Qualitative properties of a continuum theory for thin films Propriétés qualitatives d’une théorie de continuum
pour des couches minces
Bernd Schmidt
Max-Planck-Institute for Mathematics in the Sciences, Inselstr. 22, 04103 Leipzig, Germany
Received 5 December 2005; received in revised form 24 April 2006; accepted 29 September 2006 Available online 20 December 2006
Abstract
We discuss qualitative aspects of a continuum theory for thin films rigorously derived in [B. Schmidt, On the passage from atomic to continuum theory for thin films, preprint 82/2005, Max-Planck Institut für Mathematik in den Naturwissenschaften, Leipzig].
The stored energy density is examined for convexity properties and limiting behavior under large and small strains. A study of the dependence of the theory on relaxation parameters leads to the result that the scale of convergence used in [B. Schmidt, On the passage from atomic to continuum theory for thin films, preprint 82/2005, Max-Planck Institut für Mathematik in den Naturwissenschaften, Leipzig] is the only scale for which a limiting theory that also accounts for atomic relaxation effects is non-trivial.
©2007 Elsevier Masson SAS. All rights reserved.
Résumé
Nous discutons des aspects qualitatifs d’une théorie de continuum pour des couches minces, dérivée rigoureusement dans [B. Schmidt, On the passage from atomic to continuum theory for thin films, preprint 82/2005, Max-Planck Institut für Mathe- matik in den Naturwissenschaften, Leipzig]. La densité d’énergie emmagasinée est examinée pour des propriétés de convexité et comportement en limite sous des distorsions grandes et petites. Une recherche de la dépendance de la théorie à l’égard des para- mètres de relaxation mène au résultat que l’échelle de la convergence employée dans [B. Schmidt, On the passage from atomic to continuum theory for thin films, preprint 82/2005, Max-Planck Institut für Mathematik in den Naturwissenschaften, Leipzig] est la seule échelle pour laquelle une théorie limite qui inclut également des effets de la relaxation atomique est non-triviale.
©2007 Elsevier Masson SAS. All rights reserved.
Keywords:Thin films; Discrete-to-continuum limits; Effective theories
E-mail address:bschmidt@aero.caltech.edu.
0294-1449/$ – see front matter ©2007 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.anihpc.2006.09.001
1. Introduction
The aim of this paper is to examine qualitative features of a macroscopic theory for thin films that was derived as an effective continuum theory from atomic models in [21]. Deriving thin film limits from three-dimensional elasticity is still an active area of research, see, e.g., [17–19,13,14,16] and, most recently, [15] where a whole hierarchy of different scaling limits is discussed. For the more classical developments see, e.g., [20,8]. On the other hand, by now there are also rigorousΓ-convergence results for the passage from discrete to continuum theory: for suitable pair interaction models, especially in one dimension, see [4–6]; more complicated potentials under additional assumptions as, e.g., the Cauchy–Born rule are considered in [2,3].
In [12], starting from reference configurations Lk=Z3∩ [0, k] × [0, k] × [0, ν−1]
for fixedν∈N, the number of film layers, andk∈N, a limiting continuum theory for the energy of deformations was proposed in the limitk→ ∞taking into account atomistic relaxation effects. In [21], this effective theory was obtained rigorously as a variational limit of the elastic energy functionalE(y(k))of deformationsy(k):Lk →R3. This continuum theory was expressed in terms of the gradient of a mapu:[0,1]2→R3andν−1 director fields bi:[0,1]2→R3,i=1, . . . , ν−1:
Theorem 1.1.(Cf.[21])Under suitable assumptions on the energy functionE, and for an appropriate definition of convergence of deformations, there existsϕ:R3×2×(R3)ν−1→Rsuch that
E
y(k)“Γ”
→
[0,1]2
ϕ
∇u, b1, . . . , bν−1
ask→ ∞.
It is worth mentioning that the scheme described in [21] can be applied not only to thin films but to general three- dimensional bodies leading to a stored energy densityϕonly depending on the deformation gradient∇u∈R3×3, if one assumes sufficiently fast decay of atomic interactions. The main technical difficulty in fact stems from the non- local convergence of relative layer displacements to the family of vector fields(b1, . . . , bν−1)(cf. Definition 2.3). For the qualitative aspects examined in the present paper we will however make use of the ‘thin film structure’ as we allow atoms to explore regions perpendicular to the macroscopic film surface.
In Section 2, after introducing the model, we will recall the precise statements from [21]. Also we will collect some preparatory material that was proved in [21] and will be needed in the sequel.
The following sections are devoted to studying this continuum theory, i.e. the macroscopic energy densityϕqual- itatively. First, cf. Section 3, we examine the dependence ofϕon the relaxation parameterc0(cf. Definition 2.3 and Theorems 2.7, 2.8 and 2.9) and study the limiting casesc0→ ∞andc0→0. Moreover, we will see that the physi- cally motivated rate of convergence for which a continuum theory was derived in [21] is the only scale that leads to a non-trivial limiting theory.
In the following two Sections 4 and 5, we derive the limiting behavior under large tensile and compressive strains, and explore the convexity properties and symmetries of the limiting energy functional.
Finally, in Section 6, the scaling behavior of certain systems nearO(2,3), i.e.,(∇u)T∇u≈IdR2, is examined.
We still find a non-trivial energy response to compressive strains in this regime. It is, however, weaker than cal- culated without taking into account atomic relaxation effects. In order to prove this result we are led to study the one-dimensional version, an atomic chain, in detail. The results of this paragraph might be of independent interest.
2. The passage from atomic to continuum theory
We give a brief account of the results obtained in [21] on the passage from atomic models to a continuum theory for thin films. For details, motivations of the concepts, and proofs of the results of this section we refer to [21].
Fig. 1.
2.1. The model 2.1.1. Kinematics
We consider a film ofνatomic layers whose reference configuration will be Lk=L∩
Sk× [0, h] ,
whereSk:= [0, k] × [0, k]fork∈N,h:=ν−1 is the height of the film and, for sake of simplicity,L=Z3 (see Fig. 1).
The deformations of this configuration will be denoted by y=y(k):Lk→R3.
In order fory to be defined not only on the atomic positions, we will assume some interpolation between the atomic positions: for a deformationy:Lk→R3letx¯=x+(1/2,1/2)forx∈ {0, . . . , k−1}2and set
y(x, i)¯ =1 4
z∈Z2,
|z− ¯x|=1/√ 2
y(z, i), i=0, . . . , ν−1.
Now on each of the four triangles with corners(x, i), (z, i), (z¯ , i), wherez, z ∈Z2with|z− ¯x| =1/√
2,|z−z| =1 interpolate linearly to obtainy(x, i)forx∈Sk. Interpolating in between the layers is not so subtle, for definiteness we chooseyto be linear on the segments[(x, i−1), (x, i)]. By this particular choice we guarantee that (local) averages depend only on atomic positions.
Our aim being to study the limitk→ ∞, it is natural to introduce the rescaled functionsy˜defined on the common domainS1× [0, h]:
˜
y(k)(x):=1
ky(k)(kx1, kx2, x3).
Considering weak*-limiting points ofy˜as natural variables for a continuum theory, we are led to elementsuof W1,∞([0,1]2;R3)as limiting deformations. In our regime of thin films of fixed atomic height, we also introduce the quantities
iy˜(k)(xp)= ˜y(k)(x1, x2, i)− ˜y(k)(x1, x2,0), i=1, . . . , ν−1,
xp=(x1, x2), to measure the relative shift of the layers of our film. Also these have weak*-limits inL∞. As in [21] we define:
Definition 2.1.Letu∈W1,∞(S1;R3)andb=(b1, . . . , bν−1)∈L∞(S1;(R3)ν−1). We say that(u,b)is admissible (for givenc0>0), i.e.(u,b)∈A, if there existsc1>0 such that
u(x)−u(z)c1|x−z| ∀x, z∈S1 (1) (minimal strain hypothesis), and there existsb0∈L∞such that
b0
∞,bi−b0
∞c0, i=1, . . . , ν−1. (2)
The unrescaled version ofuis denotedU, i.e.U=u. An easy consequence of our interpolation is the following
Lemma 2.2.Supposeuis admissible andy:Lk→R3some deformation withsupx∈Lk|y(x)−U (xp)|c. Thenyis Lipschitz. For any(rescaled)Lipschitz interpolationy:Sk× [0, h] →R3(y˜:S1× [0, h] →R3)there are constants C1, C2, C3>0such that,
(i) supx∈S1×[0,h]| ˜y(x)|C2,
(ii) C1|x−z| −C3|y(x)−y(z)|C2|x−z| ∀x, z∈Sk× [0, h].
We next define in what sense we understand deformations to converge to the limiting quantitiesuandb.
Definition 2.3. Let u∈W1,∞(S1;R3), b∈L∞(S1;R3). Choose c0>0 a constant. We say that y(k)→(u,b) (w.r.t.c0) if
y˜(k)−uc0/k and ∀i: kiy˜(k) b∗ i inL∞ask→ ∞.
Here and in the sequel we denote byf, respectively ˜fin rescaled variables, f := sup
x∈Lk
f (x), resp. ˜f := sup
x∈Lk
f (x˜ p/k, x3).
As detailed in [21], this corresponds to a relaxation scheme where the individual atoms are allowed to move in a region comparable to atomic dimensions.
2.1.2. Energy
The energy of a system ofN atoms at positionsy1, . . . , yN∈R3will be a functionE:(R3)N→Ronly depending on atomic positions. To studyEwe will endow the configuration space(R3)N with the norm
(y1, . . . , yN)= sup
1iN
|yi|2.
The elastic energy of a deformation y, i.e. the energy of the system (y(x): x ∈Lk) respectively a subsystem M=y(K),K⊂Lk, is denoted
E(y)=E
y(x): x∈Lk
resp. E(M)=E
y(x): x∈K . We normalizeEso thatE(∅)=0.
The main two assumptions onEare firstly the following splitting estimate.
Assumption 2.4.Supposeuis admissible. There exists a functionψ:[0,∞)→Rsuch that
|ψ|M and ψ (r)Mr−q (3)
whereM, qare constants,M >0,q >3, such that for disjoint setsMandN of atoms we have E(M∪N)−E(M)−E(N)
v∈M, w∈N
ψ
|v−w| ,
whenevery −U∞C. (The function ψ may depend onC and on u throughc1 andc2 wherec1|x1−x2|
|u(x1)−u(x2)|c2|x1−x2|.)
Secondly, we need to assume some regularity ofE:
Assumption 2.5.Letube admissible. We assume thatEis locally Lipschitz and in aC-neighborhood ofU ∂
∂yiE(y) L
whereLmight depend onCand onUthroughc1,c2.
Furthermore, we assumeE to be frame indifferent and only depending on the atomic positions, i.e.E remains unchanged after renumbering of atoms and rigid motion of the configurationy(K).
For some results we will have to impose an additional restriction:
Assumption 2.6.Assume thatψ andL of Assumption 2.4 resp. 2.5 depend only onC1 andC3 wherey satisfies
|y(x)−y(z)|C1|x−z| −C3. 2.2. Convergence theorems
SupposeEsatisfies Assumptions 2.4 and 2.5, and a relaxation parameterc0>0 is chosen. The main result of [21]
is the following variational convergence result:
Theorem 2.7.There exists a macroscopic stored energy functionϕsuch that(in the spirit ofΓ-convergence, cf.[10]), (i) ify(k)→(u,b),(u,b)admissible, then
lim inf
k→∞ E y(k)
E(u,b),
(ii) and for all admissible(u,b)there exists a sequencey(k)→(u,b)such that
klim→∞E y(k)
=E(u,b).
HereE(u,b)is the macroscopic energy E(u,b)=
S1
ϕ
∇u, b1, . . . , bν−1
. (4)
To computeϕby an associated cell problem, set Nk0,1(A,b)=
y:Lk→R3: y−Ac0and 1 (k+1)2
x∈Z2∩Sk
iy(x)=bi
. (5)
Theorem 2.8.The macroscopic energy densityϕof Theorem2.9is given by ϕ(A,b)= lim
k→∞ϕk(A,b) (6)
where for later use we have introduced the quantities ϕk(A,b)= 1
νk2 inf
y∈Nk0,1(A,b)
E(y). (7)
This limit is uniform on compact subsets ofAhomand depends continuously onA,b.
Here,Ahom⊂R3×2×(R3)ν−1, the set ofadmissible(A,b), is defined by Ahom:=
A, b1, . . . , bν−1
: rank(A)=2,∃b0∈R3s.t.b0, max
1iν−1
bi−b0c0
for matricesA∈R3×2and vectorsb1, . . . , bν−1.
We also mention the following quantitative version of Theorem 2.7:
Theorem 2.9.Supposel=l(k)is such thatl(k)→0andkl(k)→ ∞ask→ ∞. Let Wkl(u,b):=
y: ˜y−uc0/k, kiy˜−bi
W−1,∞l
wherefW−1,∞:=sup
f·χ: χ∈W01,1, χW1,1
0 = ∇χL1=1
. Then
klim→∞
1 νk2 inf
y∈Wkl(u,b)
E(y)=
S1
ϕ
∇u(x),b dx.
In fact, Theorems 2.7 and 2.8 also apply to the more general case whereEis of the form E(y)=1
2
i=j
W
|yi−yj|
+E0(y) (8)
whereE0 satisfies the usual assumptions, butW (r)becomes infinitely large as r tends to zero. (In particular, the Lennard–Jones potential is covered by these energy functions.)
Theorem 2.10.For anyr0>0assume thatW is Lipschitz on[r0,∞)and there existsM=M(r0)∈Rsuch that for (a.e.)rr0
W (r)Mr−q, W(r)Mr−q+1,
for rr0. Then Theorem 2.7extends to energy functions of the form (8) where, as in Theorem2.8,ϕ:Ahom→ (−∞,∞]is given by(6), continuous as a function with values inR∪ {∞}.
As another extension we note that the above results also apply to suitable systems of distinguishable particle systems with finite range interaction. Leta >0. To eachxi∈Lkwe assign a neighborhood
Uxi=
xj∈L: |xj−xi|a
=
x1i, . . . , xri
a
where the enumeration of elements ofUxi shall be such thatxi1=xi and if(xi1)3=(xi2)3, then xji1−xi1=xji2−xi2 forj=1, . . . , ra.
LetSka= [a, k−a]2and suppose the energy of a deformationyis given by Efr(y)=
xi∈L∩(Sak×[0,h])
fxi y
x2i
−y x1i
, . . . , y xir
a
−y x1i
+O(k), (9)
wherefxi:R3(ra−1)→Rare given functions representing the energy of the interactions between thei-th atom at its positiony(xi)=y(x1i)and its neighboring atoms in their positionsy(x2i), . . . , y(xria). (The termO(k)is introduced to compensate for boundary effects, sinceUxi is not contained inSk× [0, h]for xi in a boundary layer of constant width.) We need the following periodicity assumption: there exist fixedp1, p2∈Nsuch that
f(x1+p1,x2,x3)=fx=f(x1,x2+p2,x3) (10)
forx=(x1, x2, x3)∈(Z+)2× {0, . . . , ν−1}.
Proposition 2.11.SupposeEfris defined as in(9)and(10)holds. Assume that thefxiare locally Lipschitz. Then the limitϕfrof Theorem2.8exists and we have
klim→∞
1 νk2 inf
y∈Wkl(u,b)
Efr(y)=
S1
ϕfr
∇u(x),b(x) dx
asl→0andkl→ ∞.
Remark.For such systems we do not need to suppose thatusatisfies a minimal strain hypothesis. Thus,ϕis defined on all ofR3×2×(R3)ν−1.
Fig. 2.
2.3. Technical results
We now collect some of the technical results obtained in [21] that will be useful in the following sections.
Consider deformationsy:kΩ× [0, h] →R3forΩ⊂ [0,1]2.
Lemma 2.12.Letybe a deformation satisfying| ˜y−u|c/kandK⊂L∩(kΩ× [0, h]). Then there is a constantC (not depending onK)such that ifK=K1∪K2for disjointK1andK2, then
E
y(x): x∈K
−E
y(x): x∈K1C#K2.
SupposeQ= [0, a)2,a1, is partitioned by squaresU1, . . . , Ur of side-lengthl wherec0/kla plus some restRwith|R| =O(a·l),l a, as in Fig. 2. (Thenr∼(a/ l)2.)
SetM:= {y(x): x∈L∩(kQ× [0, h])},Mi:= {y(x): x∈L∩(kUi× [0, h])}.
Lemma 2.13.Supposey:kQ× [0, h] →R3satisfies| ˜y−u|c/kfor some admissibleu. Then there existsC >0 such that
E(M)− r i=1
E(Mi) C
ka2 l +k2al
.
Remark.In both of the previous lemmas,Cwill only depend onC1andC3provided Assumption 2.6 is satisfied.
To measure local spatial averages, we define the measureρ=ρ(k)=
x∈Z2δx/ kwhereδx/ kis the Dirac measure atx/k. Also set (after extendingbiboundedly outsideS1(constantly ifbiis constant))
b¯i(x)= −
x+[−1/2k,1/2k]2
bi(z)dz. (11)
Letb0as in (2) be given. For later use we introduce the deformationsv=v(k), defined by (interpolation of) v(x1, x2, i)=
u(x1, x2)−1kb¯0(x1, x2) fori=0,
u(x1, x2)+1k(b¯i(x1, x2)− ¯b0(x1, x2)) for 1iν−1 (12) for(x1, x2)∈1kZ2∩S1. Clearly,v(k)→(u,b). Its un-rescaled version will be denotedV, i.e.V˜ =v.
Lemma 2.14.Supposey is a deformation with y−Uc0+δ1 and
−
[0,1]2
kiy˜− ¯bi dρ
δ2, δ1, δ21. Then there existsy :Lk→R3with
y −Uc0, −
[0,1]2
kiy˜ dρ= ¯bi,
and E(y)−E(y)C
δ11/5+δ21/5 k2. (This combines Lemmas 3.11 and 3.13 in [21].)
Instead ofbi, it is sometimes more convenient to work with the quantitiesBi defined by choosingb¯0minimizing max
1maxiν−1
b¯i− ¯b0,b¯0 (c0) and setting
Bi:= ¯bi−1− ¯b0 fori=2, . . . , ν, B1:= − ¯b0. (13)
3. The dependence ofϕon the relaxation scheme
Our notion of convergencey(k)→(u,b)of atomic deformations to macroscopic variables u,b depends on the constantc0(cf. Definition 2.3). (To keep track of this dependence, we will sometimes addc0as an additional sub- script as e.g. inNk,c0,10,ϕk,c0.) Our first task is to analyze this dependence of our continuum theory on the relaxation parameterc0. It will turn out that we cannot relax sendingc0 to infinity. This is due to the (physically reasonable) decay assumptions on atomic interactions. Moreover,c0/k will prove to be the only scale which both accounts for atomistic relaxation effects and yields a non-trivial continuum theory. We start by proving the following regularity result.
Proposition 3.1.Fix(A,b)∈Ahom. The mappingc0→ϕc0(A,b)is decreasing and continuous.
Proof. Supposec0> c0. By Theorem 2.8,ϕc0(A,b)ϕc
0(A,b). Conversely, giveny∈Nk,c0,10(A,b), by Lemma 2.14 we find a deformationy ∈Nk,c0,1
0
(A,b)withE(y)E(y)+C(c0−c0)1/5k2 provided(A,b)is admissible forc0 and|c0−c0|1. Thereforeϕc
0(A,b)ϕc0(A,b)+C(c0−c0)1/5. 2 3.1. The limitc0→ ∞
SupposeE is an admissible pair potential with purely attractive pair interactionW 0, W ≡0. Considering deformations with larger and larger periodic cells where every atom is mapped to a single point, we see that for all admissibleA,b,
c0lim→∞ϕc0(A,b)= −∞.
In this paragraph we will show that the limitc0→ ∞in general will be trivial if Assumption 2.6 is satisfied.
Theorem 3.2.SupposeEsatisfies Assumptions2.4, 2.5, and2.6. Defineϕ∞:=limc0→∞ϕc0.(This limit exists point- wise in [−∞,∞)by Proposition 3.1.) Then ϕ∞(A,b)=ϕ∞(A,b)for all matrices A, A of rank two and all vectorsb,b ∈(R3)ν−1.(Every such matrix resp. vector is admissible ifc0is large enough.)
Proof. Suppose first thatA =A. ByVA,bwe denote the un-rescaled version ofv(cf. (12)) corresponding tou=A andb0set to zero. Forbsuch that the projection of eachbionto graph(A)has norm less than 2|A|,
VA,b(x)−VA,b(x)=A(xp−xp)+bx3−bx3 A(xp−xp)−4|A| C1|x−x | −C3,
C1, C3 independent ofb. From Assumption 2.6 and Lemma 2.12 we then find a constantC such that for thoseb, E(VA,b)Ck2. On the other hand, if for two vectorsb1,b2and somei∈ {1, . . . , ν−1},
bj2=bj1, forj=i, and bi2=b1i +Az, z∈Z2,
Fig. 3.
thenE(VA,b1)=E(VA,b2)+O(|z|k). So for allbwe obtain limk→∞ 1
νk2E(VA,b)C, whenceϕ∞(A,·)is an upper bounded function onR3(ν−1)with values in[−∞,∞). Since it is convex (by Proposition 5.3 allϕc0(A,·)are convex), it must be constant.
For the remaining part it suffices to show that ϕ∞(A,b)ϕ∞(A,b).
We proceed similarly as in the proof of Proposition 3.16 of the existence ofϕunder homogeneous conditions in [21].
Fixc0andδ >0. Choosingk0large enough we find by Theorem 2.8y∈Nk0,10,c0(A,b)with 1
νk20E(y)ϕc0(A,b)+δ
2. (14)
We construct a deformationy :Lk→R3,kk0, by patching together appropriately translated copies ofy: let U1, . . . , Usbe translates of[0, k0+1)2as in Fig. 3.
Letz1, . . . , zs denote the lower left corners of these sets, setfi=Azi and define y(x1, x2, x3)=y
x1−zi1, x2−zi2, x3 +fi forx∈L∩(Ui× [0, h]). Then
y −A = sup
x∈Lk0
y(x)−Axp sup
x∈Lk0
y(x)+ sup
xp∈Sk0
|Axp| =:c0.
Soc0depends onk0(andA, A) but is independent ofk. Since
−
[0,1]2
kiy˜ −bi
dρ= −
Uj
kiy˜ −bi dρ+O
k02 k
= 1 sk02
s j=1
Uj
kiy˜ −bi dρ+O
k02 k
=O k02
k
(note|kiy˜|2c0), by Lemma 2.14 we find a deformation ˆ
y∈Nk,0,1c0(A,b) (15)
such that 1
νk2E(y)− 1 νk2E(y)ˆ
C(c0) k02
k 1/5
. (16)
Using Lemma 2.13 and translational invariance, we would now like to split the energy to find that 1
νk2E
y(x): x∈Lk
− 1 νk20E
y(x): x∈Lk0C 1
k0+k0
k
. (17)
Fig. 4.
If this is possible, we find that by (17), (15), (16) and (14) forkk01 ϕk,c0(A,b) 1
νk2E ˆ
y(x): x∈Lk
1
νk02E
y(x): x∈Lk0
+δ 2 ϕc0(A,b)+δ.
Letting firstk→ ∞, we deduce from Proposition 3.1 ϕ∞(A,b)ϕc0(A,b)+δ.
Sinceδwas arbitrary, we finally get sendingc0→ ∞ ϕ∞(A,b)ϕ∞(A,b).
It remains to justify the application of Lemma 2.13. The problem is thatc0depends onk0. (For nearest neighbor models as discussed in Proposition 2.11, this splitting in (17) will in general not be possible: fory as described above neglecting the bonds between setsy(Ui × [0, h])could result in neglecting an essential part of the energy.) By the remark after Lemma 2.13, however, this will be possible if we can replacey by somey such that stilly −Ac0 depends only onk0andy consists of translates ofy(Lk0), but in addition satisfies a far-field minimal strain hypothesis with constantsC1, C3independent ofk0, i.e.
y (x1)−y (x2)C1|x1−x2| −C3. (18) We re-enumerate the squaresU1, . . . , Usas depicted in Fig. 4. (r∈Nto be specified later.) Depending onA, A, k (andc0,c0) we choose a unit vectore∈R3perpendicular to the graph ofA and numbers 0< a1<· · ·< ar2 (to be specified later), and define
y (x1, x2, x3)=y(x1, x2, x3)+aje ifx∈L∩Ui,j× [0, h], j∈ {1, . . . , r2}.
We will now findC1, C3independent ofk0such that (18) holds. Since still, on each of the setsUi,j× [0, h],y is a translated copy ofy, we may replacey byy . Applying (17) then finishes the proof.
Ifx1andx2lie in the sameUi,j× [0, h], this is clear from Lemma 2.2 sincey∈Nk0,10,c0(A,b).
Now suppose this is not the case, but still|x1−x2|∞< (r−1)(k0+1). Thenx1∈Ui1,j1×[0, h],x2∈Ui2,j2×[0, h] withj1=j2. But then
y (x1)−y (x2)|aj1−aj2| −y(x1)−y(x2) |aj1−aj2| −fi1,j1−fi2,j2−y
x1−(zi1,j1,0)
−y
x2−(zi2,j2,0) |aj1−aj2| −fi1,j1−fi2,j2−2c0−A
(x1)p−zi1,j1
−A
(x2)p−zi2,j2
|aj1−aj2| −Crk0−2c0−Ck0
2rk0 for|aj1−aj2|sufficiently large |x1−x2|.
So we assume that|aj1−aj2|,j1, j2∈ {1, . . . , r2}, are large enough to justify the above calculation.
Finally, letx1∈Ui1,j1× [0, h],x2∈Ui1,j2× [0, h]and|x1−x2|∞(r−1)(k0+1). Sinceeis perpendicular to the graph ofA andy lies in ac0-neighborhood of that graph, we find that forrnot too small
y (x1)−y (x2)=(aj1−aj2)e+y (x1)−y(x2)
(aj1−aj2)e+A(x1−x2)−y(x1)−Ax1−y(x2)−Ax2 |Ax1−Ax2| −2c0
y(x1)−y (x2)−4c0
fi1,j1−fi2,j2−y
x1−zi2,j1
−y
x2−zi2,j2−4c0
fi1,j1−fi2,j2−2c0−2|A|k0−4c0
czi1,j1−zi2,j2−2c0−2|A|k0−4c0
c
2zi1,j1−zi2,j2
c
6|x1−x2|,
wherec=min|x|=1|Ax|. The last but one inequality follows from the fact that fori1=i2 c
2zi1,j1−zi2,j2c(r−1)k0
4 2c0+2|A|k0+4c0
for|x1−x2|∞> (r−1)k0if we choosersufficiently big.
Settingc0=c0+max1jr2|aje|we furthermore havey −Ac0. So by possibly enlargingc0toc0, we can indeed split the energy to obtain (17), and the proof is finished. 2
For systems that do not satisfy Assumption 2.6,ϕ∞may be non-trivial (for an example see Proposition 4.5 in [21]).
In Section 5.1 we will prove thatϕ∞is quasiconvex with respect to the first variable and convex with respect to the second.
3.2. The limitc0→0
In our definition of convergencey(k)→(u,b), it does not make sense to consider the limiting case of very restricted relaxation, i.e.c0→0, unless allbi are zero. Instead of asking ˜y−u in Definition 2.3 to be less thanc0/k one could demand that
˜y−vc0/k (19)
wherev is as in (12) corresponding to u,bwithb0 set to zero. (Condition (2) is not needed for this definition of convergence.) This alternative set-up leads to analogous results in the passage to continuum theory, as shown in [21].
It is not hard to calculate the limit ϕ0(A,b):= lim
c0→0ϕc0(A,b)
which exists in(−∞,∞]sincec0→ϕc0(A,b)is decreasing.
Proposition 3.3.LetVA,bbe as in(12)for constant∇u=Aandb. Then ϕ0(A,b)= lim
k→∞
1 νk2E
VA,b(x): x∈Lk
.
In particular, the limit on the right-hand side exists(inRunder the usual Assumptions2.4and2.5, in(−∞,∞]for energies of the form(8)).
Proof. Suppose first E is of the form (8) and there arei=j ∈ {0, . . . , ν−1}such that bi∈bj+AZ2. Then, if y−VA,br,
E(y)k2 4 inf
0<srW (s)−Ck2→ ∞
asr→0. For the remaining cases note thatE(VA,b)is bounded by Lemma 2.12 and, ify−VA,br, E(y)−E(VA,b)Lνk2r.
Therefore, lim sup
k→∞ sup
y∈Nk0,1(A,b)
1
νk2E(y)− 1
νk2E(VA,b) Lc0. Now lettingc0→0 proves the claim. 2
Example.For admissible pair potentials (i.e.Wsatisfies the conditions of Theorem 2.10) Epp(y)=1
2
i=j
W
|yi−yj|
, (20)
we get
ϕ0(A,b)= lim
k→∞
1 2νk2
x,z∈Lk
x=z
WVA,b(x)−VA,b(z).
Restricting this sum to those x such that dist(xp, ∂[0, k]2) > l where 1lk yields an error term of order O(kl/k2)=o(1). Then summing over allz∈Z2× {0,1, . . . , ν−1},z=x, instead ofLk\ {x}gives another error term of orderO(l2−q)=o(1). This sum now being independent ofxp, we obtain
ϕ0(A,b)= 1 2ν
ν−1
i=0
z∈L∩(R2×[0,h]) z=(0,0,i)
WVA,b(z)−VA,b(0,0, i)
= 1 2ν
ν−1
i,j=0
zp∈Z2 (zp,j )=(0,0,i)
WAzp+bj−bi.
The corresponding macroscopic energy functional is given by E(u,b)=
S1
1 2ν
ν−1
i,j=0
z∈Z2 (z,j )=(0,0,i)
W∇u(x)z+bj(x)−bi(x)dx.
This expression can be seen as a thin-film version with directorsb1, . . . , bν−1of a formula derived in [3].
3.3. Triviality for slowly converging deformations
By our definition of convergence, the effective continuum theory depends on the scalel1=c0/kmeasuring the rate of uniform convergence ofy˜(k)tou. This paragraph serves to prove that in fact only the physically motivated choice l1(k)=const./kyields non-trivial results. Physically this amounts to the fact that thin film configurations are expected to be only locally energy minimizing: admitting for fracture, i.e., large interatomic distances, the film could locally (3d-) crystallize. Physically reasonable decay assumptions on atomic interactions will then lead to trivial macroscopic limits.
It is easy to see that forl11/kwe reproduce the limit obtained in Proposition 3.3. So suppose nowl1=l1(k) 1/k. (Then allb∈L∞(S1;(R3)ν−1)will be admissible.) In analogy toWkl (cf. Theorem 2.9) we define
Wkl1,l2(u,b):=
y: ˜y−ul1, kiy˜−bi
W−1,∞l2 .
Theorem 3.4.SupposeE satisfies Assumptions 2.4, 2.5, and 2.6. Assumel1(k),l2(k)satisfy kl1(k), kl2(k)→ ∞.
Then for all admissibleu(cf.(1))and allbthe limit E=E(u,b)= lim
k→∞
1
νk2 inf
y∈Wkl1,l2(u,b)
E(y)
exists in[−∞,∞)and is the same for all(u,b).
Proof. Let
E(u,b):=lim inf
k→∞
1
νk2 inf
y∈Wkl1,l2(u,b)
E(y).
Suppose thatb,b ∈L∞(S1;(R3)ν−1)andu,u are admissible. The proof follows along the lines of the proof of Theorem 3.2, we indicate the necessary modifications. Choosing a suitable largek0, we findy∈Wkl10,l−21(u,b)with
1
ν(k0−1)2E(y)E(u,b)+δ/3
(resp.−1/δforE(u,b)= −∞). In addition to the setsUi=zi+[0, k0+1)2consider the subsetsUi=zi+[0, k0)2, and constructy similar as in the proof of Theorem 3.2 by
y(x1, x2, x3)=y
x1−zi1, x2−zi2, x3
+U zi
onLk∩ Ui× [0, h] ,
where U denotes the unrescaled version of u. On the remaining (2k0+1)ν atoms of Ui × [0, h] we define y appropriately such that
−
Ui/ k
kiy˜ dρ(k)= 1 (k0+1)2
xp∈Z2∩Ui
y(xp, i)−y(xp,0)= −
Ui/ k
b¯idρ(k).
We may assume that forx, x ∈Lkwithxp∈Ui\Ui andxp∈Ui,|y(x)−y(x )||xp−xp|and thaty −Uis bounded in terms ofk0independently ofk.
Considering local spatial averages, we still findyˆ∈Wkl1,l2(u,b)such that fork0fixed 1
νk2E(y)− 1 νk2E(y)ˆ
→0 ask→ ∞.
(To prove this, one may choose a scale l3 such that 1/kl3l2 and apply Lemmas 3.13 and 3.14 in [21] resp.
Lemmas 2.2.12 and 2.213 in [22] with constants depending onk0.)
In order to show that the energy splits, again we possibly have to replacey byy . For the construction ofy we can only guarantee that
y (x1)−y (x2)C1|x1−x2| −C3
withC1, C3independent ofk0andkforx1andx2that do not lie in the sameUi,j∩Ui. But Lemma 2.13 still works in this more general case.rnow might not be a fixed number, but still it only depends onk0, the same being true for a1, . . . , ar2. Also note that for the same reason and by translational invariance
E
y(x): x∈L∩
Ui× [0, h]
−E(y)Ck0. Finally sendingkto infinity gives