These Lionel Roques

***
Universit
e Pierre et Marie Curie, Paris 6
Ecole doctorale des Sciences Math
ematiques de Paris Centre
UFR 921
Equations
de r
eaction-diffusion
non-lin
eaires et mod
elisation en
ecologie
`
THESE
presentee et soutenue publiquement le lundi 5 juillet 2004
pour lobtention du
Doctorat de lUniversit
e Pierre et Marie Curie Paris 6
(sp
ecialit
e Math
ematiques Appliqu
ees)
par
Lionel ROQUES
Composition du jury
Rapporteurs :
M. Michel LANGLAIS
M. Jean-Claude SAUT
Examinateurs :
M.
M.
M.
M.
M.
Henri BERESTYCKI
Ham BREZIS
Alain FRANC
Jean-Pierre FRANC
OISE
Francois HAMEL
Directeur
Laboratoire Jacques Louis Lions UMR 7598
Mis en page avec la classe thloria.
Remerciements
Je tiens tout particulirement remercier mon Directeur de Thse, Henri Berestycki,
qui m'a propos des sujets passionnants, et avec qui j'ai grand plaisir travailler. J'apprcie normment son approche des mathmatiques, toujours claire et rigoureuse. Il m'a
beaucoup apport, et je lui en suis inniment reconnaissant.
J'exprime galement toute ma gratitude Franois Hamel, qui m'a souvent conseill et
guid, faisant toujours preuve de beaucoup de disponibilit et de gentillesse. J'ai galement
beaucoup apprci travailler avec lui.
Je suis trs sensible l'honneur que m'ont fait Michel Langlais et Jean-Claude Saut
en acceptant d'tre rapporteurs de cette thse. Je remercie galement les autres membres
de mon jury de thse, Alain Franc et Jean-Pierre Franoise, pour l'intrt qu'ils portent
mon travail.
Ham Brezis me fait un grand honneur en acceptant de participer ce jury. Je l'en
remercie vivement.
Je tiens galement exprimer ma reconnaissance envers les agents INRA avec lesquels j'ai travaill : Marie-Anne Auger-Rozenberg, Solen Boivin, Alain Chalon, Jean-Nol
Candau, Jean-Pierre Fabre et Arthur Raimbault. Ils ont su rpondre avec beaucoup de
Megastigmus schimitscheki, et m'ont fourni tous les documents qui m'taient ncessaires.
patience mes questions incessantes concernant la biologie du fameux
J'adresse un remerciement particulier Mayan Mimura, qui m'a invit un Congrs

au Japon, sjour que j'ai beaucoup apprci, et au cours duquel j'ai pu dcouvrir une
autre approche des mathmatiques.
Je suis galement redevable envers le Laboratoire Jacques-Louis Lions, et envers le
CMI-LATP, qui m'ont accueilli et m'ont permis d'eectuer ce travail dans de bonnes
conditions.
Je remercie l'imprimeur du Laboratoire, Christian David, pour la qualit de son travail.
Je voudrais galement saluer Flin, Stadium, Rodrigo, Gor, Mikos, Vith, Jrme
et Samy, qui, s'ils ne sont pas tous mathmaticiens, m'ont toujours encourag dans mon
travail.
Je tiens aussi remercier mes parents et Sarah, notamment pour leur relecture de ce
travail (mme s'ils n'ont sans doute pas tout compris).
ii
Sarah
iii
iv
Table des matires

Introduction Gnrale
1
Premire partie : Modles d'invasion biologique en environnement priodique fragment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1
Inuence des htrognits du milieu sur la survie des espces . . .
1.2
Invasions biologiques et fronts progressifs pulsatoires
. . . . . . . .
Deuxime partie : Etude d'un modle de combustion avec pertes de chaleur
Troisime partie : Modlisation de la dispersion d'un insecte invasif
Problmes ouverts et perspectives
. . . .
13
. . . . . . . . . . . . . . . . . . . . . .
14
Partie I : Analysis of the periodically fragmented environment model
17
Note au CRAS
19
Chapitre 1 : Inuence of periodic heterogeneous environment on species

persistence
27
1
2
Introduction
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Statements of the main results
. . . . . . . . . . . . . . . . . . . . . . . .
2.1
Existence and uniqueness results
2.2
Large time behaviour
2.3
28
32
. . . . . . . . . . . . . . . . . . .
33
. . . . . . . . . . . . . . . . . . . . . . . . .
34
Eects of the heterogeneity on species survival . . . . . . . . . . . .
35
Existence and uniqueness of a stationary solution
. . . . . . . . . . . . .
38
. . . . . . . . . . . . . . . . . . . . . . . . . . .
39
. . . . . . . . . . . . . . . . . . . . . . . . . .
41
3.1
Proof of existence
3.2
Proof of uniqueness
3.3
Energy of stationary states
The evolution equation
. . . . . . . . . . . . . . . . . . . . . .
46
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
Table des matires

5
Conservation of species in ecological systems
. . . . . . . . . . . . . . . .
5.1
Inuence of the amplitude" of the reaction term
5.2
Inuence of the shape" of
fu (x, 0)
. . . . . . . . . .
49
. . . . . . . . . . . . . . . . . .
50
The eect of fragmentation in bounded domain models
. . . . . . . . . .
Chapitre 2 : Biological invasions and pulsating travelling fronts

1
Introduction and main results
Existence result
53
57
. . . . . . . . . . . . . . . . . . . . . . . . .
58
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
2.1
Existence result in nite cylinders for a regularized problem
. . . .
63
2.2
Passage to the limit in the innite domains . . . . . . . . . . . . . .
69
2.3
Passage to the limit
2.4
Existence of a solution
(c1 , u1 )
2.5
(c, u)
4.3
c > c1
77
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
82
The minimal speed
for all
Existence of a positive minimal speed

Characterization of
Dependency of
74
. . . . . . . . . . . . . . . . . . . . . . . . .
4.2
. . . . . . . . . . . . . . . . . . . .
70
75
Monotonicity of the solutions
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . .
4.1
. . . . . . . . . . . . . . .
82
. . . . . . . . . . . . . . . . . . . . . . . . .
83
with respect to the nonlinearity
. . . . . . . . .
Partie II : Study of the premixed ame model with heat losses
87
95
Introduction and main results
. . . . . . . . . . . . . . . . . . . . . . . . .
96
Proof of the existence theorem . . . . . . . . . . . . . . . . . . . . . . . . .
98
2.1
Equivalence with a problem on
. . . . . . . . . . . . . . . . . .
98
2.2
Existence of solutions in bounded domains . . . . . . . . . . . . . .
99
2.3
Passage to the limit in
2.4
vi
48
Proof of theorem 2
R+
R+
. . . . . . . . . . . . . . . . . . . . . . . 109
. . . . . . . . . . . . . . . . . . . . . . . 110
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.1
An upper bound for
. . . . . . . . . . . . . . . . . . . . . . . . . 111
3.2
A lower bound for the unburnt gases
3.3
Upper bounds for the speed
. . . . . . . . . . . . . . . . 112
. . . . . . . . . . . . . . . . . . . . . 113
Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Partie III : Une premire approche de la modlisation en biologie

1
123
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Aspect pratique, site et collecte des donnes
Formulation d'un modle spcique . . . . . . . . . . . . . . . . . . . . . . 125

3.1
Hypothses comportementales utilises dans la construction du modle
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.2
Modlisation de l'volution pendant la priode de vol . . . . . . . . 127
3.3
Modlisation de la phase ovo-larvaire
Estimation des paramtres
Tests numriques
. . . . . . . . . . . . . . . . 125
. . . . . . . . . . . . . . . . 129
. . . . . . . . . . . . . . . . . . . . . . . . . . 129
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.1
Test 1 : Inuence de la dure de la priode d'mergence
5.2
Test 2 : Inuence du taux de diapause prolonge
5.3
Test 3 : Inuence des arbres isols
Discussion
. . . . . . 131
. . . . . . . . . . 131
. . . . . . . . . . . . . . . . . . 133
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.1
Validation du modle
. . . . . . . . . . . . . . . . . . . . . . . . . 133
6.2
Rsultats sur le comportement de l'insecte
6.3
Limites
6.4
Rsultats venir
6.5
Adaptation d'autres espces vivantes
. . . . . . . . . . . . . . 135
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
. . . . . . . . . . . . . . . . 137
Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Bibliographie
147
vii
Table des matires
viii
Introduction Gnrale
Introduction Gnrale
L'tude de problmes de raction-diusion non-linaires constitue l'objet de cette
thse. Elle se divise en trois parties distinctes ; la premire concerne l'analyse mathmatique de modles d'invasion biologique en environnement priodique fragment. La
deuxime tudie un modle mathmatique issu de la thorie de la combustion. La dernire partie, enn, aborde la modlisation de la dispersion d'un insecte invasif.
Le point commun entre ces trois parties est l'utilisation et l'tude d'quations aux
drives partielles en rapport la modlisation, et de la forme
u
= (A(x)u) V (x) u + f (x, u), t R+ , x RN .
t
Les premire et deuxime parties s'intressent ainsi plus spciquement aux solutions
front progressif . Les premire et troisime parties s'attachent des phnomnes d'invasion biologique modliss par ce type d'quation.
Les modles tudis dans la premire partie ont t rcemment introduits par Shigesada
et al.
[85] ; l'environnement y est prsuppos variable priodiquement, favorable
en certains endroits la survie des espces, et dfavorable en d'autres. Les quations de

raction-diusion traites auront donc des termes de diusion et de raction dpendant
priodiquement de l'espace.
Nous ferons dans un premier temps l'tude complte du problme stationnaire associ
cette quation, en donnant des conditions simples assurant l'existence et l'unicit des
solutions stationnaires, puis en tudiant l'inuence des htrognits du milieu, en particulier la rpartition et l'amplitude des zones plus ou moins favorables ou dfavorables,
sur les conditions d'existence de telles solutions.
Dans un second temps, nous nous intresserons l'existence de fronts progressifs
pulsatoires, solutions des quations de raction-diusion cites ci-dessus, dcrivant l'invasion d'un tat stationnaire
par un tat stationnaire
p,
en gnral non constant. Nous
tudierons galement l'inuence des htrognits sur la propagation des fronts.
Dans la deuxime partie, nous nous pencherons sur l'existence de ammes planes,
dans le cas d'une raction chimique exothermique non-rversible du type
A B,
avec
pertes de chaleur volumtriques. L'objectif sera donc de trouver des solutions du type
front progressif un systme de deux quations de raction-diusion. En utilisant des
outils topologiques, nous prouverons ici, sous certaines conditions, l'existence de deux
solutions distinctes, et donnerons des rsultats de non-existence ainsi que des estimations
sur la vitesse des ammes.
Dans la troisime partie, nous nous consacrerons la modlisation du processus d'extension spatio-temporelle d'une nouvelle espce d'insecte, rcemment introduite au MontVentoux (Sud-Est de la France) par le biais de graines de cdre du Liban. Les dplacements
de l'insecte au stade adulte seront modliss par une quation de raction-diusion. Le
recours des tests numriques nous permettra de mettre en vidence l'utilit de ce modle
dans l'approfondissement de la comprhension du processus d'invasion de l'insecte, et dans
la prvision de sa position pour les annes venir. Nous voquerons galement les possi-
1. Premire partie : Modles d'invasion biologique en environnement priodique fragment

bilits d'adaptation du logiciel ralis au cours de ce travail l'tude de la propagation
d'autres espces vivantes.
1 Premire partie : Modles d'invasion biologique en

environnement priodique fragment
Les quations de raction-diusion du type
ut = u + f (u)
sur
RN
(1)
ont t introduites la n des annes 30 dans des travaux de Fisher (1937) [43] et Kolmogorov, Petrovsky et Piskunov (1938) [68], pour des modles de gntique des populations.
Les non-linarits
considres alors taient du type

2
ses extensions (ex. : f (u) = u(1 u )).
f (u) = u(1 u)
(loi logistique) ou
Ces quations, avec de telles non-linarits, se retrouvent galement en cologie et en

combustion. Elles permettent de rendre compte de la propagation de fronts, solutions de
u(t, x) = U (xect), o e est la direction de propagation et c la vitesse d'invasion

de l'tat constant 0 par l'tat 1 (U () = 1 et U (+ = 0)). De tels fronts existent pour
p
toute vitesse c c = 2
f 0 (0) [5, 68]. De plus, toute solution positive qui, l'instant
initial, est non nulle et support compact, converge vers 1 en temps grand, en s'tendant
avec la vitesse asymptotique c .

la forme
Si le cas de l'quation homogne (1), o les coecients de diusion et de raction sont

constants en espace, a donn lieu de nombreuses tudes, la question de la propagation
de fronts dans le cas des extensions htrognes de (1) n'a t envisage que rcemment.
Ici les coecients de diusion et de raction varient avec les htrognits du milieu.
Dans ce cas, l'un des modles les plus simples est le patch model, dans lequel l'environnement est suppos homogne par morceaux, induisant une quation coecients
constants par morceaux. Ce type de modle a t introduit par Shigesada, Kawasaki et Teramoto [86] pour tudier des phnomnes d'invasion biologique dans des environnements
priodiques, comme dcrit dans le livre [85].
Nous nous intressons ici aux modles plus gnraux d'invasion biologique en environnement fragment, reprsents par les quations de raction-diusion
ut (A(x)u) = f (x, u), x RN
(2)
A(x) et le terme non linaire f (x, s) dpendent des variables

x = (x1 , , xN ) de faon priodique. Dans ces modles d'cologie proposs par Shigesada
o le champ de matrices
et al.
[85], l'environnement varie priodiquement comme soulign prcdemment : plus
fu (x, 0)
est ngatif, plus le milieu est dfavorable la survie des espces. L'exemple le
plus typique, utilis dans [66, 85, 86] est
f (x, u) = u((x) (x)u),
ou, plus simplement,
f (x, u) = u((x) u).

Ce travail a pour but l'analyse mathmatique rigoureuse et complte du problme (2)
et des questions de propagation dans un environnement gnral priodique. Il s'attache
en particulier l'inuence de la fragmentation et de l'amplitude des zones plus ou moins
favorables ou dfavorables sur les conditions de survie et de propagation d'une espce.
Introduction Gnrale
A notre connaissance, et mme dans le cas simpli du patch model, nous apportons
ici la premire preuve rigoureuse des rsultats noncs. Si d'autres auteurs ont obtenu
des rsultats concernant le problme (2), aucun en eet n'en a fait une tude complte.
En outre, l'intrt port au cas gnral de l'quation (2), o les coecients ne sont pas
ncessairement constants par morceaux, o
mension
fu (x, u)
peut changer de signe, et o la di-
est quelconque, constitue une spcicit de ce travail. Ainsi, l'inuence de la
position des zones favorables et dfavorables sur la survie des espces a t tudie dans
le cas born, et pour des fonctions continues par morceaux par Cantrell et Cosner [28] ;
l'existence de front progressifs pulsatoires a t montre, en dimension
1, et sans prouver
l'existence d'une vitesse minimale par Hudson et Zinner [56] ; enn, une formule pour la
vitesse minimale des fronts c peut tre dduite d'un travail de Weinberger [92], avec une
approche dirente, o il est notamment suppos
a priori
que les fronts sont croissants
en temps.
1.1
Inuence des htrognits du milieu sur la survie des espces
Dans ce chapitre, nous nous intressons l'existence et l'unicit des tats stationnaires de l'quation (2). L'tat
solutions
tant toujours un tat stationnaire, nous cherchons des
du problme
(A(x)p) = f (x, p)
p(x) > 0, x RN .
dans
RN ,
(3)
Sous certaines hypothses, nous prouvons que les solutions de (3) sont en fait priodiques
(L1 -priodicit en
x1 ,..., LN -priodicit
en
xN ).
Nous tablissons ici une condition nces-
saire et susante simple pour obtenir l'existence d'une solution de (3).

Nous dmontrons galement un nouveau rsultat d'unicit du type Liouville, d'un
intrt pouvant dpasser le cadre de ce travail.
En termes d'cologie, l'existence de solutions au problme (3) correspond la condition
de survie de l'espce. Nous obtenons en eet une description complte du comportement
asymptotique des solutions de l'quation (2), et prouvons, sous certaines hypothses, que
si une solution de (3) existe, alors les solutions de (2) convergent asymptotiquement vers
cette solution ; au contraire, si le problme (3) n'a pas de solution, on dmontre que les
solutions de (2) convergent vers
0,
et donc qu'il y a extinction de l'espce.
Ensuite, nous nous intressons au rle de la fragmentation de l'environnement. Ainsi,

dans des situations englobant le patch model, nous trouvons un rarrangement de l'environnement assurant toujours de meilleures chances de survie que l'environnement initial.
De plus, les mthodes introduites, utilisant dirents types de rarrangements, permettent
d'tendre les rsultats de Cantrell et Cosner [28] obtenus sur des domaines borns, des
dimensions suprieures et des non-linarits plus gnrales.
Nous nous penchons aussi sur les eets des grandes amplitudes de la non-linarit.
Prcisons que tous ces rsultats peuvent tre utiliss pour amliorer les chances de
survie d'une espce, mais peuvent galement tre vus en terme de lutte biologique.
1.1.1 Rsultats : Existence, unicit, comportement en temps long et eet des

htrognits
L'quation considre ici est donc (2), dont nous cherchons les solutions stationnaires,
qui vrient (3).
et le terme non linaire f (, s) sont supposs priodiques.
L1 , . . . , LN > 0, une fonction g dnie dans RN est dite
N
priodique si g(x + k) = g(x) pour tout x R et k L1 Z LN Z. Soit C la cellule
de priodicit C = (0, L1 ) (0, LN ).
On suppose qu'il existe > 0 tel que A(x) I pour tout x, au sens des matrices
2,
symtriques (I est la matrice identit). Enn, A est de classe C
( > 0), et f est de
1,
2
N
classe C
en (x, u), C en u. On suppose que f (x, 0) = 0 pour tout x R . Nous ferons
Le champ de matrices
Rappelons qu'tant donns
ventuellement les hypothses suivantes :
x RN , s 7 f (x, s)/s
est strictement dcroissante par rapport
s>0
(4)
et/ou
M 0, s M, x RN , f (x, s) 0.
Soit
(5)
(A(x) ) fu (x, 0) avec condilequel il existe une fonction , prio-
la valeur propre principale de l'oprateur
tions de priodicit, c'est--dire l'unique rel pour

2
dique, de classe C , vriant
(A(x)) fu (x, 0) = 1 ,
(x) > 0
dans
RN .
0 de (3) est instable si 1 < 0, et stable sinon. L'instabilit

fu (x, 0) 0, 6 0 (par exemple si f 0 (0) > 0 dans le cas o f ne dpend pas
Nous dirons que la solution

est vrie si
de
x),
mais cette condition n'est pas ncessaire.
Nos rsultats d'existence et d'unicit sont rsums par le thorme suivant :
Thorme 1 1) Supposons 1 < 0. Si f vrie (5), alors il existe une solution priodique
p de
(3) ; si f vrie (4), alors il existe au plus une solution borne de (3), qui est alors
priodique.
2) Si 1 0 et f vrie (4), alors il n'existe pas de solution borne de (3).
Le rsultat d'existence de la partie 1) se vrie pour des quations plus gnrales

comprenant ventuellement des termes de transport. Nous le dmontrons en adaptant la
mthode classique des sur- et sous-solutions.
Le rsultat d'unicit est de type Liouville et ne suppose pas a priori que les solutions
sont priodiques. La recherche d'une constante strictement positive minorant les solutions
y
constitue la partie dicile de la dmonstration. Pour le prouver, les solutions u du problme (3) translat dans la direction y sont compares aux fonctions propres principales
yR du linaris en 0 de ce problme sur une boule BR . Une srie de lemmes nous permet
y
de prouver que pour R assez grand, R est une sous-solution (avec une condition de nory
y
malisation bien choisie) du problme (3) translat. Ds lors, u (0) > R (0), et ensuite un
argument de priodicit permet de prouver que toute solution positive borne de (3) est
Introduction Gnrale
minore par une constante strictement positive. L'unicit est alors une consquence du
principe du maximum, et la priodicit une consquence de l'unicit.
Des conditions d'existence et d'unicit des solutions
en temps grand des solutions
u(t, x)
p de (3), dcoule le comportement

u0 (x) 0, u0 (x) 6 0,
de (2) avec donne initiale
borne et uniformment continue :
Thorme 2 Sous les hypothses (4) et (5), alors u(t, x) p(x) (unique solution borne
2
de (3)) dans Cloc
(RN ) quand t + si 1 < 0, et u(t, x) 0 uniformment en x quand
t + si 1 0.
Ainsi, sous les hypothses (4-5), la condition
1 < 0
est quivalente la survie des
espces (avec une concentration nale obtenue gale l'unique solution borne
1 0.
propre 1 , qui
de (3)),
tandis qu'il y a extinction si

Le signe de la valeur
ne dpend que de
et de
fu (, 0),
joue donc un
rle crucial, et nous prcisons dans la suite de cette section l'inuence des htrognits
du milieu sur la valeur de
1 ,
que nous notons
Thorme 3 1) On a 1 [] 1 [0 ], o
1 = 1 []
avec
(x) = fu (x, 0).
= 0 |C| et |C| est la mesure de Lebesgue

C
de C .
2) Si A = I , alors 1 [ ] 1 [], o est la symtrisation de Steiner priodique de
.
Z
3) Si 0 et maxRN > 0, alors 1 [B] < 0 pour tout B > 0, et 1 [B] est
C
strictement dcroissant par rapport B 0. Si < 0, alors 1 [B] > 0 pour B > 0
C
petit. Si maxRN > 0, alors 1 [B] < 0 au moins pour B > 0 susamment grand et
1 [B] est strictement dcroissant en B ds que 1 [B] < 0.
Ces trois rsultats sont prouvs en utilisant une caractrisation variationnelle de
1 [].
Leur originalit tient en partie l'utilisation d'une formule de B. Kawohl [65], nous permet-
||
tant d'armer que
| |2 ,
est la fonction propre principale associe
1 [].
A la lumire des thormes 2 et 3, ces rsultats ont une interprtation simple, mais
non triviale intuitivement, en termes de survie de l'espce. Ainsi, la partie 1) signie que,
moyenne gale, un taux de natalit htrogne (non constant en
x) favorise la survie des
espces, et ce quel que soit le type de diusion. Pour prciser le rsultat 2), nous devons
rappeler les notions de symtris priodique de Schwarz et de Steiner.
En dimension 1, le rarrangement priodique de Schwarz (x) est l'unique fonction
L-priodique sur R, symtrique par rapport la droite x = L/2, dcroissante sur [0, L/2],
puis croissante sur [L/2, L] et ayant la mme distribution que . En dimension N , on
6

appelle rarrangement priodique de Steiner, dans l'ordre
x 1 , , xN ,
la fonction
ob-
tenue en faisant successivement les rarrangements priodiques de Schwarz par rapport

chacune des variables
x 1 , x2 , . . . , x N .
La gure 1 ci-dessous donne un exemple de rarrangement priodique de Steiner (dans
) dans les directions
x puis y , et illustre le mode de regroupement des zones dfavorables.
Fig. 1 Les eets d'un rarrangement priodique de Steiner sur les zones dfavorables.
Rpartition des zones dfavorables : (a) pour
(x) ;
Le thorme 3, 2), montre que la conguration
(b) pour
(x).
(concentration des zones dfavo-
rables) donne plus de chances de survie l'espce. Cela implique galement que la rpartition optimale des zones est ncessairement stable par rarrangement de Steiner (quel que
soit l'ordre des variables dans lequel on fait le rarrangement). Cependant, comme nous
le soulignerons dans la section 4 de cette introduction, la recherche de cette rpartition
optimale est un problme ouvert.
D'aprs la partie 3), dans le cas d'un milieu qui n'est pas compltement dfavorable,
l'amplication susamment grande du taux de natalit est meilleure pour la survie des
espces.
Dans le cas des domaines borns, avec conditions de Dirichlet ou Neumann, nous
largissons les rsultats de Cantrell et Cosner [28], des non-linarits plus gnrales, et
des dimensions suprieures. Nous considrons en eet une non-linarit similaire celle
du cas priodique, et apportons la preuve, dans le cas Dirichlet, qu'un rarrangement de
Steiner de
donne toujours de meilleures chances de survie que
en termes de
1 .
Dans
le cas avec conditions de Neumann, on obtient un rsultat similaire au cas priodique, en

utilisant un rarrangement monotone de Steiner, faisant intervenir une ingalit dmontre
dans [12] (cf. partie I, chapitre 1, section 6).
Introduction Gnrale
1.2
Invasions biologiques et fronts progressifs pulsatoires
Dans ce second chapitre, nous examinons les phnomnes de propagation, et en particulier la propagation de fronts, associs l'quation (2).
Nous tablissons le lien entre les conditions de survie des espces (existence d'une
solution
p>0
de (3)), et les conditions de propagation de fronts progressifs pulsatoires
est envahi par l'tat htrogne p.
Nous prouvons en premier lieu l'existence d'une vitesse c telle qu'une solution du
type front progressif pulsatoire existe si et seulement si c c . Nous donnons ensuite une
formule variationnelle pour c , et analysons les eets de l'htrognit du milieu sur la

(voir dnition ci-aprs), o l'tat
vitesse de propagation. Nous dmontrons en particulier une dpendance croissante entre

la vitesse d'invasion et l'amplitude du taux de natalit.
1.2.1 Rsultats : Existence d'une vitesse minimale c , croissance des solutions, caractrisation de c et dpendance de c par rapport la nonlinarit f
Etant donns une solution
RN , on appelle front
eective c (6= 0), une solution
de (3) et un vecteur norm
pulsatoire se propageant dans la direction e avec la vitesse

u = u(t, x) (t R, x RN ) de (2) telle que
de

N
Y
ke
N
k
Li Z, x R , u t
, x = u(t, x + k),
c
i=1
u(t, x) 0, u(t, x) p(x) 0,

xe+
xe
o les limites sont prises localement en
t,
et uniformment dans les directions orthogo-
e. Cette notion de front pulsatoire gnralise

u(t, x) = U (x e ct) pour l'quation homogne (1).
nales
Thorme 4 Supposons que
(6)
et que 1 < 0. Soit p l'unique solution de

(3). Il existe c > 0 tel que l'quation (2) avec les conditions (6) a une solution (c, u) si
et seulement si c c . De plus, toute solution u est strictement croissante en t. Enn, c
est donne par c = min>0 k()/, o k() est la valeur propre de l'oprateur L dni
par
f vrie
celle de front progressif du type
(4-5)
L = (A(x)) 2 eA(x) + [ (A(x)e) + 2 eA(x)e + fu (x, 0)]
avec conditions de priodicit.

Pour prouver le rsultat d'existence, nous nous sommes inspirs des mthodes de H. Berestycki et F. Hamel [13]. Cependant, ici, la non-linarit
n'est plus ncessairement
positive, ce qui induit une nouvelle dicult. La mthode de rgularisation et d'approximation des non-linarits par cut-o ne pouvant plus s'appliquer, nous proposons une
autre mthode. Considrant des problmes rgulariss dans des domaines borns, nous
2. Deuxime partie : Etude d'un modle de combustion avec pertes de chaleur

prouvons cette fois directement que l'ensemble des vitesses pour lesquelles une solution
existe est un intervalle, et c est dni comme la borne infrieure de cet intervalle. Dans
la dmonstration, nous faisons notamment plusieurs fois appel une proprit sur le signe
de l'nergie de l'tat stationnaire positif
p, solution de (3). Cette proprit est tablie dans
le chapitre 1, au cours de l'tude du problme stationnaire (3).
Sous les hypothses du thorme 4, c ne dpend que de A, e et

sons que
et
B>0
fu (x, 0).
Nous suppo-
A et e sont xs, et que fu (x, 0) = (x), o est une fonction priodique donne
un paramtre rel. Notons c = c [].
Thorme 5 ZSupposons que f vrie (4-5) et que A est constante.
1) Si m = / |C| > 0, alors c [] c [m] = 2 eAe m.

C
2) Si m 0 et M = maxRN > 0, alors c [B] est strictement croissante en B > 0.
De plus,
2 eAe B m c [B] 2 eAe B M pour tout B > 0,
1
c [B]
c [B]
eAe M lim inf
lim sup
2 eAe M ,
B+
2
B
B+
B
et c [B]/ B 2 eAe m quand B 0+ .

Pour tablir ces rsultats, on utilise la caractrisation de
c = min
>0
k (B)
suivante :
k (B)
,
(7)
est la valeur propre principale de l'oprateur
LB, = (A) 2Ae 2 eAe B(x),

avec conditions de priodicit. En eet, cette caractrisation est quivalente la formule
obtenue dans le thorme 4 (cf. [13] et [93]). Les rsultats noncs dans le thorme 5
dcoulent des proprits de
k (B).
D'aprs ce qui vient d'tre tabli, moyenne constante, la prsence d'htrognits

dans l'environnement (taux de natalit non constant) augmente ainsi la vitesse des fronts
par rapport au cas homogne. De plus, cette vitesse augmente avec l'amplitude
taux de natalit et se comporte en
pour
du
grand et petit. Lorsque le milieu est
trs direnci (B grand), la vitesse des fronts devient de plus en plus grande, mme si
l'environnement est ainsi parfois trs dfavorable certains endroits .
2 Deuxime partie : Etude d'un modle de combustion

avec pertes de chaleur
Cette partie concerne l'existence de solutions du type front progressif pour un modle
de combustion. Rappelons que les problmes de ammes planes se propageant dans des
Introduction Gnrale
gaz prmlangs sont dcrits par des systmes scalaires d'quations de raction-diusion
non-linaires couples. La plupart des aspects de la propagation de ammes peuvent
tre tudis dans le cas du modle simpli d'une raction chimique exothermique nonrversible du type
A B.
Dans ce cas, le problme associ se rduit un systme de
deux quations de raction-diusion couples, o les inconnues sont i) la temprature

du mlange, ii) la concentration
en ractant et iii) la vitesse
de la amme. Ce type de
problme, considr dans un cylindre inni, s'crit :
u00 + cu0 = f (u, v)

v 00 + cv 0 = g(u, v)
avec des conditions aux limites en
sur
(8)
R,
Il faut souligner qu'une temprature d'ignition
peut exister (par exemple pour certains hydrocarbures), de telle sorte, que pour des tempratures trop basses, la raction n'a pas lieu (ce qui impose une condition sur
f (, v)
et
g(, v)).
De nombreux travaux se sont intresss au systme (8), avec temprature d'ignition,
en particulier dans le cas adiabatique, o les pertes de chaleur sont ngliges [23], [46],
[72], [26]. Ainsi, avec
g(u, v) = f (u, v),
certains calculs se trouvaient simplis.
Nous nous intressons au cas avec pertes de chaleur ; le systme de raction-diusion

associ est alors
u00 + cu0 = f (u, v) h(u)

v 00 + cv 0 = f (u, v)
sur
(9)
R,
avec les conditions aux limites
u() = 0, u(+) = 0,
v() = 1, v 0 (+) = 0.
(10)
Dans ce cas, Zeldovitch a prouv, dans des travaux prcurseurs [95] (1941), et en utilisant
des mthodes asymptotiques, que ce systme avait deux solutions du type
progres front
sif, dans le cas limite des hautes nergies d'activation (f (u, v)
0).
1
exp
2
u1
(u)v ,
Plus rcemment, Glangetas et Roquejore, dans [47], ont retrouv ce rsultat
comme une consquence d'une formule obtenue par Joulin et Clavin [60], [61], qu'ils ont
leur tour rigoureusement prouve dans [47]. Des non-linarits plus gnrales ont t
tudies par Giovangigli dans [46], o il dmontre l'existence d'une solution pour une
vitesse
xe (considrant le paramtre de pertes de chaleur
problme), avec un nombre de Lewis
Le = 1
(remarquons que
comme une inconnue du
Le = 1/).
Ici, nous res-
tons dans le cadre des articles de Berestycki, Nicolaenko et Scheurer [23], en conservant
comme une inconnue du problme, et tablissons l'existence de deux solutions distinctes

pour les petites valeurs de
du nombre de Lewis
Le,
De plus, nous prouvons ce rsultat quelle que soit la valeur
et obtenons ainsi un rsultat de non-unicit.
En eet, Marion [72] a montr l'unicit dans le cas adiabatique, pour des nombres de
Lewis plus grands que
1,
et Bonnet, quant lui, a mis en vidence, dans [26], des cas o
l'unicit ne se vrie pas pour des nombres de Lewis plus petits que
1.
Nous dmontrons
ici qu'il n'y a jamais unicit dans le cas non-adiabatique (pour de faibles pertes de chaleur
).
10
2. Deuxime partie : Etude d'un modle de combustion avec pertes de chaleur

De plus, nous prouvons, dans certains cas, que la solution avec la plus grande vitesse
de propagation (sur les deux solutions obtenues) converge vers la solution du problme
adiabatique quand
0,
0.
alors que l'autre vitesse converge vers
Nous tablissons galement de nouvelles bornes pour les solutions du problme (910). Dans ses travaux, Giovangigli avait prouv (dans [46]) avec
tait borne par la vitesse
pour
Le 1
cad
Le = 1,
Le = 0.4
(rappelons que
pour l'hydrogne),
est infrieure la vitesse
solution d'un problme adiabatique scalaire. De plus, pour tout

une majoration explicite de
que la vitesse
solution du problme adiabatique. Ici, nous obtenons que,
c,
qui ne dpend pas de
Le > 0,
nous donnons
Nous prouvons galement que la
raction n'est pas totale, et calculons une borne infrieure pour la quantit de ractant
non brle.
2.1
Rsultats : Existence de deux solutions, estimations sur les

solutions
Les hypothses sur la non-linarit
sont les suivantes : il existe deux fonctions
et
telles que
f (u, v) = p(u)g(v)
o la fonction
tion
est Lipschitzienne sur
R,
sur
R R,
(11)
croissante, et admet une temprature d'igni-
(0, 1)
et la fonction
tq.
p(x) = 0
si
et
p(x) > 0
est continue, strictement croissante sur
g<0
sur
et
R+
si
x > ,
(12)
et telle que
g(0) = 0;
(13)
De plus, nous sommes amens faire une hypothse technique sur la fonction g , vrie
n
par exemple pour des fonctions du type g(y) = y , avec n > 0, ou encore pour des
ieme
fonctions g , n fois drivables, et telles que la n
drive en 0 soit non nulle (n 1).
1
Nous supposons que la fonction h est borne dans C , strictement croissante, et satisfait
h(0) = 0.
Soient (uad , vad , cad )
les solutions du problme sans pertes de chaleur ((9) avec
voir [23] pour leur existence), et
(us , cs )
= 0,
l'unique solution (voir [23]) du problme adiaba-
tique suivant
u00s + cs u0s = f (us , 1 us ),
(14)
avec les conditions aux limites
us () = 0, us (+) = 1.
(15)
Nous dmontrons alors que
Thorme 6 1) Pour tout
> 0, et pour > 0 assez petit, il existe deux solutions

distinctes et non triviales (u1 , v1 , c1 ) et (u2 , v2 , c2 ) au problme (9 -10), avec c1 < c2 .
2)Pour tout > 0, c1 0 quand 0. De plus, si 1 et g(y) = y , (u2 , v2 , c2 )
converge sur tout compact vers (uad , vad , cad ) quand 0. Le mme rsultat est vrai pour
le terme de raction plus gnral f (x, y) dans le cas = 1.
11
Introduction Gnrale
Pour obtenir le rsultat d'existence, nous tudions d'abord le problme sur un ouvert
born, an d'utiliser, comme dans [23], [46] et [72] un argument de degr topologique
de Leray-Schauder (voir [79]). Cependant, aprs avoir fait les estimations
a priori
et
avoir calcul le degr, nous constatons que ce dernier est nul ( cause du nombre pair
de solutions), ce qui a conduit Giovangigli changer le rle de
et de
Pour pouvoir,
malgr tout, prouver l'existence de solutions, on doit donc obtenir une nouvelle estimation
a priori, nous permettant de sparer deux types de solutions, certaines avec une vitesse c
proche de
0,
et d'autres avec une vitesse
faisant tendre
des solutions
plus grande. Cette estimation est obtenue en
vers 0. A la dicult mentionne ci-dessus s'ajoute celle de la non-unicit

avec = 0 pour certaines valeurs de (cf. [26]), qui est rsolue par une
succession de lemmes techniques. Suite au calcul du degr sur deux ouverts distincts, nous
obtenons le rsultat, aprs passage la limite. La partie 2) du thorme 6 se dduit des
calculs eectus pour obtenir le rsultat d'existence, et des rsultats d'unicit de Marion
[72].
Les solutions obtenues ci-dessus ne sont pas ncessairement les seules, et nous n'avons
. Les rsultats prsents dans

(u, v, c) du problme (9 -10).
prouv leur existence que dans le cas de petites valeurs de

le thorme suivant sont vrais pour toutes les solutions
f (1, 1)
, le problme (9-10) n'a pas de solution.
h()
2) Si g est Klipschitzienne et si (u, v, c) n'est pas triviale, on a

Kp(1)
v(+) > exp
.
h()
Thorme 7 1) Si >
3) Pour tout 1, 0 < c cs , o (us , cs ) est la solution de (14-15).

f (s, 1 s)
et 2 = max f (1 min{1, }s, s), alors
4) Soit 1 = max
s
s(,1)
s[0,1]
p
0 < c < 2 1 pour tout 1 et 0 < c <
2
pour tout > 0.
Le premier rsultat est donc un rsultat de non-existence, dj donn par Giovangigli
[46] dans le cas = 1. Il laisse supposer l'existence d'un critique tel que, pour tout
< , il existe deux solutions, et tel que, pour tout > , il n'y a pas de solution.
Nous reviendrons sur cette question dans notre section 4.
Le deuxime rsultat donne une borne infrieure pour les gaz non brls. Nous avons d
supposer que la fonction
nous permet d'tablir que

par
v,
tait lipschitzienne ; en eet, le thorme de Cauchy-Lipschitz
v(+) 6= 0 ;
nous divisons ensuite par
l'quation satisfaite
et l'intgrons par parties pour obtenir le rsultat.
Le troisime rsultat et la premire estimation du 4) sont obtenus en s'inspirant de

calculs issus de [23], [72] ou encore d'un principe de comparaison prouv dans [20]. La
fonction
n'tant plus croissante dans le cas non-adiabatique, les mmes mthodes ne
peuvent pas s'appliquer directement. Nous contournons cette dicult en faisant l'analogie
12
3. Troisime partie : Modlisation de la dispersion d'un insecte invasif

u du cas adiabatique et la fonction 1v dans le cas avec pertes de chaleur,
0
1
introduisant la fonction j(y) = v v
(1 y).
entre la fonction
et en
Notons que la deuxime estimation du 4) s'obtient trs facilement par un principe du

maximum.
3 Troisime partie : Modlisation de la dispersion d'un

insecte invasif
Ce volet correspond la partie la plus applique de cette thse. Il est en eet consacr
la ralisation d'un modle, la rsolution numrique du problme de raction-diusion
associ, et la conception d'un logiciel permettant d'utiliser ce modle. Il s'intgre dans
le cadre du projet Caractrisation des potentialits invasives d'un nouveau ravageur des
graines de cdre rcemment introduit au Mont-Ventoux, conclu entre l'INRA et le Ministre de l'Agriculture. Ce projet a dj donn lieu la publication d'un article biologique
[40].
L'insecte tudi ici est
Megastigmus schimitscheki
Novitzky. D'autres espces de
Me-
gastigmus, dj prsentes en France, ont fait l'objet de modles, mais sur des zones moins
tendues (tude de l'impact d'un Megastigmus sur la production de graines d'un verger,
[59]). Ici, nous nous intressons un insecte dont la date et le lieu d'introduction (MontVentoux, 1994) sont connus, ce qui facilite l'tude des processus de dispersion partir du
point source, au contraire d'espces plus anciennes et maintenant rparties sur l'ensemble
du territoire. En outre, l'insecte tudi tant invasif, il n'a pas de prdateur ; remarquons
que le cdre est lui-mme une essence introduite, et n'a de ce fait galement que trs peu
de prdateurs. Ces trois points ont beaucoup contribu la possibilit de mise en place
d'un modle simple et raliste.
Une tude antrieure [40], mene dans le cadre de ce projet, nous fournit des donnes
sur la biologie de l'insecte.
Grce la synthse des caractristiques de l'insecte, qui nous a permis d'obtenir une
formulation plus claire du problme, nous avons construit un modle divis en deux parties. Dans la premire, nous simulons la phase ovo-larvaire ; dans la seconde, nous calculons
la dispersion des adultes en utilisant une quation de raction-diusion :
u
(t, x) = Du(t, x) v V (x, t) u(t, x) + f (t, x) X(t, x)u,
t
(16)
t [0, Nj ] (o Nj est le nombre de jours total o des adultes sont prsents), et

x R2 . Le terme f (t, x) modlise l'mergence progressive des insectes, tandis que X(t, x)
pour
correspond au taux de mortalit quotidien pendant la priode de vol. Le coecient de

diusion
D est constant, alors que le terme de transport est variable en temps et en espace
(des modles pour d'autres types d'insectes se sont montrs ecaces avec des valeurs de
transport variables et un terme de diusion xe, [9]). Concernant la variation de
en
espace, nous tenons compte de la distance de la fort la plus proche. Nous supposons en
eet que l'insecte ne dispose pas de critres dcisionnels lui commandant de sortir de la
fort, qu'il n'y parvient que par un mcanisme de diusion alatoire, et s'lve ensuite pour
tre davantage soumis au vent, en fonction de son loignement de la fort. Pour prciser
13
Introduction Gnrale
la force du vent laquelle est soumis l'insecte, nous utilisons une formule de [74], utilise
pour calculer la force du vent laquelle est soumise une graine pour une vgtation rase.
En eet, l'insecte s'levant susamment haut, nous pouvons supposer qu' l'extrieur de
la fort de cdre le milieu correspond plus une vgtation rase qu' une fort dense :
V (x, t) = Ve (x, t) min{ln[d(x, f oret)(e 1)/da + 1], 1},

o
Ve
est le vent enregistr et
d(x, f oret)
la distance jusqu' la fort de cdre la plus
proche.
Les modles de raction-diusion sont souvent utiliss en dynamique des populations,
en particulier dans les cas o l'on ne peut suivre les individus pour observer leurs mouvements (voir les livres de Turchin [89] et de Shigesada & Kawasaki [85]).
Nous rsolvons cette quation en utilisant une mthode de directions alternes ( pas
fractionnaires), propose par Godunov [7] et Yanenko [94] pour l'quation de la chaleur.
Nous estimons ensuite les paramtres
et
v .
Pour
D,
nous utilisons une formule
donne dans le livre de Shigesada et Kawasaki [85], nous permettant de dduire la valeur
de la diusion partir des rsultats exprimentaux dont nous disposons. Pour le calcul
de
v ,
nous utilisons une mthode gnrale dcrite par Turchin (d'aprs des travaux de
Banks et al. [8]). Cette tape consiste simplement, aprs avoir fait des hypothses initiales
sur la valeur du paramtre estimer, calculer les prdictions du modle, et optimiser
la valeur du paramtre pour ajuster ces prdictions aux valeurs exprimentales.
Nous eectuons ensuite dirents tests numriques pour vrier la validit du modle et
quantier l'inuence de certains traits adaptatifs et de certaines congurations de terrain
sur la dispersion de l'insecte. Les simulations rendent compte de rsultats proches des
mesures eectues sur le terrain. De plus, d'un point de vue qualitatif, le comportement
de l'insecte, tel que dcrit par ces tests, est cohrent avec les attentes des biologistes.
Ces calculs numriques nous ont galement permis de mieux connatre le rle la diapause
prolonge (mergence en retard d'un ou deux ans), et d'avoir une meilleure estimation de
son taux. Ils nous ont en outre donn des informations sur l'inuence de la longueur de la
priode d'mergence ou encore la prsence d'arbres isols sur la dispersion des insectes.
Nous voquons nalement les limites de ce type de modle dans le cas o les phnomnes mtorologiques ont une inuence majeure sur la dispersion de l'insecte. En eet,
l'impossibilit de les prvoir d'une anne sur l'autre rduit les possibilits de prvisions
de la dispersion long terme. C'est dans cette optique que nous avons dvelopp un lor
giciel (avec le langage de programmation Matlab ) s'adaptant d'autres congurations
de terrain et d'autres espces vivantes, dont on pourrait prdire l'volution sur de nombreuses annes de faon plus able que dans le cas trait ici. Nous pensons notamment
des espces dpourvues d'ailes, ou pourvues d'ailes susamment puissantes pour que
l'inuence du vent sur leur mouvement soit ngligeable. Le terme de transport serait alors
moins variable en temps (bassin d'attraction par exemple, ou pente pour des vgtaux).
4 Problmes ouverts et perspectives

Partie 1 :
nous avons prouv qu'un rarrangement priodique de Steiner des zones
favorables et dfavorables d'un environnement priodique orait toujours de meilleures
14
4. Problmes ouverts et perspectives

chances de survie une espce que la conguration initiale. Cela prouve que la conguration optimale de l'environnement en termes de survie est stable par symtrie de Steiner.
La recherche d'une telle conguration reste cependant un problme ouvert.
Partie 2 :
nous avons dmontr l'existence de deux solutions front progressif pour
de faibles valeurs du paramtre de pertes de chaleur
, et leur nonexitence pour les grandes
valeurs de ce paramtre. Un problme ouvert naturel est donc de dterminer une valeur
critique du paramtre de pertes de chaleur, tel que le problme considr admette
une solution si et seulement si .
Partie 3 :
une adaptation du modle construit est prvue, dans le cadre d'un autre
projet avec l'INRA, la modlisation de l'expansion spatiale d'insectes forestiers en relation avec les changements climatiques.
15
Introduction Gnrale
16
Partie I : Analysis of the periodically

fragmented environment model
17
Partie I : Analysis of the periodically fragmented environment model
18
Note au CRAS
19
Note au CRAS
20
21
Note au CRAS
22
23
Note au CRAS
24
25
Note au CRAS
26
Chapitre 1 : Inuence of periodic

heterogeneous environment on species
persistence
Ecrit en collaboration avec Henri Berestycki
et Franois Hamel
a EHESS, CAMS, 54 Boulevard Raspail, F-75006 Paris, France

b Universit Aix-Marseille III, LATP, Facult Saint-Jrme, Case cour A
Avenue Escadrille Normandie-Niemen, F-13397 Marseille Cedex 20, France
Sommaire
1
2
4
5
6
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Statements of the main results . . . . . . . . . . . . . . . . . 32
2.1
. . . . . . . . . . . . . . . . .
33
2.2
2.3
Eects of the heterogeneity on species survival
. . . . . . . . . . . . . . . . . . . . . . .
34
. . . . . . . . .
35
. . . . . . . . . . . . . . . . . . . . . . . . .
39
Existence and uniqueness of a stationary solution . . . . . . 38
3.1
Proof of existence
3.2
Proof of uniqueness
. . . . . . . . . . . . . . . . . . . . . . . .
41
3.3
Energy of stationary states . . . . . . . . . . . . . . . . . . . .
46
5.1
Inuence of the amplitude" of the reaction term . . . . . . . .
49
5.2
The evolution equation . . . . . . . . . . . . . . . . . . . . . . 47

Conservation of species in ecological systems . . . . . . . . . 48
fu (x, 0)
. . . . . . . . . . . . . . . .
50
The eect of fragmentation in bounded domain models . . . 53
Preprint 04-14 LATP 2004, soumis
J. Math. Biology.
27
Chapitre 1 : Inuence of periodic heterogeneous environment on species persistence
1 Introduction
In the last two decades, it has become increasingly clear that the spatial dimension and, in
particular, the interplay between environmental heterogeneity and individual movement, is an
extremely important aspect of ecological dynamics.
P. Turchin, Qualitative Analysis of Movement
Reaction-diusion equations of the type
ut = u + f (u)
in
RN
(1)
have been introduced in the celebrated articles of Fisher (1937) [43] and Kolmogorov, Petrovsky
and Piskunov (1938) [68]. The initial motivation came from population genetics and the scope
was to shed light on spatial spreading of advantageous genetic features. The nonlinear reaction
term considered there are that of a logistic law of which the archetype is
extensions like
f (u) = u(1
f (u) = u(1 u)
or
u2 ).
Several years later, Skellam (1951) [87] used this type of models to study biological invasions,
i.e. spatial propagation of species. With these he succeeded to propose quantitative explanations
of observations, in particular of muskrats spreading throughout Europe at the beginning of 20th
Century or the early dissemination of birch trees in Great Britain.
Since these pioneering works, this type of equation has been widely used to model spatial
propagation or spreading of biological species (bacteria, epidemiological agents, insects, plants,
etc). Systems involving this type of equations have also been proposed to model the spread of
human cultures (Compare in particular Cavalli-Sforza et al. [3, 32]).
A vast mathematical litterature has been devoted to the homogeneous equation (1). Of particular interest is to understand the structure of travelling front solutions and their stability, as well
as propagation or spreading properties that this equation exhibits. The former are solutions of
u(t, x) = U (x e ct) for any given direction e (|e| = 1, e is the direction of propagation)
and U : R (0, 1). The latter are related to the fact that starting with an initial datum u0 0,
u0 6 0 which vanishes outside some compact set, then u(t, x) 1 as t + and the set where
u is, say, close to 1 expands at a certain speed which is the asymptotic speed of spreading. The
the type
papers of Aronson and Weinberger [5], and of Fife and McLeod [42], in particular, represent two
mathematical milestones in the literature.
With the important exception of the work of Grtner and Freidlin (1979)[44] on the asymptotic speed of spreading, it is only relatively recently that the questions of travelling fronts,
propagation and spreading have been addressed within the framework of heterogeneous extensions of (1) (see e.g. [13, 56, 85, 93]).
In ecological modelling or for biological invasions, indeed, the heterogeneous character of the
environment plays an essential role. It appears that even at macroscopic scales, the medium
and its various characteristics are far from homogeneous. In the words of Kinezaki et al. [66] :
... natural environments are generally heterogeneous. For example, they are usually mosaic of
heterogeneous habitats such as forests, plains, marshes and so on. Furthermore, they are often
fragmented by natural or articial barriers like rivers, cultivated elds and roads, etc. Thus
growing attention has been paid in recent years to the question of how such environmental
1 1998,
28
Sinauer Assoc. Inc., Sunderland, Mass.
1. Introduction
fragmentations inuence the spreading and persistence of invading species". For recent works on
this aspect of ecological modelling, we refer the reader for instance to [54, 57, 62] ; compare also
the references on this model quoted in [66].
A rst approximation to heterogeneous environments, therefore, is the so-called patch model.
In it, one assumes a mosaic of dierentiated environments, each of which having a relatively
well dened structure which one might consider as homogeneous. This involves an equation
with piecewise constant coecients (compare below). This type of model has been proposed by
Shigesada, Kawasaki and Teramoto [86] to study biological invasions in periodic environments
and is described in the book [85]. In the recent years a vast literature has been devoted to the
study of the role of fragmentation of the environment on species survival. It is concerned mostly
with space dimenson
1,
and with bounded domains, which is dierent from the point of view
that one has in studying spreading (see i particular [27, 28, 29].
More generally, a periodic heterogeneous model is proposed to investigate the eect of heterogeneity of the environment for more general periodic frameworks. Equation (1) is generalized
to :
ut (A(x)u) = f (x, u), x RN ,

where the diusion matrix
x = (x1 , , xN )
A(x)
and the reaction term
f (x, u)
(2)
now depend on the variables
in a periodic fashion.
The typical example considered in [66, 85, 86] is
f (x, u) = u((x) (x)u),
(3)
f (x, u) = u((x) u).
(4)
or even, simply,
Here,
(x)
can be interpreted as an eective birth rate of the population and
(x)
reects a
saturation eect related to competition for resources.

Thus, regions of space where
(x)
is relatively high represent favourable zones where the
species can develop well. On the contrary, low

will consider cases where
(x)
(x) regions are less favourable to the species. We
can actually become negative. In such a region, if isolated from
other regions, the species would actually die out.

The way the diusion matrix
A(x)
depends on
in more or less favourable environments
varies from one case to another. As pointed out by Shigesada and Kawasaki [85], certain species,
upon arriving in unfavourable environments, speed up (meaning, say in one dimension, that
increases) while the progression of others is hindered (meaning that
A(x)
A(x)
decreases).
The periodic patch model considered by Shigesada and Kawasaki is a particular important
case of this periodic framework. As was already indicated, it consists in considering that the
environment is made up of patches each of which is homogeneous arranged in a periodic
fashion. For instance, in one dimension, with period
and

(x) =
and
in
in
in the patch model, one assumes that
(x)
and
A(x)
E + [0, L],
E = [0, L]\E + ,
a+
a
in
are extended periodically to
R.

A(x) =
Then,
L,
are piecewise constant, that is
in
E+,
E.
Here,
+ , , a+
and
are constants.
29

Saying that the environment
E+
is more favourable than
means that
+ > .
As mentioned before, there is no reason to have a general inequality between
a+
and
a .
The case of the periodic patch model above, in one dimension, is discussed in [86], relying on
numerical computations and heuristic arguments.
One can also formulate similar models in higher dimensions. For instance, in dimension two,
C = [0, L1 ] [0, L2 ] (L1 , L2 > 0) then one considers the same denitions as above for problem
+
(2) with A(x1 , x2 ) and f (x1 , x2 , u) being L1 -periodic in x1 and L2 -periodic in x2 , with (E , E )
2
being a partition of C which is extended periodically in R .
if
In this paper, we address the general case of equation (2) (not necessarily piecewise constant)
and in higher dimensions as well. The aim of the present work is twofold : (i) to give a complete
and rigorous mathematical treatment of these questions (the results are new even in the one
dimensional case and for the patch model), and (ii) to discuss these types of problems in the
framework of a general periodic environment and in higher dimensions as well. In particular,
we address here from a mathematical standpoint the question of environmental fragmentation,
which is an important issue in ecology. As far as we know, even in the simplied case of the
periodic patch model, the results are proved rigorously here for the rst time. We introduce here
the method of rearrangement. As we will see in the last section, it also allows us to simplify and
generalize the known results for bounded domains.
The present paper is the rst one in a series of two. Here we are chiey concerned with
discussing the existence of a stationary state of (2), that is a positive solution
(A(x)p) = f (x, p)
p(x) > 0, x RN .
in
p(x)
of
RN ,
(5)
Under some assumptions which will be made precise later, solutions of (5) may turn out to
be periodic. But the periodicity assumptions will not always be made a priori. Periodicity is
understood here to mean
f (x, u)
L1 -periodicity
x).
in
x1 ,..., LN -periodicity
in
xN
(assuming that
A(x)
and
have such a dependance in
In problems (2) and (5), it is not easy to understand the complex interaction between more
favourable and less favourable zones. Furthermore, how does the balance between diusion and
reaction play a role ? It is not obvious a priori and it may actually sometimes be counter-intuitive.
We establish here a simple necessary and sucient condition for such a solution of (5) to exist.
This criterion is related to existing results in the literature. In the bounded domains case, with
various boundary conditions, it is discussed in several papers of Cantrell and Cosner (see Section
6). In the unbounded domains case, with
f (x, s) = (x)s (x)s2 ,
the sucient condition
for existence follows from results of Pinsky [77] and Englnder, Kyprianou [34]. These authors
actually consider more general operators, not necessarily in the periodic framework (see Remark
1 below ).
For general equation (5), we prove some new Liouville type results of independent interest.
These results will be stated in the next section.
In the ecological context, existence of a solution of (5) should be viewed as a condition
allowing for the survival of the species under consideration. In fact more precisely, we obtain
a complete description of the asymptotic behaviour of solutions of the evolution equation (2).
Under some assumptions which are made precise in section 2, we prove that when a solution
2 We
30
would like to thank A. E. Kyprianou for pointing out these results to us.
1. Introduction
p(x) of (5) exists, then, asymptotically, at every x, the solution u(t, x) converges to that p(x)
as t +. On the contrary, as soon as p(x) ceases to exist, there is extinction, meaning that
u(t, x) 0 as t + at every point.
Thus, within the framework of model (2), we obtain a necessary and sucient condition for
the survival of this species in a periodic environment.
Having established this criterion, we proceed to discuss various conditions under which the
species survives. In particular, we treat the question of the role of fragmentation of the environment. It has been shown in [85], for the patch model in one dimension on the basis of numerical
computations, that, everything else equal, having small unfavourable zones leaves less chances
for survival than having one large zone (with the same total surface). This is a remarkable discovery in this model which sets on a rm theoretical ground the adverse eect of environment
fragmentation. A good example to see this is to wonder whether several roads accross some forest
are better or worse for species survival than one large road with the same total width. Results of
the same kind have been established but in the framework of bounded intervals with boundary
conditions, in a series of papers by Cantrell and Cosner, who have also further discussed systems
in [29].
In this paper, we prove this result rigorously. Further, we actually derive a much more general
k types of
k = 2 as before), we derive the optimal arrangement of these zones in
result. For instance, a consequence of our nding is that assuming a patch model with
habitat (including the case
order to allow for species survival. We establish a result in higher dimensions as well and prove
that symmetric connected patches are more favourable than disconnected domains. The optimal
shape, however, is not known in higher dimensions and this leads to interesting open problems.
Actually, the methods which we introduce in this paper using various rearrangements also allow
us to extend the results of Cantrell and Cosner in the framework of bounded domains to more
general nonlinearities. Precise results are discussed in Section 6.
Further, we analyze the eects of high amplitude. One of the results we establish here is that
it suces to have a very favourable (even quite narrow) zone to allow for species survival, no
matter how bad the environment may be elsewhere.
All these properties bear consequences for species survival but also shed light on conditions
needed to eradicate invading biological species.
In a forthcoming paper [18], we analyze the question of invasion for problems of the type
(2). More precisely, we connect the necessary and sucient condition for species survival to that
for propagation of pulsating fronts invading the uniform state
(see [85, 93] for the denition
and [13, 16, 56, 92, 93] for some related mathematical results). We further obtain a variational
formula for the minimal speed of propagation of such fronts.
The present paper is organized as follows. In the next section, we set the mathematical
framework and state all the main results. In Section 3, we prove uniqueness and existence results
for solutions of (5). In particular, there we establish a nonlinear Liouville theorem. Next, in
Section 4, we give some stability results concerning the long time behaviour of solutions of
problem (2) with initial data. In Section 5, we apply the general results to some special classes
of functions
arising in some biological models, and we state some species persistence results.
Lastly, Section 6 summarizes some facts on the bounded domains case with dierent types of
boundary conditions.
31
Statements of the main results

We are concerned here with equation
ut (A(x)u) = f (x, u), t R+ , x RN ,
(6)
and its stationary solutions given by
(A(x)u) = f (x, u), x RN .
(7)
L1 ,...,LN > 0 be N given real numbers. In the following, saying that a function g : RN
periodic means that g(x1 , , xk + Lk , , xN ) g(x1 , , xN ) for all k = 1, , N . Let
Let
is
R
C
be the period cell dened by
C = (0, L1 ) (0, LN ).
Let us now describe the precise assumptions. Throughout the paper, the diusion matrix
eld
A(x) = (aij (x))1i,jN
is assumed to be periodic, of class
C 1,
(with
> 0), and uniformly
elliptic, in the sense that
0 > 0, x RN , RN ,
aij (x)i j 0 ||2 .
(8)
1i,jN
f : RN R+ R is of class C 0, in x locally in u, locally Lipschitz-continuous

N
with respect to u, periodic with respect to x. Moreover, assume that f (x, 0) = 0 for all x R ,
1
N
that f is of class C in R [0, ] (with > 0), and set fu (x, 0) := lims0+ f (x, s)/s. Unless
The function
otherwise specied, the assumptions above are made throughout the paper. In Remark 2 below
we also explain how to include in our results the case of the patch model which involves terms
f (x, u)
and
A(x)
which are discontinuous with respect to
In several results below, the function
x.
is furthermore assumed to satisfy
x RN , s 7 f (x, s)/s
s>0
is decreasing in
(9)
and/or
M 0, s M, x RN , f (x, s) 0.
Examples of functions
u((x) (x)u)
(10)
satisfying (9-10) are functions of the type (3) or (4), namely
or simply
f (x, u) = u((x) u),
where
and
f (x, u) =
are periodic.
The criterion of existence (as well as uniqueness and asymptotic behaviour) is formulated
with the principal eigenvalue
of the operator
L0
dened by
L0 := (A(x)) fu (x, 0),

with periodicity conditions. Namely, we dene
exists a function
>0
Let us recall that
as the unique real number such that there
which satises
(A(x)) fu (x, 0) = 1
is periodic, > 0, kk = 1.
in
1 < 0.
The stationary state
is an unstable solution of (7)
is said to be stable otherwise, i.e. if
will be seen to be natural in view of the results we prove here.
32
(11)
is uniquely dened by (11).
With a slight abuse of denition, in the following we say that

if
RN ,
1 0.
These denitions
2. Statements of the main results

2.1
We are now ready to state the existence and uniqueness result on problem (7). Let us start
with the criterion for existence.
Theorem 1
1 < 0).
1) Assume that
satises (10) and that
is an unstable solution of (7) (that is
Then, there exists a positive and periodic solution
2) Assume that
of (7).
1 0).
is a stable solution of (7) (that is
there is no positive bounded solution of (7) (i.e.
Then
is the only nonnegative and bounded solution
of (7)).
Remark 1
For special nonlinearities
of the type
f (x, u) = u((x) (x)u),
where
and
may not be periodic anymore, and for a more general non-divergence elliptic operator like
(A(x)u)+B(x)u = f (x, u) with drift B , the above results have a probabilistic interpretation.
Such equations arise in the theory of branching processes. In this framework, and for nonlinearities
f (x, u) = u((x) (x)u),
Part 1) of Theorem 1 is due to Englnder, Pinsky [35] (see also
Englnder, Kyprianou [34] and Pinsky [77] and Remark 5 below). We are much grateful to
J. Englnder, A. Kyprianou and R.G. Pinsky for several useful comments about this literature
and on the relationship with the probabilistic point of view.
In the bounded domains case with Dirichlet, Neumann or Robin boundary conditions, the same
type of results as in Theorem 1 can be found in [70] (in dimension
1) or [27] (in higher dimensions,
with constant diusion). We refer to Section 6 for more details about the bounded domains case.
Remark 2
We have assumed that
f (x, u)
was (at least) continuous with respect to
fact, one can easily extend these results to more general classes of
and
u.
In
which cover, in particular,
the case of the patch model. The more general statement assumes the following :
f (x, s) is measurable in x and bounded,

n
(ii) f (x, s) 0 for all s M , a.e. x R ,
(i)
uniformly on compact sets of
(iii)There is some periodic bounded measurable function
(x)s
for all
L (Rn )
s [0, +),
such that
f (x, s)
RN ,
s R and a.e. x
> 0 there exists > 0,
(iv) For all
such that
f (x, s) (x)s s
for all
s [0, )
and a.e.
x RN
(v) assumption (9) is understood a.e.
x RN .
Notice that the eigenvalue problem (11) is still well dened. When
(x) = fu (x, 0)
type f (x, s) = (x)s
sarily must take

terms of the
s f (x, s)
Part 2) still holds good if, instead of (9), one only assumes that, for any
such that
f (x, s) fu (x, 0)s for

f is assumed to be
(9), the function
is
C1 ,
one neces-
L (RN ). Lastly, the asumptions are satised by nonlinear

(x)s2 , when , L (RN ).
all
such that
and (strictly) decreasing at least for some
> 0,
there is
>0
RN and
x
s . Part 2) also holds good if, instead of

f (x, s)/s is nonincreasing in s > 0 for all x RN ,
: we would like to thank R.G. Pinsky for pointing
out this fact.

In the following, for simplicity, we write the proofs under the more stringent assumptions of the
theorem, but the arguments are readily extended to handle this more general framework.
Next we state our uniqueness result.
Theorem 2
Assume that
0 is an unstable solution of (7) (that is 1 < 0).
Then, there exists at most one positive and bounded solution of (7). Furthermore, such a solution,
if any, is periodic with respect to
x. If 1 0
0.
and
satises (9), then there is no nonnegative
bounded solution of (7) other than
33
Remark 3
The last part of this theorem was already included in Theorem 1 above. We repeat
it here for the statement to cover both cases
1 < 0
and
1 0.
This theorem is a Liouville type result for problem (7). Notice that the solutions of (7) are not
a priori assumed to be periodic in
x.
The core part in Theorem 2 consists in proving that any
positive solution of (7) is actually bounded from below by a positive constant (see Proposition 2
below), which does not seem to be a straightforward property.
2.2
Let us now consider the parabolic equation (6), and let

condition
u(0, x) = u0 (x)
in
RN .
u(t, x) be a solution of (6), with initial

u(t, x) as t + is described in
The asymptotic behaviour of
the following theorem :
Theorem 3
Assume that f satises (9) and (10). Let u0 be an arbitrary bounded and continuous
RN such that u0 0, u0 6 0. Let u(t, x) be the solution of (6) with initial datum
u(0, x) = u0 (x).

2
N as
1) If 0 is an unstable solution of (7) (that is 1 < 0), then u(t, x) p(x) in Cloc R
t +, where p is the unique positive solution of (7) given by Theorems 1, part 1), and 2.
N as
2) If 0 is a stable solution of (7) (that is 1 0), then u(t, x) 0 uniformly in R
t +.
function in
Remark 4
u0 ,
um
is the (unique) minimal solution of (6)
in the sense that
u(t, x) =
where
u(t, x)
In the above statement, the solution
with initial condition
solves (6) with initial condition
lim um (t, x),
m+
u0,m ,
and the family
(u0,m )mN
is any given nonde-
u0
RN . Actually, each um (t, x) is itself the limit of um,n (t, x) as n +, where
creasing sequence of nonnegative, smooth, compactly supported functions which converge to

locally uniformly in
um,n
solves
(um,n )t (A(x)um,n ) = f (x, um,n ), t > 0, x Bn

um,n (t, x) = 0 for t > 0 and x Bn ,
n (for n in N large so that Bn
p
contains the support of u0,m ). For nonlinearities of the type f (x, s) = (x)s (x)s with p > 1,
where and are periodic and inf RN > 0, it follows from Theorem 2 of Englnder and Pinsky
[36] that this minimal solution u is the unique solution of (6) with initial condition u0 . Without

where
Bn
u0,m
in
Bn
and boundary condition
is the open euclidean ball with center
and radius
the periodicity assumption and with more general non-selfadjoint operators with unbounded
coecients, the subtle question of the uniqueness or nonuniqueness of the solutions of (6) with
a given initial condition
u0
is discussed in [36].
Let us now consider the particular case

with respect to
x.
f (x, u) = u((x) u)
for
u 0,
where
is periodic
Such nonlinearities arise in ecological models of species conservation and
biological invasions (see section 1 for the motivation and [18, 56] for propagation phenomena
related to these equations). Such a function
f (x, u) = u((x) u) fullls conditions (9) and (10).
Following Remark 2, we may actually relax the regularity assumptions.

Gathering all the previous results, the following corollary holds :
34
Corollary 1
Let
u0 0, u0 6 0
f (x, u) = u((x) u)
u 0,
where
L (RN ) and periodic.

N and let u(t, x) be
in R
is in
be a bounded and uniformly continuous function
solution of (6) with initial datum
2.3
Let
the
u(0, x) = u0 (x).
1 < 0) then there exists a unique

positive solution p of (7), and u(t, x) p(x) locally in x as t +.
2) If 0 is a stable solution of (7) (that is 1 0), then 0 is the unique nonnegative
N as t +.
solution of (7), and u(t, x) 0 uniformly in R
1) If
for
is an unstable solution of (7) (that is
bounded
bounded
Eects of the heterogeneity on species survival
Let us denote by
1 []
the rst eigenvalue of (11) with
fu (x, 0) = (x).
From the previous
results, we see that, in this model, the survival of the species or its extinction hinge on the
sign of
1 [].
Furthermore, we show in [18] that this sign also determines biological invasions
in the form of travelling front-like solutions (actually pulsating travelling fronts). Hence it is of
particular interest to investigate how the various factors such as the shape of
of unfavourable zones or large amplitude oscillations in
(x),
(x), the distribution

1 []. The next
aect the sign of
series of results discuss these eects.
2.3.1 Distribution eects

Let us rst discuss the inuence of a heterogeneous function
as compared to the case where
Proposition 1
that is that
depends on
x,
would be constant with the same average.
Under the above assumptions, one has
1 [] 1 [0 ] = 0 ,
1
|C|
(0, LN ).
where
0 =
and
|C|
denotes the Lebesgue measure of the cell of periodicity
Let us now study the inuence of the repartition of

of
C = (0, L1 )
, assuming that the distribution function
is given.
Let us rst discuss the one-dimensional periodic patch model described in the introduction.
There, we assume that

(x) =
E+ R,
E = R\E+ .
in
in
Consider now another function (x) having the same distribution function as
but where the
unfavourable zone is an interval in any periodicity cell. That is, we set
(x) =
with
being
L-periodic,
as is
and
in
in
R,
E+
= R\E ,
E
+
(0, L)
E
is a (connected) interval. The question we
want to solve is to know which of the two congurations leaves most chances for survival.
Following the metaphor of Shigesada and Kawasaki [85], one can think of a forest in which a
periodic array of parallel roads are cut through. The forest is thought of as a favourable homogeneous medium and roads as an unfavourable homogeneous medium with a constant negative
reproduction rate
< 0
(or death rate
| |).
The question here is to know whether several
35

small forest roads, say of widths
l1 ,...,lp ,
in a given periodicity cell, are better in the sense of
species survival than one big road of width
l1 + + lp .
Relying on numerical calculations,
Shigesada and Kawasaki [85] have observed that the latter leaves more chances for species survival (see also Section 6 for a discussion on such results with other boundary conditions). This
remarkable nding illustrates the adverse eect of environment fragmentation on species survi-
val which is important in ecology. Incidentally, this is also relevant to eradication techniques of
certain harmful species.
Here, we actually prove this result rigorously.
Theorem 4
With the above notations, assuming that the diusion coecient
A(x)
is a constant
positive real number, one has
1 [ ] 1 [].
Therefore, whenever
survival while
allows for survival, so does
will allow for

1 [ ] < 0 < 1 [].
but in some cases,
will not. It is indeed simple to construct examples where
We actually prove a much more general result, in arbitrary dimension, and for general reaction
terms
f (x, u).
The previous proposition is a particular case of it.
To state our result, we need to introduce the notion of Schwarz and Steiner periodic symmetrizations of a function. For more details and properties about these notions, we refer the reader
to the monograph of B. Kawohl [65].
Consider a L-periodic function (x) dened on the real line R. There exists a unique function
(x), L-periodic on R, satisfying the following properties :
(i) is symmetric with respect to x = L/2 and is nondecreasing on [0, L] away from the
symmetry center L/2, i.e.
for all
(ii)
(x) (y)
has the same distribution function as

meas
for all real
x, y [0, L],
if
|x L/2| |y L/2|
that is :
{x (0, L); (x) } = meas {x (0, L); (x) }
This function
is called the Schwarz periodic rearrangement. An example of it is given in
Figure 1 below.
Fig. 1 A function
36
and its periodic Schwarz rearrangement

RN
C = (0, L1 ) (0, LN ).
Keeping xed all other variables but xk , we can rearrange as above the function (x) with respect
to xk . This is called Steiner periodic rearrangement in the variable xk . By performing such Steiner
periodic rearrangements successively on all variables x1 , x2 , . . . , xN , we obtain a new function,
(x1 , , xN ). Thus, this function is periodic in all variables, symmetric and decreasing.
Consider now a function
Theorem 5
(x)
periodic in
with a period cell
A is the identity matrix and denote by 1 [fu (x, 0)]

fu (x, 0). Let fu (, 0) be the successive Steiner symmex1 , . . . , xN .
Assume that the diusion matrix
the principal eigenvalue of (11) involving

trizations of
fu (, 0)
in the variables
Then,
1 [fu (, 0)] 1 [fu (, 0)].

As already pointed out, this covers the case of the patch model. Even in this case, in higher
dimension, say
N = 2, this property is new. An example of how unfavourable zones are assembled
by Steiner rearrangement is described in Figure 2 below.
Fig. 2 The eect of Steiner symmetrizaton on unfavourable zones. Distribution of
unfavorable zones : (a) for
x1
and
(x)
and (b) for rearranged
(x)
successively in the variables
x2 .
Theorem 5 shows that the
conguration leaves better survival chances. Hence, the more
the unfavourable zones are concentrated, the better the chances of survival of the species. Note
that the result of the succession of Steiner symmetrizations will depend on the order in which the
variables are taken. This result supports the adverse eect of fragmentation of the environment on
species persistence. It holds not only in the periodic patch model when
for an arbitrary function
(x)
(x) takes two values, but
(also one taking several values). Note, however, that, in the patch
model, for a given total area of unfavourable environment in one periodicity cell, the optimal
shape (i.e. the one that minimizes
1 []) is not known. It could actually be like in gure 3 below.
We pose as an open problem, to determine the optimal shape and to derive its properties.
2.3.2 Eects of the amplitude of the heterogeneity

The following result is concerned with the study of the inuence of the size of the nonlinearity
f.
To stress this eect, we now call
1 (f )
the rst eigenvalue of (11) with the nonlinearity
f.
37
Fig. 3 An example of a rearrangement which is Steiner symmetric in both variables but
is not obtained by the procedure described here.
Consider the problem
(A(x)u) = Bf (x, u)
where
B > 0
is a given positive real number and
in
RN ,
(12)
satises assumptions (9) and (10). As it
follows from Theorem 1 and 2, this problem admits a positive periodic solution if and only if
an unstable solution of (12). Let us examine the eect of the amplitude factor
theorem below holds for general functions
Theorem 6
every
1) If
fu (x, 0) < 0,
The following
fu (x, 0) > 0, or if
fu (x, 0) = 0 and fu (x, 0)
C
C
>Z 0, and the function B 7 1 (Bf ) is decreasing in B 0.
2) If
B.
0 is
then
1 (Bf ) > 0
for all
B > 0
6 0,
then
1 (Bf ) < 0
for
small enough. Assume that there
x0 C such that fu (x0 , 0) > 0. Then 1 (Bf ) < 0

B 7 1 (Bf ) is decreasing in B for B large enough.
exists
for
large enough, and the function
f has a dependence with respect to one parameter B we write

fuB (x, 0) = h(x) + Bg(x) for some h, g L and assuming that g > 0 on
positive measure, we can prove the following. For large B (no matter how h and g
Likewise, if we assume that
f = f B
such that
some set of
are distributed), there is always survival. The proofs are the same as for Theorem 6 and will not
be detailed separately.
As a consequence of the last theorem, we can say that increasing the size of the nonlinearity,
assuming that the favourable region is not empty, enhances the chance of having
Hence, it increases the chance of biological survival (existence of a positive solution
0 unstable.
p of (7)). In
our forthcoming paper [18], we show that it also increases the speed of biological invasion. This
result bears consequences on species survival, as well as on techniques of eradication for harmful
species.
Existence and uniqueness of a stationary solution

We start with existence which is a simpler aspect here.
38
3. Existence and uniqueness of a stationary solution

3.1
Proof of existence
Assume rst that
is an unstable solution of (7) and that condition (10) is fullled. Let
us prove that there exists a positive and periodic solution of (7). Let
solution of
with
1 < 0.
Since
f (x, u)
(A(x)) fu (x, 0) = 1
is periodic, > 0, kk = 1,
is of class
C1
in
RN [0, ]
(with
in
be the unique positive
RN ,
> 0),
(13)
for
>0
small enough, one
gets :
f (x, ) fu (x, 0) +
in
RN .
(14)
Therefore, it follows that
(A(x)) f (x, )
and
1
0
2
in
RN ,
is a subsolution of (7) with periodicity conditions. Moreover, if
(15)
is taken as in (10),
M is an upper solution of (7) with periodicity conditions, and (for small enough)
RN . Thus, it follows from a classical iteration method that there exists a periodic
N
solution p of (7) which satises p M in R . Theorem 1, part 1) is proved.
the constant
M
classical
in
Next, assume that
0). Let be the rst

fu (x, 0)(x) for all x RN
( 1
eigenfunction of (13). From hypothesis (9), one has

and
> 0.
Hence,
(A(x)()) f (x, ) > 1 0

for all
0 is stable
f (x, (x)) <
is a nonnegative bounded solution of (7) and assume that
in
RN
(16)
is bounded from below
> 0.
Recall that
away from
and
is a nonnegative and bounded solution of (7). Since
is bounded, one can dene

= inf > 0, > p
> 0, and set z := p.
z(xn ) 0 as n +.
Assume that
such that
Then
in

RN 0.
z 0,
and there exists a sequence
Assume at rst that up to the extraction of some subsequence,

By continuity,
z(x) =
0. As is a supersolution of
(7)
(17)
xn x RN
as
xn RN
n +.
(in the sense that it satises (16)) with
periodicity conditions, it is easy to see from the strong elliptic maximum principle that
Therefore
is a positive and periodic solution of (7). Since
1 0 (0
z 0.
is assumed to be
stable), it follows from (7) and (16) that
0 = (A(x)p) f (x, p(x)) > 0.

One is thus led to a contradiction.
Q
xn xn N
i=1 Li Z. Then, up to the extraction
of some subsequence, one can assume that there exists x C such that xn x as n +.
Next, set n (x) = (x + xn ) and pn (x) = p(x + xn ). Since both A and f are periodic with respect
to x, the functions n and pn satisfy

In the general case, let
(xn ) C
be such that
(A(x + xn )( n )) f (x + xn , n ) > 0
(A(x + xn )pn ) f (x + xn , pn ) = 0
in
RN .
(18)
39

From standard elliptic estimates, it follows that (up to the extraction of some subsequences)
2
converge in Cloc to a function
satisfying
(A(x + x )p ) f (x + x , p ) = 0
while the sequence
( n )
converges to
:= ( + x ),
in
z (x) := (x) p (x).
RN ,
(19)
and
(A(x + x )( )) f (x + x , ) > 0
Let us set
pn
in
RN .
(20)
Then
z (x) = lim [ (x + xn ) p(x + xn )] ,

n+
z (x) = limn+ z(x + xn ). Therefore z 0 and z (0) = 0. It then follows from the
strong maximum principle that z = 0 and reaches a contradiction as above.
Finally, in all the cases, one has = 0, thus p 0, and the proof of Theorem 1 is complete.

whence
Remark 5
Theorem 1 holds, as such, if equation (7) is replaced by
(A(x)u) + B(x) u = f (x, u)

where
is a
C 0,
in
RN ,
(21)
periodic drift. Indeed, the proof of Theorem 1 does not rely on the variational
structure of (7).
More generally, consider the case where
A, B
and
are not periodic (with respect to
x) anymore.
One can wonder whether a result similar to Theorem 1 still holds in this case. For this purpose,
a possible generalization of the rst eigenvalue
fu (x, 0)
in
RN
of the operator
L = (A(x)) + B(x)
is
N
1 = inf { R, C 2 (RN ) L
(R ), > 0, (L ) 0
L(x)
.
=
inf
sup
(x)
C 2 (RN )L (RN ), >0 xRN
in
RN }
.
1 gives the same value as before in the periodic case. With the same arguments
as above, one can easily prove that if f satises (10) and if 0 is an unstable solution of (21)
(1 < 0), then there exists a positive solution p of (21). On the other hand, if f satises (9) and
if 0 is a strictly stable solution of (21) (1 > 0), then there is no positive bounded solution of (21)
(i.e. 0 is the only nonnegative and bounded solution of (21)). For the proof, assume indeed that
there is a positive bounded solution p of (21) ; it follows from (9) that Lp 0, whence 1 0.
This denition for
However, it is not clear whether, under assumption (9), the nonexistence of positive bounded
solutions
of (21) still holds if
1 = 0.
We mention this question as an open problem.
Another possible generalized rst eigenvalue of
in the non-periodic case is the following
01 = sup { R, C 2(RN ), > 0, (L ) 0

L(x)
.
=
sup
inf
N
(x)
C 2 (RN ), >0 xR
in
RN }
.
In the case of a bounded smooth domain, this denition reduces to the classical rst eigenvalue of
L with Dirichlet boundary conditions (see [25], [75]). In the periodic case in RN , one has 1 01 ,
but with a strict inequality in general, even in the case of constant coecients (for instance, for
40

Lu = u00 + u0
in
R,
one has
0 = 1 < 01 = 1/4),
see [2], [19], [76]. But this denition of
01
is
well-suited for a condition on the existence of other types of solutions of (21), maybe not bounded,
in the general nonperiodic case. Namely, for a function
of the type
f (x, s) = (x)s (x)s2 ,
Pinsky [77] (see also [34]) proved that the existence of a solution of minimal growth at innity
solution
01 < 0 (a solution of minimal growth at innity for (21) is a positive

u of (21) such that u v in RN \D for all bounded domain D and for all nonnegative
v of (21) in RN \D with u v on D).
3.2
Proof of uniqueness
for (21) is equivalent to

solution
For analogous problems on bounded domains with e.g. Dirichlet conditions on the boundary,
uniqueness of the positive solutions is well known (compare [10]). The diculty here arises
because of the lack of compactness and because of the fact that one does not assume a priori
that
is bounded from below away from zero.
The proof of Theorem 2 essentially relies on the following property.
Proposition 2
Assume that

u C 2 RN be a bounded
> 0 such that u(x) for all
is an unstable solution of (7). Let
nonnegative solution of (7). Then, either
u0
or there exists
x RN .
Note that periodic solutions obviously satisfy this property. But, here, we look at uniqueness
within a more general class of functions. In particular, it is not assumed a priori that
which we will now rule out.
inf u > 0,
RN
BR be the open ball of RN , with

N
centre 0 and radius R. Let y be an arbitrary point in R . It is well-known that there exist a
y
y
unique real number (principal eigenvalue) R , and a unique function R (principal eigenfunction)

2 B , satisfying3
in C
R
We prove Proposition 2 through a succession of lemmas. Let
(A(x + y)yR ) fu (x + y, 0)yR
yR
yR
y
kR k
yR yR
0
0
1.
BR ,
BR ,
on BR ,
in
in
(22)
yR are unique, standard elliptic estimates and compactness arguments imply

y
y
y
that the maps y 7 R and y 7 R are continuous with respect to y (the continuity of R is
y
understood in the sense of the uniform topology in BR ). Note that, since f is periodic in x, R
y
and R are periodic with respect to y as well.
y be the principal eigenvalue and y the principal eigenfunction of
Let
1
y y in RN ,
(A(x + y)y ) fu (x + y, 0)y =
1
(23)
y is periodic and positive in RN ,
ky k = 1.
Since both
yR
=
>
=
=
and
First, it is straightforward to observe :
Lemma 1
where
The rst eigenvalue
does not depend on
y = 1
y . In other words,
1
for all
y RN ,
is the rst eigenvalue of (11).
3 throughout
the paper, the operator always refers to the derivation with respect to the x variables
41
Proof. Set (x) := y (x y). The function satises
y
(A(x)) fu (x, 0) =
1
is periodic and positive in RN ,
kk = 1.
Therefore, by uniqueness, one has
Lemma 2
For all
y RN
and
= 0 ,
R > 0,
and
in
RN ,
(24)
y =
0 = 1 .
1
1
one has
yR > 1 .
Proof. The function yR satises
(A(x + y)yR ) fu (x + y, 0)yR
1 yR
yR
yR
y
kR k
Assume that
yR 1 .
y > 0
in
(yR 1 )yR
0
0
1.
be the solution of (23). Then
BR ,
one can assume that
yR < y
in
BR

:= sup > 0, yR < y
in
BR ,
in BR ,
on BR ,
in
(25)
satises
(A(x + y)y ) fu (x + y, 0)y = 1 y

y > 0
Since
Let
=
>
=
=
for all
in
in
>0
BR ,
BR .
(26)
small enough. Now, set

BR > 0.
yR y in BR and there exists x1 in BR such that yR (x1 ) = y (x1 ).

y
y
But, since > 0 in BR and R = 0 on BR , it follows that x1 BR .
y
On the other hand, the assumption R 1 implies, from (25), that
Then, by continuity,
(A(x + y)( yR )) fu (x + y, 0) yR 1 yR 0
Therefore, it follows from the strong elliptic maximum principle that
is impossible because of the boundary conditions on
Finally, one concludes that
[25]).
yR > 1
in
BR .
yR y
in
BR ,
which
BR .
(This can also be derived from a characterization in
Lemma 3
For all
y RN ,
the function
R 7 yR
is decreasing in
R > 0.
Proof. Let R1 and R2 be two positive real numbers with R1 < R2 . The proof of this lemma is
similar to that of Lemma 2, replacing
in
yR
by
yR1
and
by
yR2 , and using the fact that yR2 > 0
BR1 .
The next lemma is a standard result (see e.g. [30]), but we include its proof here for the sake
of completeness.
Lemma 4
42
One has
limR+ yR = 1
uniformly in
y RN .
Proof. For y RN , call Ly the elliptic operator dened by Ly u := (A(x + y)u) fu (x +

y, 0)u.
Since it is a self-adjoint operator, one has the following variational formula for
yR =
where
min
H01 (BR ), 60
yR
QyR (),
(27)

(A(x + y)) fu (x + y, 0) 2 dx
BR
Z
.
QyR () =
2
(28)
BR
(R )R2 ,
Choose a family of functions

of
R,
bounded in
C 2 RN
(for the usual norm) independently
and such that
R (x) = 1 if |x| R 1,
(x) = 0 if |x| R,
R
0 R 1.
Set
R = y R
where
is the solution of (23). Then
R H01 (BR )
(29)
and

2
R (A(x + y)R ) fu (x + y, 0)R
dx
BR
Z
.
QyR (R ) =
2
R
(30)
BR
Integrating the numerator by parts over
BR ,
and using the boundary conditions on
BR ,
one
gets

2
(A(x + y)R ) R fu (x + y, 0)R
dx
BR
y
Z
QR (R ) =
,
2
R
(31)
BR
and, by denition of
R ,
2
[ (A(x + y)R ) R fu (x + y, 0)R
]dx
BRZ
=
[ (A(x + y)y ) y fu (x + y, 0) (y )2 ]dx
Z BR1
+
[ (A(x + y)(y R )) y R fu (x + y, 0)(y R )2 ]dx.
(32)
BR \BR1
From equation (23) satised by
y
Z

with respect to
and
R,
and using that
it follows that there exists
R are bounded in C 2 RN
C 0 such that
and

2
(A(x + y)R ) R fu (x + y, 0)R
dx
BR

Z

1
(y )2 CRN 1

BR1
for all
R2
and
y RN .
, uniformly
(33)
Likewise, one has
Z

Z

2
R
(y )2 C 0 RN 1

BR

BR1
(34)
43

C 0 0,
R 2 and y RN .
y
y
But, since each function is continuous, positive and periodic, and since the functions
N
depend continuously and periodically on y (in the sense of the uniform topology in R ), there
Z
for some
exists
>0
for all
such that
y (x)
x RN
for all
and
y RN .
(y )2 2 |BR1 |.
Thus
BR1
Therefore,
2
R
BR
as
R +,
(35)
y 2
( )
BR1
y RN . Using (31), (33) and (34), one gets that QyR (R ) 1 as

R +, uniformly in y RN .
y
y
y
Next, (27) and Lemma 2 yield 1 < R QR (R ). As a consequence, R 1 as R +,
N
uniformly in y R . This completes the proof of Lemma 4.
uniformly with respect to
We are now able to complete the
Proof of Proposition 2. Let u C 2
RN
be a nonnegative and bounded solution of (7). Let
u 6 0. The strong maximum principle then implies that u > 0 in RN .

1
N
N and
Since f (x, u) is of class C in R [0, ] (with > 0), since f (x, 0) 0 in R
periodic with respect to x, one can choose 0 > 0 small enough such that
us assume that
is
f (x + y, yR ) yR fu (x + y, 0) +
for all
0 < 0 , y RN
and
R>0
(recall that
From Lemmas 3 and 4, there exists
R0 > 0
1 < 0,
1 y
R
2
and
R R0 .
Set
yR > 0
0 yR
in
(36)
BR ).
uy (x) := u(x + y).
1
< 0.
2
(37)
The function
(A(x + y)uy ) f (x + y, uy ) = 0
Furthermore,
BR ,
such that
R R0 , y RN , yR <
In the sequel, x some
in
since
in
uy
satises
RN .
(38)
satises
0 (A(x + y)yR ) = fu (x + y, 0)0 yR + yR 0 yR
in
BR .
(39)
Thus, using (36) and (37), one has
0 (A(x + y)yR ) f (x + y, 0 yR ) (yR

In other words,
0 yR
1
)0 yR 0
2
in
BR .
(40)
is a sub-solution of (38).
uy > 0 yR in BR . If not, there exists 0 < 0 and x1 BR

y
y
y
y
y
N
such that R (x1 ) = u (x1 ) and u R in BR (remember that u > 0 in R , whence
y
minBR uy > 0). Next, since R 0 on BR , it follows that x1 BR . On the other hand,
y
the computations above show that the function R is still a sub-solution of (38). The strong
y
y
maximum principle gives that R u in BR , which is in contradiction with the conditions
Let us now show that
on
44
BR .

uy > 0 yR
BR , thus uy (0) > 0 yR (0). In other words, u(y) > 0 yR (0) for
y
N
N
all y R . Since the function y 7 0 R (0) is periodic, continuous and positive over R , there
y
N
exists > 0 such that 0 R (0) > for all y R , and this completes the proof of Proposition
2.
Finally, one has
in
Let us now turn to the
Proof of Theorem 2. Let u and p C 2

Proposition 2, there exists
>0
such

RN be two positive and bounded solutions of (7). By
N
that u and p in R .
Therefore, we can dene the positive real number

= sup > 0, u > p

RN > 0.
(41)
z := u p 0. From the denition of , it follows

N such that z(x ) 0 as n +.
that there exists a sequence xn R
n
N as n +. By
Assume rst that, up to the extraction of some subsequence, xn x R
N
continuity, one has z 0 in R , and z(x) = 0. Moreover, z satises the equation
Assume that
< 1,
in
and let us set
(A(x)z) f (x, u) + f (x, p) = 0

Furthermore, by assumption (9),
one has f (x, p)
<
f (., s)/s
is decreasing in
f (x, p). Hence, (42) gives
R+
in
RN .
< 1,
(43)
such that
(A(x)z) bz > 0
Since
(42)
is locally Lipschitz continuous in the second variable, one infers from (43) that there
exists a bounded function
z 0
z 0,
RN .
and since we have assume
(A(x)z) f (x, u) + f (x, p) > 0

Since
in
and
z(x) = 0,
in
RN .
(44)
it follows from (44) and from the strong maximum principle that
which is impossible because of the strict inequality in (44).
Q
xn xn N
i=1 Li Z. Then, up to the extraction
of some subsequence, one can assume that there exists x C such that xn x as n +.
Next, set un (x) = u(x + xn ), and pn (x) = p(x + xn ). Since both A and f are periodic with
respect to x, the functions un and pn satisfy
In the general case, let
(xn ) C
be such that
(A(x + xn )un ) f (x + xn , un ) = 0
(A(x + xn )pn ) f (x + xn , pn ) = 0
in
RN .
(45)
From standard elliptic estimates, it follows that (up to the extraction of some subsequences)
and
pn
converge in
2
Cloc
to two functions
and
(A(x + x )u ) f (x + x , u ) = 0
(A(x + x )p ) f (x + x , p ) = 0
Moreover,
un
satisfying
in
RN .
(46)
u > 0 and p > 0.

z (x) := u (x) p (x). Then z 0 and z (0) = 0. Furthermore, z satises
Let us set
(A(x + x )z ) f (x + x , u (x)) + f (x + x , p (x)) = 0
in
RN .
(47)
Then, arguing as for problem (42) above, one obtains a contradiction.
45

1, hence u p. By interchanging the roles of u and p, one
can prove similarly that p u. Furthermore, if p is a positive solution of (7), so is the function
x 7 p(x1 , , xi + Li , , xN ), for each 1 i N . Hence, p is periodic. The proof of Theorem
2 is complete.
Therefore, we know that
The same arguments as above lead to the following uniqueness result for a class of solutions of
more general elliptic equations with drift terms, under a slightly stronger version of assumption
(9) :
Theorem 7
Let A = A(x) be a symmetric matrix eld satisfying (8) and assume that A is of
1,
N
class C
(R ) and that A and its rst-order derivatives are in L (RN ). Let B be a vector eld
0, (RN ) L (RN ). Let f : RN R R, (x, s) 7 f (x, s) be Lipschitz-continuous
of class C
+
0, (RN ) L (RN ) locally in s. Assume
in s uniformly in x and assume that f (, s) is of class C
that
0<s<s,
Let
and
Then
inf RN u > 0
u = v.
Remark 6
inf
xRN
f (x, s) f (x, s0 )
s
s0

> 0.
be two positive bounded solutions of
such that
(A(x)u) + B(x) u = f (x, u)

(A(x)v) + B(x) v = f (x, v)
and
in
RN ,
(48)
inf RN v > 0.
However, it is not true in general the positive solutions
of (48) are bounded from
1 < 0, where the generalized rst

1 is dened as in Remark 5 above.
Indeed, let f be a Lipschitz-continuous function dened in [0, 1], such that f (0) = f (1) = 0,
f > 0pon (0, 1), f 0 (0) > 0 and f (s) f 0 (0)s for all s [0, 1]. It is known (see [68]) that, for any
c 2 f 0 (0), there are positive solutions u of
below by a positive constant under the only assumption
eigenvalue
u00 cu0 + f (u) = 0,

with
u() = 0
and
u(+) = 1.
0<u<1
in
But, under the notations of Remark 5,
1 = f 0 (0) < 0
in
this case.
3.3
Energy of stationary states
This subsection is about an independent result, dealing with the sign of the energy associated
to a positive solution of (7), under condition (9). This result, of independent interest, will be
used in the forthcoming paper [18] on propagation phenomena.
Assume in this subsection that there exists a positive and bounded solution
condition (9) is fullled. It then follows from Theorem 1, part 2), that
of (7) (1
< 0),
p is
1 < 0
and from Theorem 2 that such a function
have seen, the existence of
is known for instance if
p of (7) and that
is an unstable solution
then unique and periodic. As we

and if condition (10) is satised
(from Theorem 1, part 1)).

Consider the energy functional
Z
E(u) :=
C
46

1
u (A(x)u) F (x, u) dx,
2
(49)
4. The evolution equation

dened on

1
1
Hper
:= Hloc
(RN )
Z
with
such that
is periodic
F (x, u) :=
f (x, s)ds.
We now prove that the energy of
is negative.
Proposition 3
solution
Assume that condition (9) is satised and that there exists a positive bounded
of (7). Then
E(p) < 0.
Proof. Under the assumptions of Proposition 3, let be the function dened in [0, 1] by
Z
t [0, 1], (t) = E(tp) =
C
The function
is of class
C1

1 2
t p (A(x)p) F (x, tp(x)) dx.
2
(50)
and
t [0, 1], 0 (t) =
Z
{tp (A(x)p) f (x, tp(x))p(x)} dx.
(51)
C
From (9) and from the positivity and periodicity of
tf (x, p(x))
in
for all
t (0, 1).
and
p in x, it follows that f (x, tp(x)) >
Therefore,
t (0, 1), (t) < t
{p (A(x)p) f (x, p(x))p(x)} dx = 0,
(52)
C
the last equality being obtained by multiplication of the equation (7) satised by
over
C.
p and integration
As a conclusion,
E(p) = (1) < (0) = E(0) = 0.
The evolution equation

This section is devoted to the
Proof of Theorem 3.
identically equal to
of
Assume that f satises (9) and (10). Let u0 be a nonnegative, not

0, bounded and uniformly continuous function, and let u(t, x) be the solution

ut (A(x)u) = f (x, u), t R+ , x RN ,
(53)
u(0, x) = u0 (x), x RN .
0 is an unstable solution of (7) (1 < 0). Let R = 0R be

y = 0, and call R = 0R . Namely, R C 2 BR and satises
Assume rst that

satisfying (22) with
(A(x)R ) fu (x, 0)R = R R in BR ,

R > 0 in BR , R = 0 on BR , kR k = 1.
the function
(54)
u(1, x) > 0 in RN . Therefore, for > 0

N by setting v (x) := (x)
chosen small enough, R < u(1, x) in BR . Let us extend R to R
0
R
N
in BR , and v0 (x) := 0 in R \BR . Dene the function v1 by

t v1 (A(x)v1 ) = f (x, v1 ), t R+ , x RN ,
(55)
v1 (0, x) = v0 (x), for x RN .
From the strong parabolic maximum principle, one has
47

As it has been done in the course of the proof of Proposition 2, using (40), for
and
>0
small enough,
subsolution of (7) in
RN
RN .
is a subsolution of (7) in
Thus
v1
large enough
BR , and therefore v0 is a "generalized"

t. Furthermore, v1 (0, x) u(1, x) in
is nondecreasing in time
implies
v1 (t, x) u(1 + t, x)
Moreover, for
> 0
small enough,
v0 (x) p(x)
R+ R N .
in
in
RN ,
1 < 0,
where
periodic, solution of (7) (the existence and uniqueness of such a

(10) and
(56)
owing to Theorems 2 and 1, part 1)). Since
is the unique positive, and
follows from assumptions (9),
is a stationary solution of (6),
one has
v1 (t, x) p(x)
in
R+ R N ,
(57)
v1 is nondecreasing in time t, standard elliptic estimates imply that v1 converges in

2
N
Cloc R
to a bounded stationary solution v ( p) of (6). Furthermore, one has v (0)
v1 (0, 0) R (0) > 0. Using the strong maximum principle, it follows that v > 0 in RN , and
Because
v p.
N for all s M
Next, from (10), there exists M > 0 such that f (x, s) 0 in R
N
Take M large enough so that M u0 in R
and let v2 be dened by

t v2 (A(x)v2 ) = f (x, v2 ), t R+ , x RN ,
v2 (0, x) = M, x RN .
we infer from Theorem 2 that
M is a supersolution of (7), v2 is nonincreasing in time t.

M u0 (x) 0 in RN ,
v2 (t, x) u(t, x) 0 in R+ RN .
Then, since
and
Besides, since
x RN .
(58)
v2 (0, x) =
(59)

N
2
v2 0 from the maximum principle, v2 converges in Cloc
R as t + to
a bounded and nonnegative stationary solution v ( M ) of (6). From Theorem 2, either v 0
or v p. Finally, one has
Furthermore, since
v1 (t, x) u(1 + t, x) v2 (1 + t, x), t > 0 x RN .
(60)
v1 (t, x) p(x)
as t +, it follows from (60) that v p, and that u(t, x) converges
2
p(x) in Cloc
RN as t +. Part 1) of Theorem 3 is proved.
Let us now assume that 0 is a stable solution of (7). Then, as carried above, there exists
M > 0 such that f (x, s) 0 for all s M and x RN . Taking M large enough so that u0 M ,
one again obtains, dening v2 as above,
Since
to
v2 (t, x) u(t, x) 0
in
R+ R N .
(61)

2
Cloc
RN to 0 as t +.
Furthermore, the convergence is uniform in x : indeed, v2 is periodic in x at each time t 0,
since it is so at t = 0, and equation (6) is periodic in x. It follows from (61) that u(t, x) converges
to 0 uniformly as t +, and this concludes the proof of Theorem 3, part 2).
But this time, from the result of Theorem 1, part 2),
v2
Conservation of species in ecological systems

In this section, we study various eects of the term
(11).
48
converges in
fu (x, 0)
on the principal eigenvalue
of
5. Conservation of species in ecological systems

5.1
Inuence of the amplitude" of the reaction term
This subsection is devoted to the stability condition of the zero steady state when the nonlinearity
Bf in (7), where B is a positive real number. Here, the function f is

1 (Bf ) the rst eigenvalue of (11) with the nonlinearity Bf , and B C 2 RN
principal eigenfunction (with the normalization condition kB k = 1) of

(A(x)B ) Bfu (x, 0)B = 1 (Bf )B ,
(62)
B is positive and periodic in RN .
is replaced by
xed. Let us call

the unique
The next two statements are concerned with the dependence of
1 (Bf )
with respect to
and correspond to parts 1) and 2) of Theorem 6 respectively.
Proposition 4
every
B>0
Proof.
fu (x, 0)
C
and the function
> 0,
If
fu (x, 0) = 0
or if
and
B 7 1 (Bf )
is decreasing in
fu (x, 0) 6 0,
1 (Bf ) < 0
then
for
R+ .
B 7 1 (Bf ) is concave. Since the operator Lu =

(A(x)u) Bfu (x, 0) is self-adjoint, the eigenvalue 1 (Bf ) is obtained from the variational
One rst shows that the mapping
characterization :
Z
1 (Bf ) =
min
1 , 60
Hper
(A(x)) Bfu (x, 0)2

C
Z
,
2
(63)
C
where
1
Hper
was dened in the previous section. Thus, it follows that
whence continuous (on
B 7 1 (Bf )
is concave,
R).
Next, integrate equation (62) by parts over
C.
Using the periodicity of
Z
B
fu (x, 0)B = 1 (Bf )

C
Take an arbitrary sequence
B ,
one obtains,
Bn 0.
B .
(64)
C
Since
1 (Bn f ) 1 (0) = 0,
standard elliptic estimates
Bn
W 2,p
= 1,
and Sobolev injections imply, up to the extraction of some subsequence, that the functions
, locally (and therefore uniformly

fu (x, 0) is in L ). Furthermore, is
converge to a nonnegative function

for all
1<p<
(we recall that
by periodicity) in
kk
such that
is periodic and satises
(A(x)) = 1 (0) = 0.
(65)
is positive and 0 1. By a classical

can then show that the whole family B converges to 1 as B 0.
Then, divide (64) by B and pass to the limit as B 0, B 6= 0. It follows that

Z
d1 (Bf )
|C| =
fu (x, 0),
dB B=0
C
From the strong maximum principle,
where
|C|
denotes the Lebesgue measure of
fu (x, 0) > 0. Since B 7 1 (Bf ) is concave, 1 (0) = 0 and

C
0, it
R+ .
follows that
1 (Bf ) < 0
(66)
C.
Z
Assume now that
argument we
for every positive
and the function
B 7 1 (Bf )
d1 (Bf )
dB B=0
<
is decreasing in
49

Z
fu (x, 0) = 0, then 1 (Bf ) 0 for every positive
C
equation (62) by B and integrating over C leads to :
Similarly, if
Z
1 (Bf ) |C| =
C
B.
Furthermore, dividing
B (A(x)B )
.
2B
(67)
1 (Bf ) = 0 for some B > 0, then B is constant, whence fu (x, 0) 0. Therefore, if one further
fu (x, 0) 6 0, then 1 (Bf ) < 0 for each B > 0, and the function B 7 1 (Bf ) is
decreasing in R+ . This completes the proof of Proposition 4.
Z
In the case
fu (x, 0) < 0, we ow prove the following result.
If
assumes that
Proposition 5
x0 C
in B .
Z
fu (x, 0) < 0,
If
1 (Bf ) > 0
then
for all
B>0
such that
fu (x0 , 0) > 0,
then, for
large enough,
1 (Bf ) < 0
Proof. From the proof of Proposition 4, it is easy to show that, if

0
for
B>0
small enough. If there exists
small enough, since
1 (0) = 0
and, from (66),
There exists a positive and periodic function
and
1 (Bf )
is decreasing
C
d1 (Bf )
dB
fu (x, 0) < 0, then 1 (Bf ) >

Z
= fu (x, 0) > 0.
B=0
such that
fu (x, 0)20 > 0.
(68)
C
Then, from (63),

0 (A(x)0 ) Bfu (x, 0)20 dx
Z
1 (Bf ) C
.
2
0
(69)
1 (Bf ) < 0 for B large enough. The concavity of B 7 1 (Bf ) and the
fact that 1 (0) = 0 then imply that B 7 1 (Bf ) is decreasing at least when 1 (Bf ) is negative,
and thus for B > 0 large enough.
Clearly, this shows that
5.2
fu (x, 0)
This section is concerned with the study of the dependence of the rst eigenvalue
on the shape of the function
of (11)
fu (x, 0). One denotes (x) = fu (x, 0) and 1 = 1 []. The following
and of its average.
proposition compares the eect of
Proposition 6
Let
be a real number. Then
1 [] 1 [0 ],
(70)
Z
= 0 |C|
as soon as
C
50
(where
|C|
is the Lebesgue measure of the set
C ).
5. Conservation of species in ecological systems
Proof. From (11), replacing fu (x, 0) by (x), one obtains

(A(x)) (x) = 1 = 1 [], x RN ,
where
(71)
> 0 is the principal periodic eigenfunction associated to 1 , with the

kk = 1. Dividing (71) by and integrating by parts over C yields
Z
Z
(A(x))
= 1 |C| = 1 []|C|.
2
C
C
normalization
condition
Clearly, clearly,
0 1
and
1 [0 ] = 0 . Therefore, it follows
Z
C
1 []
= 0 = 1 [0 ].
|C|
This completes the proof of Proposition 6.
(72)
from equation (72) that
In a sense, Proposition 6 shows that a heterogeneous medium improves the biological conservation, in comparison with a constant medium
0 .
Moreover, the medium becomes all the more conservative the more it is heterogeneous. This
statement is made precise in the following
Proposition 7
0 R and let f be such that fu (x, 0) = (x) = 0 + B(x), where has

6 0. Let 1,B = 1 [] be the rst eigenvalue of (71). Then the function
B 7 1,B is decreasing in R+ . Furthermore, 1,B is negative for all B > 0 if 0 0, and 1,B
is negative for B > 0 large enough if 0 < 0.
Let
zero average and
Proof.

As in Proposition 6, it can be shown that the function
d1,B
dB B=0
= 0, 1,0 = 0 .
B 7 1,B
is concave, and
The conclusion follows as in the proofs of Propositions 5 and 6.
Let us now turn out to the eect of rearranging the level sets of
We denote by
function obtained by performing a succession of Steiner periodic rearrangement of

to the ordered variables
Proposition 8
x1 , ..., xN

the
with respect
(see Section 2.3.1 above for the denition) .
Under the above notations, and assuming furthermore that
is the identity
matrix, the following inequality holds
1 [ ] 1 [].
(73)
Proof. The proof rests on rearrangement inequalities. Let k be a nonnegative real number such
that
+k 0
in
RN ,
normalization condition
be
kk = 1.
and let
the principal eigenfunction associated to
1 [],
with the
A classical inequality for rearrangement (Compare e.g. [65]) asserts that :
( + k) ( )
C
( + k)2 .
(74)
C
51
Since ( + k)
+ k , one infers from (74) that
( )
C
[2 ( )2 ] = 0,
the other hand,
+ k
C
[2 ( )2 ].
On
whence
2 .
( )
C
(75)
Next, it follows from Theorem 2.1 and Remark 2.6 in [65] (see also [12], [20]) that
||2
C
As already emphasized,
1 [ ]
are given by the following variational formul
Z
1 [] =
(76)
1 []
and
| |2 .
min
1 , 60
Hper

||2 2
Z
,
2
(77)
C
and
Z
1 [ ] =
min
1 , 60
Hper

||2 2
Z
.
2
(78)
C
Furthermore, the minimum in (77) is reached for
Z
1 [ ]
= .
It follows from (78) that

| |2 ( )2
Z
.
2
( )
(79)
Z
that
()2 =
( )
From (75), one has
C
2
( )
||
, and, from (76),
| |2 .
One also knows
. Finally, it follows from (79) that
Z
1 [ ]

||2 2
Z
= 1 [],
2
(80)
C
and Proposition 8 is proved.
As a conclusion, one can say that from the biological conservation standpoint, among all
having a given distribution function, the optimal one is necessarily Steiner symmetric,
xi = 0 and decreasing in xi , for xi [0, Li /2] (for each
i = 1, , N ). Note, however, that the actual optimal shape (among all Steiner symmetric
functions in all variables) is not known, even when takes only two values.
periodic
that is, symmetric with respect to
52
6. The eect of fragmentation in bounded domain models
The eect of fragmentation in bounded domain models

In this section, we summarize some properties concerned with the case of bounded domains,
and, using a symmetrization argument, we state some general results extending previous work .
Consider the equation
ut (A(x)u) = f (x, u), x ,

set in a bounded smooth domain
satises (9-10), and that
RN .
(81)
Assume, say, that the nonlinearity
is smooth and
is a smooth uniformly elliptic matrix eld.
For Dirichlet boundary conditions
u=0
there exists a solution
on
of
(A(x)p) = f (x, p), x

p(x) > 0, x
p(x) = 0, x ,
(82)
1 of the operator L = (A(x)) fu (x, 0) in (with

Dirichlet boundary conditions on ) is negative. Furthermore, if it exists, p is unique.
2
In the one-dimensional case with constant diusion duxx and f of the type f (u) = u u
u2 /(1 + u2 ), this result is due to Ludwig, Aronson and Weinberger [70] (see also Murray and
if and only if the rst eigenvalue
Sperb [73] for the two-dimensional case with additional drift terms). It was generalized to any
dimension by Cantrell and Cosner [27], in the case of a nonlinearity
m(x)u c(x)u2
(with
c(x) > 0)
and with constant diusion
of the type
f (x, u) =
du.
For the general equation (82), the results mentioned above can be proved with the same
methods as the ones used in the present paper. Notice that the case of bounded domains is
actually much simpler than the periodic case in
RN .
In particular, uniqueness of the positive
solutions, in particular, was proved in [10].
u(t, x) of (81)
u0 0, u0 6 0 converge uniformly in x as t + to the unique
p(x) if 1 < 0 ; otherwise, that is if 1 0, then u(t, x) 0 uniformly in x
Furthermore, using the same arguments as those of section 2.2, the solutions
positive solution
as
t +
(see also [27, 28, 70] for earlier results in some particular cases).
Some of the above results have been extended by Cantrell and Cosner [29] to some special
cases of systems of two equations.
For problem (81) in a bounded interval
(0, L),
the inuence of the location of the favourable
and unfavourable regions has been studied in [28], on the basis of explicit analytic calculations.
This work is restricted to the case of the patch model, that is when the birth rate
fu (x, 0)
is piecewise constant and only takes two possible values. For Dirichlet boundary conditions,
it is better for species conservation to have the most favourable region concentrated around
the middle of the interval, away from the boundary. On the contrary, in the case of Neumann
boundary conditions, it is better for species conservation to have the favourable and unfavourable
regions concentrated near each of the two boundary points of the interval.
Using symmetrization techniques as we did above for the periodic case allows us to extend
and much simplify these results. For a general
(x),
we prove the following :
53
Theorem 8
Let
direction say
x1 ,
and symmetric with respect to
RN and assume that

x1 = 0. Let be continuous
be a smooth bounded domain of
is convex in some
in
and let
1 []
be the rst eigenvalue of
(x) = 1 []
= 0 on .
in
1 [ ] 1 [], where is the Steiner symmetrization of in the variable x1 , with respect

to {x1 = 0} and nonincreasing away from {x1 = 0}.
Furthermore, equality 1 [] = 1 [ ] holds if and only if is symmetric with respect to x1

and nonincreasing away from {x1 = 0}.
Then,
in the sense of
having a given distribution function that function, namely , that minimizes
1 []. In higher dimensions, this is not known. In dimension N = 2 consider the simple patch
model when is allowed to take two values. When is say a rectangle, then the shape of the
optimal is not known. From the theorem, we know that it is doubly symmetric in the Steiner
In an interval, this theorem provides the optimal rearrangement of a function
nding, among all
sense, hence the favourable is connected. However, it is not known that this region is convex, a
property which we conjecture holds.
For Neumann boundary conditions, the situation is more delicate. One has to use monotone
rearrangement (see [65, 20] for denition and properties and [12] for the pecise inequality).
Consider now the eigenvalue problem :
(x) = 1 []
= 0 on .
in
Like for the periodic case or for the bounded domain case with Dirichlet condition, the sign of
1 []
determines the existence of stationary solutions and asymptotic bahaviour of the solution
of the nonlinear problem (81) with Neumann condition.
v(x) of one variable, on an interval (a, b), is dened

]
as the unique monotone (say) nondecreasing function v on (a, b) which has the same distribution
function as v . Then, dene the Steiner monotone rearrangement of a function v(x1 , . . . , xN ) on
]
a set {x ; xi (ai , bi ), i = 1, . . . , N }, as the function v which is obtained from v by performing
successive monotone Steiner rearrangements in each of the directions x1 , . . . , xN .
The monotone rearrangement of a function
Theorem 9
Assume that
{x ; xi (ai , bi ), i = 1, . . . , N }.
is a cube
Under Steiner monotone

1 [] ] 1 [].
rearrangement, the Neumann eigenvalue satises the following inequality :

This theorem rests on the following rearrangement inequality :
||
This is well known in dimension
|] |2 .
(83)
1 (see [65]) but somewhat delicate in dimension N . This inequa-
lity is proved in [12].

As a consequence of this result, we see that, in the simplied case of the patch model on
a rectangle, the rearranged conguration, where all the favourable patch is concentrated in one
of the corners of the domain, leaves better chances of survival than an originally fragmented
conguration. An example is given in the gure 4 below.
Remark 7
As a consequence of this theorem, in the patch case, the optimal rearrangement,
under the constraint of a given area of the unfavourable zone, is a
such that
] = .
However,
like for the other cases, in higher dimension, the question of the optimal shape of the environment
(in the patch model) is still open. This appears to be an interesting mathematical question.
54
6. The eect of fragmentation in bounded domain models
Fig. 4 (a) Initial patch, and (b) after monotone Steiner rearrangement
55
56
Chapitre 2 : Biological invasions and

pulsating travelling fronts
Ecrit en collaboration avec Henri Berestycki
et Franois Hamel
a EHESS, CAMS, 54 Boulevard Raspail, F-75006 Paris, France
b Universit Aix-Marseille III, LATP, Facult Saint-Jrme, Case cour A

Avenue Escadrille Normandie-Niemen, F-13397 Marseille Cedex 20, France
Sommaire
1
2
3
4
Introduction and main results . . . . . . . . . . . . . . . . . . . 58

Existence result . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.1
Existence result in nite cylinders for a regularized problem . .
63
2.2
Passage to the limit in the innite domains
. . . . . . . . . . .
69
2.3
. . . . . . . . . . . . . . . . . .
70
2.4
Existence of a
. . . . . . . . . . . . . . . . . .
74
c1 . . . . . . . . . . . .
75
0 . .
1 1
solution (c , u )
solution (c, u) for
all
c>
2.5
Existence of a
4.1
. . . . . . . . . . . . .
82
4.2
. . . . . . . . . . . . . . . . . . . . . . .
Characterization of c
Dependency of c with respect to the nonlinearity f

. . . . . .
83
Monotonicity of the solutions . . . . . . . . . . . . . . . . . . . 77

The minimal speed c . . . . . . . . . . . . . . . . . . . . . . . . 82
4.3
57
87
1 Introduction and main results

This paper deals with the mathematical analysis of a periodically fragmented environment
model which is given by the reaction-diusion equation
ut (A(x)u) = f (x, u), x RN

with periodic dependence in the
(1)
variables. It is the sequel to the paper [17], which focused on
some conditions for the stationary equation
(A(x)p) = f (x, p)
p(x) > 0, x RN ,
in
RN ,
(2)
to have a bounded solution, and on the eects of the heterogeneity in
x.
The present paper is
concerned with the propagation pheneomena, and especially the propagation of fronts, associated
to (1) precise denitions will be given below. Some formulas for the speeds of propagation of
fronts are proved and the dependence in terms of the coecients of (1) is analyzed.
The archetype of such reaction-diusion models is the following equation
ut u = f (u)
in
RN
(3)
which was introduced in the pioneering papers of Fisher [43] and Kolmogorov, Petrovsky and
Piscounov [68]. An example of nonlinear term is given by the logistic law
f (u) = u(1 u).
This
type of equation was rst motivated by population genetics, and, as (1), it also arises in more
general models for biological invasions or combustion.
Of particular interest are the propagation phenomena related to reaction-diusion equations
of the type (3), or (1). First, equation (3) may exhibit planar travelling fronts, which are special
u(t, x) = U (x e + ct) for some direction e (|e| = 1, e is the direction of

propagation) and U : R (0, 1) (assuming that f (0) = f (1) = 0). Such solutions are invariant
in time in the comoving frame with speed c in the direction e. Second, starting with an initial
datum u0 0, 6 0 which vanishes outside some compact set, then, under some assumptions
on f , u(t, x) 1 as t + ; furthermore, the set where u is close to 1 expands at a certain
solutions of the type
speed which is the asymptotic speed of spreading and which, in the case of equation (3) with a
nonlinearity
positive in
(0, 1),
is the minimal speed of planar fronts (see e.g. [5]).
Whereas the homogeneous equation (3) has attracted many works in the mathematical literature, propagation phenomena for heterogeneous equations of the type (1), where both the
diusion and the reaction coecients depend on the space variables
x,
were studied more re-
cently (see e.g. [13, 44, 56, 85, 93]). In models of biological invasions, the heterogeneity may be
a consequence of the presence of highly dierentiated zones, such as forests, elds, roads, cities,
etc., where the species in consideration may tend to diuse, reproduce or die with dierent rates
from one place to another.
A(x)
f (x, u) now depend on the variables x = (x1 , , xN ) in a periodic fashion.
One focuses here on periodic environments models and for which the diusion matrix
and the reaction term
As an example,
may be of the type
f (x, u) = u((x) (x)u),
(4)
f (x, u) = u((x) u),
(5)
or even, simply,
58
1. Introduction and main results

(x), which may well be negative, can be interpreted as an eective
birth rate of the population and the periodic function (x) reects a saturation eect related to
competition for resources. The lower is, the less favorable the environment is to the species.
where the periodic coecient
These models for biological invasions in unbounded domains were rst introduced by Shige-
1 and 2 (see [66, 85, 86]). In these works, the nonlinearity f is given by
A and are piecewise constant and only take two values. This model is then referred to
sada et al. in dimensions

(5), and
as the patch model. Numerical simulations and formal arguments were discussed in [66, 85, 86]
about this model in space dimensions 1 or 2. The various works of Shigesada and her collaborators have been an inspiring source for the present paper. We aim here at proving rigorously
some properties which had been discussed formally or observed numerically. The introduction of
new mathematical ideas will furthermore allow us to derive results in greater generality and for
higher dimensional problems as well.
In the paper [17] and in the present one, we discuss these types of problems in the framework
of a general periodic environment, and we give a complete and rigorous mathematical treatment
of these questions. In the rst paper [17], we discussed the existence of a positive stationary
state of (1), that is a positive bounded solution
of (2). The latter is referred to as biological
conservation. We also analyzed in [17] the eects of fragmentation of the medium and the eects
of coecients with large amplitude on biological conservation. Here, we connect the condition
p) to that for propagation of pulsating fronts for

uniform state 0. This type of question is referred
for species survival (existence of such a solution

which the heterogeneous state
invades the
to as biological invasion. We also analyze the eects of the heterogeneity of the medium on the
speed of propagation. We especially prove a monotonous dependence of the speed of invasion on
the amplitude of the eective birth rate.
L1 , , LN > 0 be N given
be periodic if g(x1 , , xk +
Let us make the mathematical assumptions more precise. Let

positive real numbers. A function
Lk , , xN ) g(x1 , , xN )
for all
RN
g :
R is meant to
k = 1, , N . Let C be the period
cell dened by
C = (0, L1 ) (0, LN ).
The diusion matrix eld
periodic, of class
C 2,
A(x) = (aij (x))1i,jN

> 0),4 and uniformly
(with
0 > 0, x RN , RN ,
is assumed to be symmetric (aij
= aji ),
elliptic, in the sense that
aij (x)i j 0 ||2 .
(6)
1i,jN
f : RN R+ R is of class C 1, in (x, u) and C 2 in u, periodic with respect

N and one sets f (x, 0) := lim
assumes that f (x, 0) = 0 for all x R
u
s0+ f (x, s)/s.
The function
to
x.
One
Furthermore, throughout the paper, one assumes that
x RN , s 7 f (x, s)/s
is decreasing in
s>0
(7)
and
M 0, s M, x RN , f (x, s) 0.
(8)
4 The
smoothness assumptions on A, as well as on f below, are made to ensure the applicability of some
a priori gradients estimates for the solutions of some approximated elliptic equations obtained from (1)
(see Lemma 6 in Section 2.3). These gradient estimates are obtained for smooth (C 3 ) solutions through
a Bernstein-type method, [14]. We however believe that the smoothness assumptions on A, as well as on
f , could be relaxed, by approximating A and f by smoother coecients.
59

Examples of functions
satisfying (7-8) are functions of the type (4) or (5), namely
u((x) (x)u) or simply f (x, u) = u((x) u), where and

Let 1 be the principal eigenvalue of the operator L0 dened
f (x, u) =
1, periodic functions.
are C
by
L0 := (A(x)) fu (x, 0),

with periodicity conditions. Namely,
function
>0
is the unique real number such that there exists a
C2
which satises
(A(x)) fu (x, 0) = 1
is periodic, > 0.
in
RN ,
(9)
1 < 0, and stable if 1 0.

We especially proved in [17] that, if 1 0, then 0 is the only nonnegative bounded solution
of (2) and any solution of (1) with bounded nonnegative initial condition u0 converges to 0
N as t + (one refers to this phenomenon as extinction). On the other
uniformly in x R
hand, if 1 < 0, then there is a unique positive bounded solution p of (2), which turns out to
5
be periodic, and the solution u(t, x) converges to p(x) locally in x as t +, as soon as
u0 0, 6 0.
One says that
is an unstable solution of (2) if
The above results motivate the following
Denition 1
We say that the hypothesis for conservation is satised if there exists a positive
bounded solution
of (2).
A simple necessary and sucient condition for the hypothesis for conservation (or survival)
1 < 0, and the solution p is then unique and periodic. This hypothesis is
fullled especially if fu (x, 0) 0, 6 0. An example of a function f satisfying the hypothesis is
the classical Fisher-KPP nonlinearity f (x, u) = f (u) = u(1 u) (see [43, 68]), where p(x) 1.
For a general nonlinearity f satisfying (7) and (8), comparison results and conditions on fu (x, 0)
for 1 to be negative are given in [17] (see also Theorem 2 below). However, it is not easy to
to be satised is that
understand in general the interaction between the heterogeneous diusion and reaction terms.
One focuses here on the set of solutions which describe the invasion of the uniform state
by the periodic positive function
p,
when the hypothesis for conservation is satised. A solution
u(t, x) of (1) is called a pulsating travelling front propagating in the direction e with the eective
speed c 6= 0 if
(t, x) R RN , ut (A(x)u) = f (x, u),

N
Y
ke
(10)
N
Li Z, x R , u t +
,x
= u(t, x + k),
k
c
i=1
with the asymptotic conditions
u(t, x)
xe
0, u(t, x) p(x)
The above limits are understood as local in
t,
xe+
0.
and uniform in the directions of
(11)
RN
orthogonal to
e.
Our rst result is the following existence theorem :
5 Notice that the periodicity is forced by the uniqueness, but was not a priori required in the formulation
of equation (2).
60
Theorem 1
Under the above assumptions on A and f , and under the hypothesis for conserva
tion, there exists c > 0 such that problem (10-11) has a classical solution (c, u) if and only if
c c . Furthermore, any such solution u is increasing in the variable t.
Lastly, the minimal speed c is given by the following variational formula
c = min {c, > 0

where
c ()
such that
c () = 0} ,
is the principal eigenvalue of the elliptic operator
Lc,
(A(x)) 2 eA(x)
[ (A(x)e) + 2 eA(x)e c + fu (x, 0)],
(12)
with periodicity conditions.

Before going further on, let us briey comment this result and recall some earlier works in the
literature. Observe rst that the formula for the minimal speed simply reduces to the well-known
p
2 f 0 (0) for the minimal speed of planar front (x e + ct) for the homogeN with f satisfying (7-8) and p(x) min {s > 0, f (s) 0}.
neous equation ut u = f (u) in R
Periodic nonlinearities f (x, u) in space dimension 1 were rst considered by Shigesada, Kawasaki
6
and Teramoto [86], and by Hudson and Zinner [56]. The case of equations ut u+vu = f (u)
with shear ows v = ((y), 0, , 0) in straight innite cylinders {(x1 , y) R } was dealt
with by Berestycki and Nirenberg [24], under the assumption that f stays positive in, say, (0, 1) ;
min-max type formulas for c were obtained in [48]. Berestycki and Hamel [13] generalized the
Fisher KPP formula
notion of pulsating fronts and got existence and monotonicity results in the framework of more
ut (A(x)u) + v(x) u = f (x, u) in periodic domains, under

the assumption that f 0 and f (x, 0) = f (x, 1), f (x, s) > 0 for all s (0, 1). A formula for
the minimal speed is given in [16] under the assumption that f (x, s) fu (x, 0)s for all s [0, 1]
and the dependence of c in terms of the diusion, advection, reaction coecients as well as the
general periodic equations
geometry of the domain, is analyzed. Some lower and upper bounds for the minimal speed when
the advection term
is large are given in [6, 11, 15, 33, 52, 67]. Lastly, let us add that some pre-
viously mentionned works, as well as other ones, [5, 13, 22, 24, 41, 42, 48, 51, 52, 53, 65, 91, 93],
were also devoted to other types of nonlinearities (combustion, bistable), for which the speed of
propagation of fronts may be unique.
One of the diculties and specicities of problem (10-11) with a nonlinearity
(7-8) is that
may now be negative at some points
x,
satisfying
whereas it is positive at other places, for
u. Besides the existence of pulsating fronts and the variational characterization

t
similar formula for c was given by Weinberger in [92], with a dierent approach,
the same value of
of the minimal speed, Theorem 1 above also gives the monotonicity of all fronts in the variable
(notice that a
but the monotonicity of the front was a priori assumed there).

Consider now a nonlinearity
and
fu (x, 0) = (x) + B(x), where

parameter. If 1 < 0 (namely if the
satisfying (7-8), and such that
are given periodic functions and
is a positive
hypothesis for conservation is satised), one calls
c (B)
the minimal speed, given in Theorem 1,
of the pulsating fronts solving (10-11). The following theorem especially gives a monotonous
dependency of
0, B
c (B)
on
as well as some lower and upper bounds for large or small
(when
can then be viewed as the amplitude of the eective birth rate of the species in
consideration). Furthermore, Theorem 2 below also deals with the inuence of the heterogeneity
of
on the minimal speed of pulsating fronts.
6 Hudson
and Zinner proved the existence of one-dimensional pulsating fronts for problems of the type
ut uxx = f (x, u), provided c c , but did not actually prove that c was the minimal speed.
61
Theorem 2
Assume that
is a constant symmetric positive matrix and assume that
(7-8) and that fu (x, 0) is of the type

C 1, functions, and B R.
a) Assume that
B>0
max > 0.
fu (x, 0) = (x) + B(x),
where
and
Then the hypothesis for conservation (1
satises
are given periodic
< 0)
is satised for
large enough and
c (B) 2
= 0
furthermore, if
is constant, then
p
eAe max( + B) ;
c (B)
is increasing in
(for
large enough so that
1 < 0).
Z
0,
b) Assume that
B>
satised for all
= 0 0
c)
C
0, and
and
max > 0.
eAe
|C|
Then the hypothesis for conservation is
c (B)
B>0
B > 0,
is increasing in
is constant. Furthermore, for all
and
Z
under the additional assumption that
p
c (B)
(B 1 (x) + (x))dx
2 eAe max(B 1 + )
B
C
1
c (B)
c (B)
eAe max lim inf
lim sup
2 eAe max .
B+
2
B
B
B+
Z
Assume that 0, fu (x, 0) = B(x) with
0, max > 0. One has
C
c (B)
lim
=2
B0+
B
Theorem 2 implies especially that, when
eAe
|C|
fu (x, 0)
Z
(x)dx.
C
is of the type
(x) + B(x),
and no matter
how bad the environment may be elsewhere, it suces to have a very favorable (even quite
narrow) zone (namely
>0
somewhere) to allow for species survival and to increase the speed
of propagation
Z of fronts. Furthermore, the speed is comparable to
for large amplitudes
as
> 0.
soon as, say,
c [] the minimal speed of pulsating travelling fronts solving he problem (10-11)

fu (x, 0) = (x), provided the assumption for conservation is satised. From the previous
Lastly, call
with
theorem, one immediately deduces the following corollary :
Corollary 1
(7-8), with
Assume that
is a constant
Z symmetric positive matrix and assume that
0 |C| with 0 > 0. Then f
fu (x, 0) = (x). Assume that
satises
satises the hypothesis
for conservation and
c [] c [0 ] = 2
p
(eAe)0 .
This corollary simply means that the heterogeneity of the medium increases the speed of
propagation of pulsating fronts, in any given unit direction of
RN .
As already underlined, the main dierence with the results in [13], in the existence and
monotonicity result (Theorem 1), is that the function
The nonnegativity of
here is not assumed to be nonnegative.
played a crucial role in [13], where the existence of the minimal speed
was proved by approximating
with cut-o functions, as in [24]. Although we solve some
regularized problems in bounded domains as in [13], the method used in this paper is rather
62
2. Existence result
dierent since we directly prove that the set of possible speeds
is an interval which is not
bounded from above, and we dene c as the minimum of this interval. Existence of pulsating
fronts is proved in Section 2. Monotonicity is proved in Section 3. Lastly, the characterization
of
is given in Section 4, as well as the eects of the heterogeneity of the medium on the
propagation speeds.
2 Existence result
This section is devoted to the proof of the existence of pulsating fronts for (10-11) for large
speed. Throughout this section, one assumes that the hypothesis for conservation is satised,
namely that there exists a (unique) positive bounded solution
2.1
of (2), which is periodic.
Existence result in nite cylinders for a regularized problem
Let us make the same change of variables as Xin [93] and Berestycki, Hamel [13]. Let
(s, x)
be the function dened by :

(s, x) = u
for all
sR
and
The function

sxe
,x
c
x RN , where u is a classical solution of (10-11).

satises the following degenerate elliptic equation
x (A(x)x ) + (eA(x)e)ss + x (A(x)es )
(13)
+s (eA(x)x ) cs + f (x, ) = 0
in
DL (R R )
together with the periodicity condition
Moreover, since
u(t, x) 0
as
is L-periodic with respect to
x.
(14)
x e and u(t, x) p(x) 0 as x e +, locally in t

RN which are orthogonal to e, and since is L-periodic with
and uniformly in the directions of

respect to
x,
one gets
(, x) = 0,
Conversely, if
then
(+, x) = p(x)
is a solution of (13-15) such that
uniformly in
x RN .
u(t, x) = (x e + ct, x)
(15)
is
C1
in
t, C 2
in
x,
is a classical solution of (10-11).
Let
and
be two positive real numbers, and set
a = (a, a) RN .
As it was done in [13],
one rst works with elliptic regularizations of (13) of the type
L + f (x, )
is L-periodic w.r.t. x,
x RN , (a, .) = 0, (a, x)
where
is the elliptic (in the
(s, x)-variables)
p(x),
in
a ,
(16)
operator dened by
x (A(x)x ) + (eA(x)e + )ss

+x (A(x)es ) + s (eA(x)x ) cs .
We will follow the scheme as in [13] to prove the existence of solutions of (16) and state some
of their properties, only indicating the dierences which may appear.
Let us establish at rst the
63
Lemma 1
For each
PROOF. Let
satises
c R,
c C 2 a
(s, x) = p(x)
be the function dened by
s+a
.
2a
of (16).
(A(x)p) f (x, p) = 0
s+a
L v = f (x, v + ) 2a f (x, p)
+(2a)1 [2A(x)e p + (A(x)e) p cp]
v is L-periodic w.r.t. x,
v(a, .) = 0, v(a, .) = 0.
Using the fact that

L-periodic and
C 1 ),
f (x, p), p, A, p
(17), in the distribution sense, in
and
Lemma 2
a .
x (Ae)
in
a ,
are globally bounded, (since
(17)
and
are
The function
of the rst equation of
Then, from the regularity theory up to the boundary, this
is a classical solution of (17) in
classical solution of (16).
2
in C
Then, since
one can follow the proof of Lemma 5.1 of [13]. Namely, using Lax-Milgram
Theorem with Schauder xed point Theorem, one can nd a solution
solution
v = .
One sets
N
in R , the problem (16) is equivalent to
a .
dened above is increasing in
= v + C 2 a
is a
and it is the unique solution of
(16)
Finally, the function
0 < c (s, x) < p(x) in a . Since f 0 in RN (, 0],

c
N
yields that > 0 in (a, a) R . Let us show that
PROOF. One has to show at rst that

the strong elliptic maximum principle
c (s, x) < p(x).

Set

= inf , p(x) > c (s, x)
for all

(s, x) a .
p > 0 and p is L-periodic with respect to x, there exists > 0 such that p > in a .
Therefore does exist. Moreover since (a, x) = p(x), 1. One has to show that = 1.
c
Assume > 1. By continuity, one has p in a . On the other hand, there exists a
c
sequence n , n < and a sequence (sn , xn ) in a such that n p(xn ) (sn , xn ). Since p
c
and are L-periodic in x, one can assume that xn C . Up to the extraction of a subsequence,
one can also assume that (sn , xn ) (s1 , x1 ) [a, a] C . Passing to the limit n , one
c
obtains p(x1 ) = (s1 , x1 ).
Next, set z = p . One has f (x, p) f (x, p) since f (., s)/s is supposed to be
c
decreasing in s. Thus L ( p) + f (x, p) 0. As a consequence, L z + f (x, p) f (x, ) 0.
Since
Hence,

where
L z + bz 0
z0
is a bounded function (because
p(x)
p(x)
in
a ,
(18)
is globally lipschitz-continuous). Moreover, one has
>
>
c (a, x) = 0,
c (a, x) = p(x),
(19)
x in RN (the last inequality follows from the assumption > 1 and from the positivity
N
of p in R ). Therefore, the point (s1 , x1 ) where z vanishes, lies in (a, a) C . Using (18) with
the strong maximum principle, one obtains that z 0, which contradicts (19).
= 1 and c p. Using again the strong maximum principle, one obtains the
Thus
c
N
inequality (s, x) < p(x) for all (s, x) [a, a) R .
for all
In order to nish the proof of the lemma, we only have to follow the proof of Lemma 5.2 given
in [13], which uses a sliding method in
64
(see [24]), replacing
c (a, x) = 1
by
c (a, x) = p(x).
2. Existence result
Lemma 3
sense : if
c
The functions are decreasing and continuous with respect to c in the following
0
> c0 , then c < c in a and if cn c R, then cn c in C 2 (a ).
PROOF. The proof is similar to that of lemma 5.3 in [13].

In the following, for any
C2
> 0, a > 0
and
c R,
we call
c,a
the unique solution of (16) in
Set
p = min p(x) = min p(x) > 0

xRN
xC
Lemma 4
There exist
a1
and
p+ = max = max p(x) > 0.

xC
(0, 1],

p
c
(c > K) max ,a (0, x) <
.
2
xC
and
such that, for all
a a1
xRN
and
n 2 be an integer and g be a C 1 function dened on [0, np+ ], such that g(0) = 0,

g(np ) = 0, g(u) > 0 on (0, np+ ), g 0 (np+ ) < 0. For n large enough, one can choose (using
N and u [0, np+ ].
hypothesis (8)) g such that f (x, u) g(u) for all x R
1
Then, using a result of [21], one can assert that there exists c such that the one-dimensional
PROOF. Let
+
problem
v 00 kv 0 +
g(v)
=0
0
in
v() = 0 < v(.) < v(+) = np+
k=
c1 , and let
that
is
and
p
v(0) =
,
2
(20)
k c1 > 0
(remember that 0 > 0 is given in (6)). Set

v = v(s) be the unique solution of (20) associated to k = c1 . It is also known
increasing in R. Take c > max {(eA(x)e + 1)k + (A(x)e)}, and (0, 1]. One has
admits a unique solution
v,
R,
for each
xRN
L v + f (x, v)
(eA(x)e + )v 00 (s) + ( (A(x)e) c)v 0 (s) + f (x, v(s))
{(eA(x)e + )k + (A(x)e) c} v 0 (s)

(eA(x)e + )g(v(s))/0 + f (x, v(s)),
from (20). Thus
L v + f (x, v) {(eA(x)e + )k + (A(x)e) c} v 0 (s) g(v(s))/0

eA(x)e 0 and f (x, u) g(u) for all (x, u) RN [0, np+ ]. Moreover g 0, v 0 > 0
(eA(x)e + )k + A(x)e c < 0, owing to the choice of c. As a consequence, one gets
since
L v + f (x, v) < 0
in
and
a .
v and c,a , and by using the monotonicity

v(a) > p+ p(x) for all x RN and for a large enough,
By using a sliding method as in Lemma 2, with

of
and the fact that
v(a) > 0
and
one can conclude that
c,a (s, x) < v(s),

for all
(s, x) a
and for
large enough.
65

Hence, it follows that
max c,a (0, x) = max c,a (0, x) < v(0) =
xRN
for
xC
p
,
2
c > max {(eA(x)e + 1)k + (A(x)e)} and a large enough, which completes the proof of the
lemma.
RN
Let us now consider the functions

Let us pass to the limit
the functions
0<
0,an
n +.
0,a ,
associated to
c = 0.
Take a sequence
an +.
From standard elliptic estimates and Sobolev's injections,
converge (up to the extraction of a subsequence) in
1, to a function 0 which satises
2,
Cloc
(R RN ),
for all
L 0 + f (x, 0 ) = 0 in R RN ,
0 is L-periodic w.r.t. x,
0
is nondecreasing w.r.t. s,
with
c = 0.
Furthermore,
0 0 (s, x) p(x)
for all
(s, x) R RN .
One then has the
Lemma 5
There exist
PROOF. Assume
x1 C
and
0 (0, x) < p(x)
N0 N
for all
such that for all
x C.
n N0 , 0,an (0, x1 ) >
p
.
2
Then
0 max 0 (0, x) < p+ .

xC
Let
(sn )nN
be a sequence in
(a, a),
such that
1
1
max 0,an (sn , x) = b := p+ + max 0 (0, x).
2
2 xC
xC
Let us show that
sn > 0
for
large enough.
Assume by contradiction that there exists a subsequence
k N.
Then, since
0,ak
is increasing in
ak +
such that
sk 0
for all
s,
b = max 0,ak (sk , x) max 0,ak (0, x),

xC
xC
while
max 0,ak (0, x) max 0 (0, x)

xC
Passing to the limit
k +,
as
k +.
xC
one obtains
b max 0 (0, x).

xC
That leads to a contradiction since
p+ > max 0 (0, x).
Therefore, one has shown that
xC
large enough.
Set
an (s, x) = 0,an (s + sn , x),
dened on
(an sn , an sn ) RN .
max an (0, x) = b.
xC
66
Then
sn > 0
for
2. Existence result
One easily sees that
an sn
as
n +.
Then two cases may occur, up to the extrac-
tion of some subsequence :
an sn +.
case 1 :
functions
an
From standard elliptic estimates and Sobolev's injections, the
converge (up to the extraction of a subsequence) in
to a nonnegative function
2,
Cloc
(RRN ), for all 0 < 1,
satisfying
L + f (x, )
max (0, x)
p(x)
b.
in
R RN
(with
c = 0),
(21)
xC
Moreover, from standard elliptic estimates, from the monotonicity of

in
x,
, and from the periodicity
it follows that
(s, x) (x)
where each function
in
C 2 (RN )
as
s ,
satises
(A(x) ) + f (x, ) = 0
is L-periodic,
0.
in
RN ,
(x) 0 or
(x) p(x). Moreover, 1 (x) 1 (0, x) because of the monotonicity of 1 with respect to s.
+
Therefore 1 (x) b < p . Thus, 1 (x) 6 p(x), and 1 0. Similarly, 1 (x) 1 (0, x). Thus
+
+
there exists x0 C such that 1 (x0 ) b > 0. As a consequence, 1 p.
From the uniqueness Theorem 2.1 of [17], one can conclude that either
Next, multiply the equation (21) (with
c = 0) by s
and integrate it over
(N, N )C , where
is a positive real number. Then,
(N,N )C
(eA(x)e + )ss s dsdx +

x (A(x)x )s dsdx
(N,N )C
Z
+
{x (A(x)es ) + s (eA(x)x )} s dsdx
(N,N )C
Z
+
f (x, )s dsdx
(22)
0.
(N,N )C
First, one has
Z
(eA(x)e + )ss s dsdx =
(N,N )C
1
2
From standard elliptic estimates, one knows that
N +
Z
C

N
(eA(x)e + )(s )2 N dsdx.
s 0
as
s +.
(23)
in (23), one obtains
Z
(eA(x)e + )ss s dsdx = 0.
(24)
RC
67

Next, using an integration by parts over
to
x,
(N, N ) C
and the periodicity of
with respect
one has
Z
x (A(x)x )s dsdx
(N,N )C
x s A(x)x dsdx
(N,N
)C
Z
=
=
since the matrix eld
A(x)
1
(x A(x)x )s dsdx
2 Z(N,N )C
1
[x A(x)x ]N
N dsdx,
2 C
is symmetric. Passing to the limit
N +
(25)
in (25), one obtains,
using standard elliptic estimates :
1
x (A(x)x )s dsdx =
2
RC
Next, from the periodicity of
with respect to
x,
Z
p (A(x)p)dsdx.
(26)
C
one can similarly show that
Z
{x (A(x)es ) + s (eA(x)x )} s dsdx = 0.
(27)
RC
Z
Set
F (x, u) =
f (x, s)ds.
Then,
f (x, )s dsdx =
RC
F (x, (s, x))s dsdx =
F (x, p(x))dx.
N + in (22), and using (24), (26), (27) and

Z
1
F (x, p(x)) p (A(x)p) dx = 0.
2
C
Moreover, using a property on the energy of
(28)
RC
p,
(28), one gets
(29)
which has been established in Proposition 3.7 of
[17], one asserts that

Z
1
F (x, p(x)) p (A(x)p) dx =: E(p) > 0.
2
C
The latter is in contradiction with (29), therefore case 1 is ruled out.
an sn d < +. Up to the extraction of some subsequence, the functions an

2,
Cloc
((, d) RN ) (for all 0 < 1) to a function satisfying (21), with c = 0,
N
in the set (, d) R . Moreover, the family of functions (an ) is equi-Lipschitz-continuous in
any set of the type [an sn 1, an sn ] C . Therefore, for all > 0, there exists > 0 such
case 2 :
converge in
that
x C, n, s [an sn , an sn ], p(x) an (s, x) p(x).

Then choose
x0 C
such that
p(x0 ) = p+ .
Formula (30) applied on
max an (0, x) = b < p+

xC
implies that
an sn >
for some
> 0.
Hence
d>0
and
max (0, x) = b.
xC
68
x0
together with
(30)
2. Existence result
Moreover (30) implies that
can be extended by continuity on
{d} RN
with
Furthermore, from standard elliptic estimates up to the boundary, the function
C 1 (, d] RN
(d, x) = p(x).
is actually in
Following the proof of case 1, one shows that
(, .) 0.
The next steps are similar to
those of case 1. One gets a contradiction.
0 (0, x1 ) = p(x1 ). Hence, taking any sequence

2,
0
an + such that the sequence (,an ) converges in Cloc
(R RN ) for all 0 < 1, there
p
0
exists N0 such that for all n N0 , ,a (0, x1 ) >
. That completes the proof of Lemma 5.
n
Therefore, there exists
x1 C
such that
Finally, one gets
Proposition 1
R, N 1 N
Fix
(0, 1]. Let an + be the sequence dened above. Then, there exist K
n N1 there exists a unique real number c = c,an such that c,an
such that for all
satises the normalization condition
max c,an (0, x) = max c,an (0, x) =
xRN
xC
p
.
2
(31)
Furthermore,
0 < 1, n N1 , 0 < c,an < K.
(0, 1]. Under the notations of the two preceding lemmas, let us
> a1 and N1 > N0 . It follows from this lemmas that for each n N1 ,
c K, max c,an (0, x) <

,
2
xRN
.
for c = 0, max 0,an (0, x) >
2
xRN
PROOF. Fix
that
aN1
On the other hand, Lemma 3 yields that, for each
n N1 ,
dene
N1
such
the function
(c) = max c,an (0, x)

xR
is decreasing and continuous with respect to
2.2
c.
Therefore the proposition follows.
Passage to the limit in the innite domains
Using the result of Proposition 1, we are going to pass to the limit

cylinder
R RN
Proposition 2
for the solutions
,an
c,an
n +
in the innite
satisfying (31).
Under the notations of Proposition 1, one has
> 0, 0 < c :=
lim inf
n+,nN1
c,an K.
69

0 c K . Up to the extraction of a subsequence, one
,an
2,
n + and c,an in Cloc
(R RN ), for all 0 < 1, where
PROOF. From Proposition 1, one has

,an
can assume
as
satises
with
c = c
L + f (x, ) = 0 in R RN ,
and
max (0, x) =
xRN
p
.
2
Then, following the calculus of Lemma 5, case 1, one can assert that
(, x) = 0, (+, x) = p(x)
Up to the extraction of some subsequence,

1), to a function such that, in R RN ,
,an
c,an
converge in
x (A(x)x ) + (eA(x)e + )ss + x (A(x)es )
+s (eA(x)x ) cs + f (x, )
max (0, x)
xRN
is increasing w.r.t. s.
Furthermore,
(, x) = 0
and
(32)
c > 0.
Proposition 3
0<
and
from Proposition 3.7 of [17]. Therefore
all
x R,

Z
1
(s ) dsdx =
F (x, p) p (A(x)p) dx = E(p) > 0,
2
RC
C
for all
(+, x) = p(x)
for all
2,
Cloc
(R RN )
(for
= 0,
=
p
,
2
x RN .
PROOF. The convergence follows from the same arguments that were invoqued in the preceding
c,an is increasing in s. The

propositions. Moreover, in nondecreasing w.r.t. s because each ,a
n
limits
(, x) = 0
and
(+, x) = p(x)
can be proved in the same way as in Lemma 5,
case 1, using
max (0, x) =
xRN
and the fact that
s.
h > 0,
p
,
2
in nondecreasing in s. The only thing that it remains to prove is that
is
increasing in
For any
the function
(s + h, x) (s, x)
is a nonnegative and nonconstant solution
of a linear elliptic equation with bounded coecients. It follows then from the strong maximum
(s + h, x) (s, x) > 0,
increasing in the variable s.
principle that
is
2.3
for all
(s, x) R RN .
Our rst aim is to prove that the real numbers

constant.
70
That proves that the function
are bounded from below by a positive
2. Existence result
Proposition 4
Under the notations of Proposition 1, one has
0 < lim inf c K.

0
> 0, one has 0 < c K . Assume that there exists a

0 as n +. In the sequel, for the sake of simplicity,
u (t, x) = (x e + c t, x). Then u is a classical solution of
PROOF. From Proposition 1, for each
sequence n 0, n > 0 such that c n
we drop the index
n.
Set

u + x (A(x)x u ) ut + f (x, u ) = 0 in R RN ,
)2 tt
(c

N
Y
ke
k
Li Z, u t +
, x = u (t, x + k) in R RN ,
i=1
u (t, x) 0 as t , u (t, x) p(x) as t +.
(33)
0 < u (t, x) < p(x) for all (t, x) RRN . Lastly, since is increasing in the variable s
> 0, each function u is increasing in the variable t.
Moreover,
and c
Up to the extraction of some subsequence, as it was said in [13] (Proposition 5.10), three
cases may occur :
(0, +), 2 +
2
(c )
(c )
0
(c )2
or
as
0+ .
Let us study the :
(0, +). Let x0 RN be such that x0 N

i=1 Li Z and x0 e > 0.
2
(c )
Since u (t, x0 ) 0 as t and u (t, x0 ) p(x0 ) as t + from our assumptions on u ,
one can assume, up to translation with respect to t, that u (0, x0 ) =

.
2
Since
, standard elliptic estimates imply that the functions u converge (up to
(c )2
2,
N
extraction of some subsequence) in Cloc (R R ) (for all 0 < 1) to a function u satisfying

utt + x (A(x)x u) ut + f (x, u) = 0 in R RN ,
0 u p, ut 0 in R RN ,
case 1 : Assume
and
u(0, x0 ) =
p
.
2
Now, x any
suciently small. Thus, as
u (0, x0 )=
p
2
B R.
is increasing in
Since
t,
. Therefore, passing to the limit
B > 0,
c 0+
one has
0,
and
u (B, 0)
x0 e
c for
x0 e
x0 e
u ( c , 0). But u ( c , 0) =
x0 e > 0, B <
one obtains
u(B, 0)
p
.
2
(34)
u+ be the function dened in RN by u+ (x) = limt+ u(t, x). This function can be dened since u is bounded and nondecreasing in t. From standard elliptic estimates, the convergence
2
N
+
holds in Cloc (R ), and u solves
Let
(A(x)u+ ) + f (x, u+ ) = 0
in
RN
(35)
71

with
0 u+ (x) p(x)
for all
x RN .
But it follows from our hypotheses on
and from Theo-
rems 2.1 and 2.3 of [17] that the equation (35) admits exactly two nonnegative solutions, which
are
and
p.
u(0, x0 ) =
+
u (0) p2 . As a
Therefore, as
However, (34) gives
p
2 and
ut 0, one has u+ (x0 ) p2 > 0, thus u+ p.

+
consequence, u cannot be equal to p and case 1 is ruled
out.
+. As it was done in [13] (Proposition 5.10),

(c )2

= (c / )t. The function v (, x) = u

,
x
satises
c
case 2 : Assume that

change of variables
one makes the
+ (A(x) v )
v
v + f (x, v ) = 0 in R RN ,
x
x

N
Y
ke
k
Li Z v + , x = v (, x + k) in R RN ,
i=1
v (, x) 0 as , v (, x) p(x) 0 as +.
Moreover,
0 < v < p
and
is nondecreasing with respect to
Furthermore, as it was
QN
p
N
done in case 1, one can assume that v (0, x0 ) =
i=1 Li Z
2 for some x0 R such that x0
and x0 e > 0. Since c / 0 , the functions v converge (up to the extraction of some
2,
N
subsequence) in Cloc (R R ) (for all 0 < 1) to a function v which satises

and
v(0, x0 ) =
v + x (A(x)x v) + f (x, v)
0 v p, v
0
0
in
in
R RN ,
R RN ,
p
2 . Moreover, as in case 1, one can show that
B > 0,
By dening
v + (x) := lim + v(, x),
v(B, 0)
p
.
2
one can also obtain a contradiction the same way as
in case 1.
case 3 : Assume that

limit
0,
0.
(c )2
The elliptic operators in (33) become degenerate at the
and one cannot use the same arguments as in cases 1 and 2.
In order to pass to the limit
0,
let us state two new inequalities on
u .
Using the same calculations as those which were used to prove (29) and (32), and making
the change of variables
t=
sxe
,
c
one obtains
(ut )2 = E(p).
(0, 1),
RC
Therefore, the periodicity condition in (33) gives us that
(ut )2 = (2n)N E(p).
(0, 1), n N,
(36)
R(nL1 ,nL1 )(nLN ,nLN )

Similarly, multiplying equation (33) by
and
u ,
one gets the existence of
Z
(0, 1), n N,
such that
f (x, u ) + |x u |2 (2n)N .
(37)
R(nL1 ,nL1 )(nLN ,nLN )

Next, using Theorem A.1 of [13] (see also [14]), one has the following a priori estimate :
72
2. Existence result
Lemma 6
M,
There exists a constant
which does not depend on
such that the function
solving (33) satises
|x u | M
for
in
R RN
(38)
small enough.
The above estimates were proved in [14] with a Bernstein-type method. This method uses the
|x u |2 , and it requires that the func3

C . The latter is true here because of the smoothness assumptions on A and f .
maximum principle applied to some quantities involving
tions u are of class
From (36) and (38), and arguing as in [13] (Proposition 5.10), we obtain that
(up to the extraction of some subsequence) almost everywhere in
1 (R
Hloc
RN ) and
(u , ut , x u ) * (u, ut , x u)
for every compact subset
2
in Lloc
RN
strong.
converges
RN to a function
L2 (K),
in
K R RN . From (36) and (38), one can actually assume that u u

Moreover, 0 u p(x), ut 0 and from (36), (37) and (38),
Z
|x u|2 + (ut )2 C(K1 ),
(39)
RK1
for every compact subset
RN .
K1
u
From parabolic regularity,
is then a classical solution of
ut x (A(x)x u) f (x, u)
0 u p, ut
Moreover, one can assume, up to a translation in
t,
0
0
in
in
R RN ,
R RN .
that
u (t, x + x0 )dxdt = |C|
> 0,
(0,1)C
p
.
2
(41)
x0 N
i=1 Li Z such that x0 e > 0. Since c 0 and u is increasing in
all B R, and for suciently small,
x0 e
(t, x) (0, 1) C, u (B + t, x) u (t + , x) = u (t, x + x0 )
c
for some
that for
(40)
(0, 1) C and passes to the limit 0+ .

u * u in L2loc R RN weak,7 one obtains
Z
p
B R,
u(B + t, x)dxdt |C| .
2
(0,1)C
from (33). Next, one integrates over

and the fact that
Using the monotonicity of
in
t,
let us dene
u+ (x) = lim u(t, x).

t+
t, it follows
By using (41)
Then, one has
(42)
u+ 0
x (A(x)x u+ ) + f (x, u+ ) = 0. Therefore, as it was stated in Theorems 2.1 and 2.3 of [17],
u+ 0 or u+ p. But (41), passing to the limit 0 and t +, rules out the case u+ 0.
+
Hence u p. Next, using (42), and since u is nonincreasing in t, one obtains
Z
p
p(x)dx |C|
2
(0,1)C
and
which is impossible. The proof of Proposition 4 is complete.
7 This
convergence is actually strong.

73

2.4
(c1 , u1 )
Let us choose a subsequence
(x e + ct, x).
such that
c c1 > 0.
For each
As it was done in case 3 of Proposition 4, the functions
1
extraction of some subsequence), in Hloc (R
1
solution u of (40).
set
u (t, x) =
converge (up to the
RN ) weak, and almost everywhere, to a classical

Q
ke
u1 t + 1 , x = u1 (t, x + k) for all k N
i=1 Li Z and for all (t, x) R
c
B > 0 and every compact set K1 in RN , one has,

2

Z
ke
u t + 1 , x u (t, x + k)
c
(B,B)K1

2
Z
ke
ke
=
u t + 1 ,x u t + ,x
,
c
c
(B,B)K1
Let us prove that
RN .
For all
whence
Z
(B,B)K1
from (39). Therefore, by
almost everywhere in

2
ke
u t + 1 , x u (t, x + k)
c

Z

ke ke 2
ke ke 2
2

(ut )

C(K1 )
c1
c
c1
c
RK1
passing to the limit 0, we obtain

ke
1
u t + 1 , x = u1 (t, x + k)
c
R RN .
u1
Since
is continuous, the equality holds for all
(43)
(t, x) R RN .
1
In order to obtain our result, one has to prove that u (t, x) 0 as x e and
1
1
u (t, x) p(x) 0 as x e +, locally in t. Since u veries (43), and c > 0, it is equivalent
1
1
to prove that u (t, x) 0 as t and u (t, x) p(x) as t +, locally in x.
As it was done in the proof of Proposition 4, case 3, one can assume, up to a translation in t,
that
u (t, x)dxdt = |C|
,
(0,1)C
Since
u1
is bounded and nondecreasing with respect to
t,
p
.
2
(44)
one can dene
u (t, x) = lim u1 (t, x).

t
As done above, one knows that u satises
(A(x)u )+f (x, u ) = 0 and one has 0 u p.
As already said, this equation admits exactly two nonnegative solutions, which are not larger
than
p,
namely
and
p.

to
t,
in (44), and using the fact that
one has
p
,
2
(45)
u (x) |C|
p
.
2
(46)
C
One then easily concludes from (45) that
from (46), u is not equal to
p, thus u
u+
is not equal to
74
0,
and therefore
u + p,
and
0.
From strong parabolic maximum principle, one obtains that

tence result follows.
is nondecreasing with respect
u+ (x) |C|
C
and
u1
u1
is increasing in
t.
The exis-
2. Existence result
2.5
Proposition 5
and
c > c1 ,
For each
for all
c > c1
of (10-11), associated to the speed
c,
t.
is increasing in
1
PROOF. Set (s, x)
(c, u)
=u
sxe
,x
c

, and, as before, dene
by
L = x (A(x)x ) + (eA(x)e + )ss + x (A(x)es ) + s (eA(x)x ) cs .

Then, as it was done in [13] (Proposition 6.3), using Krylov-Safonov-Harnack type inequalities
applied to
v = t u1 ,
one gets the existence of a constant
such that
|tt u1 | Ct u1
in
R RN ,
whence
L 1 + f (x, 1 ) = 1ss + (c1 c)1s < 0

for
>0
small enough. In what follows, let
For any
a R+
and
R,
>0
in
R RN ,
(47)
be small enough so that (47) holds.
set
h = min 1 (a + , .).
C
With a similar method as in Lemma 1, one can show the existence of a solution
C 2 a
of
the following problem :
L + f (x, )
p(x)
(a, x) = h + , (a, x)
p
p(x)
<
p+
for all
one has, from the strong maximum principle,
1 (a
in
a ,
(48)
+ , x)
(s, x) a .
> 0
in
First, since
a .
p(x)
= sup > 0, > h +
p
f (x, u) = 0
for all
u 0,
Therefore, one can dene
for all
RN .

in
a .
< 1. As in the proof of Lemma 2, using the fact that h p(x)/p+ < (a, x)
N (since p(x)/p+ 1 and 1 is increasing w.r.t. s), one gets the existence of
for all x R
(s , x ) (a, a) C such that h p(x)/p+ (s, x) for all (s, x) [a, a] RN , with

N
equality at (s , x ) (a, a) R . On the other hand,
Assume that
L (h
since
h /p+ < 1,
and since
p
h
p
) = + L (p) > f (x, h + ),
p+
p
p
f (., s)/s
is decreasing in
from our hypothesis on
f.
That leads
to a contradiction as in Lemma 2.
Therefore,
1,
whence
h p/p+ ,
and the strong maximum principle yields
(s, x) a , h
(s, x) < p(x) for all (s, x) a .

enough. Let k be the smallest k such
Similarly, one can easily show that
k, x) (s, x)
in
for
large
p(x)
< (s, x).
p+
(49)
Therefore,
1 (s + +
that the latter holds.
75

From the boundary conditions in (48), one knows that
1
necessarily follows that (s +
is increasing in
+ k, x) (s, x)
k 0.
k > 0. By continuity, it
1
point (s, x) a . Since
Assume
with equality at a
s,
1 (a + + k, ) > 1 (a + , ) h h
p()
= (a, ),
p+
1 (a + + k, ) > 1 (a + , ) = (a, ). Therefore (s, x) (a, a) C (one can assume this

1
1
using the L-periodicity in x of and ). But, from (47), it is found that (s + + k, x) is a
1
supersolution of (48). Therefore, the strong maximum principle implies that (s + + k, x) =
(s, x) in a . One gets a contradiction with the boundary condition at s = a. As a consequence,
k = 0 and one has
(s, x) a , (s, x) 1 (s + , x).
(50)
and
Since
is increasing in
s,
it also follows that
(s, x) < 1 (a + , x)
in
a .
As a conclusion, from (49) and (50), one has
(s, x) a , h
p(x)
< (s, x) < 1 (a + , x).
p+
is increasing in s and

C 2 a . Moreover, using the fact that the boundary conditions
for at s = a are increasing in , one can prove, as in Lemma 5.3 in [13], that the functions

are continuous with respect to in C 2 a and increasing in . But, since 1 (, x) = 0
1
N
and (+, x) = p(x) in R , it follows from (49) and (50) that 0 as uniformly
in a and that,
p
> 0, T, > T, (s, x) > + p in a .
p
Using the same sliding method as in Lemma 5.2 in [13], it follows that
is the unique solution of (48) in
Therefore, for each
a > 1, there exists a unique (a) R such that ,a := (a)
satises
,a (s, x)dsdx =
(0,1)C
Let
(for all
(p )2
|C| min(c, 1).
2p+
is nonincreasing with respect to
s,
satises
(s, x)dsdx =
(0,1)C
0 (s, x) p(x)
states that (s, x)
(x) as
2
in C (C). Moreover,
(A(x) ) + f (x, ) = 0
0 (x) p(x)
in
C.
in
R RN ,
and
(p )2
|C| min(c, 1).
2p+
From standard elliptic estimates, and from the monotonicity of
76
(51)
2,
an +. From standard elliptic estimates, the functions ,an converge in Cloc
(RRN )
0 < 1), up to the extraction of a subsequence, to a function satisfying

L + f (x, ) = 0 in R RN ,
(52)
is L-periodic w.r.t. x.
Moreover,
with
solves (48) and
in
with respect to
s,
are L-periodic and satisfy

C,
one
3. Monotonicity of the solutions

But, as one has said before, from Theorems 2.1 and 2.3 of [17], the former equation, together
0 (x) p(x), admits exactly two nonnegative solutions, which are
p. Since is nondecreasing in s, and from (51), one has

with the bounds
+ (x)dx
C
and
(x)dx
C
From (53) one deduces that
(p )2
|C| min(c, 1) > 0,
2p+
(p )2
|C| min(c, 1) <
2p+
+ p
and
(53)
Z
p(x)dx.
(54)
0.
u (t, x) = (x e + ct, x).
and from (54) one has
Coming back to the original variables
(t, x),
one denes
As it was
done in the proof of (36) and Lemma 6, it follows from (52) and from the limiting behavior of
as
that
. As it was done in subsection

1 (R RN ) such that (up to the extraction of a subsequence),
u Hloc
1 (R RN ). From parabolic regularity, u is then a classical solution of
Hloc

ut x (A(x)x u) f (x, u) = 0 in R RN ,
0 u p, ut 0 in R RN .
satises the estimates (39), independently of
2.4, there exists a function
u * u
weakly in
Moreover, as it was done in subsection 2.4, one still has

for all
QN
i=1 Li Z. Furthermore,
u(t + ke
c , x) = u(t, x + k)
in
R RN
satises
Z
u(t, x)dtdx =
{0<xe+ct<1, xC}
(p )2
|C| min(c, 1).
2cp+
u in t, that
u(t, x) u (x) locally in x as t , and that u solve (A(x)u ) + f (x, u ) = 0
N
in R . Moreover, 0 u p. From the monotonicity of u with respect to t, one can also assert
that
Z
(p )2
u+ (x)dx
|C| min(c, 1) > 0,
2cp+
C
One deduces from standard parabolic estimates and from the monotonicity of
and
(p )2
u (x)dx
|C| min(c, 1) <
2cp+
C
Z
p(x)dx.
C
u 0.
Finally, one deduces from the (t, x)-periodicity of u and the positivity of c that u(t, x) 0
as x e and u(t, x) p(x) 0 as x e +, locally in t. Thus (c, u) is a classical
solution of (10-11). Moreover, since ut 0, the strong parabolic maximum principle yields that
u is increasing in t.
That completes the proof of Proposition 5.
Therefore using the same argument as for
one concludes that
u+ p
and
3 Monotonicity of the solutions

We are going to establish the monotonicity result in Theorem 1, namely, each solution
of (10-11) is such that
speed
is increasing with respect to
t.
(c, u)
This will enable us to dene a minimal
in the next section.
One rst establishes the following lemma, which is close to Lemma 6.5 of [13]. Nevertheless,
its proof does not use the fact that
uses the property that
has a given sign (which is not true in general) and clearly
is an unstable solution of the stationary problem.
77
Lemma 7
Let
(c, u)
c>0
be a classical solution of (10-11). Then
0 < :=
lim inf
t, xC
and
ut (t, x)
< +.
u(t, x)

sxe
, x . From standard paPROOF. Let us rst prove that c > 0. Set (s, x) = u
c
rabolic estimates, ut (t, x) 0 as t , x u(t, x) 0 as t and x u(t, x) x p(x) as
t +. Therefore s (s, x) 0 as s , x (s, x) 0 as s and x (s, x) x p(x)
as s +. Then, arguing as in Lemma 5 (case 1), one can prove that
Z
c
(s )2 = E(p) > 0.

RC
Therefore
c > 0.
Next, as it was done in [13] (Lemma 6.5), one can assert, using standard interior estimates,
(t, x)-periodicity of u, that ut /u and u/u are globally boun be dened as in the stating of the above lemma. Then is a nite real number.
(tn , xn ) be a sequence in R C such that tn and
Harnack type inequalities and the

ded. Let
Let
ut (tn , xn )/u(tn , xn )
as
n +.
Up to the extraction of some subsequence, one can assume that

Now set
wn (t, x) =
From the boundedness of
ut /u
and
u/u,
xn x C
as
n +.
u(t + tn , x)
.
u(tn , xn )
one can assert that the functions
wn
are locally
bounded. Moreover, they satisfy
t wn (A(x)wn )
f (x, u(t + tn , x))

wn = 0
u(t + tn , x)
From standard parabolic estimates, the positive functions

of some subsequence, to a function
w ,
wn
u(t, x) 0
as
locally in
R RN .
converge, up to the extraction
which is a nonnegative classical solution of
t w (A(x)w ) fu (x, 0)w = 0

since
in
x.
Moreover,
in
R RN ,
w (0, x ) = 1, thus w is positive from

w (t + k e/c, x) = w (t, x + k) for all
the strong parabolic maximum principle. One also has
(t, x) R RN
and for all
QN
i=1 Li Z.
w (t, x)et
not depend on t. Indeed, one clearly has (w )t /w , and (w )t (0, x ) = w (0, x )
the denition of (tn , xn ). Therefore, the function z = (w )t /w satises
Then using the arguments of the Lemma 6.5 of [13], one can check that
t z (A(x)z) 2
w
z = 0
w
z and z(0, x ) = , and w /w is bounded. The

z ; in other words, w (t, x)et does not depend on t.
with
78
in
does
from
R RN
strong maximum principle yields

C 2 (RN )
Therefore the
function
(x) = w (0, x)e(xe)/c
is positive and
L-periodic.
Moreo-
ver, it satises
Lc, = 0,
where one has set
= /c,
Lc,
(55)
and
(A(x)) 2(eA(x))
[2 eA(x)e + (A(x)e) c + fu (x, 0)].
Now, from [13] (Proposition 5.7.1), one knows that for all
c () R
and a unique positive function
and
c in R,
there exists a unique
C 2 (RN ) such that
Lc, = c () in RN ,
is L-periodic, k k = 1.
That allows us to dene the function
7 c ().
(56)
Let us show that it is concave. First, using
the result 2) of Proposition 5.7 in [13], one has
Lc,
,
E

L periodic . Let E0
c () = max inf
RN
where

E = C 2 (RN ), > 0,
is
n
E0 = C 2 (RN ), E
Then,
c () = c + h()
with
be the set dened by
o
(x) = exe .
with

h() = max0 inf
E RN

(A(x))
fu (x, 0) .
h is concave. Let 1 , 2 R and t [0, 1]. Set = t1 + (1 t)2 .

One only has to show that h() th(1 ) + (1 t)h(2 ). Let 1 and 2 be two arbitrary chosen
0
0
functions in E and E respectively, and set z1 = ln(1 ), z2 = ln(2 ), z = tz1 + (1 t)z2 and
1
2
z
= e . It easily follows that E0 . Therefore
Our aim is to show that

h() inf
RN

(A(x))
fu (x, 0) .
Moreover,
(A(x))
fu (x, 0) = (A(x)z) zA(x)z fu (x, 0),
and
zA(x)z
tz1 A(x)z1 + (1 t)z2 A(x)z2

t(1 t)(z1 z2 )A(x)(z1 z2 )
since
tz1 A(x)z1 + (1 t)z2 A(x)z2 ,
0 t 1.
79

As a consequence,
(A(x))
fu (x, 0)
t[ (A(x)z1 ) z1 A(x)z1 fu (x, 0)]

+(1
t)[ (A(x)z2 ) z2 A(x)z2 fu (x, 0)]
(A(x)1 )
t
fu (x, 0)

1
(A(x)2 )
fu (x, 0) .
+(1 t)
2
Thus,

h() t inf
RN
and, since

(A(x)1 )
(A(x)2 )
fu (x, 0) + (1 t) inf
fu (x, 0) ,
1
2
RN
1 and 2 were arbitrarily chosen, one bets that h() th(1 )+(1t)h(2 ). Therefore
h is continuous. Thus 7 c () = c + h() is continuous and
is concave. This implies that
concave.
c (0) < 0. By denition, c (0) is the rst eigenvalue of the linear

(A(x)) fu (x, 0) with L-periodicity conditions. From the hypothesis for
conservation, it follows that c (0) < 0.
0
Finally, it remains to show that c (0) > 0. For each R, consider the positive and L2
N
periodic eigenfunction C (R ) for problem (56), associated to the eigenvalue c (). By
Next, let us show that
problem
denition, it satises the equation
(A(x) ) 2(eA(x) )
(2 eA(x)e + (A(x)e) c + fu (x, 0))
Multiply this equation by
periodicity of
and
0 ,
0 ,
and integrate it by parts over
and since the matrix eld
A(x)
1 () .
C.
One obtains, using the
Z
Z
(A(x)0 ) [ (A(x)e ) + eA(x) ]0
C
Z
C
2
( eA(x)e c) + fu (x, 0) 0
c ()
0 .
C
the equation satised by
0 ,
one obtains,
(57)
Z
=
C
Multiplying by
Z
[ (A(x)0 ) + fu (x, 0) 0 ] = c (0)
0 .
, one
Z
[ (A(x)e ) + eA(x) ]0
C
Z
(eA(x)e c) 0
C
Now, take an arbitrary sequence
n 0.
Since
gets
c () c (0)
c (n ) c (0),
(59)
Z
0 .
C
standard elliptic estimates,
and Sobolev injections imply, up to the extraction of some subsequence, that the functions
L-periodicity) in C 2, (for all 0 < 1)

= 1, 0 is L-periodic and satises
converge locally (and therefore uniformly by
such that
k 0 k
(A(x) 0 ) fu (x, 0) 0 = c (0) 0 .

80
(58)
Substituting (58) into (57), and dividing by
nonnegative function
L-
is symmetric,
n
to a

0 > 0, and by uniqueness (up to
0
normalization), = 0 , and the whole family n converges to 0 as n +. Therefore,
passing to the limit 0 in (59), one obtains that c is dierentiable at 0, and
Z
Z
Z
[ (A(x)e0 ) + eA(x)0 ]0 + c
02 = 0c (0)
02 .
From strong elliptic maximum principle, it follows that
C
From the
L-periodicity of 0 , and since the matrix eld A(x) is symmetric, one has
Z
Z
[ (A(x)e0 ) + eA(x)0 ]0 = [eA(x)0 A(x)e 0 ]0 = 0,
C
whence
0c (0) = c > 0.
coming back to our solution
proved.
of
7 c () is concave, with c (0) < 0 and 0c (0) > 0. Moreover,

(55), one has c (/c) = 0. Therefore > 0, and the lemma is
One can now

turn
sxe
u
, x . Then
c
to the proof of the monotonicity result in Theorem 1. Set
(s, x) =
ut (t, x)/u(t, x) = cs (x e + ct, x)/(x e + ct, x).

c>0
One knows from Lemma 7 that
and
lim inf
t, xC
ut (t, x)
> 0.
u(t, x)
Therefore,
lim inf
s, xRN
and, from the
L-periodicity
of
with respect to
s (s, x)
>0
(s, x)
x,
one can deduce that there exists
sR
such
that
s s, x RN , s (s, x) > 0.
Moreover
exists
inf
ss, xRN
BR
(s, x) > 0
such that
and
B s
(, x) = 0
uniformly in
x RN .
As a consequence, there
and
0, s B, x RN , (s, x) (s, x)
where one has dened
Fix now any
(s, x) = (s + , x).
0.
One can assume that
B 0.
Set

= inf ,
The real
(60)
is well dened since
in
is bounded and

[B, +) RN .
inf
sB, xRN
(s, x) > 0.
> 1. Since (s, x) p(x) > 0 as s + uniformly in x, with p bounded from

below, and since is L-periodic in x, there exists a point (s0 , x0 ) [B, +) C such that
(s0 , x0 ) = (s0 , x0 ).
Assume
81

in [B, +) RN by continuity and in (, B] RN by (60)

and because > 1. Coming back to the original variables (t, x), set z(t, x) = (x e + ct, x)
N
(x e + ct, x). Then z 0 in R R , moreover, z satises the following equation :
Furthermore,
zt (A(x)z) = f (x, ) f (x, ).

Therefore, using (7), one obtains
f (x, ) f (x, ).
Thus one has
zt (Az) f (x, ) f (x, ).

Therefore, there exists a bounded function
such that
zt (A(x)z) + b(x)z 0.
Furthermore, since
Besides,
(s0 , x0 ) = (s0 , x0 ),
setting
t0 =
(61)
s0 x0 e
, one has
c
z(t0 , x0 ) = 0.
is nonnegative, and satises (61) ; therefore, from the strong parabolic maximum
z(t, x) = 0 for all t t0 and x RN , whence

z(t, x) 0 in R RN since
QN
N
z t+
= z(t, x + k) for all (t, x) R R and k i=1 Li Z. One gets a contradiction
since z(t, x) ( 1)p(x) > 0 as t +.
N for all 0. One therefore gets that

Thus 1 for all 0, whence in R R
is nondecreasing with respect to s, and u is nondecreasing in t because c > 0. Finally, with the
same arguments as above, one can prove, using the strong parabolic maximum principle, that u
is increasing in t. That concludes the proof of the monotonicity result in Theorem 1.
principle, one has
ke
c ,x
Remark 1
Notice that this monotonicity result especially implies that any solution
(10-11) is such that
0 < u(t, x) < p(x)
for all
(c, u)
of
(t, x) R RN .
4 The minimal speed c

This section is devoted to the proof of the existence of a minimal speed of propagation of the
pulsating fronts, and some properties of this minimal speed with respect to the nonlinearity
4.1
As it was proved in Lemma 7, any solution
(c, u)
of (10-11) is such that
c > 0.
complete the proof of the rst part of Theorem 1 and to obtain the existence of a
that there exists a solution of (10-11) if and only if
c c .
Proposition 6
c c ,
c > 0
c < c .
There exists
while no solution exists for
such that, for
PROOF. First, assume by contradiction that there exists a sequence

functions
un
cn 0+
In order to
c > 0
such
of (10-11),
and some classical
such that
Again, Let
82
(cn , un ) is a solution of (10-11).

QN
x0 i=1 Li Z, be such that x0 e > 0. One
f.
can assume that
un (0, x0 ) =
p
.
2
4. The minimal speed c

un converge locally uniformly, up
(in t) function u, which is a classical
From standard parabolic estimates, the positive functions

to the extraction of some subsequence, to a nondecreasing
solution of
t u (A(x)u) = f (x, u) R RN .
p
.
2
+
Since u is nondecreasing in t, one can dene u (x) =: lim u(t, x), and from standard elliptic
Moreover,
satises
0up
and one has
u(0, x0 ) =
t+
(A(x)u+ ) + f (x, u+ ) = 0. Moreover 0 u+ p. Hence, as already

+
+
said (using Theorems 2.1 and 2.3 of [17]), u 0 or u p. But for every B > 0, for n large
p
p
x e
+
+
enough, un (B, 0) un ( 0 , 0) = un (0, x0 ) =
cn
2 . Therefore u (0) 2 . Thus u 0. But since
estimates,
u+
satises
is nondecreasing and
u(0, x0 ) =
p
2 ,
u+ (0)
p
2 , which is contradictory with the preceding
result.
On the other hand, the arguments used in Proposition 5 actually imply that, if
solution of (10-11) with
c0 > 0
and
(u0 )t > 0,
then there is a solution
(c, u)
(c0 , u0 )
is a
of (10-11) for each
c > c0 .
c > 0 such that for all c > c , there exists a
solution u of (10-11), while no solution exists for c < c .
In particular, there exists a sequence (cn , un ) of solutions of (10-11), such that cn c as

n +, with cn > c . As it was done in the rst part of the proof of this Proposition assume
p
that un (0, x0 ) =
2 . From standard parabolic estimates, un converge locally uniformly in (up to
the extraction of some subsequence), to a classical solution u of

Using Lemma 7, one concludes that there exits
t u (A(x)u ) = f (x, u ) R RN ,
0 u p, and ut 0. Moreover, u (0, x0 ) = p2 . Using the same arguments as those of the
beginning of this proof, one concludes that limt u (t, x) = 0 and limt+ u (t, x) = p(x),

ke
locally in x. Furthermore, by passing to the limit, u

t + c , x = u (t, x + k) for all (t, x)
QN
N
R R and k i=1 Li Z. Finally, the strong maximum principle, with ut 0, gives us that u
is increasing in t.
with
4.2
Characterization of
This section is devoted to the proof of the variational characterization of the minimal speed
c . Notice rst that the assumption (7) implies that

x RN , u 0, f (x, u) fu (x, 0)u.
(62)
Let us dene
c0 = inf {c R, > 0
where
c ()
c () = 0} ,
Lc, =
with
with
L-periodicity
Proposition 7
(A(x)) 2 eA(x)
[2 eA(x)e + (A(x)e) c + fu (x, 0)],
conditions.
One has
c = c0 .
83

The proof is divided into several lemmas.
Lemma 8
The real number
c0
does exist and
0 c0 c .
c c , and (c, u) be a solution of (10-11). Then, arguing as in the proof of Lemma 7,

one obtains a positive function , satisfying (55) with = /c > 0. In other words, c () = 0.
That yields c0 c .
Moreover, using the concavity of 7 c (), which has been shown in the proof of Lemma 7,
0
together with c (0) < 0 and (c ) (0) = c, one immediately gets that if c < 0, then c () < 0 for
all > 0. Therefore 0 is a lower bound of the set {c R, > 0 with c () = 0}.
PROOF. Let
From Lemma 8 and Proposition 6, the next lemma follows :
Lemma 9
For all
Now, for all
c < c0 ,
> 0,
problem (10-11) has no solution
let us dene
c = inf {c R, > 0
where
c ()
with
c () = 0} ,
Lc, =
with
(c, u).
L-periodicity
(A(x)) 2eA(x)
(eA(x)e + )2 (A(x)e) + c fu (x, 0),
conditions.
First, using a result of [13] (Proposition 5.7.2), one obtains that
c ()
where

E = C 2 (RN ), > 0,
= max inf
Lc,
E RN
is L-periodic .
Set

j() = max inf
E xRN

(A(x)) 2eA(x)
2 eA(x)e (A(x)e) fu (x, 0) .
Then,
c () = j() + c 2 = c () 2 .
Lemma 10
The real number
does exist for all
> 0,
and
c 0.
c () = j() + c 2 , there exists c > 0 large

>
= 1 < 0 (1 is the rst eigenvalue of (A(x))
fu (x, 0) with L-periodicity conditions), and since 7 c () = c () 2 is concave, whence
0
0
continuous, one gets the existence of such that c ( ) = 0. Therefore c < + for all > 0.
Moreover, as it was done in Lemma 8, one easily sees that 0 is a lower bound of the set
{c R, > 0 with c () = 0}.
PROOF. Let
be xed. Let
enough such that c ()
be given. Since
0. Since c (0)
Lemma 11
84
For all
c > c ,
there exists
>0
such that
c () = 0.

c > c . From the denition of c , one knows that there exists a sequence
(cn ) such that cn c as n + and, for each n, there is n > 0 with c (n ) = 0. There
fore, there exists N such that cN < c. One has c (N ) = c (N ) + (c cN )N > 0. Using the
N
same argument than this of Lemma 10, one deduces that there exists > 0 such that c () = 0.
c
PROOF. Let
be s.t.
Next, let us prove that
Lemma 12
One has
c c0
as
0.
PROOF. First, one observes that c c0 for all > 0. Indeed, for c > c , there exists, from
Lemma 11, > 0 such that c () = 0. Thus, since c () > c () = 0, arguing as in Lemma 11,
0 > 0 such that c (0 ) = 0.
Next, let us show that for any c > c0 , there exists 0 such that c c for all < 0 .
Indeed, one deduces from Lemma 11, adapted to c0 , that for each c > c0 , there exists 1 > 0
such that c+c (1 ) = 0. Then
0
one easily sees that there exists
c (1 ) = c+c0 (1 ) +
2
c c0
1 21 .
2
cc0
Thus, c (1 ) =
2 1 1 . Hence, for small enough, c (1 ) > 0. Therefore, there exists
> 0 such that c () = 0. Finally, from the denition of c , one deduces that c c , and the
lemma is proved.
Let us now turn to the
PROOF of Proposition 7. Let c be such that c > c0 . Then, from Lemma 12, one knows that for
small enough, c > c . Therefore, from Lemma 11, there exist > 0 and > 0 L-periodic,
depending on
and such that
Lc, = 0.
Now, set
1 (s, x) := (x)es ,
for all
(s, x) R RN ,
(63)
and let
be dened as in the proof of
Proposition 5. Then,
L 1 =
and, since
{ (A(x)) + 2eA(x)

+(eA(x)e + )2 + (A(x)e) c es ,
satises (63), one has
L 1 = fu (x, 0)1 .
Therefore, using (62), one obtains,
L 1 + f (x, 1 ) = f (x, 1 ) fu (x, 0)1 0.

Moreover,
is increasing in
and
L-periodic
with respect to
(64)
x.
Now, with the notations of Section 2.1, and as it was proved in Lemma 1, there exists a
solution

C 2 a of the following problem :
L + f (x, ) = 0 in a ,
is L-periodic in
x,
1
(a, x) = min inf (a + , y), p
yC
(a, x) = min{1 (a + , x), p(x)}
p(x)/p+ ,
(65)
85

First, following the proof of Proposition 5, one obtains that
(s, x) a , (a, x) < (s, x).

Next, in the case
1 (a + , x) > p(x)
for all
x C,
(66)
one obtains in the same way that
(s, x) a , (a, x) = p(x) > (s, x).

(67)
1

1
Now, in the other case, one has (a, x) = min (a + , x), p(x) . But, since (s, x) +
1
as s +, one has, for k large enough, (s + + k, x) > (s, x) in a . Let k be the smallest
k such that the latter holds. It exists since 1 (s, x) 0 as s and (s, x) > 0 in a .
1
Assume k > 0. By continuity, (s + + k, x) (s, x) with equality at a point (s, x) a .
1
1
1
Since is increasing in s, (a + + k, x) > (a + , x) (a, x) in C , and similarly,
1
1
(a + + k, x) > (a + , x) (a, x). Thus, (s, x) (a, a) C (one can assume this using
1
1
the L-periodicity in x of and ). But, from (64), it is found that (s + + k, x) is a super1
solution of (65). Therefore, the strong maximum principle implies that (s + + k, x) (s, x)
in a . One gets a contradiction with the boundary condition at s = a. As a consequence, k = 0
and, one has
(s, x) a , (s, x) 1 (s + , x)
and, since
is increasing in
s,
it follows that
(68)
(s, x) < 1 (a + , x)
in
a .
As a conclusion, from (66), (67) and (68), one has
(s, x) a , (a, x) < (s, x) < (a, x).
(69)
Using the same arguments as those of Proposition 5, it follows that

and is the unique solution of (65) in
s = a
C 2 a
is increasing in
. Moreover, since the boundary conditions for
are
. But, since 1 (, x) = 0 and
1 (+, x) = + in RN , it follows from (69) that 0 as uniformly in a and that,
at
, one
C 2 a
are nondecreasing in
continuous with respect to
in
can prove, as in Lemma 3, that the functions
and nondecreasing in
> 0, T, > T,
Therefore, for each
a > 1,
there exists
(a) R
>
such that
(a, .)
is bounded independently of
a.
Moreover,
L + f (x, )
is L-periodic w.r.t. x.
is nonincreasing with respect to
s,
Z
(0,1)C
86
,a := (a)
(s, x)dsdx =
solves (65) and satises
an
N
R ) (for all 0
= 0
in
R RN ,
and satises
0 (s, x) p(x)
and
a .
Thus, letting
2,
,a
elliptic estimates, the functions n converge in Cloc (R
extraction of a subsequence, to a function satisfying
in
(p )2
|C| min(c, 1).
2p+
,a (s, x)dsdx =
(0,1)C
Moreover
p p
p+
in
R RN ,
(p )2
|C| min(c, 1).
2p+
+, from standard
< 1), up to the

Next, passing to the limit
and using the same arguments as those of the end of the
(c, u) of the problem (10-11).

c was chosen arbitrarily such that c > c0 , one concludes that there exists a solution
(10-11) for all c > c0 . Next, using Lemma 9, one obtains that c = c0 .
proof of Proposition 5, one obtains a solution

But since
(c, u)
of
That completes the proof of Theorem 1.
4.3
Dependency of
with respect to the nonlinearity
In this section, we study the dependency of

the nonlinearity
f.
c ,
with respect to the "shape" and the "size" of
This section is devoted to the proofs of Theorem 2 and Corolarry 1. In the
A(x) = A is constant, and one considers the

fu (x, 0) is of the type fu (x, 0) = (x) + B(x),
0, functions.
where B is a positive real number and m u and are periodic C
It then follows from Theorem 2.8 of [17] that the function f satises the hypothesis for all
Z
Z
B > 0 if 0 and 0 with 6 0. It also follows from Theorem 2.8 of [17] that f satises
whole section, one assumes that the matrix eld

problem (10-11), with a nonlinearity
the hypothesis for conservation for
such that
B>0
max > 0.
characterizations of c
large enough under the only assumption
In the next propositions, we will make several uses of the following
rst, from Theorem 1,
c = inf {c, > 0

where
c,B ()
with
c,B () = 0} ,
(70)
Lc,B, = (A) 2Ae (2 eAe c) ((x) + B(x)),

on the set
of
pulsating fronts
L-periodic C 2 functions. Furthermore, as it was said

N
in R , the formula below is equivalent to the following
c = min
>0
where
k (B)
in [13] and in [93], for

one :
k (B)
,
(71)
is the principal eigenvalue of the operator
LB, = (A) 2Ae 2 eAe ((x) + B(x)),

acting on the same set
to
k (B).
of functions
We call
B,
be the principal eigenfunction associated
It satises
LB, B, = k (B)B, ,
N
is L-periodic, B, > 0 in R ,
B,
kB, k = 1 (up to normalization).
We are going to study the monotonicity of the function
(72)
B 7 c = c (B),
as soon as the
hypothesis for conservation if satised. One has the
Proposition 8
max > 0.
Z
Assume that
= 0 0
Then, the hypothesis for conservation is satised for all
increasing function of
(x)dx 0
is constant and assume that
B > 0
and
c (B)
with
is an
B > 0.
87

PROOF. As already underlined at the beginning of this section, the hypothesis for conservation
is satised for all
B > 0.
As done in the proof of Lemma 7, one has
k (B) = max0 inf

E RN
where
E0
(A)
0 B(x),
be the set dened by
n
E0 = C 2 (RN ), > 0, L-periodic
with
o
(x) = exe .
B1 , B2 R and t [0, 1]. Set B = tB1 + (1 t)B2 . Let 1 and 2 be two arbitrary
0
z
chosen functions in E , and set z1 = ln(1 ), z2 = ln(2 ), z = tz1 + (1 t)z2 and = e . It easily
0
follows that E . Therefore

(A)
k (B) inf
0 B(x) .
RN
Let
Then, arguing as in the proof of Lemma 7, one obtains that
k (B)

(A1 )
0 B1 (x)

1

(A2 )
+(1 t) inf RN
0 B2 (x) ,
2
t inf RN
2 were arbitrary chosen, one has k (B) tk (B1 ) + (1 t)k (B2 ). Therefore
the function B 7 k (B) is concave. This also implies that this function is continuous.
2
Next, one easily sees that k (0) = eAe 0 , and that the associated eigenfunction 0,
is equal to 1.
0
Now, let us calculate k (0). Let B, be the principal eigenfunction associated to k (B)
dened in (72), and let us integrate by parts the equation (72) over C . Using the L-periodicity
of B, , one obtains
Z
Z
Z
2
( eAe + 0 )
B, B
(x)B, dx = k (B)
B, .
(73)
and, since
and
k (B) k (0) as B 0. Still arguing as in the proof of Lemma 7,

2, (for all 0 < 1) to
one also knows that B, converges in C
0, 1 as B 0. Then,
dividing the equation (73) by B , one gets
Z
Z
k (B) + 2 eAe + 0
B, =
(x)B, dx.
B
C
C
By continuity, one knows that
Therefore, passing to the limit
B 0,
one obtains
k0 (0)
Z
=
(x)dx.
C
Z
(x)dx > 0,
In the case
one has
k0 (0) < 0.
From the concavity of
that this function is decreasing with respect to

concludes that the minimal speed
88
c (B)
B > 0.
B 7 k (B),
> 0,
B > 0.
Since this is true for all
given in (71) is an increasing function of
one deduces
one

Z
(x)dx = 0,
Similarly, if
and
max > 0,
divide the equation (72) by
B,
and integrate it
C . By L-periodicity, one obtains

#
Z "
Z
B, AB,
2
(x)dx = k (B)|C|,
( eAe + 0 )|C| B
2B,
C
C
by parts over
B, is not constant (because is not constant) and the matrix A is elliptic, one gets
2
0
that k (B) < ( eAe + 0 ) = k (0) for all B > 0. Hence, since k (0) = 0 and k (B) is concave
in B , one concludes that B 7 k (B) is decreasing in B > 0. Finally, it follows that c (B) is
increasing in B > 0.
and, since
The biological interpretation of this proposition is that increasing the amplitude of the favorableness of the environment increases the invasion's speed.
Remark 2
If one only assumes that
conservation for
max > 0,
then the function
satises the hypothesis for
B > 0 large enough. Furthermore, under the other assumptions of Proposition 8,

B 7 c (B) is an increasing function of B
the same arguments as above imply that the function

(for
large, as soon as the hypothesis forn conservation is satised).
In the next proposition, one assumes that
satises the hypothesis for conservation.

As one
Z
has said above, it follows from Theorem 2.8 of [17] that it is true for all
B >0
(x) 0
if
Z
(x) 0
and
with
6 0 ;
if one only has
max > 0,
it is true if
is large enough.
Proposition 9
max > 0 and that the function f

fu (x, 0) = (x) + B(x). Then
p
c (B) 2 eAe max( + B).
Assume that
conservation with
satises the hypothesis for
1 in (9) and that the hypothesis for consermax( + B) > 0.
Next, using the characterization of c given by (70), let us integrate by parts over C the
equation Lc,B, = 0. Using the L-periodicity of , one obtains the following inequalities :
PROOF. Let us rst observe that the denition of

vation (1
< 0)
imply that
c,B () 2 eAe + c max( + B).

p
c 2 eAe max( + B), there exists 0 > 0 such that c,B (0 ) 0. On the other
hand, c,B (0) = 1 < 0 from the hypothesis for conservation. By continuity, it follows that there
p
exists a solution > 0 of c,B () = 0 as soon as c 2
eAe max( + B). Thus, one nally has
p
c (B) 2 eAe max( + B).
(74)
Therefore, if
That completes the proof of Proposition 9.
Remark 3
B
A is not assumed to be uniform in the space variables

max > 0, then the hypothesis for conservation is satised
If the diusion matrix eld
anymore (but still satises (6)), and if

for
large enough and the same arguments as above imply that
c (B)
lim sup
2 max(eAe) max .
B
B+
89

Z
Proposition 10
max > 0.
Assume now that
fu (x, 0) = (x) + B(x),
0,
where
Then
s
2
and
eAe
|C|
r
Z

c (B)
+ dx
2 eAe max
+
B
B
C B
1
c (B)
c (B)
eAe max lim inf
lim inf
2 eAe max .
B+
B+
2
B
B
and
(75)
(76)
PROOF. As already underlined, the assumptions of Proposition 10 guarantee that the hypothesis
B > 0.
for conservation is satised for all
c given by (71). Let B, be dened by (72). Dividing
We will now use the characterization of

(72) by
B, |C|
and integrating by parts leads to
Z
( + B)
C
eAe +
|C|
k (B)
.
(77)
One deduces from (77) and (71) that
s
2
eAe
|C|
( + B) c (B),
(78)
and the result (75) follows from (74) and (78).

The proof of the lower bound in (76) is divided in two steps. Let
c (B)
0 := lim inf
2 eAe max
B+
B
and (Bn )nN + such that c (Bn )/ Bn as n +. First, from (71), there exits a
kn (B)
sequence (n )nN in R+ such that

as n +. Moreover, from (74), one knows
n Bn
that
where
kn (B)
2 eAe max + n ,
n Bn
n 0
as
n +.
Z
Using the equation (77), one obtains with (79)
( + B)
n eAe +
Assuming that
p
p
kn (Bn )
2 Bn eAe max + n Bn .
n
|C|n
Z
Z
0 and
0,
C
one deduces that
r
n 2
90
(79)
p
Bn
max + n Bn .
eAe
(80)

Next, consider the eigenvalue problem
(AB, ) 2Ae B,
2 eAeB, ( + B)B, = e
k (B)B, ,
B, > 0 on C, B, = 0 on C, kB, k = 1,
e
k (B) > k (B) (for all > 0 and B > 0).
e
k (B) k (B) ; then the function B, satises
and let us prove that

one has
Assume that, on the contrary,
(AB, ) 2Ae B, 2 eAeB, ( + B)B, k (B)B,

= (e
k (B) k (B))B, 0.
Since the function
for all
>0
B,
dened by (72) is positive in
C,
one can assume that
B, < B,
(81)
in
small enough. Now, set

= sup > 0, B, < B,
B, B, in C and
B, > 0 in C and B, = 0
Then, by continuity,
in
there exists

C > 0.
x1
in
such that
B, (x1 ) =
x1 C . Therefore, using
B, B, in C , which is
impossible from the boundary conditions on C . Finally, one concludes that e
k (B) > k (B).
xe
Let us now dene B, (x) = e
(x)
.
From
(81),
the
function
B,
B, satises the eigen-
B, (x1 ).
But, since
on
C ,
it follows that
(81), it follows from the strong elliptic maximum principle that
value problem
(AB, ) ( + B)B, = e
k (B)B, ,
B, > 0 on C, B, = 0 on C,
and it follows that
Z
e
k (B) =
min
H01 (C), 60
(A) ((x) + B(x)) 2

Z
.
2
C
> 0 be arbitrarily chosen. Then, there exists in H01 (C), such that k k = 1, 0
for all x C ,
(x) > 0 (max (x) < ) .
Let
and,
One then easily obtains (see the proof of Proposition 5.2 in [17])
Z
Z
(A(x) ) + (x)2
C
C
Z
+ B (max ) e
k (B)
2
(82)
C
for all
> 0.
e
kn (Bn ) > kn (Bn ) one has

kn (Bn )
1
lim inf
eAe
max
.
n+ n Bn
2
max
Hence, using (80) and (82), and since
since
kn (B) 0.
Since
>0
was arbitrary, one concludes that
c (B)
1
= lim inf
eAe max .
B+
2
B
91

The formula (76) follows.
PROOF of Corollary 1. In the special case where
fu (x, 0) = (x) ( = 0
and, say,
B = 1)
with
Z
(x)dx 0 |C| > 0,
c
it follows from the lower bound in (75) that
c [] c [0 ] = 2 eAe 0 .
In other words, an heterogeneous medium

increases the biological invasion's speed, in comZ
fu (x, 0)dx > 0.
parison with a constant medium, when
fu (x, 0) = (x) + B(x), it follows from the Proposition 10

average, it suces for to be positive somewhere for the speed
Coming back to the case where

that, even if
c (B)
and
have zero
to increase like to the square root of the amplitude of the eective birth rate.
Proposition 11
Z
Assume that
0, fu (x, 0) = B(x)
with
and
max > 0.
Then, one
has
s
Z
c (B)
eAe
lim
=2
(x)dx.
|C| C
B0+
B
PROOF. First, it follows from (75) that
s
2
for all
eAe
|C|
c (B)
(x)dx
B
C
(83)
B > 0.
In order to establish the opposite inequality at the limit

cases :
B 0+ ,
one shall consider two
Z
> 0.
Case 1 :
Let
B,
be dened by (72) and call
B = B,B
with
s
B =
B
eAe|C|
Z
(x)dx.
C
Z
Multiply (72) by
and integrate it by parts over
C.
Dividing by
2B ,
Z
kB (B) =
2B
B AB
2
C
Z
B eAe B Z
,
2
2
B
B
C
and
Z
kB (B)
B C
Z
B eAe +
B
B
2B
C
92
.
2B
one obtains

Moreover, observing that
knows that
converges
can write
where
B 0
as
B 0.
B 0 as B + and arguing as in the proof of

2, (RN ) (for all 0 < 1) to 1 as B 0.
in C
0
Z
B
B
kB (B)
B eAe +
+
B ,
B
B |C| C
B
Replacing
B by
s
kB (B)
2
B
Lemma 7, one
Therefore, one
(84)
its value in (84), one obtains
B
eAe
|C|
v
u
u BeAe|C|
+u Z
B .
t
From the characterization (71) of
c (B),
c (B)
2
B
where
eB 0
as
B 0.
one then obtains
eAe
|C|
Z
C
+ eB ,
Using (83), one concludes that
s
Z
c (B)
eAe
lim
=2
.
B0
|C| C
B
Z
= 0.
Case 2 :
Choose now
B =
for arbitrary
> 0.
The above arguments imply
C
that
and the conclusion follows.
c (B)
lim sup
eAe
B
B0+
Finally, Theorem 2 follows from the last four propositions.
93
94
Partie II : Study of the premixed ame

model with heat losses
Sommaire
1
2
Introduction and main results . . . . . . . . . . . . . . . . . . 96

Proof of the existence theorem . . . . . . . . . . . . . . . . . . 98
2.1
. . . . . . . . . . . . . . . .
98
2.2
Existence of solutions in bounded domains . . . . . . . . . . . .
99
2.3
. . . . . . . . . . . . . . . . . . . .
109
2.4
. . . . . . . . . . . . . . . . . . . .
110
. . . . . . . . . . . . . . . . . . . . . . .
111
R+ .
0.
R+
Proof of theorem 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.1
An upper bound for
3.2
3.3
. . . . . . . . . . . . . .
112
. . . . . . . . . . . . . . . . . .
113
Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Preprint 04-15 LATP 2004
95
1 Introduction and main results

Zeldovitch has shown in [95], using asymptotic methods, that the reaction-diusion system
modelling the propagation of a premixed laminar ame with heat losses has got two travelling
wave solutions in the case of high activation energies. More recently, Glangetas and Roquejore
in [47] have demonstrated the same result, as a consequence of the dispersion relation which has
been obtained by Joulin and Clavin ([60],[61]), and proved rigorously in [47]. The case of a more
general nonlinearity has been studied by Giovangigli in [46], where the author proves the existence
c (considering in his work that the minimum heat loss rate

Le equal to 1. In this paper,
we follow the framework of Berestycki, Nicolaenko and Scheurer [23], keeping c as a solution of
the problem, and we prove the existence of two solutions for small values of . Moreover, the
of a solution for a xed reaction speed

parameter
is an unknown of the problem), with a Lewis Number
existence result is established for every positive Lewis Number, hence one also solves a nonuniqueness problem. Indeed, in the adiabatic case, Marion [72] has proved the uniqueness of
ames when the Lewis number (1/ in this paper) is greater than
1,
and Bonnet has shown in
[26] that when the Lewis number is less than unity, uniqueness cannot be generally assumed.
Here, we prove that uniqueness never holds in the non-adiabatic case (for small heat losses
).
Besides, we establishe that, in some cases, the solution with the greatest speed converges to the
solution of the adiabatic problem as
0,
whereas the other speed converges to
0.
Moreover, we also compute some new bounds for the solutions. In particular, Giovangigli has
proved in [46] that, when
speed
cad
Le = 1,
the reaction speed
was bounded from above by the reaction
of the adiabatic problem. Here, we extend his result to
meaningful since e.g.
Le = 0.4
for hydrogen), by showing that
of an adiabatic scalar problem. Moreover, for all
Le > 0,
which does not depend on the heat loss rate parameter
Le 1
(which is physically
is inferior to the reaction speed
we give an explicit upper bound for
c,
We also give a lower bound for the
unburnt gases after the reaction.

Let
and
be two positive real numbers. The aim of this work is to prove existence and
nonexistence results for the following problem :

Finding two nonnegative classical functions
and
and a nonnegative real number
which
satisfy
u00 + cu0 = f (u, v) h(u)

v 00 + cv 0 = f (u, v)
on
(1)
R,
with the boundary conditions
u() = 0, u(+) = 0,
v() = 1, v 0 (+) = 0.
The following assumptions will be made on
(2)
in the sequel : there exist two functions
and
such that
f (u, v) = p(u)g(v)
where the function
is in
s.t.
C 0 (R),
p(x) = 0
for all
increasing on
g<0
96
R R,
(3)
p is globally Lipschitz continuous on R, nondecreasing and of ignition
(0, 1)
and the function
in
on
R+
and
p(x) > 0
for all
x > ,
type :
(4)
and such that
and
g(0) = 0;
(5)

moreover, for all
> 0,
let us set
g(s)
,
k () := max sup
s(0,1) g(s)
and
)
,

g(s)
k () := min
inf
, ,
s(0,1) g(s)
one then assumes that

for all
> 0, 0 < k () k () < +.
(6)
g of the type g(y) = y n , with n > 0. It also

n 1, n N such that g (n) (0) 6= 0, where g (n)
Hypothesis (6) is for instance satised by functions

works with
C (R)
functions
th derivative of
is the n
The function
such that it exists
g.
is supposed to be in
C 1 (R),
strictly increasing. Moreover, it satises
h(0) = 0, h(1) = 1, , R
One call
(uad , vad , cad )
s.t.
0 < h0 .
(7)
the solutions of the following problem without heat loss (see [23] for
the existence of such solutions) :
u00ad + cad u0ad = f (uad , vad )

00 + c v 0
vad
ad ad = f (uad , vad )
on
(8)
R,
Remark 1
uad () = 0, uad (+) = 1,

vad () = 1, vad (+) = 0.
(9)
The reaction term used here is more general than this which was used in [23],
nevertheless, the results of [23] can be easily adapted to our case.
Remark 2
In the case
g(y) = y
and
the solution
(uad , vad , cad )
of (8-9) is shown to be
unique (up to translation) in [72]. Moreover, [26] proves that uniqueness does not hold in the
general case for
Let
(us , cs )
> 1.
be the unique solution (see [23]) of the following adiabatic problem
u00s + cs u0s = f (us , 1 us ),
(10)
us () = 0, us (+) = 1.
(11)
Remark 3
Notice that in the case
= 1,
the problem (8-9) reduces to the scalar case (10-11),
and therefore, uniqueness always holds.
Under this hypothesis, one has the following results :
97
Theorem 1
> 0, if is suciently small, there exist two distinct and nontrivial

solutions (u1 , v1 , c1 ) and (u2 , v2 , c2 ) of the problem (1 -2), with c1 < c2 . Moreover, vi (i =
1, 2) is nonincreasing and, setting = min{1, }, one has the following estimate on vi (+) :
1
, (i = 1, 2).
vi (+) <
1) For all
> 0, c1 0
0.
g(y) = y on R,
(u2 , v2 , c2 ) converges locally uniformly to (uad , vad , cad ), the unique solution of (8-9), as 0.
The same result holds in the case = 1, for the more general reaction term f (x, y).
2) For all
as
Moreover, if one assumes that
The next Theorem gives new bounds for every solution
Theorem 2
1) If
f (1, 1)
,
h()
>
2) Assume that
(u, v, c)
is the Lipschitz constant, then
of (1-2), one has
Kp(1)
v(+) > exp
h()
(u, v, c)
of (1-2).
the problem (1-2) has got no solution.
is globally Lipschitz continuous, where
for all nontrivial solution
3) Let
(u, v, c)
and
be a solution of (1-2), then, for all

.
1, 0 < c cs ,
where
(us , cs )
is the
solution of (10-11).
4) Let
(u, v, c)
be a solution of (1-2), and set
1 = max
s(,1)
0<c<2
f (s, 1 s)
s
and
2 = max f (1 s, s),
s[0,1]
for all
1and 0 < c <
for all
then
> 0.
2 Proof of the existence theorem

2.1
R+
In this section, we will recall some results of Giovangigli. He establishes in [46], Proposition
2.2, that, under conditions (4) and (7),
Proposition 1
For all
c0
is a solution of
and
0,
every solution (u,v) of (1-2), after a shift of the origin,
u00 + cu0 = f (u, v) h(u)

v 00 + cv 0 = f (u, v)
on
(0, +) ,
(12)
where
u(0) = ,
u0 (0) = c +
u(+) = 0, v 0 (+) = 0.
h(u ),
v 0 (0) =
c
(v(0)
1),
(13)
is the unique (see [46]) solution of
Conversely, every solution

nontrivial solution of (1-2).
98
R0
u00 + cu0 + h(u ) = 0

u () = 0, u (0) = .
(u, v)
on
R ,
of (12-14) can be extended to
(14)
in such a way that it is a
2. Proof of the existence theorem
Remark 4
This proposition clearly uses the fact that
in [46] with
g(y) =
p is of ignition type. It was demonstrated

1/ equal to 1, but it is straightforwardly still valid
f (x, y) and with 1/ 6= 1.
y n and a Lewis number
with our more general reaction term
Finally, Proposition (1) shows that solving (1-2) is equivalent to nding a solution
on
R+
(u, v, c) of (1)
with
u0 (0) = cu(0) + k(, c, ), u(+) = 0, u(0) = ,

v 0 (0) = c(v(0) 1),
v 0 (+) = 0,
R0
k(, c, ) = h(u ).

where one has set
2.2
(15)
Existence of solutions in bounded domains
In order to be able to use a topological degree argument, we will study the system (1) on a
bounded interval of
R+ .
Namely, for each
a > 0,
one sets
Ia = (0, a) ,
and one seeks solutions
(u, v, c) in C 2 (Ia , [0, 1])2 R+ of

u00 + cu0 = f (u, v) h(u)
v 00 + cv 0 = f (u, v)
on
Ia ,
(16)
u0 (0) = cu(0) + k(, c, ), u(0) = , u(a) = 0

v 0 (0) = c(v(0) 1),
v(a) = 0.
(17)
Let us dene the Banach space
Xa = C 1 (Ia , [0, 1]) C 1 (Ia , [0, 1]) R,

with the norm
k(u, v, c)kXa = kukC 1 (Ia ) + kvkC 1 (Ia ) + |c|,
and let
be the application dened
by
J : Xa Xa , (u, v, c) 7 (U, V, u(0) + c ),

where
(U, V )
is the unique solution of the linear problem
on
Ia = (0, a)
U 00 + cU 0 = [f (u, v) h(u)] + (1 )(v u)

V 00 + cV 0 = f (u, v) (1 )v,
with the boundary conditions,
where
(18)
U 0 (0) = cU (0) + k(, c, ), U (a) = 0

V 0 (0) = cV (0) c,
V (a) = 0,
(19)
= + (1 ).
Remark 5
This mapping
is close to the one which was used in [46]. One cannot use here
the homotopic transformation of [23]. Indeed, we will see in the proof of Proposition 3 that the
right-hand side plays a crucial role in the denition of a positive real number
solution
(u, v, c)
exists with
c (, c )
for
small enough (and
such that no
arbitrarily small).
99

(u , v , c )
Let
[0, 1]. In order to compute a

estimates on (u, v, c). More precisely, we are going
J ,
be a xed point of the application
topological degree, we will need some a priori
with
to prove the following proposition :
Proposition 2
Let
(u , v , c ) be a xed point of the application J , with c 0. Then, it satises
the following estimates :
1 u
,
min{1,
r }

c
f (1, 1) + 1
1
c < u0 < c 1 +
,
< v0 < 0, 0 < c
.
0 < u < 1,
0 < v < 1,
(20)
PROOF : In order to improve the readability of the following calculations, we will skip the
index
replacing
Lemma 1
(u , v , c )
One has
PROOF. For each
v>0
on
by
(u, v, c).
Ia .
(0, 1), v
satises the following equation
v 00 + cv 0 = f (u, v) (1 )v
on
Ia ,
(21)
therefore, one has
v 00 + cv 0 + v = 0
on
Ia ,
(22)
where one has set
(x) :=
f (u, v) + (1 )v
v
From our hypothesis (3-6), one has
if
(x) 0
v(x) 6= 0,
on
Ia .
and
(x) := 1
if
v(x) = 0.
Therefore, using a result of [45], one can use
a maximum principle, which gives us
inf v(x) = min {v(0), v(a)} .
xIa
(23)
v(0) < 0, then, from (19), one has v 0 (0) 0. Since v(a) = 0, we necessarily
= 0, which implies c = 0 from (19). Therefore, from (21), v is concave
v(x) 0, which is impossible since v(a) = 0. It follows that v 0. From the strong
principle applied to (22), one has v > 0 on Ia .
Let us assume that
0
have, from (23), v (0)
as long as
maximum
Lemma 2
v<1
and
c
< v0 < 0
PROOF. From (21), and since
Ia .
on
v > 0, one has

0
c
e x v 0 0
thus, integrating this equation between
x Ia
and
on
a,
Ia ,
(24)
one obtains,
e (ax) v 0 (a) v 0 (x).

v > 0 on Ia and v(a) = 0, hence v 0 (a) 0. One deduces v 0 0. Now, let x1 Ia be such
0
0
that v (x1 ) = 0. Then, integrating (24) between x1 and x for x x1 , one nds v (x) 0. Thus,
But
100

v is constant on [x1 , a]. This is in contradiction with v > 0. Therefore v 0 < 0,
v(0) < 1, hence v < 1 on Ia .
0
Moreover, from (21), v cv is nondecreasing on Ia , thus
c
on
Ia ,
1u
min{1, }
on
Ia .
v 0 (x) >
from (19), and since
Lemma 3
On has
and, using (19),
(25)
v > 0.
0<u<1
and
PROOF. From the strong maximum principle, and using the hypothesis (7), one easily obtains
u>0
on
Ia .
w = u + v.
As it was done in the proof of Lemma 14, set
Then
satises
w00 + cw0 = ( 1)v 00 [ h(u) + (1 )u].

An integration between
and
x Ia
gives, using (19),
w0 (x) + cw(x) c + k ( 1)v 0 (x)

since
h(u) 0.
Thus, since
k 0,
and
a,
Ia ,
one has
(wecx )0 ecx ( 1)v 0 (x)

and integrating between
on
on
Ia ,
one obtains
cx
w(x) ( 1)e
v 0 (t)ect dt + 1,
(26)
x
thus, if
> 1,
one has
u+v 1
since
v0 0
from the above calculations. Next, if
1,
we
deduce from (26) that
w (1 )v + 1,
therefore,
u 1 v . Thus, since v > 0, in both cases, u < 1. Besides, one has v
(27)
1u
.
min{1, }
Lemma 4

c < u0 < c 1 +
on
Ia .
PROOF. Let us add the equations satised by
v,
and
and integrate the sum between
and
x.
Using the boundary conditions (19), one obtains,
u0 cu c + k + c v 0 > c.
It follows that
Moreover,
u0 > c.
setting y = u + v ,
one has
(y 0 ecx )0 ecx v 0 c( 1),

and an integration from
x Ia
0
to
gives
cx
y (x)e
Z
c( 1)
ect v 0 (t)dt,
x
101

since
y 0 (a) 0.
Thus, since
v 0 < c
(from (25)), one deduces that
u (x)c + c( 1)
ec(xt) v 0 (t)dt
on
Ia ,
(28)
x
therefore, if
thus

u0 < c 1 +
Lemma 5
< 1, from (28) and using again v 0 < c,

Z
c cx a ct
c
u0 (x) < c +
e
ce dt < c +
(1 ec(xa) ),
1,
then
u0 < c,
one has

r
One has
and for
0<c
PROOF. One knows that
f (1, 1) + 1
.
satises,
u00 + cu0 = [f (u, v) h(u)] + (1 )(v u)

Let us multiply (29) by
and integrate it from
u(a)u0 (a) + u(0)u0 (0) +
to
a.
on
Ia .
(29)
One obtains

u0 2 + c/2 u(a)2 u(0)2
Z a
<
uf (u, v) + (1 )uv,
(30)
0
therefore, from (19),
0
but, since
0 2
c
< ( + k) +
2
uf (u, v) + (1 )uv,
(31)
u < 1 and from (21), one has,

Z a
Z a
uf (u, v) + (1 )uv
f (u, v) + (1 )v,
0
0
Z a
=
v 00 cv 0 ,
0
= v 0 (a) v 0 (0) cv(a) + cv(0),

c.
Therefore, combining this inequality with (31), we obtain,

u0 2 < c,
(32)
0
therefore, one has
c > 0.
In order to obtain an upper bound for
c,
let us consider a function
of the following boundary value problem, with
which is the solution
M := f (1, 1) + 1,
w00 + cw0 = M on Ia ,
w0 (0) = cw(0) + k,
w(a) = 0.
102
w,
(33)

w is a super-solution of the equation satised by u. Therefore, from the boundary
conditions of w , we deduce, with the maximum principle, that w u on Ia . A direct computation
The function
gives
w(0) =
But, since
w(0) u(0)

1
ca
ca
M
4
kc
+
4
kce
M
(1
+
ac)e
.
c2
one has
M
c2
4 ck (1 eca )
thus
k
M
+ 4 (1 eca ) .
2
c
c
Finally,
M
.
Proposition (2) is proved.
In order to compute a topological degree, we need an other estimate, that will be established
.
(u , v , c ) Xa
for small parameters

Let
Proposition 3
For all
be a xed point of
> 0,
there exist
J .
1 > 0
and
a1 > 0
such that for all
a > a1 ,
for all
[0, 1],
( < 1 ) = (c
/ (, c )),
where
is a positive real number dened at the end of the proof.
PROOF : Let (un , vn , cn , n ) be a sequence of xed points of Jn , with cn (, c

), n [0, 1], an +, n 0 and > 0. Then, from Proposition (2), the sequence

2
(un , vn , cn , n )nN is bounded in C 2 (0, an ) R [0, 1]. By compactness, we obtain the conver1
2
gence (up to the extraction of some subsequence), in Cloc (R+ ) R [0, 1] of (un , vn , cn , n ) to
(u, v, c, ) which is a classical solution of

u00 + cu0 = f (u, v) + (1 )v
on R+ ,
(34)
v 00 + cv 0 = f (u, v) (1 )v
with
u0 (0) = c, v 0 (0) = cv(0) c, u(0) = ,

since
k(, c, 0) = 0
(35)
(see (19)), and
c [, c ].
Let us set
= min(1,
1
)
and
= max(1,
(36)
1
).
One has the following
Lemma 6
(1 u) v (1 u).
PROOF :
one has
v (1u) is a consequence of Proposition (2). Setting w = u+v and y = u+ v ,

w00 + cw0 = ( 1)v 00
on R+ .
y 00 + cy 0 = ( 1)cv 0
103

Integrating the former expressions from
to
x R+
and using (35), we obtain
w0 (x) + cw(x) + (1 )v 0 (x) = c

y 0 (x) + cy(x) + c(1 )v(x) = c
thus
on
(w(x)ecx )0 = (v 0 (x)(1 ) c)ecx

(y(x)ecx )0 = (v(x)(1 ) 1)cecx
R+ ,
(37)
R+ ,
on
hence,
Z +
cx
v 0 (t)ect dt
w(x) = 1 + ( 1)e
Zx +
cx
y(x) = 1 + c( 1)e
v(t)ect dt
on
R+ ,
therefore,
Z +
cx
v(x)
=
1
u(x)
+
(
1)e
v 0 (t)ect dt
Zx
1 u(x) ( 1) cx +
v(x) =
+
e
v(t)ect dt
x
Therefore, if
1, then v 1 u since v 0 0, and if > 1, then v
concludes the proof of the lemma.
Lemma 7
on
One has
R+ .
1u
since
v > 0. That
u0 (+) = v 0 (+) = u00 (+) = v 00 (+) = 0.
PROOF : Let us integrate the equation satised by
v (x) cv(x) = c +
from
to
x R+ .
One has
[ f (u, v) + (1 )v(x)] dx.
(38)
0
Since
is nonincreasing and nonnegative,

Z it has got a nite limit as
f (u, v) + (1 )v(x) 0,
Thus, from (38)
v 0 (x)
x +.
[ f (u, v) + (1 )v(x)] dx
the integral
Moreover, since
converges (even to
+).
has a limit as
x +.
Since
is bounded, this limit is equal to
0.
Moreover,
Lemma 8
The function
u+v1
has a constant sign on
R+ .
(1 u)( 1) u + v 1 (1 u)( 1). Thus, if > 1, = 1

1 = 1 and one has u + v 1 0.
PROOF : From Lemma 6,

and
u + v 1 0,
and if
Using the equation satised by
in (37), we obtain
v 0 (x) + u0 (x) = c(u(x) + v(x) 1).

(39)
Rx
Therefore, v(x) + u(x) = c
0 (u(t) + v(t) 1)dt + v(0) + . But, from Lemma 8 the integral
of the right-hand side converges, hence u admits a limit at +. Arguing as for v we deduce that
u0 (+) = 0. It then follows from (34) that u00 (+) = v 00 (+) = 0. That completes the proof
of Lemma 7 .
104
Lemma 9
u0 0
on
R+
PROOF : The equation satised by
in (34) is equivalent to
(u0 (x)ecx )0 = ecx [ f (u(x), v(x)) + (1 )v(x)] 0.

Therefore, integrating this expression from
sought result.
Lemma 10
x R+
u(+) = 1
and
+,
and using Lemma 7 one obtains the
v(+) = 0.
PROOF : From (34) and (39),
u(+)
v(+)
and
satisfy
f [u(+), v(+)] + v(+)(1 ) = 0

u(+) + v(+) = 1.
c > , one has u0 (0) = c > 0.

v(+) = 0 and u(+) = 1.
Thus, since one has

follows that
to
Therefore, from Lemma 9,
u(+) > .
It
Using Lemma 6, we obtain :
u00 + cu0 = f (u, v) + (1 )v
(40)
f (u, (1 u)) + (1 ) (1 u).

Moreover, from the hypothesis (6) and from the denition of
k ( )f (u, 1
u),
(41)
k ( ), one has f (u, (1u))
and
k ( ) < +,
(42)
u00 + cu0 k ( )[ f (u, 1 u) + (1 )(1 u)].
(43)
hence, from (41) and (42),
Let us now multiply the inequality (43) by
u and integrate over (0, +). One then obtain, using
again Lemma 7,
Z
0
since
u 1.
c
(u ) + (1 + 2 ) k ( )
2
0 2
[ f (u, 1 u) + (1 )(1 u)],
(44)
Next, still using Lemma 6, one obtains
u00 + cu0 = f (u, v) + (1 )v

f (u, (1 u)) + (1 ) (1 u).
Again, from the denition of
k ( ),
one has
f (u, (1 u)) k ( )f (u, 1 u)
0 < k ( ) .
Then, integrating (45) over
(45)
and
(46)
(0, +), and using (46), one has, from the limiting behaviors obtained
in the previous lemmas,
Z
c k ( )
[ f (u, 1 u) + (1 )(1 u)],

0
105

hence, with (44), one obtains the inequality
Z
0
which is equivalent to
Z
0
c
k ( )
(u0 )2 + (1 + 2 )
c,
2
k ( )
c
(u )
2
0 2

k ( )
2
2
1 .
k ( )
u0
Next, let us multiply the equality (40) by
(47)
and integrate it over
(0, +).
Using again the
lemma 6 and the result (46) above, one gets
2 c2
+c
2
0 2
u0 [ f (u, 1 u) + (1 )(1 u)],
(u ) k ( )
0
which gives, using Lemma 9 and inequality (47),
c2
2
k ( )
2
1
k ( )
k ( )
[ f (s, 1 s) + (1 )(1 s)]ds
k ( )m ,
Z
where
is dened by
f (s, 1 s)ds, and m2 =
m = min(m1 , m2 ), with m1 :=
(1 s)ds.
Therefore, one has
k ( )2
m > 0
c 2
2k ( ) k ( )
2
from
(42)
and
(46),
(48)
1
k ( ) = 1. Furthermore, , which implies from the

1
denition of k that k ( ) k
. Similarly, for 1, = 1 and k ( ) = 1. Moreo

1
1
ver,
, which implies that k ( ) k
. Besides, it follows from (42) and (46) that

1
1
1
1
k
< + and 0 < k
. Hence one can set
s
2

m if < 1,
c :=
2k 1 1
thus, if
< 1
then
= 1
and
and,
v
u

u 2k 1 2
t

m
c :=
2 k 1
Then
one has
is positive and independent of
c c ,
(although
if
1.
and
were depending on
).
Moreover,
which is in contradiction with (36). That completes the proof of Proposition 3.

In order to compute a topological degree, we need to investigate the case
The 3-uplet
(u, v, c)
is a xed point of
106
J0
in
Xa
= 0.
if and only if it satises
u00 + cu0 + u = v
v 00 + cv 0 + v = 0
in
Ia ,
(49)

u0 (0) = cu(0),
u(a) = 0,
0
v (0) = cv(0) c, v(a) = 0,
(50)
and
u(0) = .
(51)
Proposition 4
For small enough and a large enough, (49-51) admits exactly two solutions
(u2 , v 2 , c2 ). Moreover, there exists c > 0, such that for all > 0, 2 > 0 and
0, such that for all < 2 and for all a > a2 , one has 0 < c1 < and c2 > c .
(u1 , v 1 , c1 ) and
a2 >
(u, v) which satises (49) and (50). One

(c) := u(0)(c). In order to solve the equation
(c) = , we study the function c 7 (c) on R+ . The results of Appendix show that for small
1
2
enough and a large enough, the equation (c) = admits exactly two solutions c and c , with
1
1/3
2
0 < c < (2)
and c > c , where c is positive and does not depend on and a. The proof
of Proposition 4 is then complete.
PROOF. For each
there is only one couple of function
can compute this functions explicitly. Let us set
Using Propositions 2, 3 and 4, we are now able to dene topological degrees. First, set
K I J ,
Xa . Coming back to the denition of J , with

V are in bounded in C 2 (Ia ), if u and v are in
C 1 (Ia ). Moreover C 2 (Ia ) is compactly embedded into C 1 (Ia ). Therefore the mappings J and K
are compact. Similarly, the mapping F : Xa [0, 1] (Xa 7 Xa ) dened by (u, v, c, ) 7 K
is compact and uniformly continuous with respect to . Moreover, one notices that (u, v, c) is a
classical solution of (16-17) if and only if K1 (u, v, c) = 0, therefore, we will look for the solutions
of this equation. From the properties of K , and using the homotopic invariance property of the
topological degree (see [79]), we will only need to compute the degree for = 0. At rst let us
prove that a degree can be dened. Set := min(, 1), M := f (1, 1) + 1 and
where
is the identity mapping of
(18) and (19), we can easily check that
and
(
=
(u, v, c) Xa : k(u, v, c)k(Xa ) < 2 + 2
)
M
1
(1 + ) .
(52)
From Proposition 2, one knows that
Proposition 5
For all
[0, 1],
and for
large enough,
0 6 K ().
(53)
Next, we are going to dene two opennsets of o

Xa , in which the topological degree will be calculated.
Let us set
c
, c where c and c are dened in Propositions 3 and
8
a := max{a1 , a2 } and := min{1 , 2 }, where a1 , a2 , 1 and 2 are dened in
>0
4 respectively,
such that
< min
Propositions 3 and 4.
Now, set
O1a := {(u, v, c) Xa s.t. c < 2} , and

O2a := {(u, v, c) Xa s.t. c > c 2} .
(54)
107

Notice that
O2a 6= {}
: since
< c ,
one deduces from Propositions 3 and 4 that
c2 > c (c2
is dened in Proposition 4), thus Proposition 2 ensures that
M
> c2 > c > c 2.
(55)
One has the following
Proposition 6
For all
< ,
for all
[0, 1]
one has
0 6 K (Oia ),
for
i = 1, 2,
and for
a a .
(ui , vi , ci ) be a solution of K = 0 in Oia (i = 1, 2). Then one deduces from
/ (, c ) (i = 1, 2),
Proposition 5 that c1 = 2 (and c2 = c 2). But, from Proposition 3, ci
thus one gets a contradiction, and the proposition is proved.

PROOF : Let
Proposition 6 enables us to dene the Leray-Schauder degree
deg(K , Oia , 0),
for
i = 1, 2.
Proposition 7
Therefore, one can compute its value :
Let us assume that

a
a
has deg(K , O1 , 0) = deg(K0 , O1 , 0)
< and a a ; for i = 1, 2, and for all [0, 1],

= 1 and deg(K , O2a , 0) = deg(K0 , O2a , 0) = 1.
PROOF. From the compactness of the mapping
K ,
one
and its uniform continuity with respect to
one obtains, using the homotopic invariance of the Leray-Schauder degree (see [79]),
deg(K1 , Oia , 0) = deg(K0 , Oia , 0).

K0
Furthermore,
is known explicitly :
K0 : Xa Xa : (u, v, c) 7 (u U0 (c), v V0 (c), u(0) )

where
U0
and
V0
are the solutions of (18-19). Moreover, this mapping is homotopic to
: Xa Xa : (u, v, c) 7 (u U0 (c), v V0 (c), (c) ),

where
< 2 , c1 < and, as

2
c are dened in Proposition 4). Hence
is dened as in the proof of Proposition 4. Besides, since
2
it was noticed in (55), c
> c > c 2
1
(c and
q
(0, 2) (resp. in (c 2, M
)). Using the
a
a
multiplicative property of the degree, one nds that deg(, O1 , 0) = 1 and deg(, O2 , 0) = 1
0
for a a (the sign of the degree is given by the sign of , see [23]).
(c) =
the equation
Remark 6
The real
admits exactly one solution in
can be chosen as small as one wants, provided
is suciently small.
As a consequence of Proposition 7, it follows that :
Corollary 1
For < , and for a > a , the problem (16-17) admits at least two classical
c
3c
a a a
a
a
a a
solutions (u1 , v1 , c1 ) and (u2 , v2 , c2 ), with 0 < c1 <
and
< ca2 . Moreover, from Remark 6,
4
4
a
one can assume that c1 < r , where r does not depend on a and r 0 as 0.
108

2.3
R+
< , and for a large enough, let (ua1 , v1a , ca1 ) and (ua2 , v2a , ca2 ) be the solutions obtained in
a a a
a a a
Corollary 1. Using Proposition 2, one nds that, for a large enough, (u1 , v1 , c1 ) (resp. (u2 , v2 , c2 ))
2
2
is bounded (independently of a) in C (Ia )C (Ia )R+ . By compactness, there exists a sequence
1 (R ) C 1 (R ) R
(an )nN such that (ua1n , v1an , ca1n ) (resp.(ua2n , v2an , ca2n )) converges in Cloc
+
+
+
loc
to (u1 , v1 , c1 ) (resp. (u2 , v2 , c2 )). One may use the general notation (u, v, c) for (u1 , v1 , c1 ) and
(u2 , v2 , c2 ) in the sequel, when the distinction between the two solutions is not needed. (u, v, c)
satises :

u00 + cu0 = f (u, v) h(u)
on R,
(56)
v 00 + cv 0 = f (u, v)
For
with the boundary conditions,
u0 (0) = cu(0) + k(, c, ), u(0) = , v 0 (0) = cv(0) c.

Moreover, since the real numbers
, c
and
0 c1 r
(57)
of Corollary 1 are independent of

and
a,
one has
3c
c2 .
4
(58)
Now let us prove that
Lemma 11
The two solutions found above satisfy
u(+) = 0, u0 (+) = v 0 (+) = 0
PROOF : Passing to the limit
a +
and
u00 (+) = v 00 (+) = 0.
in Proposition 2, we nd that
nonincreasing. Let us integrate the equation satised by
v 0 (x) cv(x) = v 0 (0) cv(0) +
v in
Z x
(56) from
v
to
is nonnegative and
x > 0.
One obtains
f (u, v)(s)ds.
(59)
0
Therefore, since
x +.
Next, let us
from
to
f (u, v) 0 on R+ , the right-hand side of the above equation converges as

v 0 (+) is dened. Since v is bounded, v 0 (+) = 0.
add the equations satised by u and v in (56). That gives, after an integration
Hence
x>0
v 0 (x) u0 (x) + cu(x) + cv(x) =

Z x
0
0
v (0) + u (0) cu(0) cv(0)
h(u)(s)ds.
(60)
(61)
0
is nonnegative (because
u 0), one
u(+) = 0. Now, using
has
follows that
the
v 0 (+), the equation (61) implies that u0 (+) exits. Since u is bounded,
= 0. Then u00 (+) = v 00 (+) = 0 immediately follows from (56).
one
Since the left-hand side is bounded, and since
lim h(u)(x) = 0.
x+
From the hypothesis (7) on
h(u)
h, it
above results on
0
nds u (+)
From Proposition 1, we have found two distinct solutions of (1-2). In order to complete the
u(0) =
u(+) = 0, u is not a trivial solution. Let us assume that v is a constant. Then, it follows
00
0
from (56) that f (u, v) = 0 on R, hence u + cu + h(u) = 0 on R. Since u() = 0, one
obtains that u 0 from the maximum principle, which is in contradiction with u(0) = .
proof of Theorem 1, it only remains to prove that this two solutions are nontrivial. Since
and
109

The upper bound of
1 u(0)
v(0)
,
min{1, }
1
.
v(0)
min{1, }
2.4
v(+)
and since
in Theorem 1 follows from Proposition 2. Indeed, one has
is nonincreasing, nontrivial, and
u(0) = ,
one has
v(+) <
Part 1) of Theorem 1 is then proved.
In this subsection, one studies the behavior of the two solutions

found above. Let us recall that, besides satisfying (1-2),
(ui , vi , ci )
(u1 , v1 , c1 )
and
(u2 , v2 , c2 )
veries
u0i (0) = ci ui (0) + k(, ci , ), ui (0) = , vi0 (0) = ci vi (0) ci ,

with
0 c1 r
and
r
kukC 1 (Ia ) 1 +
for
i = 1, 2.
Let 0.
(ui , vi , ci )
Since
and
3c
c2 ,
4
(63)
M
1
M
(1 + ), kvkC 1 (Ia ) 1 + ,
is bounded independently of
in
C 2 (R) C 2 (R) R+
(ui , vi , ci ) converges, up
0 0 0
a solution (ui , vi , ci ) of
(1) and (64), by compactness it follows that

subsequence, in
(62)
1 (R) C 1 (R) R to
Cloc
+
loc

(u0i )00 + c0i (u0i )0 = f (u0i , vi0 )
(vi0 )00 + c0i (vi0 )0 = f (u0i , vi0 )
on
(64)
from
to the extraction of some
R,
(65)
with
(u0i )0 (0) = c0i , (vi0 )0 (0) = c0i vi0 (0) c0i , u0i (0) = ,
since
k(, c, 0) = 0
(see (19)), and
c01 = 0,
since
(66)
lim r = 0. Therefore, since 0 <
3c
c02 ,
4
(67)
3c
c02 , one can apply Lemma (10) for (u02 , v20 , c02 ), which
4
gives
u02 (+) = 1
Besides, since for each
and
v20 (+) = 0.
> 0, u2 on R , one has

(u02 )00 + c02 (u02 )0 = 0
(v20 )00 + c02 (v20 )0 = 0
on
R .
(68)
(69)
R and v2 is nonincreasing on R ; it follows that u02 is non0

decreasing on R and v2 is nonincreasing on R . Therefore, from (66) and (69) one has
Moreover,
u2
is nondecreasing on
u02 () = 0
Hence,
(u02 , v20 , c02 )
and
v20 () = 1.
(70)
is a solution of (8-9).
(u01 , v10 , c01 ). As seen in (67), c01 = 0. Since u1 0 and v1 0 for all ,
0
0
0
0
one has u1 0 and v1 0. It then follows from (65) that u1 is concave. Since u1 is bounded, it
0
0
0
follows that u1 u1 (0) = . Similarly, v1 K1 , where K1 is an unknown constant.
Let us study now
110
3. Proof of theorem 2
Let us assume now that
g(y) = y
and
on
R.
Then, as mentioned in Remark 2 the
problem (8-9) has a unique solution. Since one has just demonstrated that for all converging sub-
(u2 , v2 , c2 ), (u2 , v2 , c2 )k
converges to a solution of (8-9), this uniqueness
result allows us to say that the whole sequence
(u2 , v2 , c2 ) converges to the solution of (8-9).

= 1, for the general reaction term f (x, y),
sequence
(u2 , v2 , c2 )k
of
Similarly, it follows from Remark 3 that, when

the whole sequence
(u2 , v2 , c2 )
converges to the solution of (8-9).
That concludes the proof of of Theorem 1, part 2).
3 Proof of theorem 2
In this section, we give the proof of Theorem 2, establishing some results about the general
solutions of (1-2).
3.1
An upper bound for
Let us prove that problem (1-2) has no solution for
Lemma 12
Let
(u, v, c)
large enough.
be a nontrivial solution of (1-2) with
PROOF : Assume by contradiction that
c = 0.
u0
and
v 0.
Then
c > 0.
Then, from (1), one has
v 00 = f (u, v) 0
on
(71)
R,
v is a convex function. From (2), it follows that the function v is constant. Hence, (71) gives
f (u, v) = 0, thus, from (1), one has u00 = h(u) 0. Similarly, it follows from the boundary
conditions (2) that u is a constant function. The solution (u, v, c) is then trivial.
thus
Lemma 13
and
Let
(u, v, c)
be a solution of (1-2) with
u0
and
v 0.
Then
is nonincreasing
v 1.
PROOF : The equation satised by

Integrating this equation between
ve
is equivalent to
c
x
and
0
+,
f (u, v)
Lemma 14
R.
one obtains, using (2) and
v0e x 0
therefore,
on
is nonincreasing, and from (2) one has
on
v 0,
(72)
R,
v 1.
Under the assumptions of Lemma 13, one has
u 1.
w = u+v then w satises the following equation : w00 +cw0 = (1)v 00 h(u).
between and x R, one obtains, using (2),
PROOF : Set
Integrating it
w0 (x) + cw(x) c ( 1)v 0 (x)
on
R,
111

since
h(u) 0.
x and +, one has

Z +
cx
w(x) ( 1)e
v 0 (t)ect dt + 1,
Therefore, integrating between
(73)
x
then, if
> 1,
one has
u+v 1
since
v0 0
from Lemma 13. Next, if
1,
we deduce from
(73) that
w (1 )v + 1,
thus,
u 1 v .
Finally, in both cases,
u 1,
(74)
and the lemma is proved.
u between and + ;
Z
+
+
f (u, v)
h(u) = 0.
Now, let us integrate the equation satised by
we obtain, from (2),
(75)
u0 () = 0, see for instance Lemma 11). Moreover, from Lemmas 13

0 v 1 and u 1, therefore, using the hypothesis (3-7), one obtains that
(Indeed,
f (u, v) f (1, 1)
and 14, one has
h(u)
.
h()
Therefore, from (75), we deduce that
h(u)
f (1, 1)
h()
thus
h(u)
f (1, 1)
.
h()
That completes the proof of Theorem 2, part 1).
3.2
In order to establish this lower bound, we need some more computations. Assume that the
R, and let (u, v, c) be a nontrivial solution of (1-2).

Let us show at rst that v(+) 6= 0. Since the solution (u, v, c) is nontrivial, one can assume
that it satises u(0) = (see Proposition 1). Thus one can dene x0 , as the smallest y [0, +)
such that u(y) = and u for all x y . Hence (u, v, c) satises the following problem

u00 + cu0 + h(u) = 0
on (x0 , +),
(76)
v 00 + cv 0 = 0
function
is Lipschitz-continuous on
u(x0 ) = ,
u(+) = 0,
0
v (+) = 0.
v(+) on (x0 , +). Moreover, using Proposition

Rv+
x0 h[u+ (s)]ds, where u+ is the unique solution of
It immediately follows that

knows that
u0 (x0 ) = c

112
u00+ + cu0+ + h(u+ ) = 0

u+ (x0 ) = , u+ (+) = 0.
on
(x0 , +),
(77)
1 (14), one
(78)
Besides, if
v(+) = 0,
one has
v(x0 ) = 0, v (x0 ) = 0, u(x0 ) =
u (x0 ) = c
and
h[u+ (s)]ds.
x0
It follows from the Cauchy-Lipschitz uniqueness theorem that
v := v(+) > 0.
Now, dividing by v the equation satised by v in (1),
0
obtains, using v > 0, v 0 (see Lemma 13) and (2),
v 0,
hence a contradiction.
and integrating by parts over
Z 0 2
Z
v
f (u, v)
+ cln(v ) =
,
v
R v
R
R,
one
(79)
and it follows that
Z
cln(v )
R
Let
be the Lipschitz constant of
one obtains
f (u, v)
.
v
(80)
g(v)
K on R,
v
Z
cln(v ) K
p(u).
g.
One has
thus, since
f (u, v) = p(u)g(v),
(81)
R
Since
p(u) h(u)
p(1)
,
h()
it follows from (81) that
p(1)
cln(v ) K
h()
and adding the equations in (1) and integrating over
Z
h(u) =
R
Z
h(u),
(82)
one obtains
c
(1 v ).
(83)
From (82) and (83), one deduces that
ln(v ) K
p(1)
Kp(1)
(1 v ) >
.
h()
h()
(84)
Finally one deduces that
Kp(1)
> exp
h()

,
and part 2) of Theorem 2 is proved.
3.3
Let
x0
be dened as in the subsection 3.2. One can again assume (up to translation) that
every solution
(u, v, c)
of (1-2) satises
u(0) =
and
on
(, 0).
113
3.3.1
Comparison with an adiabatic problem

1. As
solution (us , cs ) :
In this subsection, one assumes that

following problem admits a unique
it was proved in [23], one knows that the
u00s + cs u0s = f (us , 1 us ),
(85)
us () = 0, us (+) = 1
Moreover,
us
is strictly increasing, therefore, setting
k(y) := w0 w1 (1 y),
moreover,
and
us (0) = .
w = 1 us ,
k C 1 (0, 1), k > 0
on
one can dene a function
(0, 1)
and
Then one need the
Lemma 15
on
Let
(u, v, c)
be a solution of (1-2) with
c > 0, u 0
(86)
and
by
cs
k() = u0s (0) =
.
v 0.
Then
is decreasing
(, x0 ).
PROOF : It is similar to that of Lemma 13, using
v 0 (x0 ) = 0.
j(y) := v 0 v 1 (1 y). Then, from Lemma 15, j

C 1 ([1 v(0), 1 v ]). Moreover, j > 0 on (1 v(0), 1 v ), and
Let us set
j(1 v(0)) =
The equation satised by
is well dened and
c
(1 v(0)), j(1 v ) = v 0 (x0 ) = 0.
(87)
in 1 veries
v 00 (v 1 (1 y)) + cv 0 (v 1 (1 y)) = f (u(v 1 (1 y)), 1 y),

for
y in (1v(0), 1v ). But, since 1, one knows from the proof of Lemma 14 that u 1v
thus
v 00 (v 1 (1 y)) + cv 0 (v 1 (1 y)) f (1 v v 1 (1 y), 1 y),

for
in
(1 v(0), 1 v ),
which nally gives
(jj 0 cj)(y) f (y, 1 y)
(1 v(0), 1 v ).
(88)
(kk 0 cs k)(y) = f (y, 1 y) in (0, 1).

cs y 0
Moreover, since k > 0, one deduces from (89) that
k(y)
< 0,
between and 1 v(0), one gets
(89)
in
Similarly, one has
k(1 v(0)) <
therefore integrating
cs
cs
(1 v(0)) = j(1 v(0)),
by (87),thus
cs
k(1 v(0))
< .
j(1 v(0))
c
(90)
Then, subtracting the equations (88) and (89), one obtains,

0
2
cs
j k2 c j k
2
c
114
in
(1 v(0), 1 v ).
(91)
Now, assume that
cs
< 1,
c
then, from (90) and (91),
0
j 2 k 2 (1 v(0)) > 0, and

j 2 k 2 (1 v(0)) > 0.
(92)
(93)
{y (1 v(0), 1 v ) s.t. (j 2 k 2 )(y) = 0} is nonempty and

2
2
2
2 0
admits a lower bound y1 . Then (j k )(y1 ) = 0 and, from (91), it follows that (j k ) (y1 ) > 0
2
2
which is impossible from the denition of y1 . Therefore, j > k on (1 v(0), 1 v ), which is
impossible since (87) gives j(1 v ) = 0, and k > 0 in (0, 1). Finally, one obtains
Let us assume now that the set
cs
1,
c
(94)
and Theorem 2, part 3), is proved.
3.3.2
Computation of explicit upper bounds for c
Let us assume that
1,
and set
f (s, 1 s)
,
s
s(,1)
1 = max
one obtains, using (89),
k0
cs
1 y
k(y)
for
y (0, 1).
csp> 2 1 . Now, as done in [72]

cs + c2s 41
r=
. Then one has the
2
Let us assume that
m(y) = ry ,
with
Lemma 16
(95)
k(y) > m(y)
for all
(96)
on bounded intervals, one can set
y (, 1).
PROOF : First, let us notice that
m0 (y) =
Therefore, since
cs
,
m() = r <
1 y
cs
m(y)
and
for
y (0, 1).
k() = w0 (0) =
cs
,
(97)
one has
k() > m().
Moreover,
from (96) and (97),
1 y
(k m) (y)
It follows that
k(y) > m(y)
Hence, for all
on

for
x (0, +), m(1 w(x)) < k(1 w(x))

r<
us
1
1
m(y) k(y)
y (, 1).
(, 1).
equivalent to
where
(us )0 (x)
us (x)
for
ln
us (a)
r(1 w(x)) < w0 (x)
x (0, x0 ),
is dened by (85-86). Integrating (98) between
thus
and
which is
(98)
a > 0,
one obtains

> ra.
(99)
115

It follows from the denition of
which is in contradiction with
r that, for a chosen large

wants,
enough, cs can be as small as one
the hypothesis cs 2 1 . Finally, it follows that cs < 2 1 ,
and from (94), one has
p
c < 2 1 .
c, which depends on , but works for all 0.
Let us now compute another upper bound for

Let us set
2 = max f (1 s, s),
s[0,1]
and let
be the solution of
w00 + cw0 = 2 on (0, a),

w0 (0) = cw(0) + k(, c, ),
w(a) = 1,
where
k(, c, )
is dened as in (15). The function
can be computed explicitly :
w(x) = X1 (a) + X2 (a)ecx +

where
X1 , X2
2
x,
c
(101)
are two real valued functions.
Moreover, the equation

since
(100)
w(a) = 1.
w0 (0) = cw(0)+k(, c, ) gives X1 (a) =
Finally, one gets
Since the function
w(0) = X1 (a) + X2 (a)
2 k
c2
c
2 k
, and lim X2 (a) = 0
a+
c2
c
as
a +.
u, one deduces from the

that w u on (0, a), hence
is a super-solution of the equation satised by
w together with the maximum principle,

w(0) = X1 (a) + X2 (a) u(0) = . Therefore, taking a large enough,
boundary conditions of
r
c<
one deduces that
2
.
The proof of Theorem 2 is complete.
4 Numerical results
In this section we give some numerical approximations of the two solutions obtained in
[0, a]. Numerically, no other solution has been found. A

c and v(0) has been used to compute these approximations, and the solutions
extended to [a, a].
Corollary 1, on a bounded domain

shooting method on
have been
We have taken here
h(u) = u.
116
The values of
f (u, v) = vH(u )(u )2 , where H is the Heaviside function,

the parameters are : = 1, = 0.001, = 0.2 and a = 20.
and
4. Numerical results
One can notice that the second solution is close to the adiabatic one.
Appendix
Let us recall the notations of Proposition 4 :
(c) =
c
1
1
( a
+ a
1 e
r2 + r2 c e
r 1 + r1 c
1
1
),
ea z2 + z2 c ea z1 + z1 c
117

and
:=
c2 + 4, :=
It is straightforward that
Proposition 8
For
<1
c
c +
c
, r2 =
, z1 =
,
2
2
2
C 2 ([0, +)) function. Set 0 :=

.
c
c2 + 4, r1 =
is a
and
large enough,
vanishes only once on
and
z2 =
c+
.
2
(0, +).
PROOF : Let us show at rst that one has
Lemma 17
Let be the function such that, for a large enough and for all c in (0, +),
2
c
c
2
(
1 c+
c+ ) + (c, a). Then the function a (c) := c 7 (c, a) is in C ([0, +)) for
enough and
| ((c, a))(n) |
where
((c, a))(n)
is the
nth
a
4
e 2
1
for all
derivative of the function
(c) =
a large
n {0, 1, 2},
with respect to the variable
c.
c
c
2
(c, a) := (c) 1
( c+
c+
), it easily follows that c 7 (c, a)
C ([0, +)) function for a large enough. Moreover, one can notice that

(n)

a
c

a
e 2 , for all n {0, 1, 2}.
e

r 2 + r2 c
PROOF : Setting
2
is a
One similarly has

(n)

a
c

a
e 2 ,
e z2 + z 2 c

for all
n {0, 1, 2}.
Furthermore, one can also easily check that

(n)

a
2c
c

+ a
e 2 ,

r + r c
c + e

1
1
for all

(n)

a
c
2c

+ a

e 2 ,

c+ e
z1 + z 1 c
n {0, 1, 2}
for all
and
n {0, 1, 2}.
Finally, the lemma 17 follows from the above calculations.
0 (c, a) for
(c, a),
c
c
c
b(c) :=
.
c + c +
In the sequel, one writes

Let us set
Lemma 18
The rst derivative of
PROOF : Let us notice at rst that
and
00 (c, a)
vanishes only once on
0 =
for
(c, a).
c2
(0, +).
c
1
c2
c
, 0 = , 00 =
3
and
00 =
Then one has

2
c
c
c + c(1 + 0 ) c + c(1 + 0 )
.
b (c) =
(c + )2 (c + )2
(c + )2
(c + )2
0
118
c2
1
3.

Setting
k(c) = b0 (c)(c + )2 (c + )2 ,

k(c) =
c2

one obtains
c2
(c + )
(c + )2 ,

2
2
k 0 (c) = 2(c + )(1 + c ) c 2(c + )(1 + c ) c

i
h
2
2
+c 1 + c 3 (c + )2 + 1 c 3 (c + )2 . Hence,

1
1
0
3
+
k (c) = 5c + 4c ( ) + 6c2 + c5 13 13 + 2c4 3
3

2
2

2
0
2
+ c
c3
. Therefore, with obvious notations, one can write k (c) = (t1) +
3 3
thus one has
(t2) + (t3) + (t4) + (t5) + (t6) + (t7).

Moreover,
(t2) =
(t3) =
=
(t4) =

1
1
4c
,

1
1
6c2
2 ,
2

1
1
2
6c ( + )
,

1
1
1
1
1
+
c5
+
,

2 2

1
1
(t5) = 2c4
,
4 4

1
1
1
1
4
= 2c
+
+ +
,

2 2

1
1
3 2 2
(t6) = c
,
5 5
2

2
1
1
3
= c
+
+1+
+ 2

2

1
1
2 2
(t7) = c
,
3 3

1
1
= c 2 + + 2
.

Therefore, one can write
k 0 (c) = c
I(c),
with
1
1
1
+
+ 2
2

I(c) = 5c 4 6c( + ) + c

2

1
1
2
3
2
+ 2c
+
+ +
+c
+
+1+
+ 2
2 2
2
(2 + + 2 ).
119

Moreover, since
I(c) 8c2 . Finally,

1
1
0
3
k (c) 8c
< 0.

c < < ,
one has
k(c)
Moreover, from a series expansion of
at
c = +,
(102)
one obtains
lim k(c) = 8( 1) < 0,
(103)
c+
and, since

k(0) = 8 8 = 8 1 > 0,
k
one deduces from (102-104) that

the lemma 18 is proved.
vanishes only once, and since
(104)
k(c) = b0 (c)(c + )2 (c + )2 ,
Let us now prove Proposition 8.

One has
0 (c) =
2 0
1 b (c)
+ 0 (c, a),
thus, from a straightforward computation,
1
(0) =
+ 0 (0, a).
(1 )
0
Therefore, one deduces from Lemma 17 that for
large enough,
0 (0) >
1 1
> 0.
2 (1 )
(105)
Besides, it follows from (103) that
b0 (c)
Thus it follows from Lemma 17 that
0 vanishes at least once on
1
0
2c2
0 (c) <
(0, +).
as
c +.
1
<0
2c2
large enough. Hence, by continuity,
b0 (c) < 0 for c large enough. Therefore,

0
0
it follows from Lemma 18 that there exists a unique c0 > 0 such that b > 0 on [0, c0 ) b (c0 ) = 0
0
0
0
et b < 0 on (c0 , +). Let c1 and c2 , be two points in R+ such that (c1 ) = (c2 ) = 0. Then
From (104) and (106), one knows that
b0 (0) > 0,
for
(106)
|b0 (c1 )|
and
1 0
| (c1 , a)|.
2
Using Lemma 18, and from the continuity of
b0 , |c0 c1 | 0
(107)
as
b0 (c ) 0.
Hence Lemma 17
and (107) give
|c0 c1 | 0
Assume that
there exist
c1 6= c2
and set
c3 [c1 , c2 ]
such that
(108)
1 0
(c)(c + )2 (c + )2 , then l(c1 ) = l(c2 ) = 0
2
l0 (c3 ) = 0, since c1 6= c2 . Thus, since
and
1
(c + )2 (c + )2 0 (c, a),
2
k 0 (c3 ) = t1 (c3 )0 (c3 , a)+t2 (c3 )00 (c3 , a), where t1 and t2 are two functions with polynomial
increasing.
120
a +.
l(c) :=
l(c) = k(c) +
one has
as
From (108), one can take
large enough such that
k 0 (c) 8c3 (
Thus, since (102) gives

3
8c
k (c3 ) max
c
0
c>
0
2
1
1
),

1
1
(c) (c)
But, from Lemma 17, one can assume for
|c0 c1 | <
c0
,
2
and since
c0 > 0, c1 >
c0
.
2
one has

c30
1
1
c0
( 2 ) ( c20 )

< 0.
(109)
large enough that
3
0

k (c3 ) = t1 (c3 )0 (c3 , a) + t2 (c3 )00 (c3 , a) < c0
2
1
1
c0
( 2 ) ( c20 )

,
diction, and one

1
1
c30
1
1
<
. Hence one gets a contra ( c20 ) ( c20 )

2 ( c20 ) ( c20 )
concludes that c1 = c2 . Proposition 8 is then proved.
Proposition 9
There exists c
3
therefore, (109) gives c0
(c) =
admits exactly two
> 0 such that for small enough and a large enough, the equation
1
1/3
1
2
2
solutions c and c , with 0 < c < (2)
and c > c .
PROOF : Let us begin with the lemma
Lemma 19
The function
admits a unique maximum.
PROOF : One has seen in (105) that
(+) = 0.
0 (0) > 0
for
large enough. Moreover,
(0) = 0
and
The proof of Lemma 19 then follows from Proposition 8.
Let us now compute

(2)1/3 .
A series expansion at
=0
gives
h
i
3
(2)1/3 = 1 2/3 1/3 + ( , a).
2

1/3 can be as close as one wants to 1.
Therefore, for a large enough and small enough, (2)
It then follows from Lemma 19, with (0) = 0, (+) = 0 and < 1, that the equation (c) =
1
2
1
1/3 .
admits exactly two solutions c and c , with 0 < c < (2)
Next, a series expansion of (c) at = 0 gives
c2 + 4 c
(c) =
+ O() + ( , a),
c2 + 4 + c
and, with another series expansion at
c = 0,
one obtains
(c) = 1 c + O(c2 ) + O() + ( , a),

< 1, there exists c > 0 independent of and a, such that for small enough and
2
enough, (c ) > . It then follows from Lemma 19 that c > c . Proposition 9 follows.
thus, since
large
121
122
Partie III : Une premire approche de

la modlisation en biologie
Sommaire
1
2
3
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Aspect pratique, site et collecte des donnes . . . . . . . . . 125
Formulation d'un modle spcique . . . . . . . . . . . . . . . 125
3.1
Hypothses comportementales utilises dans la construction du

modle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
5
3.2
Modlisation de l'volution pendant la priode de vol
3.3
125
. . . . .
127
. . . . . . . . . . . . . .
129
Estimation des paramtres . . . . . . . . . . . . . . . . . . . . 129

Tests numriques . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.1
. . . .
131
5.2
. . . . . . . .
131
5.3
. . . . . . . . . . . . . . . .
133
6.1
Validation du modle
. . . . . . . . . . . . . . . . . . . . . . .
133
6.2
6.3
Limites
6.4
Rsultats venir
. . . . . . . . . . . . . . . . . . . . . . . . . .
136
6.5
Adaptation d'autres espces vivantes . . . . . . . . . . . . . .
137
Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
. . . . . . . . . . .
135
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
136
Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Ce travail s'intgre dans le cadre du projet Caractrisation des potentialits invasives d'un
nouveau ravageur des graines de cdre rcemment introduit au Mont-Ventoux conclu entre
l'INRA et le Ministre de l'Agriculture (convention DERF-INRA Num. 61.45.19/01).
123
1 Introduction
L'insecte dont nous allons modliser la dispersion, un ravageur au nom de Megastigmus schi-
mitscheki Novitzky, a vraisemblablement t introduit depuis le Proche-Orient avec des graines

de cdre du Liban (Cedrus libani A. Rich.).
Dans une situation marque au plan international par la multiplication des invasions biologiques, l'tude de ce processus invasif prsente de nombreux intrts, tant au plan fondamental
qu'appliqu. Il s'agit d'une invasion rcente, de dispersion encore trs limite, les tudes menes
au Mont-Ventoux ([37], [38], [40]) ayant permis de dater avec prcision l'anne d'introduction
1994. Cela facilite l'tude des processus de dispersion spatio-temporelle partir du point source,
au contraire d'espces plus anciennes et maintenant rparties sur l'ensemble du territoire.
L'tude prcdente eectue dans le cadre de ce projet ([40]) avait pour but, entre autres,
d'estimer l'impact de M. schimitscheki sur la rgnration naturelle des forts de cdre, et de comparer les caractristiques biologiques des populations de la zone d'invasion avec celles observes
dans leur milieu naturel (Asie Mineure). Elle nous fournit ainsi des donnes pour la construction
du modle. Par ailleurs, si le processus d'invasion et d'installation de M. schimitscheki n'tait
que peu document, sa biologie a t relativement bien tudie dans les rgions d'origine ([50],
[63], [81], [82]). Ainsi, chaque insecte se dveloppe dans une seule graine et, dans les rgions
d'origine, chaque espce est spcialise dans l'exploitation d'essences du mme genre vgtal. La
reconnaissance de l'hte est sous-tendue par des signaux visuels et olfactifs mis par les cnes
de cdre ([71], [83]). Par ailleurs, les
Megastigmus disposent d'une batterie de traits adaptatifs
permettant leurs populations de faire face aussi bien l'absence de graines sur une ou plusieurs
annes (diapause prolonge larvaire [90]), que de s'aranchir de la rencontre des sexes (parthnognse [81]). L'ensemble de ces mcanismes d'chappement constitue un avantage vident dans
les zones d'invasion, et contribue l'intrt de la construction d'un modle.
Nous savons de plus que le cdre est galement une espce vgtale introduite, qui n'a t
encore que peu colonise par la faune autochtone (un seul autre prdateur, M. pinsapinis, qui
semble avoir des capacits invasives infrieures celles de M. schimitscheki [40]). Cela contribue
la simplicit du modle tudi, tout comme le fait que M. schimitscheki, tant une espce
introduite, n'a pas de prdateur.
Nous utilisons ici deux modles imbriqus. L'un sert simuler la ponte et la phase larvaire,
en particulier la diapause prolonge, l'autre permet de calculer la dispersion des adultes. C'est un
modle de raction-diusion, avec un terme de transport. Les modles de raction-diusion sont
souvent utiliss en dynamique des populations, en particulier dans les cas o l'on ne peut suivre
les individus et observer leur mouvement (voir les livres de P. Turchin, [89] ou N. Shigesada et
K. Kawasaki [85]). Rappelons que la dispersion passive se distingue du mouvement actif par la
source d'nergie utilise pour accomplir les dplacements. Ici, on parlera de dispersion semipassive, car l'insecte peut interfrer avec la dispersion passive, en modiant son altitude de vol,
s'il se trouve prs d'un bassin d'attraction (une fort). Le mouvement se fait dans un espace
trois dimensions, (on se place sur une carte en deux dimensions, et on peut faire intervenir
l'altitude sur des paramtres tels que la date d'mergence).
Nos objectifs initiaux taient (i) de construire un modle reproduisant dlement le comportement de M. schimitscheki, (ii) d'utiliser ce modle pour approfondir la comprhension des
processus d'invasion, et (iii) de construire un modle pouvant ventuellement s'adapter plusieurs espces (le projet concernant en eet galement l'aspect comptition cf. [40]), pour dnir
des critres plus ou moins favorables l'expansion.
124
2. Aspect pratique, site et collecte des donnes
Aspect pratique, site et collecte des donnes

Des collectes annuelles standardises de cnes mrs ont t organises par les agents de
l'INRA Avignon depuis 1994 dans la plupart des cdraies du Sud-Est de la France, le nombre
de cnes collects variant de 8 139 selon l'importance de la fructication. Cette dernire a t
estime par les services de l'Oce National des Forts.
Aprs rcolte, les cnes ont t dsarticuls et leurs graines radiographies an d'valuer les
proportions respectives de graines pleines, attaques et vides, qui se sparent aisment par radiographie. A part M. schimitscheki, une autre espce introduite, M. pinsapinis, peut infester les
graines et apporter une confusion l'examen des radiographies. Les graines attaques seront donc
surveilles en closoir jusqu' l'mergence des adultes, an d'apprcier la proportion respective
de chaque espce, et d'estimer les dgts imputables M. schimitscheki.
Le calcul du pourcentage d'attaque se fait de la faon suivante :
%attaque
nombre de graines attaques

nombre de graines attaques + pleines
100,
(1)
on fait donc implicitement l'hypothse, discutable (voir la section 6.3), que les insectes n'attaquent que les graines pleines (pollinises).
Les donnes concernant les vents sont fournies par des stations Mto France rparties sur le
territoire tudi.
Formulation d'un modle spcique
3.1
Hypothses comportementales utilises dans la construction

du modle
3.1.1
Hypothses simplicatrices
Nous avons mis en lumire les aspects importants du problme, et ceux qui sont d'une importance moindre, pour tablir quels sont les paramtres qui doivent tre pris en compte dans la
construction du modle, et ceux qui peuvent tre carts. Cela nous permet nalement d'avoir
une formulation plus claire du problme (tape dcrite par Turchin [89]).
Ainsi, les donnes dont nous disposons [40] rendent compte d'un sex-ratio peu prs constant
au moment de l'closion (environ 50% de mles, 50% de femelles), et le mcanisme de parthnognse ne rend pas indispensable la fcondation des femelles [81]. Nous ne modlisons donc
pas les phnomnes de fcondation et les interactions mle/femelle. Par ailleurs, si les variations
des dates d'mergence en fonction de l'altitude pourraient facilement tre intgres, nous ne les
prenons pas en compte dans les tests numriques, an de rduire le nombre de donnes entrer,
et de clarier le modle.
Nous reprsentons ci-dessous un cycle de vie simpli de l'insecte, sur deux ans, tel que nous
l'avons pris en compte pour l'laboration du modle.
3.1.2
Comportement
Les insectes closent progressivement sur une priode de 12-15 jours [40]. Le temps de vol
estim d'un insecte est d'environ 10 jours [59] ; en outre, les
Megastigmus ne volent que durant
la journe [80] ce qui ramne environ 7 jours de temps de vol eectif. Nous supposons qu'ils
125
Juillet anne n-2- avril anne n
Mai anne n juin anne n
Une partie des insectes merge. Lautre

partie attendra lanne n+1 ou n+2. Cest la
diapause prolonge , qui permet la survie
de lespce dans le cas dune anne n
fructification nulle.
Les insectes sont au stade

larvaire ; ils se
dveloppent dans les
graines.
n
devient
n+1
ou
ou
Lmergence stale sur
une priode de 15 jours.
Elle est quasi-constante.
La dure de vie
moyenne des adultes est
de 10 jours.
n devient n+2
Les adultes
meurent, les
ufs se
transforment
en larves.
Juin,
anne n
Si la femelle a
t fertilise, elle
pond environ
50% de mles et
50% de femelles.
Si la femelle na
pas t fertilise,
elle pond 100%
de mles.
Chaque
femelle pond
environ 10
ufs dans les
10 graines
trouves.
A lintrieur de la fort
dmergence, les insectes
diffusent alatoirement.
Certains sont ainsi amens
sortir de la fort.
A lextrieur des forts, les

insectes prennent de laltitude
avec leurs ailes, pour tre
davantage soumis aux vents.
Soit linsecte
arrive dans
une fort
avec une
fructification
non nulle, et
trouve une
graine
inoccupe.
Soit linsecte ne
trouve pas de
graine disponible
(chaque graine
ne contient
quune seule
larve)

Fig. 1 Cycle de vie simpli
126
3. Formulation d'un modle spcique

diusent au sein de la fort puis, quand les mcanismes de diusion les amnent en sortir,
ils se servent de leurs ailes pour s'lever et tre ainsi soumis des vents plus forts. Il est en
eet vident que le transport de l'insecte par le vent n'est pas passif (mme si ses ailes ne lui
permettent pas d'aller son encontre). En eet, on note exprimentalement (A. Chalon, donnes
non publies) qu'un grand nombre d'insectes est retrouv de grandes distances (100 km) du
point source d'anne en anne, ce qui n'est pas compatible avec une dispersion passive sur une
longue distance par le vent, en comparaison avec les distances parcourues par des graines (voir
par exemple l'article de Nathan et al. [74]). Par ailleurs, cette hypothse d'lvation des insectes,
et les distances parcourues releves exprimentalement, concordent avec les rsultats dcrits par
Tackenberg et al. dans [88] pour les graines. Il est en eet soulign dans cet article que la dispersion
sur une longue distance est en grande partie conditionne par les vents ascendants. Arriv une
certaine distance
da
d'une fort l'insecte rduit alors son altitude an d'tre moins soumis au
vent (attraction visuelle et olfactive [71], [83]) et de s'approcher de la foret par diusion. Au
cours de la priode de vol, nous supposerons que le coecient de mortalit,
est constant.
Le cycle de dveloppement simpli est rsum la gure 1.
3.2
Modlisation de l'volution pendant la priode de vol
3.2.1
Equation de raction-diusion
An de modliser le dplacement dcrit dans les sections prcdentes, nous utilisons l'quation
de raction-diusion suivante
u
(t, x) = Du(t, x) v V (t, x) u(t, x) + f (t, x) X(t, x)u,
t
pour
t [0, Nj ]
(o
Nj
est le nombre de jours total o des adultes sont prsents), et
est un ouvert susamment grand de
R2 ,
(2)
x ,
et
u0
sur
Explicitons chaque terme de l'quation (2), et donnons sa valeur.

Le terme
f (t, x)
correspond un terme de source, modlisant l'mergence progressive des
insectes. Les rsultats de [40] (Fig. 2), montrent un taux d'mergence quotidien des femelles
(rappelons que nous simulons le dplacement des femelles) peu prs constant sur la priode
Nem .
Soit
u0 (x)
le nombre d'insectes prts clore, avant l'mergence, la position
x R2 .
Alors,
f (t, x) =
u0 (x)
Nem
pour
t [0, Nem ],
et
f (t, x) = 0
pour
Nous devons ensuite ajouter un terme de mortalit quotidienne
t > Nem .
X(t, x)
(3)
durant la priode de
vol. Ce taux tant suppos constant, nous avons
X(t, x) =
1
dure moyenne de vol
pour
l'extrieur d'une fort,
(4)
nous supposons de plus que
X(t, x) = 0
pour
dans une fort,
(5)
la ponte n'tant en eet prise en compte qu' la n de la priode de vol, nous devons faire comme
si les insectes arrivs dans une fort y restaient jusqu' la n de la priode de vol.
127

Nous avons donc prcis le terme de raction. Venons-en la diusion. La valeur de
tant
ici dicile estimer pour des insectes qui sont trop petits pour tre marqus et recapturs, il
ne nous parat pas judicieux de faire varier
avec
et
(au moins dans un premier temps).
De plus, des modles pour d'autres types d'insectes se sont montrs ecaces avec des valeurs de
transport variables et un terme de diusion xe [9].
Le terme de transport est variable en temps et en espace, la variabilit en temps correspondant
aux variations quotidiennes des vents. Concernant la variation en espace, nous prendrons en
compte, en sus des vents mesurs, le critre d'loignement d'une fort. Nous supposerons en eet
que l'insecte ne dispose pas de critres dcisionnels lui commandant de sortir de la fort quand
celle-ci est sature en insectes, et qu'il n'en sort que par un mcanisme de diusion alatoire, et
qu'ensuite, comme prcis dans la section 3.1.2, il s'lve pour tre davantage soumis au vent,
en fonction de son loignement de la fort. Nous supposons de plus qu'il prend de l'altitude de
faon proportionnelle sa distance de la fort, jusqu' une distance maximale,
da .
Pour prciser
la force du vent laquelle il est soumis, nous utilisons un rsultat dcrit par Nathan et al. [74],
et utilis pour calculer la force du vent laquelle est soumise une graine pour une vgtation
rase. En eet, l'insecte s'levant susamment haut, nous pouvons supposer qu' l'extrieur de
la fort de cdre, le milieu correspond plus une vgtation rase qu' une fort dense. Il est
ainsi tabli dans [74] que le prol du vent, qui dcrit sa moyenne horizontale chaque altitude,
est logarithmique. La formule utilise pour calculer
est donc :
V (t, x) = Ve (t, x) min{ln[d(x, f oret)(e 1)/da + 1], 1},

o
Ve
est le vent enregistr, et
(6)
d(x, f oret) la distance la fort de cdre la plus proche. S'impose

v . En eet, si la distance parcourue par les insectes est
alors l'introduction d'un autre coecient
trs suprieure la dispersion longue distance passive de graines [74], elle est cependant infrieure la distance totale parcourue par l'air en mouvement (le vent). Cela pourrait s'expliquer
de direntes faons : pauses faites par l'insecte durant sa priode de vol ou encore turbulences
verticales du vent rduisant l'altitude de vol de l'insecte. Cependant, ces hypothses ne sont pas
vriables, et nous serons amens estimer le coecient
3.2.2
dans la section 4.
Rsolution numrique
La mthode utilise est du type directions alternes : c'est la mthode des pas fractionnaires,
avec un schma d'Euler implicite et un dcentrage en amont pour le terme d'ordre 1. Cette
mthode est propose par Godunov dans [7] et Yanenko dans [94] pour l'quation de la chaleur.
Cela s'crit
n+1/2
ui,j
uni,j
dt
n+1/2
un+1
i,j ui,j
dt
128
n+1/2
=D
n+1/2
ui+1,j 2ui,j
n+1/2
+ ui1,j
h21

un+1/2 un+1/2
i+1,j
i,j
h
h
+v Vi,j |Vi,j |
2h1

un+1/2 un+1/2
i,j
i1,j
n+1/2
h
h
n
v Vi,j
+ |Vi,j
|
+ fi,j
Xi,j ui,j , et
2h1
n+1
n+1
n+1
un+1
un+1
i,j+1 2ui,j + ui,j1
i,j+1 ui,j
v
v
=D
+ v Vi,j |Vi,j |
2h2
h22
n+1
n+1
u
u

i,j
i,j1
v
v
n
v Vi,j
+ |Vi,j
|
+ fi,j
Xi,j un+1
i,j ,
2h2
(7)
4. Estimation des paramtres

V h et V v sont respectivement
(6), dt, h1 , h2 les pas de temps et
o
les composantes horizontales et verticales du vent dni par

d'espace.
Il est prouv qu'un tel schma avec seulement des termes d'ordre 2 est inconditionnellement
stable au sens de Fourier [84], un calcul de la stabilit avec des termes d'ordre 1 tant plus
fastidieux. De plus, si la stabilit n'est pas prouve thoriquement, les dirents tests numriques
eectus au cours de ce travail, et dans [84], ont suggr que la mthode tait stable.
3.3
Soit
u(x)
n. Chaque insecte arriv

d'oeufs dans la limite des ressources disponibles. Ainsi,
le nombre d'insectes la n de la priode de vol, l'anne
dans une fort de cdre pond un nombre
o(x) = min{g(x), u(x)},

o
o(x)
est le nombre d'oeufs pondus et
de diapause prolonge 1 an et
d2
nouveaux adultes en mai de l'anne
g(x)
le nombre de graines disponibles. Soit
d1
le taux
le taux de diapause prolonge 2 ans. Alors le nombre de
n+2
sera
u(0, x) = o(x)(1 d1 d2 ) + o1 d1 + o2 d2 ,
o
o1
et
o2
sont le nombre d'oeufs pondus les 2 annes prcdentes.
La mortalit durant cette phase est donc suppose nulle, comme cela a t montr pour
des espces voisines [59]. Il n'existe en eet actuellement ni parasite ni prdateur dans la zone
d'invasion.
Estimation des paramtres

Certains paramtres vont pouvoir tre estims directement, d'autres vont ncessiter un re-
cours au calcul numrique.

Calculons d'abord la valeur du coecient de diusion. Dans [59], pour une espce proche,
on a
D = 1km2 /jour.
Dans le livre de N. Shigesada et K. Kawasaki [85], pour d'autres espces
d'insectes (piride du chou, qui est cependant un meilleur voilier), des valeurs direntes sont
cites :
D = 0.08 0.46km2 /jour. An de calculer la valeur de D
la plus en adquation avec nos
rsultats exprimentaux, nous allons utiliser une formule de [85] (p. 38), valable dans le cas o
est le seul paramtre inuant sur la dispersion des insectes. Rappelons que dans notre modle
les vents n'ont pas d'incidence sur le mouvement des insectes l'intrieur des forts. La formule
cite ci-dessus est
D=
o
< r >
< r >2
,
t
(8)
est le rayon moyen parcouru par les insectes, partir du point source, pendant le
temps t. En utilisant les mesures eectues dans la fort du Lubron, entre Oppede et Mnerbes,
nous estimons les dplacements moyens des insectes
< r >= 1km/jour.
Nous retrouvons des
valeurs similaires sur les autres sites de mesure. Ainsi, la formule (8) donne
D = 0.32km2 /jour,
valeur proche de celle mentionne dans [85] pour la piride du chou (d'aprs des travaux de
Andow et al. [4]).
129

X(t, x), peut tre calcule directement
avec les formules (4) et (5), en utilisant la dure de vol moyenne de 10 jours [59]. Le paramtre
Nem est galement connu, et sa valeur est comprise entre 6 et 11 jours [40]. Nous prendrons en fait
pour les calculs 2/3 de la dure de vol moyenne et 2/3 du nombre de jours d'mergence pour tenir
compte de l'activit uniquement diurne de l'insecte [80]. Le taux de fcondit est voisin de 10
La valeur du taux de mortalit durant la priode de vol,
oeufs par femelle ( voir [59] et rsultats rcents de A. Chalon, donnes non publies). Les taux de
diapause prolonge que nous utiliserons sont de
5%
un an et
15%
deux ans. Ces valeurs nous
ont t transmises par les biologistes. Il semble cependant que ces taux varient d'une anne sur
l'autre, en fonction de la fructication, et puissent prendre des valeurs plus importantes (voir la
section 6.1.2). Des expriences sont en cours pour tenter de mieux comprendre comment varient
ces paramtres.
Au sujet des vents, rappelons que la direction ainsi que la vitesse du vent (donc les paramtres
Vh
et
Vv )
proviennent de mesures directes de stations Mto France releves chaque jour ( 10
mtres du sol) durant la priode de vol. Nous devons donc uniquement estimer le coecient
et le paramtre
da .
La valeur du paramtre
da ,
correspondant la distance partir de laquelle
les forts n'ont plus d'inuence sur l'altitude de l'insecte, sera xe ici
10km
(aprs discussion
avec les biologistes).

Pour estimer
v ,
nous allons utiliser une mthode gnrale dcrite par P. Turchin dans son
livre [89], d'aprs les travaux de Banks et al. [8]. Nous devons commencer par faire une hypothse
initiale sur la valeur de
v .
Nous calculons pour cela le rapport entre la distance maximale
mesure d'avance du front d'une anne sur l'autre et la distance maximale parcourue par les
vents dans une mme direction pendant une priode de
des insectes). Nous obtenons ainsi
v = 0.012
15
jours (dure de vol eective de
6.5%
(les donnes utilises pour ce calcul sont relatives
aux sites d'Aurel et de Sisteron).

Aprs implmentation du modle (avec Matlab
r ), et rsolution numrique, nous confrontons
les prdictions aux donnes exprimentales pour la distance maximale parcourue par le front
d'une anne sur l'autre, an d'ajuster la valeur du coecient
v .
A partir des donnes dont nous disposons, et en les confrontant aux prdictions du modle,
nous pouvons donner l'intervalle suivant :
v = 0.04 0.08.
Ne disposant pas de la valeur de la
fructication sur les nouvelles zones d'invasion, utilises pour ce calcul, nous ne pouvons donner
de valeur plus prcise.
Tests numriques
Nous nous penchons sur l'inuence de direntes caractristiques biologiques d'une espce
sur sa propagation, sous plusieurs conditions de vents et de rpartition des forts de cdre.
Ainsi, sont tudis i) l'inuence de la dure de la priode d'mergence, ii) l'inuence du taux de
diapause prolonge et iii) l'eet de la prsence de cdres isols entre les forts. Cela nous permet
de souligner le rle de ces facteurs, et de voir dans quelles situations ils sont avantageux.
Pour chacun de ces points, nous eectuons plusieurs tests numriques, qui nous permettront
de rpondre aux questions ci-dessus, et de vrier dans la section 6.1.1 que les rsultats fournis
par le modle ont une interprtation logique du point de vue du biologiste.
Pour les dirents tests numriques eectus dans cette section, nous n'utilisons pas les
coordonnes relles des plantations de cdre, mais des forts modles, permettant une interprtation plus facile des tests. De mme, les vents seront choisis indpendamment des vents rels. Par
ailleurs, nous faisons varier certaines caractristiques des insectes, telles que le taux de fcondit
ou la diusion, an de souligner l'inuence des autres paramtres sur la propagation.
130
5. Tests numriques
5.1
Nous tudions ici l'inuence de la longueur de la priode d'mergence sur l'tendue de la

propagation des insectes, suivant notre modle.
5.1.1 Vents constants

Dans un premier temps, les vents sont supposs constants durant l'mergence et d'anne en
anne, et les forts sont disposes comme sur la gure 2.
La fructication est uniforme (10hl de graines par plantation), et le pourcentage d'attaque
initial est nul partout sauf en
d7
(che sur la gue 2), o il est de
50%.
La gure 3 prsente la
dispersion suivant la longueur de la priode d'mergence, sur une dure totale de
ans.
Conclusion : la longueur de la priode d'mergence n'a pas d'inuence signicative sur la

dispersion dans le cas de vents constants.
5.1.2 Direction des vents variable

Ici, les vents sont choisis constants orients vers le Nord pendant
jours, puis vers le Sud
(avec la mme vitesse moyenne). Les forts et les vents sont reprsents sur la gure 4.
A nouveau, le pourcentage d'attaque initial est nul partout sauf en
Le taux de fcondit
de
= 30
d13 (che sur la gure 4).
pour souligner les carts. Les simulations (toujours sur une priode
ans) sont prsentes sur la gure 5.
Conclusion : dans le cas d'une mergence sur
jours, progression vers le nord lgrement
plus importante ; cependant, dans le cas d'une mergence sur

contre
14
jours,
parcelles sont infestes
dans le premier cas.
5.1.3 Force des vents variable

Cette fois, les vents sont orients vers le Nord, mais souent la deuxime semaine vitesse
moyenne double de la premire. En eet, le coecient pour le terme de transport est
3km/jour
la premire semaine et
v kV k = 6km/jour
v kV k =
la deuxime semaine (voir la gure 6 pour
le domaine utilis).
Le pourcentage d'attaque initial est nul partout sauf en
le taux de fcondit est de
= 30.
d2 (che sur la gure 6). A nouveau,
Les simulations sont donnes sur la gure 7.
Conclusion : Alors que les insectes dont la priode d'mergence ne s'tale que sur
jours ne
parviennent pas atteindre les parcelles au Nord, ceux qui ont une priode d'mergence de
14
jours colonisent deux parcelles supplmentaires au Nord.
5.2
Dans ce test, nous nous intressons la propagation sur
ans d'une espce, en fonction de
l'existence d'une priode de diapause prolonge.
131
5.2.1 Cas de la fructication constante

La fructication est ici constante en temps (d'anne en anne) et en espace. C'est une hypothse hautement improbable en milieu naturel, mais elle va nous permettre de voir si la diapause
prolonge fait varier la dispersion spatio-temporelle indpendamment de la fructication. Nous
sommes dans les conditions dcrites par la gure 2. Le taux de fcondit sera suppos gal
la diusion
D = 0.25km2 /jour
et le coecient de transport
30,
v kV k = 2km/jour.
Les tests numriques sont reprsents sur la gure 8.

Conclusion : Dans le cas sans diapause prolonge, la zone envahie est lgrement plus importante.
5.2.2 Cas de la fructication variable

Dans ce test, certaines parcelles pourront avoir une fructication nulle suivant l'anne. Nous
nous plaons toujours dans la conguration dcrite par la gure 2. Pour la fructication, l'hypothse faite est dcrite par la gure 9. Ainsi, la fructication est suppose uniforme (10hl par
parcelle) la premire anne puis nulle sur certaines parcelles les annes suivantes. Nous verrons
dans la section 5.2.4 que cette hypothse est plausible d'un point de vue biologique. Les rsultats
sont prsents sur la gure 10.
Conclusion : Cette fois, la propagation est plus tendue notamment au Nord dans le cas de
l'espce avec diapause prolonge.
5.2.3 Accidents avec fructication nulle partout pendant 1 an ou 2 ans

Le devenir de l'espce qui ne fait pas de diapause prolonge est vident. Le cycle du M.
2 ans, si la fructication est nulle l'anne n, il n'y aura pas d'insecte pour
n + 2k , k N. Si elle est nulle l'anne n et l'anne n + 1, l'espce dispa-
schimitscheki tant de
toutes les annes
rat. Le modle retranscrit dlement ces rsultats. Intressons-nous maintenant l'inuence de

l'existence d'une telle anne fructication nulle dans le cas d'une espce avec diapause prolon-
D = 0.25km2 /jour et le terme de transport

v kV k = 2km/jour. Le taux de diapause prolonge est suppos gal 5% un an et 15% 2 ans.
ge. Le taux de fcondit vaut
= 30,
la diusion
Nous sommes encore dans la situation dcrite par la gure 2, avec une fructication uniforme de
10hl
par parcelle. La propagation est tudie sur une dure de
ans. Le test numrique eectu
est reprsent sur la gure 11. On y a suppos que le cycle de vie de l'insecte tait de
lieu de
an (au
dans le cas de M. schimitscheki ), an d'obtenir des dirences plus signicatives (cela
revient supposer que la fructication est nulle pendant
2 ans, sur une priode totale de 12 ans)
Conclusion : la prsence d'une anne o la fructication est nulle n'a que peu d'incidence sur
la propagation d'une espce avec diapause prolonge.
5.2.4 Cas concrets

Nous prsentons ici quelques donnes mesures dans les cdraies au Mont-Ventoux an d'illustrer l'intrt des tests sur la diapause prolonge et les arbres isols. D'aprs les rsultats de la
gure 12, la fructication peut en eet tre nulle certaines annes sur une zone tendue (fort
800ha),
20ha).
du Ventoux :
d'Oppede :
132
ou encore nulle partir d'une anne sur zone ici moins tendue (fort
6. Discussion
5.3
Deux congurations direntes de parcelles sont tudies (cf. gure 13). Ainsi, nous nous
penchons sur la question de l'inuence de la prsence d'arbres isols entre deux zones forestires.
Les vents sont donns par la gure 2 ; le taux d'attaque initial est suppos nul partout sauf en
d2
o il est de
50%.
5.3.1 Cas o le transport est grand devant la diusion (rgions venteuses)

Dans ce cas, le terme de transport a une norme
D=
0.25km2 /jour. De plus,
= 15.
v kV k = 5km/jour,
et la diusion vaut
Les rsultats obtenus sont prsents sur la gure 14.
Conclusion : la propagation est plus rapide dans le cas sans arbre isol.
5.3.2 Cas o le rapport diusion/transport est plus grand

Cette fois, le terme de transport a une norme
D = 3km2 /jour.
Ici,
= 50.
v kV k = 2km/jour,
et la diusion vaut
Les rsultats obtenus sont prsents sur la gure 15.
Conclusion : ici, la propagation est plus importante et les insectes arrivent atteindre les
parcelles situes au Nord, dans le cas avec arbres isols ; ils n'y parviennent pas dans l'autre cas.
Discussion
6.1
Validation du modle
6.1.1 Interprtation des rsultats donns par les tests.

Avant d'accorder de l'importance aux rsultats dcoulant de la section 5, nous pouvons nous
demander si le comportement dcrit par le modle est raisonnable ou non. Nous allons donc
vrier que chacun des rsultats numriques obtenus dans la section 5 a une interprtation
biologique logique.
Intressons-nous d'abord aux rsultats numriques obtenus concernant la priode d'mergence. Pour des vents constants (section 5.1.1), on comprend aisment que seul le nombre total
d'insectes mergents joue un rle sur la dispersion. De mme, pour les vents variables en direction (section 5.1.2), dans le cas d'une priode de
jours, l'envol d'un nombre plus important
d'insectes permettrait d'obtenir une colonisation plus signicative au Nord ; ensuite, quand la
deuxime semaine les vents seraient dirigs vers le Sud, les insectes encore en vie seraient redplacs vers le Sud, mais ne vivraient pas assez longtemps pour coloniser des parcelles au Sud.
Dans le cas d'une mergence sur
14
jours, le vent souant vers le Sud amnerait les insectes
ayant rcemment merg vers les parcelles situes au Sud. De plus certains insectes, d'abord
dports vers le Nord, puis vers le Sud seraient amens par diusion vers les parcelles l'Est et
l'Ouest de la zone initiale d'envol. Finalement, nous pouvons galement interprter de la faon
suivante les rsultats obtenus pour un vent variable en intensit (section 5.1.3) : les insectes qui
mergeraient la deuxime semaine bncieraient de vents plus forts qui leur permettraient de
franchir la zone sans cdre situe entre les parcelles
d2
et
d22.
Ensuite, concernant la diapause prolonge, dans le cas d'une fructication constante (section
5.2.1), les rsultats obtenus peuvent s'interprter de la faon suivante : au niveau de la zone de
133

front, chaque anne, la diapause prolonge ferait perdre
50%
des adultes mergents (avec les
donnes utilises dans ce test). On comprend donc pourquoi la zone d'invasion serait lgrement
en recul dans le cas de la population avec diapause prolonge. Maintenant, dans le cas d'une
fructication variable (section 5.2.2), les rsultats peuvent s'interprter comme suit : la premire
anne, les individus se propageraient sur les premires parcelles situes au Nord (d11, d12, d13).
100% des insectes mergeraient et, comme

(d11, d12, d13 et d16, d17, d18), les insectes
L'anne suivante, dans le cas sans diapause prolonge,

la fructication serait nulle sur une bande au Nord
mourraient sans y avoir pondu. L'anne suivante, les insectes qui seraient entrs en diapause pro-
d11, d12, d13),

(d16, d17, d18), alors
longe les deux annes prcdentes mergeraient comme adultes (sur les parcelles
et le vent leur permettrait d'aller envahir les parcelles situes plus au Nord
que dans le cas o la diapause prolonge serait nulle le vent trop faible rendrait la zone fructication nulle (d11, d12, d13) infranchissable. On peut noter une progression lgrement plus
importante au Sud dans le cas avec diapause prolonge, due au phnomne dcrit dans le cas de
la fructication constante. Pour le cas o la fructication prsenterait des accidents (section
5.2.3) on comprend que l'existence d'une anne avec fructication nulle, reviendrait, ici, diviser
le nombre d'adultes total par
5,
ce qui n'aurait que peu d'inuence sur la position du front. De
plus, cela ferait perdre un an en terme de propagation.

Finalement, au sujet des arbres isols, quand le terme de transport est important devant
la diusion (section 5.3.1 ), les rsultats obtenus s'interprtent comme suit : dans le cas avec
arbres isols, les insectes rduiraient leur altitude pour s'approcher des arbres, et seraient de
ce fait soumis des vents moins forts, ce qui pourrait expliquer une propagation moins rapide.
Enn, avec un rapport diusion/transport plus grand (section 5.3.2), la zone sans arbre serait
trop grande pour tre franchie en une seule anne au vu de la force des vents. Les arbres isols
serviraient donc de relais pour l'invasion des zones plus au Nord. Cependant, pour obtenir une
dirence, nous avons d choisir un taux de natalit
= 50,
car le faible nombre d'insectes
mergents sur les arbres isols mettrait du temps avant de former une population signicative
(donc mesurable) envahissant les parcelles situes au Nord.
Tous les rsultats numriques obtenus dans la section 5 peuvent donc tre relis une interprtation logique du point de vue biologique. De plus, ces rsultats concordent avec les prvisions
des biologistes (avant la construction du modle). Cela nous permet de penser que le comportement qualitatif donn par le modle est raisonnable. Pour tester le modle de faon quantitative,
nous devons comparer les rsultats mesurs sur la zone tudie avec les prvisions du modle.
C'est le but de la section suivante.
6.1.2 Comparaison des prvisions avec les mesures eectues

Les donnes dont nous disposons nous permettent uniquement de tester le modle pour la
priode 2001-2003.
Nous connaissons le pourcentage de graines desquelles des adultes de M. schimitscheki mergent en mai 2001 (i) partir de la rcolte
rcolte
1999 (un an
5 et 7 mai
entre les
2000
(sans diapause prolonge), et (ii) partir de la
de diapause prolonge). Nous savons que la priode d'mergence a dbut

2001, et s'est tale sur une priode d'environ
15
jours (mesures eectues
en laboratoire, pouvant varier suivant l'altitude). La connaissance de la fructication et des

pourcentages d'attaques nous permet d'estimer le nombre d'insectes mergents sur cette priode.
Ces paramtres, ainsi que la donne des vents, vont permettre de calculer la position et le
nombre thorique d'insectes correspondant l'attaque de la rcolte
2002.
Nous allons comparer
ces rsultats avec ceux mesurs (obtenus lors de l'mergence en laboratoire : en mai
134
2003).
6. Discussion
Position du front
Nous ne connaissons pas la fructication dans les zones nouvellement envahies. Nous ne pouvons donc pas faire de comparaison en termes de nombre d'insectes ou de pourcentage d'attaque
sur ces zones. Nous comparons donc la position du front d'invasion donne par le modle avec
celle observe exprimentalement. Les rsultats sont prsents sur la gure 16.
Conclusion : les parcelles nouvellement envahies, d'aprs le modle, sont Sisteron et Lure.
Exprimentalement, seule la parcelle de Sisteron est nouvellement envahie en
cisment, des graines attaques ont t rcoltes en
l'attaque de
2004
en mai
2001),
2002
2001.
Plus pr-
Lure (correspondant, rappelons-le,
mais uniquement des M. pinsapinis en ont merg en
2003.
L'mergence
nous dira si des M. schimitscheki en diapause prolonge taient prsents dans ces
graines. D'aprs les prvisions du modle, les parcelles de Gap, Saou, Barjac, Moulin, Saumon ne
sont pas envahies. Ces rsultats sont en adquation avec les faits. Du point de vue de la position
du front, les rsultats fournis par le modle que nous proposons semblent donc cohrents.
Taux d'attaque
Rappelons que nous ne disposons de la fructication que sur les zones dj colonises par
M. schimitscheki en
mesurs en
2001,
1999
et en
2001.
La fructication en
1999,
ainsi que les taux d'attaque
nous permettent d'estimer le nombre d'insectes prts s'envoler en mai
2001.
Nous utilisons notre modle pour prdire le nombre d'insectes prsents sur chaque zone lors de la
rcolte
2003. La carte gographique telle que nous l'avons numrise pour les calculs est reprsen-
te sur la gure 17. Les rsultats du modle (M ), ainsi que le pourcentage de graines desquelles
mergent des M. schimitscheki en
2003 (S )
et le taux d'attaque mesur la radiographie (R)
sont prsents sur la gure 18.

Le nombre de parcelles sur lesquelles il est possible de comparer le taux d'attaque prvu
avec celui mesur n'est pas susant pour permettre une relle analyse statistique. Cependant,
nous pouvons tirer de ces rsultats quelques conclusions. Ainsi, la courbe reprsentant les taux
d'attaque prvus par le modle et celle reprsentant les taux d'mergence de M. schimitscheki
ont une allure trs semblable. Globalement, la courbe
conservation des proportions. Rappelons que
est au dessus de
S,
avec une certaine
correspond au taux de graines attaques, et
inclut donc les insectes en diapause prolonge. Cela pourrait justier la relation
eet,
est trs suprieur
S,
M > S.
En
et si cette dirence correspond en partie la prsence de l'autre
ravageur des graines de cdre, M. pinsapinis, on peut supposer qu'elle est galement due un
nombre consquent, et suprieur aux valeurs que l'on a utilises, de M. schimitscheki en diapause
prolonge. Ainsi, nous observons que, l o les taux d'attaque ne sont pas trop faibles, le rapport
M/S
est d'environ
3,
ce qui correspondrait, en supposant que les rsultats donns par notre
modle sont vrais, un taux total (sur deux ans) de diapause prolonge d'environ
courbe
60%. La
2001.
prsente les rsultats que nous aurions obtenus avec un tel taux de diapause en
Cette fois, on note une relation troite avec les rsultats mesurs.
6.2
Dans cette section, nous rsumons les rsultats obtenus l'aide tests numriques de la section 5, qui vont permettre de mieux comprendre le rle jou par certaines caractristiques de
M. schimitscheki, et les avantages qu'elles peuvent ventuellement lui confrer par rapport
d'autres espces de Megastigmus.
135

D'aprs la section 5.1, si la longueur de la priode d'mergence n'a pas d'impact signicatif
sur la propagation des insectes indpendamment des vents, son inuence est importante dans le
cas de vents variables. Ainsi, dans une situation relle, o les vents sont variables en direction
et en vitesse, une priode d'mergence plus longue permet une dispersion sur un territoire plus
tendu. D'un point de vue de la comptition entre plusieurs espces pour une mme niche, on
peut ajouter, au sujet de la priode d'mergence, que la date d'mergence joue galement un rle
important, l'insecte mergeant le plus tt ayant un avantage sur les autres.
On constate galement, d'aprs les tests numriques de la section 5.2 que, comme pour la
longueur de la priode d'mergence, la diapause prolonge est un avantage en terme de propagation (et de survie) dans le cas d'un environnement prsentant une certaine variabilit. Par
contre, dans le cas d'un environnement (trs peu probable) o la fructication serait constante,
elle fait prendre un lger retard la position du front des insectes.
Les rsultats de la section 5.3 nous montrent galement que la prsence d'arbres isols n'est
pas toujours un avantage. Cela dpendra de la rgion (venteuse ou non), et des caractristiques
de l'insecte (diusion plus ou moins leve).
Enn, les tests numriques eectus pour valider le modle (section 6.1.2) nous laissent penser
que le taux de diapause prolonge peut tre suprieur celui qui nous avait t initialement
transmis par les biologistes. Des expriences en cours devraient bientt permettre d'aner la
valeur de ce taux de diapause.
6.3
Limites
Si les rsultats des sections prcdentes semblent encourageants, nous devons malgr tout
souligner les limites du modle.
D'abord en termes de prvisions. La modlisation du nombre d'insectes mergents l'anne
ncessite de connatre (i) la fructication des annes
mtorologiques du mois de mai de l'anne
n 2.
n2
et prcdentes et (ii) les donnes
Si l'on peut esprer estimer la fructication
l'aide d'un autre modle, il est impossible de prvoir la direction et la force des vents d'une
anne sur l'autre (sauf peut-tre pour certaines congurations de terrain). Ainsi, les prvisions
d'invasion ne peuvent se faire que sur deux ans.
D'autres limites proviennent d'un manque de connaissances de certaines caractristiques de
l'insecte. Ainsi, nous n'avons pas encore d'estimation able des taux de diapause prolonge. De
mme, des expriences sont en cours pour apprcier si les insectes peuvent pondre dans les graines
non pollinises et s'y dvelopper jusqu' l'ge adulte, en bloquant le processus de dgnrescence
observ dans les graines non pollinises (comme c'est le cas pour d'autres espces). Les taux
d'attaque seraient donc plus faibles que ceux utiliss ici.
Enn, nous devons noter que certains facteurs n'ont volontairement pas t pris en compte
dans ce travail de modlisation. Par exemple, la competition interspcique avec M. pinsapinis,
qui peut jouer un rle dans les forts o le taux d'attaque est proche de la saturation. Cependant,
les rsultats rcents de [40] tendent montrer que M. schimitscheki est un meilleur comptiteur
que M. pinsapinis, avec notamment une mergence plus prcoce. L'inuence de la comptition
interspcique sur la dispersion de M. schimitscheki s'en trouverait donc diminue.
6.4
Rsultats venir
Outre les expriences cites dans la section 6.3, l'utilisation de marqueurs molculaires hypervariables, comme les microsatellites, qui est en cours l'INRA, devrait permettre de typer les
136
7. Figures
populations de chaque site d'attaque, et d'apprcier les zones d'origine des individus considrs.
Ceci devrait apporter des lments nouveaux pour la comprhension des processus de dispersion.
6.5
Adaptation d'autres espces vivantes
Le modle que nous avons dvelopp, ainsi que le logiciel associ (voir sur la gure 19 un
aperu de l'interface graphique) peuvent aisment s'adapter d'autres espces, notamment aux
autres espces invasives, s'attaquant des vgtaux exotiques introduits, comme la mineuse du
marronnier d'Inde,Cameraria oridhella, qui tend envahir l'ensemble de l'Europe occidentale
l'heure actuelle.
En outre, les limites cites dans la section 6.3 pourraient dans certains cas tre attnues.
Par exemple, pour des insectes rampants ou des rongeurs, le terme de transport pourrait tre
moins variable (ex. : une pente ou une zone attractive). Le modle pourrait galement s'adapter
au milieu marin, o le terme de transport (les courants marins) serait galement plus facilement
prvisibles que dans le cas que nous avons tudi (ex. : la Caulerpe). Ce type de modle, avec
quelques modications quant la prise en compte des variables, pourrait galement s'appliquer
l'analyse et la prvision des phnomnes d'expansion de certains insectes en relation avec le
rchauement climatique. Nous prvoyons ainsi de tester notre modle dans le cadre d'un autre
projet INRA en cours, s'intressant la progression en latitude et en altitude de la processionnaire
du pin en France.
7 Figures
Fig. 2 vents utiliss pour les tests 1.1, 2.1, 2.2, 2.3, 3.1, 3.2 et domaine utilis pour les
tests 2.1, 2.2, 2.3
137
Fig. 3 (a) : mergence sur 2 jours, (b) : mergence sur 5 jours, (c) : mergence sur 10
jours (test 1.1)
Fig. 4 Vents pendant la premire et deuxime semaine et domaine utilis (test 1.2)
138
7. Figures
Fig. 5 (a) : mergence sur 7 jours, (b) : mergence sur 14 jours (test 1.2)
Fig. 6 Direction des vents et domaine utilis (test 1.3)
139
Fig. 7 (a) : mergence sur 7 jours, (b) : mergence sur 14 jours (test 1.3)
Fig. 8 (a) : pas de diapause prolonge, (b) : diapause prolonge de 20% 1 an, 30%
2 ans (test 2.1)
140
7. Figures
Fig. 9 (a) : fructication la
1ere
anne, (b) : fructication les annes 2 et 3, (c) :fructi-
cation les annes suivantes. Les parcelles marques d'un rond blanc ont une fructication
nulle. Ailleurs la fructication est de 10hl. (test 2.2)
Fig. 10 (a) : pas de diapause prolonge, (b) : diapause prolonge de 20% 1 an, 30%
2 ans (test 2.3)
141
Fig. 11 (a) : la fructication est constante d'anne en anne, (b) : la fructication est
nulle la
2eme
anne (test2.3)
Fig. 12 Fructication en
d'Oppede
142
hl
de 1983 1998 sur (a) : la fort du Ventoux, (b) : la fort
7. Figures
Fig. 13 (a) : domaine sans arbre isol, (b) : domaine avec arbres isols (tests 3.1 et 3.2)
Fig. 14 (a) : propagation sans arbre isol, (b) : avec arbres isols (test 3.1))
143
Fig. 15 (a) : propagation sans arbre isol, (b) : avec arbres isols (test 3.2)
Fig. 16 (a) : estimation exprimentale de l'invasion en mai 2001, (b) : invasion simule
par le modle
144
7. Figures
Fig. 17 Carte gographique numrise, et utilise pour la validation du modle
90
M : Modle
R : Radios
80
S : schimi. merg.
D : Modle diap. 60%
pourcentages d'attaque
70
60
50
40
30
20
10
b.
1
parcelles
S
t.
Lu
La
m
ro
es
M
n
e
rb
ed
e
O
pp
es
qu
V
n
as
ns
Fl
as
sa
B
d
o
in
Fig. 18 Comparaison entre le taux d'attaque prvu par le modle, le taux d'attaque
enregistr par radiographie, le taux d'mergence de

ce mme modle avec une diapause prolonge de
M. schimitscheki, et le taux prvu par
60%
145
Fig. 19 Interface graphique du programme permettant d'exploiter notre modle
146
Bibliographie
Mathematical model of stagnation point ames, preprint.
[2] S. Agmon, Dierential equations (Birmingham, Ala., 1983), North-Holland Math. Stud. 92, 1984,
[1] A. Aftalion, H. Berestycki,

pp. 7-17.
[3] A.J. Ammerman, L.L. Cavalli-Sforza, The Neolithic Transition

Europe, Princeton Univ. Press, Princeton, NJ, (1984).
[4] D. Andow, P. Kareiva, S. Levin, A. Okubo
(1990), pp 177-188.
Spread of invading organisms, Landscape Ecology 4
[5] D.G. Aronson, H.F. Weinberger, Multidimensional

tics, Adv. Math. 30 (1978), pp 33-76.
nonlinear diusions arising in population gene-
[6] B. Audoly, H. Berestycki, Y. Pomeau, Raction-diusion

Acad. Sci. Paris 328 II (2000), pp 255-262.
[7] K.A. Bagrinovskii, S.K. Godunov,
Nauk USSR, 115 (1957) p.431.
and the Genetics of Populations in
en coulement stationnaire rapide, C. R.
Dierence schemes for multidimensional problems, Dokl. Acad.
[8] H.T. Banks, P.M. Kareiva, and K.A. Murphy, Parameter estimation techniques for interaction
redistribution models : a predator-prey example, Oeco-logia (Berlin), 74 (1987) pp 356-362.
[9] H.T. Banks, P.M. Kareiva, and L. Zia, Analyzing eld studies of insect
dimensional transport equations Environ. Entomol., 17 (1988) pp 815-820.
[10] H. Berestycki, Le nombre
Anal. 40 (1981), pp 1-29.
and
dispersal using two-
de solutions de certains problmes semi-linaires elliptiques, J. Func.
[11] H. Berestycki, The inuence of advection on the propagation of fronts in reaction-diusion equations, In : Nonlinear PDE's in Condensed Matter and Reactive Flows, H. Berestycki and Y. Pomeau
eds., Kluwer Academic Publ., 2002.
[12] H. Berestycki, F. Brock, In preparation.
[13] H. Berestycki, F. Hamel,
55, 2002, pp 949-1032.
Front propagation in periodic excitable media, Comm. Pure Appl. Math.
Gradient estimates for elliptic regularizations of semilinear parabolic and

degenerate elliptic equations, Preprint LATP/UMR 6632, 03/07, 2003.
H. Berestycki, F. Hamel, N. Nadirashvili, Elliptic eigenvalue problems with large drift and applications to nonlinear propagation phenomena, Preprint LATP/UMR 6632, 03/13, 2003.
H. Berestycki, F. Hamel, N. Nadirashvili, The speed of propagation for KPP type problems in
periodic and general domains, submitted.
H. Berestycki, F. Hamel, L.Roques, Analysis of the periodically fragmented environment model :
I - Inuence of periodic heterogeneous environment on species persistence, preprint.
H. Berestycki, F. Hamel, L. Roques, Analysis of the periodically fragmented environment model :
II - Biological invasions and pulsating travelling fronts, preprint.
H. Berestycki, F. Hamel, L. Rossi, Liouville theorems for semilinear elliptic equations in unbounded
domains, preprint.
[14] H. Berestycki, F. Hamel,

[15]
[16]
[17]
[18]
[19]
147
Bibliographie
[20] H. Berestycki, T. Lachand-Robert, Some properties of monotone rearrangement
to elliptic equations in cylinders, Math. Nachr. 266 (2004), pp 3-19.
[21] H. Berestycki, B. Larrouturou,
preprint.
with applications
Planar travelling front solutions of reaction-diusion problems,
[22] H. Berestycki, B. Larrouturou, P.-L. Lions, Multidimensional traveling-wave

propagation model, Arch. Ration. Mech. Anal. 111 (1990), pp 33-49.
[23] H. Berestycki, B. Nicolaenko, B. Scheurer, Traveling wave solutions
their singular limits SIAM J. Math. Anal., 16 (1985), pp 1207-1242.
[24] H. Berestycki, L. Nirenberg, Travelling
Linaire 9 (1992), pp 497-572.
solutions of a ame
to combustion models and
fronts in cylinders, Ann. Inst. H. Poincar, Anal. Non
[25] H. Berestycki, L. Nirenberg, S.R.S. Varadhan, The principal eigenvalue and maximum principle for
second order elliptic operators in general domains, Comm. Pure Appl. Math. 47 (1994), pp 47-92.
[26] A. Bonnet, Non-uniqueness for ame propagation
Applied Mathematics 6 (1995), pp 287-306.
when Lewis number is less than 1 Euro Jnl of
[27] R.S. Cantrell, C. Cosner, Diusive logistic equations with indenite weights :
disrupted environments, Proc. Roy. Soc. Edinburgh 112 (1989), pp 293-318.
[28] R.S. Cantrell, C. Cosner, The eects of
29 (1991), pp 315-338.
spatial heterogeneity in population dynamics, J. Math. Biol.
[29] R.S. Cantrell, C. Cosner, On the eects of spatial

species, J. Math. Biol. 37 (1998), pp 103-145.
[30] Y. Capdeboscq, Homogenization of a neutronic
Soc. Edinburgh 132 A (2002), pp 567-594.
heterogeneity on the persistence of interacting
critical diusion problem with drift, Proc. Royal
[31] S. Cano-Casanova, J. Lpez-Gmez, Permanence under strong

H. Poincar, Anal. Non Linaire 20 (2003), pp 999-1041.
[32] L.L. Cavalli-Sforza, M.W. Feldman, Cultural Transmission

proach, Princeton Univ. Press, Princeton, NJ, (1981).
[33] P. Constantin, A. Kiselev, L. Ryzhik,
Math. 54 (2001), pp 1320-1342.
population models in
aggressions is possible, Ann. Inst.
and Evolution : A Quantitative Ap-
Quenching of ames by uid advection, Comm. Pure Appl.
[34] J. Englnder, A.E. Kyprianou, Local extinction versus local exponential growth for spatial branching
processes, Ann. Probab. (2004), in press.
[35] J. Englnder, R.G. Pinsky, On the construction and support properties of measure-valued diusions
on D Rd with spatially dependent branching, Ann. Probab. 27 (1999), pp. 684730.
Uniqueness/nonuniqueness for nonnegative solutions of second-order

parabolic equations of the form ut = Lu + V u up in Rn , J. Di. Equations 192 (2003), pp. 396-
[36] J. Englnder, R.G. Pinsky,

428.
[37] JP. Fabre, Dynamique des populations de Megastigmus suspectus var. pinsapinis Ho. (Hym. Torymidae) dans les cdraies du Sud-Est de la France, Proceedings of the 2nd Conference of the IUFRO
Cone and Seed Insects Working Party S2.07.01, publi par l'INRA (1986) pp 127-145.
[38] JP. Fabre, Distribution
spatiale de Megastigmus suspectus var. pinsapinis Ho. (Hym. Torymidae)

sur Cedrus atlantica Manetti dans un peuplement du Sud-Est de la France, Proceedings of the 3rd
Conference of the IUFRO Cone and Seed Insects Working Party S2.07.01, publi par l'INRA (1989)
pp 101-116.
Importance des attaques sur cdre du Liban, Cedrus libani,

en Turquie par Megastigmus schimitscheki (Hymenoptera Toymidae) et risques d'extension de cet
insecte au cdre de l'Atlas, Cedrus atlantica, Annales des Sciences Forestires du Maroc, 27 (1994)
[39] JP. Fabre, CU. Alpekin, A. Chalon

pp 566-575.
[40] JP. Fabre, MA. Auger-Rozenberg, A. Chalon, S. Boivin, A. Roques, Competition between exotic and
native insects for seed resources in trees of a mediterranean forest ecosystem, Biological Invasions
00 (2003) pp 1-12.
148
[41] P.C. Fife, Mathematical aspects of reacting and diusing systems, Lecture Notes in Biomathematics
28, Springer Verlag, 1979.
[42] P.C. Fife, J.B. McLeod, The approach of solutions of non-linear

front solutions, Arch. Ration. Mech. Anal. 65 (1977), pp 335-361.
diusion equations to traveling
The advance of advantageous genes, Ann. Eugenics 7 (1937), pp 335-369.

[44] J. Grtner, M. Freidlin, On the propagation of concentration waves in periodic and random media,
[43] R.A. Fisher,
Sov. Math. Dokl. 20 (1979), pp 1282-1286.
[45] D. Gilbarg, NS. Trudinger, Elliptic Partial Dierential Equations of Second Order, Springer-Verlag,
1977.
[46] V. Giovangigli, Nonadiabatic
21 (5) (1990), pp1305-1325.
plane laminar ames and their singular limits, SIAM J. Math. Anal.,
[47] L. Glangetas, J.M. Roquejore, Rigorous derivation of the dispersion relation in

model with heat losses, Publications du Laboratoire d'Analyse Numrique (1995).
[48] F. Hamel, Formules min-max pour les vitesses
Fac. Sci. Toulouse 8 (1999), pp 259-280.
a combustion
d'ondes progressives multidimensionnelles, Ann.
[49] G.H. Hardy, J.E. Littlewood, G. P'olya, Inequalities, Cambridge University Press, Cambridge, 1952.
[50] A. F. Hedlin, H.O. III Yates, D. Cibrian-Tovar, B.H. Ebel, T.W. Koerber, E.P. Merker Cone ann
seed insects of North American conifers, Ottawa : Environment Canada/Canadian Forestry Service,
Washington D.C. : US Forest Service, Mexico : Secretaria de Agricultura y Recursos Hidraulicos,
(1980) 122pp.
Homogenization of ame fronts, Preprint IWR, Heidelberg, 1993.

[52] S. Heinze, The speed of travelling waves for convective reaction-diusion equations, Preprint MPI,
[51] S. Heinze,
Leipzig, 2001.
[53] S. Heinze, G. Papanicolaou, A. Stevens, Variational principles

geneous media, SIAM J. Appl. Math. 62 (2001), pp 129-148.
for propagation speeds in inhomo-
Populations on Fragmented Landscapes with Spatially Structured Heterogeneities :

Landscape Generation and Local Dispersal, Ecology 81(6) (2000), pp 1629-1641.
[55] Y. Hosono, B. Ilyas, Traveling waves for a simple diusive epidemic model, Mathematical Models
[54] D.E. Hiebeler,
and Methods in Applied Sciences, 5 (7) (1995), pp 935-966.
[56] W. Hudson, B. Zinner, Existence of travelling waves for reaction-diusion equations of Fisher
type in periodic media, In : Boundary Value Problems for Functional-Dierential Equations, J.
Henderson (ed.), World Scientic, 1995, pp 187-199.
[57] J. Hughes, Modeling
University (2002).
the eect of landscape pattern on mountain pine beetles, Thesis, Simon Fraser
[58] V. Hutson, K. Mischaikow, P. Polacik, The Evolution of dispersal

periodic environment, J. Math. Biol. 43 (2001), pp 501-533.
rates in a heterogeneous time-
Monitoring the impact of a seed chalcid, Megastigmus

spermotrophus Wachtl (Hymenoptera : Torymidae) on seed production in a Douglas-r seed orchard
in France : modelling of the orchard invasion by non-resident chalcids, The Canadian Entomologist
[59] M. Jarry, JN. Candau, A. Roques, B. Ycart
129 (1996) pp 7-19.
[60] G. Joulin, P. Clavin, Analyse asymptotique

Acta Astronautica, 3 (1976), pp 223-240.
des conditions d'extinction des ammes laminaires,
[61] G. Joulin, P. Clavin, Linear stability of non adiabatic ames : diusional thermal model, Combust.
Flame, 35 (137) (1979).
[62] J.P. Kaipio, J. Tervo, M. Vauhkonen, Simulations of the heterogeneity
element methods, Math. Comput. Simul. 39 (1995), pp 155-172.
of environments by nite
[63] K. Kamijo A revision of the Megastigminae occuring in Japan (Taxonomic studies on the Toymidae
in Japan, I), Insecta Matsumunara, 25 (1962) pp 18-40.
149
Bibliographie
Certain problems of burning-theory equations, Sov. Math. Dokl. 2 (1961), pp 48-51.
B. Kawohl, On the isoperimetric nature of a rearrangement inequality and its consequences for
some variational problems, Arch. Ration. Mech. Anal. 94 (1986), pp 227-243.
N. Kinezaki, K. Kawasaki, F. Takasu, N. Shigesada, Modeling biological invasion into periodically
fragmented environments, Theor. Population Biol. 64 (2003), pp 291-302.
A. Kiselev, L. Ryzhik, Enhancement of the traveling front speeds in reaction-diusion equations
with advection, Ann. Inst. H. Poincar, Analyse Non Linaire 18 (2001), pp 309-358.
A.N. Kolmogorov, I.G. Petrovsky, N.S. Piskunov, tude de l'quation de la diusion avec croissance
de la quantit de matire et son application un problme biologique, Bulletin Universit d'tat
[64] Ya.I. Kanel',

[65]
[66]
[67]
[68]
Moscou (Bjul. Moskowskogo Gos. Univ.), Srie internationale A 1 (1937), pp 1-26.
[69] P.L. Lions On the existence of positive

Anal., 24 (4) (1982), pp 441-467.
solutions of semilinear elliptic equations, SIAM J. Math.
[70] D. Ludwig, D.G. Aronson, H.F. Weinberger,

Biol. 8 (1979), pp 217-258.
Spatial patterning of the spruce budworm, J. Math.
Olfactory responses of seed wasps Megastigmus pinus Partt and

Megastigmus rafni Homeyer (Hymenoptera Torymidae) to host tree odours and some monoterpenes, Journal of Applied Entomology, 123 (1999) pp 561-567.
[72] M. Marion, Qualitative properties of a nonlinear system for laminar ames without ignition temperature, Nonlinear Analysis, Theory, Methods and Applications, 9 (11) (1985), pp 1269-1292.
[73] J.D. Murray, R.P. Sperb, Minimum domains for spatial patterns in a class of reaction-diusion
equations, J. Math. Biol. 18 (1983), pp 169-184.
[74] R. Nathan, H. S. Horn, J. Chave, and S. Levin Mechanistic models for tree seed dispersal by wind
in dense forests and open landscapes, In Seed Dispersal and Frugivory : Ecology, Evolution, and
[71] A. Luik, P. Ochsner, T.S. Jensen
Conservation. CAB International Press, Oxfordshire, Great Britain, (2002) pp 69-82.
Positive harmonic functions and diusion, Cambridge University Press, 1995.

[76] R.G. Pinsky, Second order elliptic operators with periodic coecients : criticality theory, perturbations, and positive harmonic functions, J. Funct. Anal. 129 (1995), pp. 80-107.
[77] R. Pinsky, Transience, recurrence and local extinction properties of the support for supercritical
nite measure-valued diusions, Ann. Probab. 24 (1996), pp 237-267.
[78] M. H. Protter, H. F. Weinberger, Maximum Principles in Dierential Equations, Prentice Hall,
[75] R.G. Pinsky,
1967.
[79] P. H. Rabinowitz, Pairs of positive solutions

Univ. Math. J. 23 (1974), pp 729-754.
of nonlinear elliptic boundary value problems, Indiana
Biologie et rpartition de Megastigmus spermotrophus Wachtl (Hymenoptera TChalcidoidea, Torymidae) et des autres insectes lis aux cnes dans les peuplements forestiers et vergers
graines franais du sapinde Douglas, Pseudotsuga menziesii, Acta Oecologica, Oecologica Appli-
[80] A. Roques
cata, 2 (1981) pp 161-180.

[81] A. Roques Les
(1983) 135pp.
insectes ravageurs des cnes et graines de conifres en France, (Versailles INRA),
Contribution to the knowledge of seed chalcids,

(Megastigmus spp. Hymenoptera Torymidae) in China, with the description of a new species, Mitt.
Schweitz. Entomol. Gesell., 68 (1995) pp 211-223.
[83] A. Roques, JP. Raimbault Response of Megastigmus spermotrophus to colour traps baited with
volatile constituents of Pseudotsuga menziessi seed cones, Proceedings of the 5th Conference of the
[82] A. Roques, J.H. Sun, Y.Z. Pan, X.D. Zhang
IUFRO Cone and Seed Insects Working Party S7.03.01, Padova University (1998) pp 101-107.
Application des mthodes de dirences nies au pricing des options qui sont solutions
de l'quation de Black et Scholes, Mmoire de DEA, non publi (2000).
[84] L. Roques,
150
[85] N. Shigesada, K. Kawasaki, Biological invasions : theory

and Evolution, Oxford : Oxford University Press, 1997.
[86] N. Shigesada, K. Kawasaki, E. Teramoto, Traveling
Theor. Population Biol. 30 (1986), pp 143-160.
and practice, Oxford Series in Ecology
periodic waves in heterogeneous environments,
Random dispersal in theoretical populations, Biometrika 38 (1951), pp 196-218.

O. Tackenberg, P. Poschlod, and S. Kahmen Dandelion Seed Dispersal : The Horizontal Wind Speed
Does Not Matter for Long-Distance Dispersal - it is Updraft !, Plant Biology, 5 (2003) 451-454.
P. Turchin Quantitative Analysis of Movement : Measuring and Modeling Population Redistribution
in Animals and Plants , Sinauer Associates, Sunderland, MA (1998).
J.J. Turgeon, A. Roques, P. De Groot Insect fauna of coniferous seed cones : Diversity, Host plan
interactions, and Management, Annual Review of Entomology, 39 (1994) pp 179-212.
A.I. Volpert, V.A. Volpert, V.A. Volpert, Traveling wave solutions of parabolic systems, Transla-
[87] J.G. Skellam,

[88]
[89]
[90]
[91]
tions of Math. Monographs 140, Amer. Math. Soc., 1994.
[92] H. Weinberger, On spreading speed and travelling waves

periodic habitat, J. Math. Biol. 45 (2002), pp 511-548.
for growth and migration models in a
[93] X. Xin, Existence of planar ame fronts in convective-diusive periodic media, Arch. Ration. Mech.
Anal. 121 (1992), pp 205-233.
The method of fractional steps, Springer, Berlin (1991).

[95] Ya. B. Zel'dovich, A theory of the limit of slow ame propagation, Zh. Prikl. Mekh. i Tekhn. Fiz.
[94] N.N. Yanenko,
11 (1) (1941), pp 159-169. (Russian)
151

These Lionel Roques

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

These Lionel Roques

Hochgeladen von

Copyright:

Verfügbare Formate

***

Laboratoire Jacques Louis Lions UMR 7598

Mis en page avec la classe thloria.

J'adresse un remerciement particulier Mayan Mimura, qui m'a invit un Congrs

Table des matires

Premire partie : Modles d'invasion biologique en environnement priodique fragment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Inuence des htrognits du milieu sur la survie des espces . . .

Invasions biologiques et fronts progressifs pulsatoires

Deuxime partie : Etude d'un modle de combustion avec pertes de chaleur

Troisime partie : Modlisation de la dispersion d'un insecte invasif

Problmes ouverts et perspectives

Partie I : Analysis of the periodically fragmented environment model

Chapitre 1 : Inuence of periodic heterogeneous environment on species

Statements of the main results

Existence and uniqueness results

Large time behaviour

Eects of the heterogeneity on species survival . . . . . . . . . . . .

Existence and uniqueness of a stationary solution

Energy of stationary states

The evolution equation

Table des matires

Conservation of species in ecological systems

Inuence of the amplitude" of the reaction term

Inuence of the shape" of

The eect of fragmentation in bounded domain models

Chapitre 2 : Biological invasions and pulsating travelling fronts

Introduction and main results

Existence result in nite cylinders for a regularized problem

Passage to the limit in the innite domains . . . . . . . . . . . . . .

Passage to the limit

The minimal speed

Existence of a positive minimal speed

Monotonicity of the solutions

with respect to the nonlinearity

Partie II : Study of the premixed ame model with heat losses

Introduction and main results

Proof of the existence theorem . . . . . . . . . . . . . . . . . . . . . . . . .

Equivalence with a problem on

Existence of solutions in bounded domains . . . . . . . . . . . . . .

Passage to the limit in

Passage to the limit

An upper bound for

A lower bound for the unburnt gases

Upper bounds for the speed

Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

Partie III : Une premire approche de la modlisation en biologie

Aspect pratique, site et collecte des donnes

Formulation d'un modle spcique . . . . . . . . . . . . . . . . . . . . . . 125

Hypothses comportementales utilises dans la construction du modle

Modlisation de l'volution pendant la priode de vol . . . . . . . . 127

Modlisation de la phase ovo-larvaire

Estimation des paramtres

Test 1 : Inuence de la dure de la priode d'mergence

Test 2 : Inuence du taux de diapause prolonge

Test 3 : Inuence des arbres isols

Rsultats sur le comportement de l'insecte

Adaptation d'autres espces vivantes

Table des matires

[85] ; l'environnement y est prsuppos variable priodiquement, favorable

en certains endroits la survie des espces, et dfavorable en d'autres. Les quations de

par un tat stationnaire

en gnral non constant. Nous

tudierons galement l'inuence des htrognits sur la propagation des fronts.

1. Premire partie : Modles d'invasion biologique en environnement priodique fragment

Inuence des htrognits du milieu sur la survie des espces . . .

Chapitre 1 : Inuence of periodic heterogeneous environment on species

Eects of the heterogeneity on species survival . . . . . . . . . . . .

Inuence of the amplitude" of the reaction term

Inuence of the shape" of

The eect of fragmentation in bounded domain models

Existence result in nite cylinders for a regularized problem

Passage to the limit in the innite domains . . . . . . . . . . . . . .

Partie II : Study of the premixed ame model with heat losses

Formulation d'un modle spcique . . . . . . . . . . . . . . . . . . . . . . 125

Test 1 : Inuence de la dure de la priode d'mergence

Test 2 : Inuence du taux de diapause prolonge

Test 3 : Inuence des arbres isols

tudierons galement l'inuence des htrognits sur la propagation des fronts.

Si le cas de l'quation homogne (1), o les coecients de diusion et de raction sont

est quelconque, constitue une spcicit de ce travail. Ainsi, l'inuence de la

Inuence des htrognits du milieu sur la survie des espces

saire et susante simple pour obtenir l'existence d'une solution de (3).

1.1.1 Rsultats : Existence, unicit, comportement en temps long et eet des

Le rsultat d'existence de la partie 1) se vrie pour des quations plus gnrales

tant d'armer que

La gure 1 ci-dessous donne un exemple de rarrangement priodique de Steiner (dans

Le thorme 3, 2), montre que la conguration

formule variationnelle pour c , et analysons les eets de l'htrognit du milieu sur la