
= 2
dr = I1 + I2 + I3
|x|
r
0
(3.30)
where
Z
I1 = 2
0
sin(|x|r)
|x|
Cts A0 Cs (r)dsdr
|x|
I2 = 2
0
I3 = 2
|x|
Z
sin(|x|r) t
(Cts A0 Ds (r) Dts A0 Cs (r) + Dts A0 Ds (r)) dsdr
|x|
0
Z t
sin(|x|r)
(Cts A0 Ds (r) Dts A0 Cs (r) + Dts A0 Ds (r)) dsdr
|x|
0
54
2k
)esr
2j
i, j {1, ..., m}
k=1
where ckij (t, s) is some integrable function on s [0, t] arising from products of the entries of
e(ts)B and esB , which have the form (m1+emLt ) and (1+emLt ). Thus, set I1 = ((I1 )ij )m
i,j=1
and take the change of variables u = r2k |x|2k , we obtain
Z
m Z
X
2j
sin(|x|r) t k
2
k
2k 1 (ts)r
sr
|(I1 )ij | =
cij (t, s)r
e
e
dsdr
|x|
0
k=1 0
Z
Z
m
i|x|r
t
X
e
2k sr 2j
k
2
1
(ts)r
k
Im
c
(t,
s)r
e
e
dsdr
ij
|x|
0
0
k=1
Z t
Z
m
X
(2k )1
iu1/2k (ts)u|x|2k suj /k |x|2j
k
e
e
e
duds
=
cij (t, s)
Im
1+2
k
|x|
0
0
k=1
denoting
eiu
k =
1/2k
e(ts)u|x|
2k
j /k
esu
|x|2j
du
we have
Z t
m
X
(2k )1
k
|(I1 )ij | =
Im
c
(t,
s)
ds
k
ij
|x|1+2k
k=1
m
X
k=1
(2k )1
|x|1+2k
The integral k can be simplified by rotating the interval of integration by < min
thus, we get
Z
1/2
i/2k (ts)r|x|2k ei sr j /k |x|2j eij /k i
k =
eir k e
e
e
e dr
0
55
, 2, 21j
2
hence
Z
|k |
0
er
1/2k
er
1/2k
sin(/2k )
dr
dr = C(k )
Z
m
X
(2k )1 C(k )
k=1
|x|1+2k
C(t)
|x|1+2
|x| 0
0
(3.31)
computing each norm and since r [0, |x| ], we have that kA(r)k r2 and kA0 (r)k
Cr21 , thus
Z s
0
(ts)B (ts)A(r) 0
e(sw)(A(r)+B) [ewB , A(r)]ewA(r) dwk
kCts A Ds k = ke
e
A (r)
0
0
(kA(r)k+kBk)t
2skA (r)kkA(r)ke
2Cse(1+kBk)t r41
in the same way we prove that kDts A0 Cs k 2C(t s)e(1+kBk)t r41 and kDts A0 Ds k
4C(t s)se(1+kBk)t r61 .
By (3.31) and since (1/2, 1), we conclude
|(I2 )ij |
2C
t3 (1+kBk)t
C(t)
2
(t
+
)e
1+2
1+4
|x|
3
|x|
56
Now, we compute the bound for the first term of I3 . Rotating the interval of integration by
< min(, 4 1 ), we get
Z i|x|r Z t
Z
Z
sin(|x|r) t
e
0
0
1
Cts A Ds dsdr = 2Im
Cts A Ds dsdr
I3 = 2
|x|
|x|
0
|x| |x|
0
!
Z i|x|1 ei Z t
e
= 2Im
Cts A0 Ds (|x| ei )|x| iei dsd
|x|
0
0
Z
+2Im
i|x|rei
|x|
|x|
:= 2Im(1 ) + 2Im(2 )
Hence, to get a bound for 1 = ((1 )ij )m
i,j=1 , we need to state a bound for each term that appear
in (1 )ij . By abuse of notation, we call to each one of these terms. Hence, the general form
for is
Z Z t Z s i|x|1 ei
2j i2j
e
e
|x|(2j 1) ei(2j 1)
C(t s)C(w)e(ts)|x|
=
|x|
0
0
0
hp,q (s w, |x| ei )|x|2l ei2l ew|x|
2k ei2k
where C(t s) and C(w) are positive and integrable functions. Now, since |x| > 1 and by
(3.26) of Lemma 3.5
Z Z t Z s |x|1 sin()
2j
e
cos(2j )
|x|(2j 1)
C(t s)C(w)e(ts)|x|
|Im()|
|x|
0
0
0
2
|x|1+2
then, we conclude
C(t)
|x|1+2
As in the proof of 1 , we continue calling by each term that appears in (2 )ij , thus, the general
form for is
Z Z t Z s i|x|rei
2j i2j
e
=
C(t s)C(w)e(ts)r e
r2j 1 ei(2j 1)
|x|
|x|
0
0
|Im((1 )ij )|
57
2k ei2k
ei dwdsdr
then
|Im()|
e|x|
sin()
|x|
Z tZ
0
C(t s)C(w)e(ts)r
2j
cos(2j ) 2j 1
2k
cos(2k )
dwdsdr
now, breaking the first integral in two terms and bounding, we have by (3.26) in Lemma 3.5 that
Z tZ
0
C(t s)C(w)e(ts)r
2j
cos(2j ) 2j 1
0
2
0
(cr21 cos(21 ))(sw) wr2k cos(2k )
e
Z
Z t Z se
c(sw)
C(t s)C(w)e
dwds
dwdsdr
r2(j +l )1 dr
= C(t)
and applying (3.27) of Lemma 3.5, we have
Z Z tZ s
2j
C(t s)C(w)e(ts)r cos(2j ) r2j 1
1
0
2
0
(cr2 cos(21 ))(sw) wr2k cos(2k )
e
Z
Z t Z se
c(sw)
C(t s)C(w)e
dwds
0
C(t)
Hence,
1
C(t)e|x|
|Im()|
|x|
sin()
58
C(t)
|x|1+2
dwdsdr
r2(j +l )1 etr
cos(21 )
dr
Now, we state the proof of Lemma 3.6 in the higher space dimension case, i.e. when d > 1
and := m ... 1 1. Let note that this case requires the use of Bessel functions of first
and third kind.
Proof. As in the previous proof, we only need to prove that
|x|d+2 |ij (t, x)| Cij (t),
d
Cd
=
e(A(r)+B)t J d 1 (|x|r)r 2 dr
d
2
|x| 2 1 0
setting = |x|, doing a change of variables and integrating by parts
Z
d
d
d+2 1 (A(||)+B)t
+2+1
|x|
F (e
) = Cd 2
e(A(r)+B)t J d 1 (r)r 2 dr
2
Z 0
r
d
d
= Cd 2
e(A( )+B)t J d 1 (r)r 2 dr
2
0
Z
(A(s)+B)t
d
d
e
= Cd 2 1
r 2 J d (r)
dr
2
r
s
0
s=
Z
(A(s)+B)t
d
d
e
(1)
= Re Cd 2 1
r 2 H d (r)
s
2
0
(1)
dr
s= r
where H = J + iY is the Bessel Function of third kind [1]. Now, using (3.29) and the
notation given in (3.30), we have
|x|d+2 |ij (t, x)| Cd [|(I1 )ij | + |(I2 )ij | + |(I3 )ij | + |(I4 )ij |]
59
In =
d
1
2
d
2
(1)
r H d (r)
2
Jn (r1 )dsdr
(I1 )ij
m Z
X
k=1 0
m Z t
X
:=
21
d
2
(1)
r H d (r)
2
2k
es(r/)
2j
dsdr
k=1
where ckij (t, s) is some integrable function on s [0, t] formed by the product of the terms of
e(ts)B and esB . The integral term k can be estimated by rotating the interval of integration by
< min( 2 , 2 1 ), thus, we get
k =
2(k )
2k ei2k
(1)
2j i2j
es(r/)
ei dr
(1)
the last integral is finite since H d (z) i (d/2)(z/2)d/2 on the interval [0, R] and also
2
(1)
|H d (rei )|
2
c er sin()
r
|(I1 )ij |
m
X
k=1
Z
C(k )
60
Now, let continue with the bound for I2 . Hence, by rotating the interval of integration by
< min( 2 , 4 1 ), we get
Z
Z t
(1)
21
i d2
i
I2 =
(re ) H d (re )
Cts A0 Ds (r1 ei )ei dsdr
2
0
0
i(2j 1)
2
2k ei2k
ei dwdsdr
then
||
2(j l )
Z tZ
0
0
2j 1
2j
(1) i
r H d (re ) C(t s)C(w)e(ts)(r/) cos(2j )
d
2
2k
cos(2k )
dwdsdr
now, doing a similar procedure as in the last part of the previous proof, by (3.26), (3.27) of
Lemma 3.5 and since = |x| > 1, we have
Z
Z tZ s
d
c(sw)
i
+2(j +l )1 (1)
2
r
C(t s)C(w)e
dwds
|| 2
H d (re ) dr
0
eij (t)
= C
Hence, we get
|(I2 )ij | Cij (t)
To conclude, we can do a similar proof for I3 and I4 .
Now, we present an alternative simple proof of Lemma 3.6, for the particular case in which
:= i < 1 for all i {1, ..., m}, this proof is done for d 1. In this case, since we are
working with a unique index , we can bound directly in the iteration process (3.32), to prove
that the solution of the system (3.1)-(3.2) decay as |x|d2 when |x| is large enough for all
t > 0.
Proof. As in the previous cases, we have that
Z 1
X
Z 1
m
X
m
f
i
|fi (u)| =
Dfi (u)d u
uj
(u)d L
|uj |
u
j
0
0
j=1
j=1
61
where L = maxi{1,...,m} {Lip(fi )}. Following (3.14), we have that u = limi+ ui where
ui = Nui 0 (u0 ) = Nu0 (ui1 ),
(3.32)
Using the iterative process (3.32) and the semigroup properties of the operator Tt , we have that
for all i {1, ..., m} and n N
|uni (t, x)|
m
(mLt)n X
(mLt)2
+ ... +
Tt u0j (x)
1 + (mLt) +
2!
n!
j=1
(3.33)
where un = (uni )m
i=1 . Also, we know that
kun ukC([0,),X)m 0,
when n +
when n +
for all (t, x) [0, ) Rd and i {1, ..., m}. Taking the limit when n + in (3.33), we
conclude that
m
X
ui (t, x) emLt
Tt u0j (x),
(t, x) [0, ) Rd
(3.34)
j=1
Now, by hypothesis (3.3), we have that there exists ri > 0 large enough, such that
u0i (x) Ci |x|d2 ,
if |x| > ri
also we know that 0 u0i ai . Thus, if |x| > 2ri and t > 0
d
t 2 Bu0i (y)
Z
Tt u0i (x)
Rd
1 + (t 2 |x y|)d+2
dy
t 2 Bu0i (y)
dy
1
1 + (t 2 |x y|)d+2
Z
d
t 2 Bu0i (y)
+
dy
1
d+2
{|y|>|x|/2} 1 + (t 2 |x y|)
{|y||x|/2}
:= I1 + I2
If |y| |x|/2, we have |x y| |x| |y|
Z
I1
{|y||x|/2}
|x|
,
2
t 2 Bu0i (y)
then
2d+2 tB
dy
1
|x|d+2
1 + (t 2 |x|
)d+2
2
62
Z
u0i (y)dy
{|y||x|/2}
moreover,
{|y|<ri }
{ri |y||x|/2}
{ri |y||x|/2}
Ci
Ci
dy
d+2
|y|
2ri2
then,
2d+2 tB
I1
|x|d+2
Ci
ai (0, ri ) +
2ri2
d+2
Ci
Ci
2|x|d+2
, hence
Now, if |y| > |x|/2, we have that u0i (y) |y|d+2
Z
ds
2d+2 Ci B
2d+2 Ci B
I2
:=
C
|x|d+2 Rd 1 + |s|d+2
|x|d+2
t 2 Bu0i (y)
Z
Tt u0i (x)
1 + (t 2 |x y|)d+2
ai C B
ai C B(2ri )d+2
|x|d+2
dy
Rd
3.3.2.
Pm
i=1
C(t)
,
|x|d+2
(t, x) (0, ) Rd
Ci .
Now, we state the following comparison principle for classical solutions. This result will be
useful to deal with sub and super solutions. Indeed, we have not devised a mild representation
for them, so we can not apply Theorem 3.3 directly.
63
1
d m
m
Theorem 3.7 Let u = (ui )m
i=1 and v = (vi )i=1 functions in C ([0, T ]; C0 (R )) , fi satisfies
(3.4) and
t ui + (4)i ui fi (u), t vi + (4)i vi fi (v)
x Rd
also
ui (t, x) = O(|x|(d+2) ) and vi (t, x) = O(|x|(d+2) ) as |x|
for all t [0, T ]. Then
(t, x) [0, T ] Rd .
u(t, x) v(t, x)
(3.35)
(3.36)
(3.37)
for all (t, x) [0, T ] Rd . Moreover, by the decay assumptions on the functions u and v, we
have that
wi (t, x) = O(|x|(d+2) )
t [0, T ] as |x|
t [0, T ]
(3.38)
(3.39)
Rd
where Cij (T ) are constants that depend of T . Now, let wi+ be the positive part of wi , we want
to prove that
Z
Z
+ 2
t
(wi ) dx =
t (wi+ )2 dx
(3.40)
Rd
Rd
64
which is quite simple, because, from (3.36), we deduce that (wi+ )2 and t (wi+ )2 are continuous in (0, T ) Rd and
+ 2
t (w ) = 2 w+ t wi 2C2 (T ) |wi | Cg(x)
(3.41)
i
i
the last inequality and the existence of the integrable function g follows from (3.37) and (3.38),
thus we conclude (3.40). Now, multiplying each term of the equation (3.35) by the positive part
of wi and integrating over Rd , we have that
Z
Z
Z 1
Z
+
+
i
0
wi (4) wi dx
wi
fi ( )d.wdx
wi+ t wi dx
(3.42)
Rd
Rd
Rd
Now, since all the above integral exist and having in mind that fi C 1 (Rm ) for all i
{1, ..., m}, from (3.42), we deduce
Z
Z
+
wi+ (4)i wi dx
wi t wi dx +
Rd
Rd
Z
Z 1
+
wi
fi ( )d.wdx
Rd
Z
=
Rd
wi+
Z
m
X
i fi ( )dwi dx +
0
Rd
i fi ( )d(wi+ )2 dx
j=1,i6=j
m
X
Rd
wi+
j fi ( )dwj dx
0
d
j=1,i6=j R
Z
m
X
j fi ( )dwi+ wj+ dx
j=1,i6=j
Rd
j fi ( )dwi+ wj dx
i fi ( )d(wi ) dx +
j fi ( )d(wi+ )2 dx
2
d
d
R
0
0
j=1,i6=j R
Z
Z
m
1
1 X
+
j fi ( )d(wj+ )2 dx
2 j=1,i6=j Rd 0
m Z
X
C
(wj+ )2 dx
j=1
Rd
65
Rd
j=1
Rd
Rd
so,
(wj+ )2 dx
Rt
Cds
m Z
X
Rd
j=1
Z
Rd
(wj+ )2 dx = 0,
j {1, ..., m}
then, we conclude
wj (t, x) 0,
(t, x) [0, T ] Rd , j
Remark 3.8 Note that as a consequence of Lemma 3.6 (in the case d = 1 and d > 1), we have
the enough regularity to apply Theorem 3.7 to the solution of the problem (3.1)-(3.2).
3.3.3.
Lower bound
The following is an important result needed to prove the Theorem 3.1, which sets an algebraically lower bound for the solutions of the cooperative system (3.1)-(3.2). This result is valid for
any dimension d and for any index i .
Lemma 3.9 Let u = (ui )m
i=1 the solution of the system (3.1)-(3.2), with initial condition u0
satisfying (3.3) and fi satisfying (3.4). Then, there exist constants i > 0 and 1 > 0 such that
ui (t + t0 , x)
Ci tei t
d
t 2 +1 + |x|d+2
i {1, ..., m}
66
j=1
fi
( )d
uj
fi
where = u [0, u(t, x)] [0, M ] and u
: [0, M ] R is continuous for all i, j
j
{1, ..., m}, using the fact that the system is cooperative, there exist constants ij > 0 such that
fi
ii and ij fi ( ) for all i 6= j
(
)
ui
uj
t um + (4)
Z
um = fm (u)
0
fm
( )dum mm um
um
for all x Rd and t t0 , by the maximum principle of reaction diffusion equations, we have
that
um (t + t0 , x) emm t (pm (t, ) um (t0 , ))(x), for all t 0
Since um (t0 , ) 6 0 is continuous and nonnegative, we can find Rd fixed such that
um (t0 , y) C for all y BR () for some R > 0 and C > 0.
Taking |x| > R, t 1 and using that := m < 1
Z
(pm (t, ) um (t0 , ))(x) =
pm (t, x y)um (t0 , y)dy
Rd
Z
B 1 t
Z
C
|y|R
Z
= C
t 2 +1 + |x y|d+2
B 1 t
d
|z|R
dy
t 2 +1 + |x z|d+2
dz
67
Now, if |x| R and t 1, taking > 0 such that supp(um (t0 , )) B (0) 6=
Z
B 1 tum (t0 , y)
(pm (t, ) um (t0 , ))(x)
dy
d
+1
+ |x y|d+2
B (0) t 2
Z
B 1
um (t0 , y)dy
d
t 2 +1 + (R + )d+2 B (0)
Cet
for some small constant C > 0. Moreover, since t 1
(pm (t, ) um (t0 , ))(x) Cet
Ctet
d
t 2 +1 + |x|d+2
Cm tem t
t
d
+1
2
|x|d+2
x Rd , t 1
with m = mm + 1.
Step 2. To finish the proof is necessary to find a lower bound of the convolution between
pi (1, ) and d +1 t d+2 for all t 1, Thus, by the same computations as above, it is possible
+|xy|
t 2
Z
Rd
d
+1
2
|y|2
4
+ |x
y|d+2
dy
Ctet
t
d
+1
2
|x|d+2
x Rd , t 1
and
Z
Rd
1
1 + |y|d+2i
"
t
t
d
+1
2
+ |x
y|d+2
dy
Ctet
t
d
+1
2
|x|d+2
x Rd , t 1
68
for all x Rd and t t0 , where i max(ii , m + 2). Then, by the maximum principle of
reaction diffusion equation and Duhamel formula, we have that
ui (t + t0 , x) ei t (Hi (t, ) ui (t0 , ))(x)
Z tZ
i t
+im e
Hi (t s, y)um (s + t0 , x y)ei s dyds
0
Rd
1
d
(4t) 2
Hi (t, x) =
|x|2
4t
if i = 1
if i (0, 1).
pi (t, x)
So, taking t 1 for any 1 3
i t
Z tZ
Rd
Z
1
t1
e(i m )s
Z
Rd
1
1 + |y|d+2i
"
s
d
s 2 +1 + |x y|d+2
#
dyds
x Rd , t 1
d
+1
d+2
t 2 + |x|
with 1 larger if necessary and taking i := i .
Then, we proved that there exist constants Ci > 0 and i > 0 such that
Ci tei t
ui (t + t0 , x) d
,
i {1, ..., m}
t 2 +1 + |x|d+2
for all x Rd and t 1 .
69
3.4.
In order to prove Theorem 3.1, we need to construct explicit sub and super solution, which will
have the form of the following vector field
1
(3.43)
v(t, x) = a 1 + b(t)|x|(d+2)
where b(t) is a time continuous function and Rm is the normalized principal eigenvector of
DF (0) associated to the principal eigenvalue 1 and R+ is taken as in the hypothesis (H2)
f
and (H3). Note that, since the system is cooperative, uji (0) > 0 for all i 6= j {1, ..., m}, by
Perron-Frobenius Theorem, we can take > 0.
The following result allow us to understand the behavior of the fractional laplacian (4)i
on the function v defined by (3.43). The proof of this result is based in a result proved by
Bonforte and vazquez in [13], also a similar result was announced in [16].
Lemma 3.10 Let v be defined as in (3.43). Then, there exist a constant D > 0 such that
2i
in Rd
2
,
d+2
i
(3.44)
1
where w(x) = (1+ | x |(d+2) ) . Indeed, note that v(t, x) = aw(b(t) (d+2) x), also, since
w C 2 (Rd ) L (Rd ) we have that (4)i w L (Rd ) and it is 2i -homogeneous, thus
1
2i
where y = b(t) (d+2) x. Moreover w(x) |x|(d+2) and |D2 w(x)| c|x|(d+2+2) for |x|
large enough.
Since (4)i w L (Rd ), taking D large enough, it is sufficient to prove the result for
large values of |x|. We have to estimate
Z
w(x) w(y)
i
| (4) w(x) |= C(d, i ) P.V.
dy
d+2i
Rd |x y|
hence
Z
w(x) w(y)
dy
d+2i
Rd |x y|
Z
=
|y|>3|x|/2
Z
w(x) w(y)
w(x) w(y)
dy +
dy
d+2
d+2i
i
|x y|
{|x|2|y|3|x|}\B|x|/2 (x) |x y|
Z
Z
w(x) w(y)
w(x) w(y)
+
dy
+
dy
d+2i
d+2i
B|x|/2 (x) |x y|
|y||x|/2 |x y|
:= I1 + I2 + I3 + I4
70
In the first integral, since |y| > 3|x|/2 we have w(y) w(x) and then |w(x) w(y)| w(x),
hence
Z
Z
w(x)
w(y)
1
C1
|I1 | =
dy Cw(x)
dr d+2+2i
d+2
1+2
i
i
|x|
|y|>3|x|/2 |x y|
3|x|/2 r
since w(x) |x|(d+2) for |x| large enough.
Following a similar idea as in the above computation, since |y| > 3|x|/2 we have w(y)
w(x/2) and then |w(x) w(y)| w(x/2), also note that w(x/2) |x/2|(d+2) for |x| large
enough, hence
Z
Z 3|x|/2
Cw(x/2)
C2
w(x)
w(y)
d1
dy
r
dr
|I2 | =
{|x|2|y|3|x|}\B|x|/2 (x) |x y|d+2i (|x|/2)d+2i |x|/2
|x|d+2+2i
since |x y| |x|/2.
Before to continue, note that
Z
P.V.
B|x|/2 (x)
w(x) (x y)
dy = 0
|x y|d+2
moreover, if |x y| < |x|/2 then |x|/2 < |y| < 3|x|/2, therefore
|ij2 w(y)|
c
|y|d+2+2
c2d+2+2
|x|d+2+2
dr
d+2+2
2
|x|
r i 1
0
22i
C
|x|
d+2+2
|x|
2
C3
=
d+2
i +2
|x|
71
It only remains to estimate the fourth integral. Note that |y| |x|/2 implies w(x) w(y)
which gives |w(x) w(y)| w(y), thus
Z
Z
|w(x) w(y)|
2d+2i
C4
|I4 |
dy d+2i
w(y)dy d+2i
d+2
i
|x|
|x|
|y||x|/2 |x y|
Rd
since |x y| |x|/2 and |x| is large enough. Therefore, we conclude that
| (4)i w(x) |
C5
Dw(x)
|x|d+2
2
,
d+2
| 4w(x) | Dw(x)
(3.45)
1
(1 + |x|(d+2) )
Dv(x)
since |x| is large and for some constant D.
In what follows, we will use the results of previous sections to obtain appropriate classical
sub and super solutions of the system (3.1)-(3.2) with the form of the vector field (3.43), then
we use the comparison principle to obtain the desired results.
We divide the proof of Theorem 3.1 in two lemmas.
Lemma 3.11 Let d 1 and assume that F satisfies (3.4), (H1) and (H2). Let u be the solution
to (3.1)-(3.2) with u0 satisfying (3.3). Then, for every = (i )m
i=1 > 0, there exist c > 0 such
that, for all t >
o
n
1
x Rd | |x| > ce d+2 t x Rd | u(t, x) <
with > 0 large enough.
72
1
+1
1 (d+2)
2
2
1 (d+2)
21
e d+2 t )
2
1 (d+2)
1 (d+2)
2
1 (d+2)
2
, note that, 0
ai b (t)|x|1 (d+2)
2i
+1
Db(t) 1 (d+2) ui 1 ui +
1 1+1
c1+
a
i
1
+1
1 (1 + b(t)|x|1 (d+2) ) 1
(1 + b(t)|x|1 (d+2) ) 1
o
n
2
ai
0
+1
1 (d+2)
b (t) 1 Db(t)
1 1 b(t) |x|1 (d+2)
1
+1
(d+2)
1
1
1 (1 + b(t)|x|
)
n
o
2
ai
1
1 (d+2)
+
1 + ci a
Db(t)
(3.46)
1
+1
(1 + b(t)|x|1 (d+2) ) 1
in the last inequality we use that i and b(t) 1. Thus, taking
1
(D + 1 ) 1
1
c 1 mini (i )
the right hand side of inequality (3.46) is bigger than or equal to 0 for all t > 0.
To end the proof, for any t0 > 0 fixed, we can take a satisfying the above condition and t1 > t0
such that
ui (t1 , x) ui (t0 , x),
x Rd , i {1, ..., m}
note that, this is possible due to Lemma 3.6 (in the case d = 1 and d > 1). Therefore, we
conclude that u is a supersolution to (3.1)-(3.2) for all t t1 . Thus, using Theorem 3.7, we get
for all t t0
ui (t + t1 t0 , x) ui (t, x),
x Rd , i {1, ..., m}
(3.47)
73
ai e1 (t1 t0 )
1
i B 1
and since Be1 1 (t+t1 t0 ) b(t + t1 t0 ), we have that, if
1
a1 i 1 1 1 (t+t1 t0 )
e
i 1 B
Therefore
a1 i 1
i 1
ai
1
< i
74
1
2
Taking for the moment B (L1
with any constant L max{D, 1 }. We consider
1 )
21
2
2 (d+2)
1
2 (d+2) d+2 t
2
)
the function b(t) = (L1 + B
e
which a solution of the problem
2
+1
2
2 (d+2)
2 (d+2)
2
ui 1 ui +
+ Db(t)
1
1
+1
+1
2 (1 + b(t)|x|2 (d+2) ) 2
(1 + b(t)|x|2 (d+2) ) 2
o
n
2
ai
+1
0
2 (d+2)
(t)
+
Lb(t)
|x|2 (d+2)
b(t)
2
2 1
1
+1
2 (1 + b(t)|x|2 (d+2) ) 2
2
2
ai
ca
+
Lb(t) 2 (d+2) 1 +
(3.48)
1
+1
i
(d+2)
2
2
(1 + b(t)|x|
)
in the last inequality we use that i and b(t) 1. Thus, taking
a
mini {i }1
2c
1
the right hand side of inequality (3.48) is less than or equal to 0 for all t > 0.
Since, ui (0, ) u0i may not hold for all i {1, ..., m}, we look for a time t1 max{1 , 2L1
1 },
where 1 was defined in Lemma 3.9. Moreover, for any t0 > 0 fixed, we know that
ui (t1 + t0 , x)
ci t1 ei t1
d
+1
2
t1
|x|d+2
75
Consequently, we choose
a=
2 (d+2)
2
2
B=
t1
2maxi {i }t12
mini {i }1
2c
1
and
B (L1
1 )
2 (d+2)
2
are satisfied.
By the election of a and B, we deduce that
since
a
ci ei t1
d
2
ai t12
then
+1
2i t1
ci i t1
t1 e
2
B 2 (d+2) =
2
1
t1
t1 L1
1
therefore,
1
2
b(0)
"
=
and
L1
1 +B
2
(d+2)
2
2 (d+2)
2
# 1
d+2
t1 2
1
2ai
1
2
d+2 b(0)
t
ci e i 1 t1
t1 2
Then, we deduce
h
i
d
1
ci i t1
ci
+1
t1 e
ai t12 0 ai b(0) 2 ei t1 t1 |x|d+2
2
2
so,
d
1
ci i t1
+1
t1 e
(1 + b(0) 2 |x|d+2 ) ai (t12 + |x|d+2 )
2
but
2 (d+2)
(1 + b(0)|x|
1
2
1 + b(0) 2 |x|d+2
76
Therefore,
ci t1 ei t1
d
+1
2
t1
ai
+ |x|d+2
(1 + b(0)|x|2 (d+2) ) 2
x Rd , i {1, ..., m}
ai
2
1
2
x Rd , t t1 + t0
(3.49)
Then, if t t1 + t0 and |x| Ce d+2 t , we get that ui (t, x) i for all i {1, ..., m}.
1
ai
(1 + b(t t1 t0 )|x|2 (d+2) )
1
2
ai
1
= i
2 2
To conclude this section, the proof of Theorem 3.1 follows directly from Lemmas 3.11 and 3.12.
3.5.
In this subsection, we prove that under some appropriate assumptions on the nonlinearity and
the initial datum, it is possible to state that the solution u(t, x) to (3.1)-(3.2) tends, as t +,
to the smallest constant positive steady solution of (3.1)-(3.2). In order to prove this result, let
77
define by the positive constant eigenvector of DF (0) associated to the first eigenvalue 1 ,
where F = (fi )m
i=1 . Thus 1 and > 0 satisfy
(L DF (0)) = 1
> 0, kk = 1
where L = diag((4)1 , ..., (4)m ). Now, let consider the problem
(t) = F ( (t))
(0) =
(3.50)
thus, there exists 0 > 0 such that, for each (0, 0 ) we can find a constant u+
> 0 satisfying
+
+
(t) % u as t +, also F (u ) = 0. We define
u+ = inf 0 u+
(0, )
since F is continuous, we deduce that F (u+ ) = 0. Also, since the function F is positive in a
small ball near to zero, we deduce that u+ > 0.
Moreover we assume that the initial condition u0 satisfies
u0 u+
in Rd
We prove Theorem 3.2 through a succession of lemmas. Let BR (0) be the open ball of Rd ,
with center 0 and radius R. Let us call uR the unique solution of the elliptic system
(4)i uR
= fi (uR ),
i
uR = 0
uR > 0
x BR (0)
on Rd \ BR (0)
on BR (0)
(3.51)
Lemma 3.13 Let d 1, > 0 and assume that F satisfies (3.4), (H1), (H2), (H3) and (H4).
There exists R > 0 such that the solution v R of the system
t viR + (4)i viR = fi (v R ),
v R (t, x) = 0
0 < v R (0, x) min(, uR ),
t > 0, x BR (0)
on t 0, x Rd \ BR (0)
on x BR (0)
satisfies
lim v R (t, x) = uR (x)
t+
78
x B1 (0)
(3.52)
Proof: Let R be the positive eigenvalue associated to R in the ball BR (0), thus R and R
satisfy
(L DF (0))R = R R
in BR (0)
R
R
d
> 0 in BR (0), = 0 in R \ BR (0), kR k = 1
Now, following the same computations as in [10], we can deduce that R given by the minimum
of
R
Pm R R (i (x)i (y))2
1
dy
dx
t > 0, x BR (0)
on t 0, x Rd \ BR (0)
on x BR (0)
(3.53)
t > 0, x BR (0)
79
t > 0, x BR (0)
Since wR is nondecreasing in time t, standard elliptic estimates imply that wR converges locally
to a stationary solution w ( uR ) of (3.53). But since uR is the unique solution of (3.51), we
conclude that w = uR in B1 (0).
Remark 3.14 Let note that for each y Rd , if x B1 (y) then x y B1 (0). Thus taking
= ()m
i=1 > 0, as a consequence of Lemma 3.13, there exist R > 0 and T > 0 that not
depend of y, such that, for all t T
|viR (t, x y) uR
i (x y)| i
x B1 (y)
i {1, ..., m}
(3.54)
t > 0, x BR (0)
on t 0, x Rd \ BR (0)
on x BR (0)
(3.55)
(t, x) [0, ) Rd
(3.56)
Now, by Lemma 3.13, there exists R > 0 large enough such that wR (t, x) converges to uR (x),
as t + for all x B1 (0). Hence, taking the limit when t tends to + in (3.56), we have
that
v(x + y) uR (x)
x B1 (0)
80
y Rd
Now, we establish the following result in which we state that u+ is the smaller stationary
solution of (3.1)-(3.2).
Lemma 3.16 Let d 1 and assume that F satisfies (3.4), (H1), (H2), (H3) and (H4). If v is a
positive continuous solution of (3.54) such that v u+ in Rd , then v u+ .
Proof: First, since F (u+ ) = 0, then u+ satisfies the system (3.54). Now, by Lemma 3.15, there
exists > 0 such that v in Rd . By Theorem 3.3, we deduce that
v(x) (t)
t 0, x Rd
(3.57)
In the following result we state a relation between the stationary solution in the ball BR (0)
and the stationary solution in the whole space.
Lemma 3.17 Let d 1 and assume that F satisfies (3.4), (H1), (H2), (H3) and (H4). Let uR
be the unique solution of the system
(4)i uR
= fi (uR ),
i
uR = 0
uR > 0
x BR (0)
on Rd \ BR (0)
on BR (0)
(3.58)
81
d
Remark 3.18 As a consequence of Lemma 3.17, for each = ()m
i=1 > 0 and y R , there
exists R > 0 that not depend of y, such that, for all R R
+
|uR
i (x y) ui | i
x B1 (y)
1
,
d+2
then
lim
t+ |x|ect
b) If c >
1
,
d+2
then
lim sup ui (t, x) = 0
t+ |x|ect
(3.59)
where u = (ui )m
i=1 is the solution of (3.1)-(3.2).
Let > 0, by the Remarks 3.14 and 3.18, we can find R > 0 and T > 0 large enough
such that for R R and s T , we have
i
i
+
|uR
and |viR (s, x y) uR
(3.60)
i (x y) ui |
i (x y)|
2
2
for all y Rd , x B1 (y) and i {1, ..., m}.
In what follows, taking R R and large if necessary such that
ec1 T < e(c1 c)
82
(3.61)
Furthermore, since S
{z : |z| ec1 s } is a compact set, we can find a finite number of vectors
y1 , ..., yk , such that ki=1 B1 (yi ) cover {z : |z| ec1 s }. Thus, we have
c1 s
ui (s + T , x) u+
i i for all |x| e
Then, taking t = s + T + T
c1 T c1 t
ui (t, x) u+
e
i i for all |x| e
thus, we conclude the proof of part a), taking := + T and by election of , we have that
ct
ui (t, x) u+
i i for all |x| e
83
Bibliography
[1] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions With Formulas,
Graphs, and Mathematical Tables. Dover Publications, New York. 1972.
[2] D.G. Aronson and H.F. Weinberger, Multidimensional nonlinear diffusions arising in
population genetics, Adv. Math. 30 (1978), 33-76.
[3] B. Baeumer, M. Kovacs and M. M. Meerschaert, Fractional reproduction-dispersal
equations and heavy tail dispersal kernels. Bull. Math. Biology 69 No.7 (2007), 22812297.
[4] G. Barles, L. C. Evans, and P. E. Souganidis, Wavefront propagation for reactiondiffusion systems of PDE, Duke Math. J. 61 (1990), no. 3, 835-858.
[5] H. Berestycki, Le nombre de solutions de certains problemes semi-lineaires elliptiques,
J. Funct. Anal, 40 (1981), 1-29.
[6] H. Berestycki and F. Hamel and G. Nadin, Asymptotic spreading in heterogeneous diffusive excitable media, Journal of Functional Analysis, vol. 255 (2008), pp. 2146-2189.
[7] H. Berestycki, F. Hamel and L. Roques, Analysis of the periodically fragmented environment model. I. Species persistence, J. Math. Biol, 51 (2005), 75-113.
[8] H. Berestycki, F. Hamel and L. Roques, Analysis of the periodically fragmented environment model. II. Biological invasions and pulsating travelling fronts, J. Math. Pures
Appl, 84 (2005), 1101-1146.
[9] H. Berestycki, L. Nirenberg, Travelling fronts in cylinders, Ann. Inst. H. Poincare,
Anal. Non Lineaire 9 (1992), pp. 497-572.
[10] H. Berestycki, J.-M. Roquejoffre, and L. Rossi, The periodic patch model for population dynamics with fractional diffusion, Discrete Contin. Dyn. Syst. Ser. S 4 (2011),
1-13.
84
Bibliography
85
Bibliography
[24] L. C. Evans et P.E. Souganidis, A PDE approach to certain large deviation problems
for systems of parabolic equations. Annales de lnstitut Henri Poincare (C) Analyse non
lineaire S6 (1989): 229-258.
[25] P. Felmer and M. Yangari. Fast Propagation for Fractional KPP Equations with Slowly
Decaying Initial Conditions. SIAM J. Math. Anal., 45(2), 662-678.
[26] R.A. Fisher, The advance of advantageous genes, Ann. Eugenics 7 (1937) 335-369.
[27] J. Garnier, Accelerating solutions in integro-differential equations, SIAM J. Math.
Anal. 43 (2011), 1955-1974.
[28] J. Gartner and M.I. Freidlin, On the propagation of concentration waves in periodic
and random media, Doklady Akademii Nauk SSSR, vol. 20 (1979), pp. 1282-1286.
[29] F. Hamel and L. Roques. Fast propagation for KPP equations with slowly decaying
initial conditions. Journal of Differential Equations 249 (2010) 1726-1745.
[30] S. Hayek, Advanced Mathematical Methods in Science and Enginering. New York:
Marcel Dekker. 2001.
[31] C.K.R.T. Jones, Asymptotic behavior of a reaction-diffusion equation in higher space
dimensions, Rocky Mountain J. Math. 13 (1983), 355-364.
[32] Y. Kametaka, On the nonlinear diffusion equation of Kolmogorov-Petrovskii-Piskunov
type, Osaka J. Math. 13 (1976) 11-66.
equation de la di
[33] A.N. Kolmogorov, I.G. Petrovsky and N.S. Piskunov, Etude
de lO
ffusion avec croissance de la quantite de matiere et son application a un probleme
Moscou Ser. Inter. A 1 (1937) 1-26.
biologique, Bull. Univ. Etat
[34] V.N. Kolokoltsov, Symmetric stable laws and stable-like jump-diffusions, London Math.
Soc. 80 (2000), 725-768.
[35] D.A. Larson, Transient bounds and time-asymptotic behavior of solutions to nonlinear
equations of Fisher type, SIAM J. Appl. Math. 34 (1978) 93-103.
[36] M. Lewis, B. Li and H. Weinberger. Spreading speed and linear determinacy for twospecies competition models. J. Math. Biol. 45, 219-233 (2002).
[37] B. Li, H. Weinberger and M. Lewis. Spreading speeds as slowest wave speeds for
cooperative systems. Math. Biosci. 196, 82-98 (2005).
[38] R. Lui. Biological growth and spread modeled by systems of recursions. I. Mathematical theory. Math. Biosci. 93(2), 269-295 (1989).
86
Bibliography
[39] R. Mancinelli, D. Vergni, and A. Vulpiani. Front propagation in reactive systems with
anomalous diffusion, Phys. D 185 (2003), 175-195.
[40] A. Pazy, Semigroups of Linear Operators and Applications to Partial Differential Equations. Applied Mathematical Sciences, 44. Springer-Verlag, New York, 1983.
[41] D.H. Sattinger, Stability of waves of nonlinear parabolic systems, Adv. Math. 22 (1976)
312-355.
[42] D. Stan and J. Vazquez. The Fisher-KPP equation with nonlinear fractional diffusion.
To appear.
[43] A. I. Volpert, V. A. Volpert and A. I. Volpert, Traveling wave solutions of parabolic systems. Translations of Mathematical Monographs, Volume 140, American Mathematical
Society, 1994.
[44] H.F. Weinberger, M. Lewis and B. Li. Analysis of linear determinacy for spread in
cooperative models. J. Math. Biol. 45, 183-218 (2002).
[45] H.F. Weinberger, M. Lewis and B. Li. Anomalous spreading speeds of cooperative recursion systems. J. Math. Biol. 55, 207-222 (2007)
[46] H.F. Weinberger, On spreading speeds and traveling waves for growth and migration
models in a periodic habitat, Journal of Mathematical Biology, vol. 45 (2002), pp. 511548.
[47] H. F. Weinberger, Long-time behavior of a class of bioloical models, SIAM J. Math.
Anal., 13: 353-396, 1982.
[48] R.M. Wilcox, Exponential operators and parameter differentiation in quantum physics.
J. Math. Phys., 8, 962-982.
87