-
This article was downloaded by: [Universita di Padova]On: 12
April 2013, At: 06:13Publisher: Taylor & FrancisInforma Ltd
Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House,37-41 Mortimer Street, London W1T
3JH, UK
Communications in Partial Differential EquationsPublication
details, including instructions for authors and subscription
information:http://www.tandfonline.com/loi/lpde20
Long-Time Behavior of the Mean Curvature Flow withPeriodic
ForcingAnnalisa Cesaroni a & Matteo Novaga aa Dipartimento di
Matematica Pura e Applicata, Università di Padova, Padova,
ItalyAccepted author version posted online: 05 Feb 2013.Version of
record first published: 10 Apr2013.
To cite this article: Annalisa Cesaroni & Matteo Novaga
(2013): Long-Time Behavior of the Mean Curvature Flow with
PeriodicForcing, Communications in Partial Differential Equations,
38:5, 780-801
To link to this article:
http://dx.doi.org/10.1080/03605302.2013.771508
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use:
http://www.tandfonline.com/page/terms-and-conditions
This article may be used for research, teaching, and private
study purposes. Any substantial or systematicreproduction,
redistribution, reselling, loan, sub-licensing, systematic supply,
or distribution in any form toanyone is expressly forbidden.
The publisher does not give any warranty express or implied or
make any representation that the contentswill be complete or
accurate or up to date. The accuracy of any instructions, formulae,
and drug doses shouldbe independently verified with primary
sources. The publisher shall not be liable for any loss, actions,
claims,proceedings, demand, or costs or damages whatsoever or
howsoever caused arising directly or indirectly inconnection with
or arising out of the use of this material.
http://www.tandfonline.com/loi/lpde20http://dx.doi.org/10.1080/03605302.2013.771508http://www.tandfonline.com/page/terms-and-conditions
-
Communications in Partial Differential Equations, 38: 780–801,
2013Copyright © Taylor & Francis Group, LLCISSN 0360-5302
print/1532-4133 onlineDOI: 10.1080/03605302.2013.771508
Long-Time Behavior of theMean CurvatureFlow with Periodic
Forcing
ANNALISA CESARONI AND MATTEO NOVAGA
Dipartimento di Matematica Pura e Applicata, Università di
Padova,Padova, Italy
We consider the long-time behavior of the mean curvature flow in
heterogeneousmedia with periodic fibrations, modeled as an additive
driving force. Underappropriate assumptions on the forcing term, we
show existence of generalizedtraveling waves with maximal speed of
propagation, and we prove the convergenceof solutions to the forced
mean curvature flow to these generalized waves.
Keywords Asymptotic behavior; Heterogeneous media; Mean
curvature flow;Traveling waves.
Mathematics Subject Classification 35K55; 49L25; 53C44.
1. Introduction
We are interested in the long-time behavior of the mean
curvature flow in aperiodic heterogeneous medium. The evolution law
can be written as a forced meancurvature flow
v = � − gwhere v denotes the inward normal velocity of the
evolving hypersurface, � its meancurvature (with the convention
that � is positive on convex sets) and g is a periodicforcing term.
In our model, we assume that the hypersurfaces are graphs
withrespect to a fixed hyperplane and that the forcing term g does
not depend on thevariable orthogonal to such hyperplane (fibered
medium). Under these assumptionsthe evolving hypersurface coincides
with the graph of the solution to the Cauchyproblemut =
√1+ �Du�2 div
(Du√
1+�Du�2
)+ g√1+ �Du�2 in �0�+��×�n
u�0� ·� = u0 in �n�(1)
Received October 12, 2011; Accepted January 13, 2013Address
correspondence to Matteo Novaga, Dipartimento di Matematica Pura
e
Applicata, Università di Padova, via Trieste 63, Padova 35121,
Italy; E-mail: [email protected]
780
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 781
We are particularly interested in the asymptotic behavior as t→
+� of solutionsto (1), where the initial data u0 and the forcing
term g are assumed to be Lipschitzcontinuous and �n-periodic.
The expected result is that, under appropriate assumptions on g,
there exists aunique constant c ∈ � and a periodic function � such
that
u�t� y�− ct − ��y�→ 0� as t→ +�� uniformly in �n�
This is a result on the asymptotic stability of special
solutions to (1), called travelingwave solutions, which are of the
form �+ ct. The constant c and the function � arerespectively the
propagation speed and the profile of the wave.
The first question we address in Section 3 of this paper is
about existence oftraveling wave solutions to (1). We provide a
construction of such solutions usinga variational approach
developed in [23] (see also [24]). In particular, our solutionsare
critical points of appropriate functionals, which are exponentially
weightedarea functionals with a volume term, depending on the speed
of propagation c.Exploiting this variational structure, we show
existence of traveling waves underrather weak assumptions on the
forcing term g, i.e.
∃A ⊆ �0� 1�n s.t.∫Ag�y�dy > Per�A��n�
where Per�A��n� is the periodic perimeter of A (see Section 2).
Notice that, if∫�0�1�n g > 0, then the previous condition holds
true by taking A = �0� 1�n.
As our solutions are in general not globally defined, we call
them generalizedtraveling waves. In Propositions 3.7 and 3.10 we
discuss the regularity of thesesolutions and of their support.
Moreover, in Section 3.1 we list some strongerconditions on the
forcing term, involving only the oscillation and the norm of
g,under which we show existence of classical (i.e. globally
defined) traveling waves (seeProposition 3.15).
We point out that the variational method selects the fastest
traveling wavesfor (1) which are bounded above, in particular it is
uniquely defined the speed ofpropagation c of such waves and it
holds c ≥ ∫
�0�1�n g (see Corollary 3.2).We recall that the problem of
existence of classical traveling waves for the
forced mean curvature flow has already been considered in the
literature, underdifferent assumptions on the forcing term [9, 14,
19]. We also mention [22], wherethe authors construct V -shaped
traveling waves in the whole space for a constantforcing term (see
also [11, 20, 26] for similar results in the planar case).
Theconstruction of the traveling fronts in these papers relies
mainly on maximumprinciple type arguments, while we use here a
variational approach.
The second question of interest is about the convergence, as t→
+�, ofthe solution to (1) to a traveling wave solution. We point
out that the long-time behavior of solutions of parabolic problems
using viscosity solutions typearguments has been extensively
considered in the literature: see [7, 25] for the caseof semilinear
and quasilinear parabolic problems in periodic environments,
[13]where the author considers uniformly parabolic operators in
bounded domains withNeumann boundary conditions, and [6] for the
case of viscous Hamilton-Jacobiequations in bounded domains with
Dirichlet boundary conditions. However, noneof these results
applies to mean curvature type equations such as (1).
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
782 Cesaroni and Novaga
In Section 4 we prove a convergence result under the assumption
that thereexists a global traveling wave solution. In particular,
in Corollary 4.9 we show thatthe solution u�t� y� to (1)
satisfies
u�t� y�− ct→ ��y� in �1+���n�� as t→ +��
where �+ ct is the traveling wave, which in this case is unique
up to an additiveconstant.
In the general case, we obtain a weaker convergence result.
First, inProposition 4.6 we describe the asymptotic behavior as t→
+� of the maximum ofthe function u�t� ·�. Namely, letting Q �= �0�
1�n, we show that there exists a constantK > 0 such that
minQu0 + ct ≤ max
Qu�t� y� ≤ ct + K + log�1+ t�
c�
Then, in Theorem 4.7 we show that, along a sequence tn → +�,
u�tn� y�−maxQu�tn� ·� −→
{��y� locally in �1+��E�−� locally uniformly in Q\E
for all � ∈ �0� 1�, where �+ ct is a generalized traveling wave
supported in E ⊂ Q,possibly depending on the sequence tn.
We point out that the proof of the convergence result, as well
as the proofof existence of generalized waves, essentially uses
variational methods, rather thanmaximum principle based
arguments.
2. Notation and Preliminary Results
We refer to [2] for a general introduction to functions of
bounded variation and setsof finite perimeter. Letting Q �= �0�
1�n, it is a classical result that any u ∈ BV�Q�admits a trace uQ
on Q (see e.g. [2, Theorem 3.87]). Let 0Q �= Q ∩ y �
∏ni=1 yi =
0� and let � � 0Q→ Q be the function ��y� �= y +∑ni=1 i�y�ei,
where i�y� = 1
if yi = 0 and i�y� = 0 otherwise. We consider the space BVper�Q�
of functionswhich have periodic bounded variation in Q, where the
periodic total variation ofu ∈ BV�Q� is defined as
�Du�per�Q� �= �Du��Q�+∫0Q
�uQ�y�− uQ���y���d�n−1�y�� (2)
The space BVper�Q� is the space BV�Q� endowed with the norm
�u�BVper�Q� �= �u�L1�Q� + �Du�per�Q��
Observe that BVper�Q� coincides with BV��n�, where �n �= �n/�n
is the
n-dimensional torus. For every E ⊆ Q we define the periodic
perimeter of E as
Per�E��n� �= �D�E�per�Q� (3)
where �E is the characteristic function of E. We recall the
isoperimetric inequality [2]:
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 783
Proposition 2.1. There exists Cn > 0 such that
Per�E��n� ≥ Cn�E� n−1n (4)
for all E ⊆ Q of finite perimeter and such that �E� ≤ 1/2.
Remark 2.2. Notice that C1 = 2.
In this paper we always make the following regularity assumption
on the initialdatum and on the forcing term:
u0� g are Lipschitz continuous and �0� 1�n-periodic� (5)
Using the comparison principle [4] and (5), we get that there
exists a uniquecontinuous solution u to (1) with periodic boundary
conditions. Moreover, thissolution is locally Lipschitz continuous
[12, 15] and hence smooth for all positivetimes, due to the
regularity theory for parabolic problems.
Theorem 2.3. Under assumption (5), problem (1) admits a unique
solution
u ∈ ���0�+��×Q� ∩�1+ �2 �2+���0� T�×Q�
for every � ∈ �0� 1� and T > 0, with periodic boundary
conditions on Q. Moreover
ut ∈ L2��0�+��×Q� and Du�t� x� ∈ L���0� T�×Q� for every T >
0�
We need another condition on the forcing term g, in order to
prove existenceof generalized traveling wave solutions to (1),
namely we assume that
∃A ⊆ Q such that∫Ag�y�dy > Per�A��n�� (6)
Note that condition (6) implies maxQ g > 0, and is fullfilled
for instance if∫Qg > 0.
Remark 2.4. In [5] (see also [10]) we considered a sort of
complementary conditionto (6). Indeed it is proved that, if g has
zero average and there exists � ∈ �0� 1� suchthat ∫
Ag�y�dy < �Per�A��n� ∀A ⊆ Q� (7)
then there exists a periodic stationary solution of (1).
We conclude this section by recalling a classical result about
the regularityof hypersurfaces of prescribed bounded mean curvature
[21, Theorem 4.1], [27,Theorem 1].
Theorem 2.5. Let K be a Caccioppoli set with bounded prescribed
mean curvatureA�x� ∈ L�, x ∈ K. Then �k�K\�K� = 0 for every k >
n− 8, and there exists � > 0,such that for every x ∈ �K we get
that K ∩ B�x� �� = �K ∩ B�x� �� and K ∩ B�x� ��
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
784 Cesaroni and Novaga
is a �1+� hypersurface for any � ∈ �0� 1�. Moreover, letting
�Kn�n be a sequence ofCaccioppoli sets such that:
i) every Kn is a locally minimizer of the functional
Per�V�+∫VAn�y�dy, with �An�� ≤
A independent of n,ii) Kn converges to K� locally in the
L1-topology,
and letting xn ∈ Kn, with xn → x� as n→ +�, we have x� ∈ K�. If
x� ∈ �K�,then xn ∈ �Kn for all n > n0, and the unit outward
normal to �Kn at xn converges tothe unit outward normal to �K� at
x�.
3. Existence and Regularity of Generalized Traveling Waves
We now show existence of special solutions to (1), which we call
generalizedtraveling waves. They are solutions of the form ��x�+
ct, where the graph of � iscalled the profile of the traveling wave
and c is called the traveling speed. Observethat to prove the
existence of a traveling wave solution it is sufficient to
determinec ∈ � such that the equation
−div(
D�√1+ �D��2
)= g�y�− c√
1+ �D��2 (8)
admits a �n-periodic solution � � �n → �. In the following we
will show that itis always possible to define a unique traveling
speed c for the problem under ourassumption (6) on the forcing, but
in general, the previous equation does not admita global solution.
We will prove that there exists a maximal set E ⊆ Q, which is
asufficiently regular domain, and a function � � Q→ �−��+�� (which
is defined upto additive constants) such that E = � > −��, � ∈
�2+��E� and solves
−div(
D�√1+ �D��2
)= g�y�− c√
1+ �D��2 � in E (9)
with the boundary conditions
��x�→ −� as dist�x� E\S�→ 0 (10)
where S ⊂ E is a closed set with � ��S� = 0 for all � > n− 8
(see Proposition 3.10).Moreover we will show that the solutions we
construct satisfy also a strongerboundary condition, more natural
in viscosity solutions theory, say
for every � ∈ �1per�Q�� �− � achieves its minimum in E� (11)
First of all we note that the equation (8) can be interpreted,
for any c > 0, as theEuler-Lagrange equation associated to the
functional
Fc��� =∫Qec��y�
(√1+ �D��y��2 − g�y�
c
)dy � ∈ �1per�Q�� (12)
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 785
Using the change of variable ��y� �= ec��y�c
, we can rewrite the functional Fc as
Fc��� = Gc��� �=∫Q
√c2� 2�y�+ �D��y��2 − g�y���y�dy� (13)
which can extended as a lower semicontinuous functional on
BVper�Q�, see [2]. UsingGc, we can extend the functional Fc to all
measurable functions � � Q→ �−�� 0�such that ec��y� ∈ BVper�Q�
(where we use the notation e−� = 0) by setting
Fc��� �= Gc(ec��y�
c
)� (14)
In particular, for all such � the following representation
formula holds (cfr. [18,Section 12]):
Fc��� = sup{∫Qec��y�
(div�′
c+ �n+1
)dy � ��′� �n+1� ∈ �1per�Q��n+1�� ��′�2 + �2n+1 ≤ 1
}−∫Q
ec��y�
cg�y�dy (15)
which can be easily checked on smooth functions, and then
extends by relaxationto all � such that ec��y� ∈
BVper�Q�.Proposition 3.1. Under the standing assumption (6) there
exists a unique constantc > 0, with
∫Qg ≤ c ≤ maxQ g, such that
• if 0 < c < c, then infGc��� �� ∈ BVper�Q��� ≥ 0� = −�,•
if c > c, then infGc��� �� ∈ BVper�Q��� ≥ 0� = 0, and Gc��� >
0 for every� �≡ 0,
• minGc��� �� ∈ BVper�Q��� ≥ 0� = 0, and there exists � �≡ 0
s.t. Gc��� = 0.Proof. As Gc is positively one-homogeneous, it
follows that inf�∈BVper�Q���≥0Gc���can be either 0 or −�. By
definition of Gc, if c > maxQ g, then Gc��� ≥ 0 for every� ≥ 0,
so that inf�≥0Gc��� = 0.
Now we claim that under condition (6) there exists a function �
such thatGc��� < 0 if c is sufficienlty small. Let A ⊆ Q be the
set appearing in (6) and take� = �A. If A �= Q, then by condition
(6) there exists k > 1 such that
Gc��A� = Per�A��n�+ c�A� −∫Ag < −�k− 1�Per�A��n�+ c�A��
By choosing 0 < c < �k− 1�Per�A��n�/�A�, we then obtain
Gc��A� < 0, whichimplies inf�≥0Gc��� = −�. If A = Q, then
∫Qg > 0 by (6), so that
Gc��Q� < 0 for every 0 < c <∫Qg� (16)
For c > 0 we consider the constrained problem
inf{Gc��� �� ∈ BVper�Q�� � ≥ 0�
∫Qg� = 1
}� (17)
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
786 Cesaroni and Novaga
By the direct method of the Calculus of Variations, one can
easily show that thisproblem admits a (possibly nonunique)
minimizer �c [18]. We define the functionminimum value as
c ∈ �0�+�� �→ �c �= Gc��c�
and we claim that this function is continuous and strictly
increasing. Notice that, byminimality of �c, we have∫
Qc�cdy ≤ Gc��c�+
∫Qg�cdy = �c + 1� (18)
The monotonicity of �c is due to the fact that c �→ Gc��c� is an
increasing function.Indeed, for c1 < c2 we have
�c2 − �c1 = Gc2��c2�−Gc1��c1� ≥ Gc2��c2�−Gc1��c2�=∫Q
√c22�
2c2+ �D�c2 �2 −
√c21�
2c2+ �D�c2 �2
=∫Q
�c22 − c21�� 2c2√c22�
2c2+ �D�c2 �2 +
√c21�
2c2+ �D�c2 �2
> 0 �
Let us now show the continuity of c �→ �c. For c1 < c2 we
get
��c2 − �c1 � = �c2 − �c1 = Gc2��c2�−Gc1��c1�≤
Gc2��c1�−Gc1��c1�
=∫Q
�c22 − c21�� 2c1√c22�
2c1+ �D�c1 �2 +
√c21�
2c1+ �D�c1 �2
≤∫Q
�c22 − c21�� 2c1�c1 + c2��c1
dy = �c2 − c1�∫Q�c1dy
≤ �c2 − c1��c1 + 1c1
(19)
where the last inequality follows from (18). Recalling that, for
all c > 0,
�c + 1c
≤∫Q
√� 2 + �D� �
2
c2
for any fixed � ∈ BVper�Q� with � ≥ 0 and∫Qg� = 1, we have that
the function
c �→ ��c + 1�/c is locally bounded on �0�+��, that is, for all c
∈ �m�M� there existsC > 0 such that ��c� ≤ C. Letting now c1� c2
∈ �m�M�, from (19) it then follows
��c2 − �c1 � ≤ C�c1 − c2��
which shows the continuity of c �→ �c.
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 787
Since this function is continuous and strictly increasing, it is
possible to definec > 0 as the unique constant for which �c =
Gc��c� = 0. From (16) it followsc ≥ ∫
Qg.
Observe that, due to the constraints, �c �≡ 0 and, due to the
positiveone-homogeneity of Gc, k�c is also a minimizers of Gc for
every k ≥ 0.
Finally, observe that necessarily if c > c and � �≡ 0, then
Gc��� > 0. Onthe contrary, if Gc��� = 0 and � �≡ 0, then
∫Qg� = > 0. So −1� would be a
minimizer to (17), and �c = 0, for c > c, in contradiction
with the monotonicity ofthe value function. �
Recalling (14), it is immediate to state the analogous result
for the functional Fc.
Corollary 3.2. There exists a unique constant c > 0 with∫Qg ≤
c ≤ maxQ g such that
• if 0 < c < c, then infFc��� � ec� ∈ BVper�Q�� = −�,• if
c > c, then infFc��� � ec� ∈ BVper�Q�� = 0, and Fc��� > 0 for
all � �≡ −�,• there exists � � Q→ �−��+�� such that � �≡ −�, ec� ∈
BVper�Q� andFc��� = 0.
Remark 3.3. Notice that Proposition 3.1 and Corollary 3.2,
assuring the existenceof generalized traveling waves solutions,
requires only g ∈ L��Q�.
We now analyze the regularity of the minima of Fc (or
equivalently of Gc).We first give a geometric representation of the
functional Fc (cfr. [18,
Theorem 14.6]). Given c > 0 and � ⊂ Q×� we define a weighted
perimeter
Perc����n �� �= sup
{ ∫�ecz �div��y� z�+ c�n+1�y� z�� dy dz �
� ∈ �1per�Q×���n+1�� ���2 ≤ 1}� (20)
Notice that, for all � ⊂ Q×� of locally finite perimeter we
have
Perc����n �� =
∫∗�ecz d�n +
∫�ect∫0Q
��Q��y�− �Q����y���d�n−1�y�dt
where � is as in (2).
Proposition 3.4. Let � � Q→ �−��+�� be such that ec� ∈ BVper�Q�.
Then
Fc��� = �c���� �= Perc�����n ×��−∫��
eczg�y�dy dz (21)
where �� �= �y� z� ∈ Q×� � z < ��y�� is the epigraph of
�.
Proof. By exploiting formula (15) and the definition of Perc in
(20), it is possibleto check that Fc��� ≤ �c����. For the reverse
inequality, we observe first of allthat (21) holds on smooth
functions � ∈ �1per�Q� and then the inequality extends toall �’s by
relaxation. For a similar argument see [18, Theorem 14.6]. �
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
788 Cesaroni and Novaga
Lemma 3.5. Let � � Q→ �−��+�� be a non trivial minimizer of Fc,
then theepigraph �� of � is a minimizer, under compact
perturbations, of the functional �cdefined in (21).
Proof. We reason as in [18, Theorem 14.9]. Given F ⊂ Q×� such
that∫Fecz dy dz < +�, we consider �F � Q→ �−��+�� be such
that
ec�F �y�
c=∫ �F �y�−�
eczdz =∫Fy
eczdz for a.e. y ∈ Q�
where Fy �= z ∈ � � �y� z� ∈ F�. Observe that, by definition,
ec�F ∈ BV�Q� and∫Feczg�y�dy dz =
∫Qec�F �y�
g�y�
cdy� (22)
Moreover, by definition of Perc, for all � = ��′� �n+1� ∈
�1per�Q��n+1� we have
Perc�F��n ×�� ≥
∫Fecz �div�′ + c�n+1� dy dz =
∫Qec�F
(div�′
c+ �n+1
)dy� (23)
By taking the supremum over all �’s in (23), and using the
representationformula (15) and (22), we then get
�c�F� ≥ Fc��F� ≥ Fc��� = �c����
where the last equality follows from Proposition 3.4, thus
proving the claim. �
Notice that if � is a minimizer of �c, then �+ �0� z� is also a
minimizer for allz ∈ �, that is, the class of minimizers is
invariant by vertical shifts. Reasoning as in[18, Proposition 5.14]
(see also [1]) one can prove a density estimate for minimizersof
�c.
Lemma 3.6. There exist constants � r0 > 0, depending only on
n and �g��, such thatfor all minimizers � of �c, x ∈ � and r ∈ �0�
r0� the following density estimate holds:
�� ∩ Br�x�� ≥ rn+1� (24)
Proposition 3.7. Let � � Q→ �−��+�� be a non trivial minimizer
of Fc. Then �� �=�� is a �
2+� hypersurface for all � < 1, out of a closed singular set
S� ⊂ �� ofHausdorff dimension at most n− 7. Moreover, letting E� �=
��n ���\S�� the projectiononto �n of ��\S�, we have that1. E� is a
open set and E� = int�E�� = int���n���,2. � ≡ −� a.e. on Q\E�,3. �
∈ �2+�loc �E�� for all � < 1,4. � solves (9) in E� with boundary
conditions (10).
Finally, letting �̃ another minimizer of Fc, for every connected
component Ei of E� thereexists ki ∈ � such that �̃ = �+ ki.
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 789
Proof. By Lemma 3.5 �� is a minimizer of �c under compact
perturbations.Classical results about regularity of minimal
surfaces with prescribed curvature[1, 21] then imply that �� is
�
2+� for all � < 1, out of a closed singular set S�
ofHausdorff dimension at most n− 7.
Recalling that g is Lipschitz continuous and Perc�����n �� <
+�, we can
reason as in [18, p. 168 and Proposition 14.11] (see also [17])
to obtain that�n+1 �= 0 on ��\S�, where � = ��1� � � � � �n+1�
denotes the exterior unit normal to ��.Reasoning as in [18, Theorem
14.13] it then follows E� = int�E�� = int���n��� and� ∈ �2+�loc
�E��. From the density estimate (24) we can derive that � ≤ C for
someC > 0, using the same argument as in Theorem 14.10 in [18].
So, this implies that �solves (9) in E� with boundary conditions
(10).
To prove the last assertion we notice that, letting �̃ be
another minimum of Fc,by convexity we have Fc��+ �1− ��̃� = 0 for
every ∈ �0� 1�. By definition of Fcwe then get
0 = Fc(
�+ �1− ��̃
)= Fc���+ �1− �Fc
(�̃)
if and only if
�D�̃ = �̃D� on E� ∩ E�̃�
which implies the assertion. �
Remark 3.8. Integrating (9) on E� and using (10) we obtain
Per�E���n� = −
∫E�
div
(D�√
1+ �D��2)dy =
∫E�
(g�y�− c√
1+ �D��y��2)dy� (25)
which implies that E� has finite perimeter.
Corollary 3.9. Let � as in Proposition 3.7. Then � satisfies the
boundaryconditions (11) on E�.
Proof. Let � ∈ �1per�Q�. By Proposition 3.7, minQ��− �� =
minE���− ��. Assumeby contradiction that �− � attains its minimum
at y0 ∈ E�. Without loss ofgenerality, we can assume that z0 �=
��y0� = ��y0� and that ��y�− ��y� > 0 forevery y �= y0. Again by
Proposition 3.7, we have x0 �= �y0� z0� ∈ S�, where S� is
thesingular set of ��.
Let us now blow-up the sets �� and the subgraph �� of � around
x0 If we let
�s� �= x ∈ �n+1 � sx ∈ �� − x0��s� �= x ∈ �n+1 � sx ∈ �� −
x0��
by standard arguments of the theory of minimal surfaces [18,
Chapter 9], one canprove that along a subsequence si → 0, �si�
converges to a half-space H ⊂ �n+1, and�si� converges to a minimal
cone C. From the inclusion �� ⊆ �� it follows C ⊆ H ,
but this implies that C = H , thus leading to a contradiction
since the cone C issingular. �
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
790 Cesaroni and Novaga
We now define the maximal support E for minima of the functional
Fc, andstudy the regularity of such set.
Proposition 3.10. There exists a set E = ∪ki=1Ei ⊆ Q, where Ei
are connectedcomponents, such that the support of every minimum �
of Fc is given by the union ofsome connected components of E. In
particular, if E is connected, then there exists aunique nontrivial
minimizer � of Fc, up to an additive constant.
Moreover, there exists a closed set S ⊂ E such that E\S is a
�2+� hypersurface,with � ��S� = 0 for every � > n− 8, and
satisfies the geometric equation
� = g on E\S� (26)
Proof. Let �1� �2 be two minima of the functional Fc and E1� E2
be the respectivesupports. By Proposition 3.7, if Ei1 and E
j2 are connected components respectively
of E1 and E2 then either Ei1 ∩ Ej2 = ∅ or Ei1 = Ej2. In this
case there exists a constant
k such that �1 = �2 + k on Ei1 = Ej2. We then define E as the
union of all theconnected components of the supports of the minima
of the functional Fc.
We claim that the connected components of E are finite. Fix Ei
connectedcomponent of E and �i solution to (9) with support Ei.
From (25) we obtain thatPer�Ei��
n� ≤ maxQ g�Ei�. This, combined with the isoperimetric
inequality (4), givesthat �Ei� ≥ �Cn/maxQ g�n, which implies our
claim.
If E is connected, the uniqueness up to addition of constants of
the minimizersis a consequence of Proposition 3.7.
We now show the regularity of E. Let � ≥ 0 be a minimizer of Fc
and assumewithout loss of generality that E = E�. Since � = �+ is
also a minimizer forall ∈ �, from the proof of Proposition 3.7 we
know that the subgraphs � =
�y� z� ∈ Q×� � z < ��y�� (locally) minimize the functional �c
defined in (21), forall ∈ �. In particular, since � → E ×� locally
in the L1-topology, as → +�,by compactness of quasi minimizers of
the area functional [1] we have that E �is also a minimizer of �c
under compact perturbations. The thesis then follows byclassical
regularity theory for minimal surfaces with prescribed curvature
[1, 21]. �
Remark 3.11. When n = 1, (26) reduces to
g = 0 on E�
In particular, E �= Q necessarily implies minQ g ≤ 0.
Remark 3.12. Let � � E→ � be a minimizer of Fc with maximal
support, as in theproof of Proposition 3.10, and let � = ec�
cbe the corresponding minimizer of Gc.
Since Gc is a convex functional on L2�Q�, by the general theory
of subdifferentials
in [3, 8] there exist a vector field �� = � � Q→ �n, with ��� ≤
1 and div��� ∈ L2�Q�,and a function h� = h � Q→ �, with 0 ≤ h ≤ 1,
such that∫
Q�−div ��y�+ ch�y�− g�y�� �w −��dy ≥ 0 in Q� (27)
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 791
for all w ∈ BVper�Q� such that w ≥ 0. Moreover, for all y ∈
E�,
h�y� = c��y�√c2� 2�y�+ �D��y��2
��y� = D��y�√c2� 2�y�+ �D��y��2
�
If we apply inequality (27) to w = � + �F , where F ⊆ Q is a set
of finite perimeter,we obtain
Per�F�Q�+∫F�ch�y�− g�y�� dy ≥ 0� (28)
In particular, (25) and (28) imply that E is a minimum for the
functional
��F� = Per�F��n�+∫V�ch�y�− g�y�� dy F ⊆ Q�
Remark 3.13. We observe that, if � a solution to (8) such that
ec� ∈ BVper�Q� forsome c > 0, which by regularity amounts to say
that � is bounded from above,then necessarily Fc��� = 0 so that c ≤
c (see Corollary 3.2). Moreover, if c < c, theninfE� � = −�.
This means that our variational method selects the fastest
travelingwave solutions to (1) which are bounded from above [23].
However, there mightexist other traveling wave solutions with c
> c, which are not in BVper�Q� (see forinstance [22]).
3.1. Existence of Classical Traveling Waves
In this subsection we state some condition on the forcing term g
under whichequation (9) admits a bounded solution � in Q. This
problem can be restatedas following: find sufficient conditions on
g, under which the maximal support Edefined in Proposition 3.10
coincides with Q.
Remark 3.14. Observe that a first necessary condition on g,
under which equation(9), with c > 0, admits a bounded solution �
in Q is that
∫Qg > 0. In fact, if
∫Qg = 0
and � is a bounded solution to (8), then c = 0. In [5] we show
that condition (7) issufficient to get the existence of a bounded
smooth solution to (8) on Q with c = 0.Proposition 3.7 shows that
this condition is essentially optimal for the existence
ofstationary wave solutions.
We consider a solution � to (9) with boundary conditions (10)
and maximalsupport E. Let � = ec�
c. We recall that by (25)
Per�E��n� =∫E�g�y�− ch�y�� dy ≤ max
Qg�E�� (29)
where h = h� is the function defined in (27), In Remark 3.12.
Since by (28)∫Qch�y�− g�y�dy ≥ 0�
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
792 Cesaroni and Novaga
we also have
Per�E��n� ≤∫Q\Ech�y�− g�y�dy� (30)
From inequality (30), recalling 0 ≤ h ≤ 1 and that ∫Qg ≤ c ≤
maxQ g, it follows
Per�E��n� ≤(maxQg −min
Qg
)�Q\E�� (31)
Assume now �Q\E� > 0. Recalling the isoperimetric inequality
(4), from (29)and (31) we get(
maxQg −min
Qg
)1
21n
≥(maxQg −min
Qg
)�Q\E� 1n ≥ Cn or �Q\E� >
12�
In particular, if
maxQg −min
Qg < Cn 2
1n (32)
we necessarily have �E� ≤ 1/2 and, from (29),
maxQg ≥ Per�E��
n�
�E� ≥ Cn�E�− 1n ≥ Cn2 1n � (33)
If minQ g ≤ 0, then (32) implies that , in contradiction with
(33).If minQ g > 0, from (33) we get
12≥ �E� ≥
(Cn
maxQ g
)n�
From (31) it then follows(maxQg −min
Qg
)(1−
(Cn
maxQ g
)n)≥(maxQg −min
Qg
)�1− �E��
≥ Cn�E� n−1n
≥ Cn(Cn
maxQ g
)n−1�
So if minQ g, we necessarily have E = Q if either maxQ g <
Cn2 1n or maxQ g ≥ Cn2 1nand maxQ g −minQ g < Cn
(Cn
maxQ g
)n−1(1− ( CnmaxQ g )n)−1.
Collecting the previous results above and recalling Remark 3.11
we get thefollowing proposition.
Proposition 3.15. Assume that∫Qg > 0. Then equation (9)
admits a bounded solution
� in Q if one of the following conditions is verified.
– minQ g ≤ 0 and maxQ g −minQ g < Cn21/n;
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 793
– g > 0 on Q and maxQ g < Cn21/n;
– g > 0 on Q, maxQ g ≥ Cn21/n and maxQ g −minQ g < maxQ
g((maxQ g
Cn
)n − 1)−1;– n = 1 and g > 0 on Qwhere Cn is the isoperimetric
constant appearing in (4) (and C1 = 2).
Remark 3.16. Observe that the assumptions in the previous
Proposition assurethe existence of classical traveling wave
solutions to (1), i.e. solutions of the formct + ��x�, where � is a
smooth, �n-periodic solution to (9).
Remark 3.17. In [19] Lions and Souganidis showed that (9) admits
a (periodic)solution over all Q if g does not change sign and
satisfies the condition
∃� ∈ �0� 1� s.t. minx∈Q
(�g2�x�− �n− 1�2�Dg�x��) > 0�
In [9] Cardaliaguet et al. proved that, when n = 1 and ∫ 10
g�y�dy > 0, thefollowing condition implies the solvability of
the cell problem:
0 ≤∫ 10g�y�dy − min
z∈�0�1�g�z� < 2� (34)
4. Stability and Long-Time Behavior
If u is a solution to (1), then w�t� y� = u�t� y�− ct is a
solution to
wt = tr[(
I − Dw⊗Dw1+ �Dw�2
)D2w
]+ g√1+ �Dw�2 − c in �0�+��×Q (35)
with periodic boundary conditions and initial datum w�0� y� =
u0�y�. Note that wis the unique solution to (35), and it is also a
classical solution, see Theorem 2.3.Standard comparison gives that
�min g − c�t − �u0�� ≤ w�t� x� ≤ �max g − c�t +�u0�� for every t ≥
0, x ∈ �n. Moreover, under the assumption (6), w is bounded(from
below) uniformly in t.
Lemma 4.1. Let w be the solution to (35) and � be any solution
to (9), then
w�t� y�− ��y� ≥ minQ�u0 − �� ∀ t ≥ 0� y ∈ Q� (36)
Moreover, if there exists a solution � to (9) in Q, then there
exists a constant M ,depending only on �u0�� such that �w�t� y�� ≤
M for every t ≥ 0 and y ∈ Q.
Proof. We fix a � solution to (9), and let E = E� (see
Proposition 3.7). We recallthat by Corollary 3.9, � satisfies the
boundary conditions (11) on E�.
We shall prove that
m�t� �= minx∈Q�w�t� x�− ��x��
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
794 Cesaroni and Novaga
is nondecreasing in t. Obviously this is sufficient to prove
that minx∈E�w�t� x�− ��x��is nondecreasing in t. We fix s ≥ 0 and
observe that w�t + s� x� is the solution to
vt�t� x� = tr[(
I − Dv⊗Dv1+ �Dv�2
)D2v
]+ g �x�√1+ �Dv�2 − c in �0�+��× E
with initial datum v�0� x� = w�s� x�, and with boundary
conditions v�t� x� = w�t +s� x� on E for all t ≥ 0. Notice that
��y�+minŷ∈Q�w�s� ŷ�− ��ŷ�� is a regular(stationary) subsolution
to the same problem. Moreover by Corollary 3.9 we havethat w�t + s�
x�− ���x�+miny∈Q�w�s� y�− ��y��� can attain its minima only in
theinterior of E. So we can apply comparison principle arguments
(see [4]) to concludethat w�t + s� x�− ��x� ≥ miny∈Q�w�s� y�− ��y��
for every t ≥ 0 and x ∈ Q.
Finally, if there exists a solution � to (9) in the whole Q,
then ��x�+ �u0�� +���� and ��x�− �u0�� − ���� are, respectively, a
supersolution and a subsolutionto (1) and we conclude by the
standard comparison principle. �
Remark 4.2. Note that if there is a solution to (9) in the whole
Q, a similarargument gives that the quantity
maxx∈Q�w�t� x�− ��x��
is nonincreasing in t.
Lemma 4.3. Let w be a solution to (35). Then for all � > 0
there exists a constantC > 0, depending on u0, g and �, such
that �wt� ≤ C for all t ≥ �.Proof. Recalling Theorem 2.3, we
define
C �=∥∥∥∥tr[(I − Dw��� ·�⊗Dw��� ·�1+ �Dw��� ·��2
)D2w��� ·�
]+ g�x�√1+ �Dw��� ·��2 − c∥∥∥∥
L��Q�< +�
Then S�t� x� = Ct + w�t� ·� is a supersolution to (35) and s�t�
x� = −Ct + w�t� ·� isa subsolution for all t > �. Then by
comparison [4] we obtain −Ct ≤ w�t� x�−w��� x� ≤ Ct. Moreover, for
every fixed s > �, we get that w�t� x�+ supx �w�s� x�−w��� x��
and w�t� x�− supx �w�s� x�− w��� x�� are respectively a
supersolution and asubsolution to (35) with initial data w�s� x�.
So, again by comparison, and recallingthe previous estimate, for
every � ≤ s ≤ t we obtain
−Cs ≤ w�t + s� x�− w�t� x� ≤ Cs��
The estimate in Lemma 4.3 imlies that, for all t > 0 the
function w�t� ·� satisfiesin the viscosity sense
−C − g�x� ≤ div(
Dw�t� x�√1+ �Dw�t� x��2
)≤ C + c − g�x� in �n ∀t ≥ �� (37)
So, this gives in particular that the curvature of the graph of
w�t� ·� is uniformlybounded with respect to t ∈ ���+��.Proposition
4.4. Let �w�t� ⊂ Q×� be the graph of w�t� ·�. Then, for all � >
0, �w�t�are hypersurfaces of class �1+�, for all � ∈ �0� 1�,
uniformly in t ∈ ���+��.
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 795
Proof. Assume by contradiction the statement to be false. Then
we can find�xn� tn� ∈ Q× �0�+�� such that, for all � > 0, the
hypersurfaces �w�tn� ∩ B��xn� tn�are not uniformly �1+�. Letting
w̃n�x� �= w�x� tn�− w�xn� tn�, from (37) we have that
−div(
Dw̃n�x�√1+ �Dw̃n�x��2
)= hn�x�� (38)
with �hn�� ≤ C̃ for some C̃ independent of n. As a consequence
w̃n is a minimizerof the prescribed curvature functional∫
Q
(√1+ �Du�2 − hnu
)dy�
By the compactness theorem for quasi minimizers of the perimeter
[1] the graphs�w̃n of w̃n converge locally in the L
1-topology, up to a subsequence, to a limithypersurface �� of
class �1+�. We can also assume that xn → x for some x ∈ Q,and let
�� be the normal vector to �� at �x� 0�. However, by Theorem 2.5
thereexists � > 0 such that �w̃n ∩ B��x� 0� and �� ∩ B��x� 0�
can all be written as graphsin the direction given by ��.
Therefore, by elliptic regularity for minimizers of theprescribed
curvature functional [21], the sets �w̃n ∩ B��x� 0� are uniformly
of class�1+� for all � ∈ �0� 1�, thus leading to a contradiction.
�
The following lemma that will be useful in the following.
Lemma 4.5. Let Fc�v� =∫Qecv�y�
(√1+ �Dv�y��2 − g�y�
c
)dy the functional defined
in (12). Then for every (smooth) solution w to the equation in
(35),
0 ≤ Fc�w�t� ·�� ≤ Fc�u0� for all t > 0� (39)
Proof. For every solution w to (35), using the definition of the
functional Fc, we get
dFc�w�t� ·��dt
=∫Qecwwt
[−div
(Dw√
1+ �Dw�2)− g + c√
1+ �Dw�2]
= −∫Q
ecww2t√1+ �Dw�2 ≤ 0� (40)�
The first result on the asymptotic behavior of the solutions u
to (1) is about theconvergence of u�t�x�
tas t→ +�.
Proposition 4.6. Let u be the solution to (1) and E be the
maximal support defined inProposition 3.10. Then
limt→+�
maxx∈Q u�t� x�t
= c� and limt→+�
u�t� x�
t= c locally uniformly in E�
Moreover if there exists a bounded solution to (9),
limt→+�
u�t� x�
t= c uniformly in Q�
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
796 Cesaroni and Novaga
In particular there exists a constant C ∈ � such that
minQu0�x� ≤ M�t� �= max
Q�u�t� x�− ct� ≤ C + log�1+ t�
c� (41)
Proof. Recall that if the stationary problem (9) has a bounded
solution, Lemma 4.1gives an uniform bound on w�t� x� = u�t� x�− ct,
and then we obtain the result.
We observe, recalling Lemma 4.1, that to prove the general
statement it issufficient to prove (41). The lower bound on M is an
immediate consequence ofLemma 4.1, just by choosing � as the
maximal nonpositive solution to (9).
We define f�t� x� �= 2cecw�t�x�
2 , so that f 2t �t� x� = w2t �t� x�ecw�t�x�. Integrating
(40)between 0 and T , we obtain
C ≥ Fc�u0� ≥ Fc�u0�− Fc�w�T� ·� =∫ T0
∫Q
f 2t �t� x�√1+ �Dw�t� x��2 dx dt (42)
for some constant C > 0 depending only on u0 and g.Given �
> 0 we let
M��t� �= maxx
1�B��
∫B��x�f�t� y�dy�
where we identify f with its periodic extension on �n. Given a
point x̄ where M��t�attains its maximum, thanks to Proposition 4.4
we can choose � < 2/c, independentof t and x̄, such that �Dw�x�
t�� ≤ 1 for every x ∈ B��x̄�. Notice that
2cec�M�t�−��
2 ≤ M��t� ≤2cecM�t�2 for all t ≥ 0�
so that, in order to prove the second inequality in (41), it is
enough to show that
M��t� ≤ K(1+√t
)� (43)
for some constant K > 0 (possibly depending on �). Given t ≥
0, we let
��t� �={x ∈ Q � M��t� =
1�B��
∫B��x�f�t� y�dy
}�
Using the fact that �Dw�x� t�� ≤ 1 on B��x̄�t��, from (42) and
Jensen’s inequalitywe get
C̃ = 2C�B��≥∫ T0
maxx̄�t�∈��t�
1�B��
∫B��x̄�t��
f 2t �t� x�dx dt
≥∫ T0
(maxx̄�t�∈��t�
1�B��
∫B��x̄�t��
ft�t� x�dx
)2dt (44)
=∫ T0M ′��t�
2dt ≥ 1T
(∫ T0
�M ′��t��dt)2�
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 797
From (44) we then have
M��T� ≤ M��0�+∫ T0
�M ′��t��dt ≤ M��0�+√C̃T
which gives (43). �
We now prove our main convergence result.
Theorem 4.7. Let u�t� x� be the unique solution to (1) with
periodic boundaryconditions, let M�t� �= maxQ w�t� y�, and let
w̃�t� x� �= w�t� x�−M�t� = u�t� x�−maxx∈Q�u�t� x�� ≤ 0�
Then, for any sequence tn → +� there exist a subsequence tnk and
a function � � E� →� (possibly depending on the subsequence tnk)
such that
w�tnk� x� −→{��x� locally in �1+��E��−� locally uniformly in
Q\E�
(45)
as k→ +�, for all � ∈ �0� 1�. Moreover � is a generalized
traveling wave solutionto (9).
Proof. We let
W�t� y� �= ecw�t�y�
c� W̃ �t� y� �= e
cw̃�t�y�
c= e−cM�t�W�t� y� ≤ 1
c�
Notice that from (35) it follows that W satisfies the
equation
Wt =√c2W 2 + �DW �2
div DW√c2W 2 + �DW �2
+ g− c2W in �0�+��×Q�
(46)
By (39) and (41), for all t ≥ 0 we haveGc�W̃ �t� ·�� = Fc�w̃�t�
·�� = e−cM�t�Fc�w�t� ·�� ≤ e−c�minQ u0�Fc�u0��
In particular,∫Q
√c2W̃ 2�t� y�+ �DW̃�t� y��2dy = Gc�W̃ �t� ·��+
∫g�y�W̃ �t� y�dy ≤ C
for all t ≥ 0, where C depends only on u0 and g. Hence, up to
extracting asubsequence tnk , W̃ �tnk � ·� ⇀ W� weakly* in
BVper�Q�, as k→ +�. Notice that, asin the previous section, the
epigraph of w̃�t� ·� is, for every t > 0, a minimizer of
theprescribed curvature functional
� �→ Perc����n ×��−∫�eczg̃t�y�dy dz
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
798 Cesaroni and Novaga
where g̃t is an appropriate bounded function, depending on t. It
therefore satisfiesthe lower density bound (24), which implies W�
�≡ 0. We claim that
Gc�W�� = 0� (47)
We introduce the modified functional, for t > 0,
G̃c�t�W� �=∫Q
(√c2W 2 + �DW �2 − g̃tW
)dy
where
g̃t�y� �= g�y�−Wt�t� y�√
c2W 2�t� y�+ �DW�t� y��2∈ L��Q�� �g̃t�� ≤ C�
with C independent of t. From (46) it follows by direct
computation thatG̃c�t�W�t� ·�� = 0, hence also G̃c�t�W̃ �t� ·�� =
0. Recalling (40), up to extracting afurther subsequence, we can
assume that
tGc�W�tnk� ·�� = tFc�w�tnk � ·��
= −∫Q
ecw�tnk �y�w2t �tnk � y�√1+ �Dw�tnk� y��2
dy
= −∫Q
W 2t �tnk � y�√c2W 2�tnk � y�+ �DW�tnk� y��2
dy
→ 0 (48)
as k→ +�.Since Gc�v� ≥ 0 for every v, to prove the claim (47) it
is sufficient to show that
Gc�W�� ≤ 0. We get, using the convexity of Gc and the definition
of the modifiedfunctional G̃c�t,
Gc�W�� ≤ lim infk→+�
Gc�W̃ �tnk � y��
= lim infk→+�
G̃c�tnk �W̃ �tnk � y��− ∫Q W̃�tnk � y�Wt�tnk � y�√c2W 2�tnk �
y�+ �DW�tnk� y��2dy
= lim infk→+�
−∫Q
W̃�tnk � y�Wt�tnk � y�√c2W 2�tnk � y�+ �DW�tnk� y��2
dy
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 799
since G̃c�W̃ �tnk � y�� = 0. Using the Hölder inequality, (48)
and the definition of W̃ ,we obtain
lim infk→+�
∫Q
−W̃ �tnk � y�Wt�tnk � y�√c2W 2�tnk � y�+ �DW�tnk� y��2
dy
≤ lim infk→+�
∫Q
W 2t �tnk � y�√c2W 2�tnk � y�+ �DW�tnk� y��2
dy
12 (∫
Q
e−cM�t�
c2dy
) 12
= 0
which proves our claim. In particular, � �= log�cW��/c � E� →
�−��+�� is atraveling wave solution of (9) with c = c.
Let us now prove (45). Given y ∈ E�, by Theorem 2.5 there exists
r > 0 suchthat Br�y� ⊂ E� and �Dw̃�tnk � y��L��Br �y�� is
uniformly bounded in k. By standardelliptic regularity [16] it then
follows that the functions w̃�tnk � ·� are uniformlybounded in
�1+��Br�y�� for all � ∈ �0� 1�, so that they converge to � locally
in�1+��E��.
Fix now y ∈ Q\E� and take r > 0 such that Br�y� ⊂ Q\E�.
Assume bycontradiction that there exist c ∈ � and yk ∈ Br�y�, k ∈ ,
such that w̃�tnk � yk� ≥ cfor all k. By the density estimate (24)
this would imply
∫QW̃�tnk � y�dy ≥ c′ for some
c′ ∈ �, contradicting the fact that W̃ �tnk � y�→ W� in L1�Q�,
with W� ≡ 0 in Br�y�.We thus proved (45). �
Remark 4.8. If the functional Fc admits a unique minimizer �̄ �
E� → � up toan additive constant (for instance if the maximal
support E is connected, seeProposition 3.10), then instead of (45)
we have
limt→+�w�t� x� =
{��x�−maxE� �̄ locally in �1+��E��−� locally uniformly in
Q\E�
(49)
for all � ∈ �0� 1�.Corollary 4.9. Let u�t� x� be the unique
solution to (1) with periodic boundaryconditions, and assume that
there exist bounded solutions to (9) in Q (seeProposition 3.15).
Then
u�t� x�− ct −→ ��x� in �1+��Q�� as t→ +��where � is a bounded
solution to (9).
Proof. By Lemma 4.1 and Remark 4.2, it is enough to prove that
w�tn� x�→ ��x�uniformly along a subsequence tn → +�. This result
can be obtained by repeatingthe same argument as in the proof of
Theorem 4.7. �
Remark 4.10. A straightforward adaptation of the argument in
Corollary 4.9 givesthat, under assumption (7),
u�t� x�→ ��x� in �1+��Q�� as t→ +��
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
800 Cesaroni and Novaga
where � is a stationary solution of the parabolic equation (1)
(whose existence hasbeen shown in [5]).
Remark 4.11. The results of this paper can be easily extended to
equation (1)considered on a bounded open set ! ⊂ �n with Lipschitz
boundary, and withNeumann boundary conditions on !.
Acknowledgments
The authors thank Guy Barles and Cyrill Muratov for inspiring
discussions on thisproblem.
References
[1] Ambrosio. L. (1997). Corso Introduttivo Alla Teoria
Geometrica Della Misura EdAlle Superfici Minime [Introductory
Course on Geometric Measure Theory andMinimal Surfaces]. Pisa,
Italy: Edizioni della Scuola Normale Superiore.
[2] Ambrosio, L., Fusco, N., Pallara, D. (2000). Functions of
Bounded Variation andFree Discontinuity Problems. Oxford: Oxford
Mathematical Monographs.
[3] Andreu-Vaillo, F., Caselles, V., Mazón, J. (2004). Parabolic
QuasilinearEquations Minimizing Linear Growth Functionals. Basel:
Birkhäuser Verlag.
[4] Barles, G., Biton, S., Bourgoing, M., Ley, O. (2003).
Uniqueness results forquasilinear parabolic equations through
viscosity solutions methods. Calc. Var.Partial Differential
Equations 18:159–179.
[5] Barles, G., Cesaroni, A., Novaga, M. (2011). Homogenization
of fronts inhighly heterogeneous media. SIAM J. Math. Anal.
43:212–227.
[6] Barles, G., Porretta, A., Tabet Tchamba, T. (2010). On the
large time behaviorof solutions of the Dirichlet problem for
subquadratic viscous Hamilton-Jacobiequations. J. Math. Pures Appl.
94:497–519.
[7] Barles, G., Souganidis, P.E. (2001). Space-time periodic
solutions and long-timebehavior of solutions to quasi-linear
parabolic equations. SIAM J. Math. Anal.32:1311–1323.
[8] Brezis, H. (1973). Operateurs Maximaux Monotones [Maximal
MonotoneOperators]. Amsterdam: North-Holland.
[9] Cardaliaguet, P., Lions, P.-L., Souganidis, P.E. (2009). A
discussion about thehomogenization of moving interfaces. J. Math.
Pures Appl. 91:339–363.
[10] Chambolle, A., Thouroude, G. (2009). Homogenization of
interfacial energiesand construction of plane-like minimizers in
periodic media through a cellproblem. Netw. Heterog. Media
4:127–152.
[11] Chen, X., Lou, B. (2009). Traveling waves of a curvature
flow in almostperiodic media. J. Differential Equations
247:2189–2208.
[12] Chou, K.-S., Kwong, Y.-C. (2001). On quasilinear parabolic
equations whichadmit global solutions for initial data with
unrestricted growth. Calc. Var.Partial Differential Equations
12:281–315.
[13] Da Lio, F. (2008). Large time behavior of solutions to
parabolic equations withNeumann boundary conditions. J. Math. Anal.
Appl. 339:384–398.
[14] Dirr, N., Karali, G., Yip, N.-K. (2008). Pulsating wave for
mean curvature flowin inhomogeneous medium. European J. Appl. Math.
19:661–699.
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13
-
Long-Time Behavior 801
[15] Ecker, K., Huisken, G. (1989). Mean curvature evolution of
entire graphs.Annals of Math. 130:453–471.
[16] Gilbarg, D., Trudinger, N.S. (1983). Elliptic Partial
Differential Equations ofSecond Order. Berlin: Springer.
[17] Giusti, E. (1978). On the equation of surfaces of
prescribed mean curvature.Invent. Math. 46:111–137.
[18] Giusti, E. (1984). Minimal Surfaces and Functions of
Bounded Variation.Monographs in Mathematics, Vol. 80. Basel:
Birkhäuser Verlag.
[19] Lions, P.-L., Souganidis, P.E. (2005). Homogenization of
degenerate second-order PDE in periodic and almost periodic
environments and applications.Annales IHP – Analyse Nonlinéaire
22:667–677.
[20] Lou, B. (2009). Periodic traveling waves of a mean
curvature flow inheterogeneous media. Discrete Contin. Dyn. Syst.
25:231–249.
[21] Massari, U. (1974). Esistenza e regolarità delle
ipersuperfice di curvatura mediaassegnata in �n. Arch. Rational
Mech. Anal. 55:357–382.
[22] Monneau, R., Roquejoffre, J.M., Roussier-Michon, V. (2013).
Travellinggraphs for the forced mean curvature motion in an
arbitrary space dimension.Ann. Sci. Ec. Norm. Supér 46:215–246.
[23] Muratov, C. (2004). A global variational structure and
propagation ofdisturbances in reaction-diffusion systems of
gradient type. Discrete Cont. Dyn.Syst. B 4:867–892.
[24] Muratov, C., Novaga, M. (2008). Front propagation in
infinite cylinders. II.The sharp reaction zone limit. Calc. Var.
Part. Diff. Eqs. 31:521–547.
[25] Namah, G., Roquejoffre, J.M. (1997). Convergence to
periodic fronts in aclass of semilinear parabolic equations.
Nonlinear Differential Equations Appl.4:521–536.
[26] Ninomiya, H., Taniguchi, M. (2001). Stability of traveling
curved fronts in acurvature flow with driving force. Methods Appl.
Anal. 8:429–450.
[27] Tamanini, I. (1982). Boundaries of Caccioppoli sets with
Hölder-continuousnormal vector. J. Reine Angew. Math.
334:27–39.
Dow
nloa
ded
by [
Uni
vers
ita d
i Pad
ova]
at 0
6:13
12
Apr
il 20
13