Multiplicity of Entire Solutions For a Class of Almost Periodic Allen-Cahn Type Equations
Francesca ∗ Alessio
Dipartimento di Scienze Matematiche ,
Universit` a Politecnica delle Marche , via Brecce Bianche, I-60131 Ancona e-mail: alessio@dipmat.univpm.it
Piero † Montecchiari
Dipartimento di Scienze Matematiche ,
Universit` a Politecnica delle Marche , via Brecce Bianche, I-60131 Ancona e-mail: montecchiari@dipmat.univpm.it
Received 28 August 2005 Communicated by Antonio Ambrosetti
Abstract
We consider a class of semilinear elliptic equations of the form
−∆u(x, y) + a(εx)W
�(u(x, y)) = 0, (x, y) ∈ R
2(0.1) where ε > 0, a : R → R is an almost periodic, positive function and W : R → R is modeled on the classical two well Ginzburg-Landau potential W (s) = (s
2− 1)
2. We show via variational methods that if ε is sufficiently small and a is not constant then (0.1) admits infinitely many two dimensional entire solutions verifying the asymptotic conditions u(x, y) → ±1 as x → ±∞ uniformly with respect to y ∈ R.
1991 Mathematics Subject Classification . 35J60, 35B05, 35B40, 35J20, 34C37. .
Key words . Heteroclinic Solutions, Elliptic Equations, Variational Methods.
∗
Supported by MURST Project ‘Metodi Variazionali ed Equazioni Differenziali Non Lineari’
†
Supported by MURST Project ‘Metodi Variazionali ed Equazioni Differenziali Non Lineari’
515
1 Introduction
In this paper, pursuing a study started in [3] and [5], we consider semilinear elliptic equa- tions of the form
−∆u(x, y) + a(εx)W � (u(x, y)) = 0, (x, y) ∈ R 2 (1.2) where we assume ε > 0 and
(H 1 ) a : R → R is H¨older continuous, almost periodic and satisfies 0 < a ≡ inf
x∈R a(x) < sup
x∈R a(x) ≡ a;
(H 2 ) W ∈ C 2 (R) satisfies W (s) ≥ 0 for any s ∈ R, W (s) > 0 for any s ∈ (−1, 1), W ( ±1) = 0 and W �� (±1) > 0.
Classical examples of potentials which satisfy the above assumptions are the double- well Ginzburg-Landau potential, W (s) = (s 2 − 1) 2 , used in the study of phase transitions in binary metallic alloys, and the Sine-Gordon potential, W (s) = 1 + cos(πs), applied to several problems in condensed state Physics. Equation (1.2) with these potentials is respec- tively known as Allen-Cahn and Sine-Gordon stationary equation. The global minima of W represent pure phases of the material and they are energetically favorite.
We look for existence and multiplicity of two phases solutions of (1.2), i.e. solutions of the boundary value problem
� −∆u(x, y) + a(εx)W � (u(x, y)) = 0, (x, y) ∈ R 2
x →±∞ lim u(x, y) = ±1, uniformly w.r.t. y ∈ R. (1.3) One interesting feature regarding problem (1.3) is described by N. Ghoussoub and C. Gui in [14]. Indeed, in [14] the authors, dealing with the study of a De Giorgi conjecture (see [11]), proved in particular that in the autonomous case, i.e. when a(x) = a 0 > 0 for any x ∈ R, if u ∈ C 2 (R 2 ) is a solution of (1.3), then u(x, y) = q(x) for all (x, y) ∈ R 2 , where q ∈ C 2 (R) is a solution of the ordinary differential equation −¨q(x) + a 0 W � (q(x)) = 0, x ∈ R, satisfying lim x→±∞ q(x) = ±1. In other words, in the autonomous case, the problem (1.3) is one dimensional and the set of its solutions consists of one dimensional heteroclinic solutions. We quote also the papers [8], [9], [12], [2], [6] (and the reference therein) where the results in [14] are extended to higher dimensions and to more general settings. We recall moreover that the De Giorgi conjecture has been proven to hold in R n for any n ≤ 8 in the recent paper by O. Savin, [22].
As shown in [3], the one dimensional symmetry disappears in the non autonomous case, even considering, as in Problem (1.3), potentials depending periodically only on the single variable x. We mention also the paper by Alama, Bronsard and Gui, [1], where the authors exhibit systems of autonomous Allen Cahn equations which admit solutions without one dimensional symmetry.
The major difference between the case of autonomous equations and the cases consid-
ered in [3] and [1] lies in the different structure of the set of solutions of the corresponding
one dimensional problems. While in the case of a single autonomous equation the set of one dimensional solutions is a continuum homeomorphic to R, being constituted by the translations of a single heteroclinic solution (see Lemma 2.1 below), in both the works [3]
and [1] the construction of two dimensional solutions is strongly related with the discrete structure of the set of one dimensional solutions.
To be more precise, let us briefly describe the case in which the function a is a reg- ular, T -periodic, positive and not constant function, as in [3]. Let us consider the action functional
F (q) =
�
R 1
2 | ˙q(x)| 2 + a(εx)W (q(x)) dx on the class
Γ = {q ∈ H loc 1 (R) / �q� L
∞( R) = 1 and q − z 0 ∈ H 1 (R)}.
where z 0 ∈ C ∞ (R) is any fixed increasing function such that z 0 (x) → ±1 as x → ±∞
and |z 0 (x)| = 1 for any |x| ≥ 1. Set c = inf Γ F and K = {q ∈ Γ / F (q) = c}, noting that K is the minimal set of F on Γ consisting of one dimensional solutions of (1.3). It can be proved that, when a is not constant and ε is sufficiently small, the following non-degeneracy condition holds:
(∗) there exists ∅ �= K 0 ⊂ K such that, setting K j = {q(· − j T ε ) | q ∈ K 0 } for j ∈ Z, there results
(i) K 0 is compact with respect to the H 1 (R) topology,
(ii) K = ∪ j ∈Z K j and there exists d 0 > 0 such that if j �= j � then d(K j , K j
�) ≥ d 0
where d(A, B) = inf{�q 1 (x) − q 2 (x)� L
2( R) / q 1 ∈ A, q 2 ∈ B}, A, B ⊂ Γ.
The existence of solutions depending on both the planar variables can be achieved pre- scribing different asymptotic behaviors of the solutions as y → ±∞. Indeed, in [3], so- lutions to problem (1.3) which are asymptotic as y → ±∞ to different minimal sets K j
−, K j
+are found. More precisely, letting
H = {u ∈ H loc 1 (R 2 ) / �u� L
∞( R
2) ≤ 1 and u − z 0 ∈ ∩ (ζ
1,ζ
2) ⊂R H 1 (R × (ζ 1 , ζ 2 ))}, in [3], it is shown that fixed j − = 0 there exists at least two different values of j + ∈ Z\{0}
for which there exists a solution u ∈ H of (1.3) such that d(u(·, y), K j
±) → 0, as y → ±∞
(in fact, as shown in [17], there results j + = ±1).
The variational argument used in [3] is related to a renormalization procedure intro- duced by P.H. Rabinowitz in [19], [20]. The solutions are found as minima of the renor- malized action functional
ϕ(u) =
�
R 1
2 �∂ y u( ·, y)� 2 L
2( R) + (F (u(·, y)) − c) dy on the set
M j
−,j
+= {u ∈ H | lim
y→±∞ d(u( ·, y), K j
±) = 0}.
Note that, since c is the minimal level of F on Γ, the function y → F (u(·, y)) − c is non negative and the functional ϕ is well defined with values in [0, +∞] for any u ∈ M j
−,j
+. Moreover, the property (∗) of K implies that if u ∈ M j
−,j
+and ϕ(u) < +∞
then sup y ∈R d(u( ·, y), K j
−) < +∞. This is basic to avoid ‘sliding’ phenomena for the minimizing sequences (see [1]) possibly due to the unboundedness of the domain in the x-direction and so to recover for suitable values of the index j + the sufficient compactness to obtain the existence of a minimizer of ϕ in M j
−,j
+.
In the present paper we generalize the result in [3]; studying the existence of two di- mensional solutions to (1.3) in the case in which a is almost periodic, as stated in (H 1 ).
The structure of the set of one dimensional solutions of (1.3) when a is almost periodic is different from the one of the periodic case and the arguments used in [3] are no more useful in the almost periodic context.
To describe these differences, we first note that if a is almost periodic, setting c = inf Γ F (q) , we cannot say in general that c is attained, or, in other words, that the set K is not empty. Anyhow, thanks to the almost periodicity of the function a, if we look at the set of local minima of F on Γ, we see that it has discreteness and local compactness properties similar to the ones described by (∗) for the periodic case. In section 2 we prove in fact that there exist c ∗ > c, δ 0 ∈ (0, 1/4) and ε 0 > 0 such that for any ε ≤ ε 0 there is a family of open intervals {A j = (t − j , t + j ) / j ∈ Z} verifying that
there exists T − , T + > 0 such that T ε
−≤ t + j − t − j ≤ T ε
+and t + j = t − j+1 for any j ∈ Z, and for which, setting for j ∈ Z
Γ j = {q ∈ Γ / 1 − |q(t)| ≤ δ 0 for any t ∈ R \ A j }, c j = inf
Γ
jF (q) and K j = {q ∈ Γ j / F (q) = c j }
we have
i) {q ∈ Γ / F (q) ≤ c ∗ } = ∪ j ∈Z Γ j and Γ j �= ∅ for any j ∈ Z,
ii) ∃d 0 , D > 0 such that if j �= j � then d(Γ j , Γ j
�) ≥ d 0 and sup q
1,q
2∈Γ
j�q 1 −q 2 � L
2( R) ≤ D for any j ∈ Z,
iii) K j is not empty, is compact, with respect to the H 1 (R) topology, and constituted by one dimensional solution of (1.3).
In other words, if ε is sufficiently small, the sublevel set {q ∈ Γ / F (q) ≤ c ∗ } splits in the disjoint union of the uniformly separated and uniformly bounded subsets Γ j . Each set Γ j contains a nonempty set K j , compact with respect to the H 1 (R) topology, consisting of local minimizers of F at the level c j = inf Γ
jF (q) ∈ [c, c ∗ ). For our purposes this is a good generalization of the property (∗) and, following [3], we can try to look for two dimensional solutions of (1.3) prescribing to the solution to be asymptotic as y → ±∞ to different minimal sets K j
±or, more simply, to different sets Γ j
±.
Following this line, fixed j ∈ Z, we may look for this kind of solutions in the set M ∗ j = {u ∈ H / lim
y→−∞ d(u( ·, y), K j ) = 0, lim inf
y→+∞ d(u( ·, y), K j ) > 0}.
Differently from the periodic case, since the set K j is not minimal for F on Γ, if u ∈ M ∗ j , the function y → F (u(·, y)) − c j is indefinite in sign and the natural renormalized functional
ϕ j (u) = �
R 1
2 �∂ y u( ·, y)� 2 L
2( R) + (F (u(·, y)) − c j ) dy
is not a priori well defined. To overcome this difficulty we take advantage of a tech- nique already introduced in [5] to find, in the periodic case, two dimensional solutions emanating as y → −∞ from k-bump one dimensional solutions. As in [5] we observe that if u ∈ H is a solution of (1.2) on the half plane R × (−∞, y 0 ), which satisfies
�
R×(−∞,y
0) |∂ y u(x, y) | 2 dx dy < + ∞ and d(u(·, y), K j ) → 0 as y → −∞, then the energy E u (y) = 1 2 �∂ y u( ·, y)� 2 L
2( R) − F (u(·, y)) is conserved along the trajectory y ∈ (−∞, y 0 ) → u(·, y) ∈ Γ and precisely the following identity holds true
F (u( ·, y)) = c j + 1
2� ∂ y u( ·, y)� 2 L
2( R) , ∀y ∈ (−∞, y 0 ). (1.4) In particular (1.4) implies that F (u(·, y)) ≥ c j for any y ∈ (−∞, y 0 ) and suggests that we consider as variational space the set
M j = {u ∈ M ∗ j / inf
y ∈R F (u( ·, y)) ≥ c j }
on which we can look for minima of the natural renormalized functional ϕ j which is well defined there.
We note that if u ∈ M j and ϕ j (u) < +∞ we have lim inf
y →+∞ F (u( ·, y)) = c j and lim inf
y →+∞ d(u( ·, y), K j ) > 0.
As showed in Section 3, thanks to the properties (i) − (ii) of the sublevel set {q ∈ Γ / F (q) ≤ c ∗ }, this implies that the trajectory y ∈ R → u(·, y) ∈ Γ definitively en- ters in a set Γ j
∗as y → +∞ for an index j ∗ �= j such that c j
∗≤ c j . That condition on the index j is therefore necessary for the minimum problem inf M
jϕ j (u) to be meaningful and we denote the set of admissible indices
J = {j ∈ Z / there exists i ∈ Z \ {j} such that c i ≤ c j }.
Note that J is an infinite set being either J = Z or there exists an index j 0 ∈ Z such that c j
0< c j for any j ∈ Z \ {j 0 }, i.e., J = Z \ {j 0 }.
In section 3, for fixed ε ∈ (0, ε 0 ) and j ∈ J , setting m j = inf M
jϕ j (u), we show
that M j �= ∅ and m j < + ∞, and so that the minimum problem is well posed. As
in the periodic case, thanks to the discreteness property (i) − (ii) of the sublevel set
{q ∈ Γ / F (q) ≤ c ∗ } we can prove moreover that if u ∈ M j and ϕ j (u) < +∞ then,
sup y ∈R d(u( ·, y), K j ) < +∞. That allows us to avoid sliding phenomena for the minimiz-
ing sequences of ϕ j on M j obtaining suitable convergence properties, up to subsequences
and y-translations, to two dimensional functions. In fact, as limit of minimizing sequences,
we prove the existence of functions u j ∈ H verifying
y→−∞ lim d(u j (·, y), K j ) = 0 and there exists j ∗ ∈ Z \ {j} such that
y→+∞ lim d(u j (·, y), Γ j
∗) = 0.
A problem arises since we can not say that the limit functions u j are minima of ϕ j
on M j , indeed the constraint F (u j (·, y)) ≥ c j is not necessarily satisfied for any y ∈ R.
However, as limit of a minimizing sequence, u j inherits sufficient minimality properties to construct from it a two dimensional solution of (1.3). More precisely, we consider the restriction of u j on the domain R × (−∞, y 0 ) where y 0 is defined by
y 0 = sup{¯y ∈ R / u j (·, y) ∈ K j or F (u j (·, y)) > c j for any y < ¯y}.
If y 0 = +∞, then inf y ∈R F (u j (·, y)) ≥ c j and we conclude that u j is actually a minimum of ϕ j in M j . It is not difficult to show that in this case, u j ∈ C 2 (R 2 ) is a classical entire solution to (1.2). Concerning the asymptotic behavior of u j as y → +∞
we can say moreover that there exists
K ˜ j
∗⊂ Γ j
∗∩ {q ∈ Γ / F (q) = c j },
compact with respect to the H 1 (R) topology, consisting of one dimensional solutions of (1.3) and such that lim y →+∞ d(u( ·, y), ˜ K j
∗) = 0. We refer to this kind of solutions as heteroclinic type solutions.
If y 0 ∈ R, we prove that
� y
0−∞
1
2 �∂ y u j � 2 L
2( R) + (F (u j (·, y)) − c j ) dy = m j , lim y →y
−0
F (u j (·, y)) = c j , and u j (·, y 0 ) ∈ Γ j
∗.
Then, we derive that u j is minimal for ϕ j with respect to small variations with compact support in R × (−∞, y 0 ). That says that u j solves (1.2) in the half plane R × (−∞, y 0 ).
Using the conservation of Energy (1.4), since lim y→y
−0