www.elsevier.com/locate/anihpc
The relaxed energy for S 2 -valued maps and measurable weights L’énergie relaxée des applications à valeurs dans S 2
et poids mesurables
Vincent Millot
Laboratoire J.L. Lions, université Pierre et Marie Curie, B.C. 187, 4, place Jussieu, 75252 Paris cedex 05, France Received 26 October 2004; accepted 3 February 2005
Available online 10 May 2005
Abstract
We compute explicitly a relaxed type energy for mapsu:Ω⊂R3→S2. The explicit formula involves the length of a minimal connection relative to some specific distance connecting the topological singularities ofuand associated to a measurable weight function. This result generalizes a previous result of F. Bethuel, H. Brezis and J.M. Coron.
2005 Elsevier SAS. All rights reserved.
Résumé
Nous calculons explicitement une énergie de type relaxée pour des applicationsu:Ω⊂R3→S2. La formule explicite fait intervenir la longueur d’une connexion minimale relative à une certaine distance, connectant les singularités topologiques de uet associée à une fonction de poids mesurable. Ce résultat généralise un résultat antérieur de F. Bethuel, H. Brezis et J.M.
Coron.
2005 Elsevier SAS. All rights reserved.
MSC: 49D20; 49F99
1. Introduction and main results
LetΩ be a smooth bounded and connected open set ofR3and letw:Ω→Rbe a measurable function such that
0< λwΛ a.e. inΩ (1.1)
E-mail address: millot@ann.jussieu.fr (V. Millot).
0294-1449/$ – see front matter 2005 Elsevier SAS. All rights reserved.
doi:10.1016/j.anihpc.2005.02.003
for some constantλandΛ. We setHg1(Ω, S2)= {u∈H1(Ω, S2), u=gon∂Ω}, whereg:∂Ω→S2is a given smooth boundary data such that deg(g)=0. Our main goal in this paper is to obtain an explicit formula for the relaxed functional
Ew(u)=Inf
lim inf
n→+∞
Ω
∇un(x)2w(x)dx, un∈Hg1(Ω, S2)∩C1(Ω), u n uweakly inH1
,
defined foru∈Hg1(Ω, S2). By a result of F. Bethuel (see [1]),Hg1(Ω, S2)∩C1(Ω) is sequentially dense for the weak topology inHg1(Ω, S2)and then the functionalEwis well defined.
In [4], F. Bethuel, H. Brezis and J.M. Coron have proved that forw≡1, E1(u)=
Ω
∇u(x)2dx+8π L(u),
where L(u) denotes the length of a minimal connection relative to the Euclidean geodesic distance dΩ in Ω connecting the singularities ofu(see also M. Giaquinta, G. Modica, J. Souˇcek [12]). Ifu∈Hg1(Ω, S2)is smooth onΩexcept at a finite number of points inΩ, the length of a minimal connection relative todΩ connecting the singularities ofuis given by
L(u)= Min
σ∈SK
K i=1
dΩ(Pi, Nσ (i)),
where(P1, . . . , PK)and(N1, . . . , NK)are respectively the singularities of positive and negative degree counted according to their multiplicity (since deg(g)=0, the number of positive singularities is equal to the number of negative ones) and SK denotes the set of all permutations ofK indices. For the definition ofL(u) when u is arbitrary in Hg1(Ω, S2), we refer to (1.6), (1.7) below. The notion of length of a minimal connection between singularities has its origin in [8]. We also refer to the results of J. Bourgain, H. Brezis, P. Mironescu [5] and H. Brezis, P. Mironescu, A.C. Ponce [9] for similar problems involvingS1-valued maps.
Foru∈H1(Ω, S2), the vector fieldD(u)first introduced in [8] and defined by D(u)=
u· ∂u
∂x2∧ ∂u
∂x3, u· ∂u
∂x3∧ ∂u
∂x1, u· ∂u
∂x1∧ ∂u
∂x2
(1.2) plays a crucial role. Indeed, if u is smooth except at a finite number of points(Pi, Ni)Ki=1 inΩ, then (see [8], Appendix B)
divD(u)=4π K i=1
(δPi−δNi) inD(Ω) (1.3)
and if in additionu|∂Ω=g, we have (since deg(g)=0, see [8], Section IV) L(u)=Sup
K
i=1
ζ (Pi)−ζ (Ni)
, (1.4)
where the supremum is taken over all functionsζ:Ω→Rwhich are 1-Lipschitz with respect to distancedΩ i.e.,
|ζ (x)−ζ (y)|dΩ(x, y). Note that for any real Lipschitz functionζ, K
i=1
ζ (Pi)−ζ (Ni)= 1 4π
Ω
divD(u)ζ= − 1 4π
Ω
D(u)· ∇ζ + 1 4π
∂Ω
D(u)·ν
ζ, (1.5)
whereν denotes the outward normal to ∂Ω. We recall thatD(u)·ν is equal to the 2×2 Jacobian determinant ofurestricted to∂Ω and then it only depends ong. In view of (1.4) and (1.5),L(u)has been defined in [4] for u∈Hg1(Ω, S2)by
L(u)= 1 4πSup
T (u), ζ
, ζ:Ω→R1-Lipschitz with respect todΩ
, (1.6)
whereT (u)∈D(Ω)denotes the distribution defined by its action on real Lipschitz functions through the formula:
T (u), ζ
=
Ω
D(u)· ∇ζ −
∂Ω
D(u)·ν
ζ. (1.7)
In a previous paper [13], we have studied the following variational problem: given two distinct pointsP andN inΩ,
Ew(P , N )=Inf
Ω
∇v(x)2w(x)dx, v∈E(P , N )
,
where
E(P , N )=
v∈H1(Ω, S2)∩C1 Ω\ {P , N}
, v=const on∂Ω, T (v)=4π(δP −δN)inD(Ω) . In the casew≡1, H. Brezis, J.M. Coron and E. Lieb have shown that (see [8])
E1(P , N )=8π dΩ(P , N ).
For an arbitrary functionw, we have proved (see [13]) thatEw(·,·)defines a distance function satisfying
8π λdΩ(·,·)Ew(·,·)8π ΛdΩ(·,·). (1.8)
From (1.8), we infer thatEw extends toΩ× Ω into a distance onΩ. In what follows, we set for x, y∈ Ω, dw(x, y)= 1
8πEw(x, y).
Whenwis continuous, we also have shown that the distancedwcan be characterized in the following way: for any x, y∈ Ω,
dw(x, y)=Min 1
0
w γ (t )γ (t )˙ dt,
where the minimum is taken over all Lipschitz curve γ:[0,1] → Ω verifyingγ (0)=x andγ (1)=y. For an arbitrary measurable functionw, the previous formula is meaningless sincewis not well defined on curves but a similar characterization ofdw actually holds. We refer to [13] for more details. We also recall the general result in [13]:
Theorem 1.1. Let(Pi)Ki=1and(Ni)Ki=1be two lists of points inΩand consider E (Pi, Ni)Ki=1
=
v∈H1(Ω, S2)∩C1 Ω\
(Pi, Ni)Ki=1 ,
v=const on∂ΩandT (v)=4π K i=1
δPi−δNi inD(Ω)
.
Then we have Inf
Ω
∇v(x)2w(x)dx, v∈E (Pi, Ni)Ki=1
=8π Lw,
whereLw is the length of a minimal connection relative to distancedw connecting the points(Pi)and(Ni)i.e., Lw= Min
σ∈SK
K i=1
dw(Pi, Nσ (i)).
By analogy with the casew≡1, we define foru∈Hg1(Ω, S2), Lw(u)= 1
4π Sup T (u), ζ
, ζ:Ω→R1-Lipschitz with respect todw
(note that any real functionζ which is 1-Lipschitz with respect todw, is a Lipschitz function with respect todΩ sincedw is strongly equivalent todΩ and thenT (u), ζ is well defined). Whenu is smooth except at a finite number of points(Pi, Ni)Ki=1inΩ, it follows as in [8] thatLw(u)is equal to the length of a minimal connection relative to distancedwconnecting the points(Pi)and(Ni). Our main result is the following.
Theorem 1.2. For anyu∈Hg1(Ω, S2), we have Ew(u)=
Ω
∇u(x)2w(x)dx+8π Lw(u).
The proof of Theorem 1.2 is presented in Section 3 and is based on a method similar to the one used in [4]
and on a Dipole Removing Technique exposed in the next section. This technique is mostly inspired from [1] but involves some tools developed in [13] in order to treat the problem for a non smooth functionw.
In Section 4, we prove a stability property of Ew. More precisely, we give some conditions on a sequence (wn)n∈Nunder which one can conclude that the sequence of functionals(Ewn)n∈Nconverges pointwise toEw on Hg1(Ω, S2). The results are obtained using previous ones in [13]. In Section 5, we present similar results for a relaxed type functional in which we do not prescribed any boundary data.
Throughout the paper, a sequence of smooth mollifiers means any sequence(ρn)n∈Nsatisfying ρn∈C∞(R3,R), Suppρn⊂B1/n,
R3
ρn=1, ρn0 onR3.
2. The dipole removing technique
In this section, we first give a technical result which will be used for the dipole removing technique in Section 2.2.
2.1. Preliminaries
Letαandβ be two distinct points inΩ. We denote bypα,β(ξ )the projection ofξ∈R3 on the straight line passing byαandβ andrα,β(ξ )=dist(x,[α, β]),where “dist” denotes the Euclidean distance inR3. Form∈N∗, we set
aα,βm =|α−β|
m and sjα,β=j amα,β forj=0, . . . , m.
Forξ∈R3such thatpα,β(ξ )∈ [α, β], we define hα,βm (ξ )= min
0jm
pα,β(ξ )−α−sjα,β, and we set
Θm [α, β]
=
ξ∈R3, pα,β(ξ )∈ [α, β]andrα,β(ξ )amα,βhα,βm (ξ ) .
For two points x and y in Ω, we consider the class Q(x, y) of all finite collections of segments F = ([αk, βk])n(k=F1) such that βk=αk+1,α1=x,βn(F)=y,[αk, βk] ⊂Ω andαk =βk. We define the “length” of an elementF∈Q(x, y)by
¯w(F)=
n(F) k=1
lim inf
m→+∞
1 π
Θm([αk,βk])∩Ω
εαm
k,βk(ξ )w(ξ )dξ with
εαm
k,βk(ξ )= (hαmk,βk(ξ ))2(amαk,βk)4 ((hαmk,βk(ξ ))2(amαk,βk)4+rα2
k,βk(ξ ))2 .
Lemma 2.1. LetPbe a finite collection of distinct points inΩorP= ∅. For any distinct pointsx0, y0inΩ\Pand δ >0, there existsFδ=([α1, β1], . . . ,[αn, βn])∈Q(x0, y0)such that(P∪ {y0})∩(n−1
k=1[αk, βk] ∪ [αn, βn[)= ∅ and
¯w(F)dw(x0, y0)+δ.
Proof. Step 1. Assume thatwis smooth onΩ. We are going to prove that for every elementF=([α1, β1], . . . , [αn, βn])∈Q(x, y), we have
¯w(F)=
n k=1[αk,βk]
w(s)ds.
It suffices to prove that for any distinct pointsα, β∈Ω,
m→+∞lim 1 π
Θm([α,β])∩Ω
εmk(ξ )w(ξ )dξ=
[α,β]
w(s)ds. (2.1)
Without loss of generality, we may assume that[α, β] = {(0,0)} × [0, R]and we drop the indices αandβ for simplicity. We set forj=0, . . . , m−1,
Cmj+=
ξ=(ξ1, ξ2, ξ3)∈Θm [α, β] , ξ3∈
sj, sj+am 2
, and forj =1, . . . , m,
Cmj−=
ξ=(ξ1, ξ2, ξ3)∈Θm [α, β] , ξ3∈
sj−am
2 , sj
.
Forξ∈Cmj+∪Cmj−, we havehm(ξ )= |ξ3−sj|and we get that formlarge enough,
Θm([α,β])∩Ω
εmk(ξ )w(ξ )dξ=
m−1 j=0
Imj++ m j=1
Imj− (2.2)
with Imj+=
Cmj+
|ξ3−sj|2a4mw(ξ )
(|ξ3−sj|2a4m+r2(ξ ))2dξ forj=0, . . . , m−1,
Imj−=
Cmj−
|ξ3−sj|2a4mw(ξ )
(|ξ3−sj|2a4m+r2(ξ ))2dξ forj=1, . . . , m.
Using the change of variablez1=|ξ3ξ−1sj|,z2=|ξ3ξ−2sj| andz3=ξ3, we derive that
Imj+=
sj+am/2 sj Bam(0)
am4w(|z3−sj|z1,|z3−sj|z2, z3) (am4 +z21+z22)2 dz1dz2
dz3
=
sj+am/2 sj
w(0,0, z3)+O(am)
Bam(0)
a4m
(a4m+z12+z22)2dz1dz2
dz3
=π
sj+am/2 sj
w(0,0, z3)dz3+O(a2m).
By similar computations we get that Imj−=π
sj
sj−am/2
w(0,0, z3)dz3+O(a2m).
Combining this equalities with (2.2), we obtain that
Θm([α,β])∩Ω
εmk(ξ )w(ξ )dξ =π R 0
w(0,0, z3)dz3+O(am) which ends the proof of (2.1).
Step 2. We fix two distinct pointsx0, y0∈Ω\P. For any pointsx, yinΩ\(P∪ {y0}), letQ(x, y)be the class of elementsF=([α1, β1], . . . ,[αn, βn])∈Q(x, y)such that
n k=1
[αk, βk] ⊂Ω\ P∪ {y0} .
We consider the functionDw:Ω\(P∪ {y0})×Ω\(P∪ {y0})→R+defined by Dw(x, y)= Inf
F∈Q(x,y)
(¯F).
We are going to show thatDwdefines a distance function which can be extended toΩ× Ω. Letx, y∈Ω\(P∪{y0}) and letF=([α1, β1], . . . ,[αn, βn])be an element ofQ(x, y). Assumption (1.1) and similar computations to those in Step 1 lead to
λ n k=1
|αk−βk|¯w(F)Λ n k=1
|αk−βk|.
Taking the infimum over allF∈Q(x, y), we infer that
λ dΩ(x, y)Dw(x, y)Λ dΩ(x, y). (2.3)
From (2.3), we deduce that Dw(x, y)=0 if and only if x =y. Let us now prove that Dw is symmetric. Let x, y∈Ω\(P∪ {y0})andδ >0 arbitrary small. By definition, we can findFδ=([α1, β2], . . . ,[αn, βn])inQ(x, y) satisfying
¯w(Fδ)Dw(x, y)+δ.
Then forFδ=([βn, αn], . . . ,[β1, α1])∈Q(y, x), we have Dw(y, x)¯w(Fδ)= ¯w(Fδ)Dw(x, y)+δ.
Sinceδis arbitrary, we obtainDw(y, x)Dw(x, y)and we conclude thatDw(y, x)=Dw(x, y)inverting the roles ofxandy. The triangle inequality is immediate since the juxtaposition ofF1∈Q(x, z)withF2∈Q(z, y)is an element ofQ(x, y). HenceDwdefines a distance onΩ\(P∪ {y0})verifying (2.3). Therefore distanceDwextends uniquely toΩ× Ω into a distance function that we still denote byDw. By continuity,Dw satisfies (2.3) for any x, y∈ Ω.
Step 3. We consider the functionζ:Ω→Rdefined by ζ (x)=Dw(x, x0).
Note that functionζ is 1-Lipschitz with respect to distance Dw and thereforeΛ-Lipschitz with respect to the Euclidean geodesic distance onΩby (2.3). We fix an arbitrary pointz0∈Ω\(P∪ {y0})and someR >0 such that B3R(z0)⊂Ω\(P∪ {y0}). Let(ρn)n∈Nbe a sequence of smooth mollifiers. Forn >1/R, we consider the smooth functionζn=ρn∗ζ :BR(z0)→R.We write
ζn(x)=
B1/n
ρn(−z)ζ (x+z)dz
and therefore for allx, y∈BR(z0), ζn(x)−ζn(y)
B1/n
ρn(−z)ζ (x+z)−ζ (y+z)dz
B1/n
ρn(−z)Dw(x+z, y+z)dz
B1/n
ρn(−z)¯w [x+z, y+z] dz.
We remark thatΘm([x+z, y+z])=z+Θm([x, y]). Formlarge enoughz+Θm([x, y])⊂B3R(z0)and then for any vectorξ∈Θm([x, y]), we haveεmx+z,y+z(ξ+z)=εx,ym (ξ ). Hence we obtain for allz∈B1/n(0),
¯w [x+z, y+z]
=lim inf
m→+∞
1 π
Θm([x,y])
εx,ym (ξ )w(ξ+z)dξ.
Using Fatou’s lemma, we get that ζn(x)−ζn(y)
B1/n
ρn(−z)
lim inf
m→+∞
1 π
Θm([x,y])
εmx,y(ξ )w(ξ+z)dξ
dz
lim inf
m→+∞
1 π
B1/n
Θm([x,y])
ρn(−z)εmx,y(ξ )w(ξ+z)dξdz.
For eachm∈Nsufficiently large we have 1
π
B1/n
Θm([x,y])
ρn(−z)εmx,y(ξ )w(ξ+z)dξdz= 1 π
Θm([x,y])
εmx,y(ξ )ρn∗w(ξ )dξ, and sinceρn∗wis smooth, we obtain as in Step 1,
1 π
Θm([x,y])
εx,ym (ξ )ρn∗w(ξ )dξ →
[x,y]
ρn∗w(s)ds asm→ +∞. Thus for eachx, y∈BR(z0)we have
ζn(x)−ζn(y)
[x,y]
ρn∗w(s)ds.
Then forx∈BR(z0),h∈S2fixed andδ >0 small, we infer that
|ζn(x+δh)−ζn(x)|
δ 1
δ
[x,x+δh]
ρn∗w(s)ds −→
δ→0+
ρn∗w(x)
and we conclude, letting δ→0, that |∇ζn(x)·h|ρn∗w(x)for each x∈BR(z0)andh∈S2 which implies that|∇ζn|ρn∗wonBR(z0). Since∇ζn→ ∇ζ andρn∗w→wa.e. onBR(z0)asn→ +∞, we deduce that
|∇ζ|wa.e. onBR(z0). Sincez0is arbitrary inΩ\(P∪ {y0}), we derive
|∇ζ|w a.e. onΩ.
By Proposition 2.3. in [13], it follows that|ζ (x)−ζ (y)|dw(x, y)for anyx, y∈ Ω and in particular, we obtain choosingy=x0,
Dw(x, x0)dw(x, x0) for allx∈ Ω.
Step 4. End of the Proof. Letδ >0 be given. We choose somey˜0∈Ω\(P∪ {y0})such that[ ˜y0, y0] ⊂Ω\Pand
| ˜y0−y0|3Λδ . By the previous step, we can find an elementF=([α1, β1], . . . ,[αn, βn])∈Q(x0,y˜0)verifying ¯w(F)dw(x0,y˜0)+δ
3.
Then we considerF=([α1, β1], . . . ,[αn, βn],[ ˜y0, y0])∈Q(x0, y0). We have ¯w(F)¯w(F)+Λ| ˜y0−y0|dw(x0,y˜0)+2δ
3 dw(x0, y0)+dw(y0,y˜0)+2δ
3 dw(x0, y0)+δ and thenFsatisfies the requirement. 2
2.2. The dipole removing technique
We first present the dipole removing technique for a simple dipole. We then treat the case of several point singularities.
Lemma 2.2. Let P and N be two distinct points in Ω and consider u∈H1(Ω, S2)∩C1(Ω\ {P , N}) with deg(u, P )= +1 and deg(u, N )= −1. Let F =([α1, β1], . . . ,[αn, βn]) be an element of Q(P , N ) such that N /∈n−1
k=1[αk, βk] ∪ [αn, βn[. Then for anyδ >0 small enough, there exists a mapuδ∈C1(Ω, S 2)such that:
Ω
∇uδ(x)2w(x)dx
Ω
∇u(x)2w(x)dx+8π¯w(F)+δ anduδcoincides withuoutside aδ-neighborhood ofn
k=1[αk, βk]included inΩ.
Proof. Let F=([α1, β1], . . . ,[αn, βn])∈Q(P , N ) such thatN /∈n−1
k=1[αk, βk] ∪ [αn, βn[and fix some δ >0 small. We proceed in several steps.
Step 1. We consider a small 0< r0< δ verifyingBr0(α1)⊂Ω \ {N}. By Lemma A.1 in [1], we can findv∈ C1(Ω\ {α1, N}, S2)∩H1(Ω)(recall thatα1=P) satisfying
v(x)=
u(x) onΩ\Br0(α1), R
x−α1
|x−α1|
onBr0(α1), (2.4)
for some rotationRand
Ω
∇v(x)2w(x)dx
Ω
∇u(x)2w(x)dx+δ. (2.5)
LetW= {x∈R3,dist(x,[α1, β1]) < δ}. Forδ small enough, we haveW⊂Ω\ {N}. We setd= |α1−β1|. We choose normal coordinates such that α1=(0,0,0)and β1=(0,0, d). Let 0< r < r20. Since v is smooth on W\Br0(α1), we can find a constantσ (r)such that|∇v|σ (r)onW\Br0(α1). Form∈N∗, we consider
Km=
−amα1,β1 2 ,aαm1,β1
2 2
×
−amα1,β1
2 , d+aαm1,β1 2
.
Formlarge enough, we haveΘm([α1, β1])⊂Km⊂W. As in [1], we are going to construct in the next step a map v1∈C1(W\ {β1}, S2)∩H1(W )verifyingv1=vin a neighborhood of∂W and deg(v1, β1)= +1. For simplicity, we drop the indicesα1andβ1.
Step 2. We divideKminm+1 cubesQjmdefined by Qjm=
−am 2 ,am
2 2
×
j−1 2
am,
j+1 2
am
forj=0, . . . , m.
Arguing as in [1], we infer from (2.4) that m
j=0
∂Qjm
|∇v|2C r
am +mσ (r)2am2
. (2.6)
We are going to make use of a mapωm:Ba2
m(0)⊂R2→S2defined by ωm(x1, x2)= 2am2
am4 +x12+x22(x1, x2,−a2m)+(0,0,1)
(ωm was first introduced in [7] and we refer to the proof of Lemma 2 in [7] for its main properties). For j = 1, . . . , m, we choose an orthonormal direct basis(ej1, ej2, ej3)ofR3such that
v 0,0, (j−1/2)am
=(0,0,1) in the basis(ej1, ej2, ej3), and we define the mapvm1 :m
j=0∂Qjm→S2by (1) for(x1, x2, x3)∈(m
j=0∂Qjm)\(m
j=1B2
a2m(0)× {(j−1/2)am}), v1m(x1, x2, x3)=v(x1, x2, x3),
(2) forj=1, . . . , mand(x1, x2, x3)∈B2
a2m/2(0)× {(j−1/2)am}, v1m(x1, x2, x3)=ωm
2x1 am,2x2
am
in the basis(ej1, e2j, e3j),
(3) forj =1, . . . , m, for(x1, x2, x3)∈(B2
am2(0)\B2
a2m/2(0))× {(j−1/2)am}and using cylindrical coordinates (x1, x2, x3)=(ρcosθ, ρsinθ, z),
v1m(x1, x2, x3)= A1ρ+B1, A2ρ+B2,
1−(A1ρ+B1)2−(A2ρ+B2)2
in the basis(ej1, ej2, ej3), whereA1, A2, B1, B2are determined to makev1mcontinuous. More precisely, if we write v=v1e1j+v2e2j+v3e3jthen
a2mAi(θ )+Bi(θ )=vi a2mcosθ, a2msinθ, (j−1/2)am
, i=1,2, am2
2 A1(θ )+B1(θ )= 2a3m
a4m+am2 cosθ, am2
2 A2(θ )+B2(θ )= 2a3m a4m+am2 sinθ.
The mapv1msatisfies by constructionvm1 =von∂Km. Moreover, it follows exactly as in the proof of Lemma 2 in [1] that deg(vm1, ∂Qjm)=0 forj=0, . . . , m−1 and deg(vm1, ∂Qmm)= +1. Then we extendvm1 on each cubeQjm by setting
v1m(x)=v1m
am(x−bj) 2x−bj∞+bj
onQjmforj=0, . . . , m,
wherebj =(0,0, sj)is the barycenter ofQjmandx−bj∞=max(|x1|,|x2|,|x3−sj|). We easily check that vm1 ∈H1(Km, S2),v1m=von∂Km,v1mis continuous except at the pointsbj and Lipschitz continuous outside any small neighborhood of the pointsbj. We also get that
deg(v1m, bm)= +1 and deg(v1m, bj)=0 forj=0, . . . , m−1. (2.7) We remark that if we set
Dmj =Ba22
m/2(0)×
(j−1/2)am
∪Ba22
m/2(0)×
(j+1/2)am
forj=1, . . . , m−1, Dm0 =Ba22
m/2(0)× {1/2am} and Dmm=Ba22
m/2(0)×
(m−1/2)am , then we have
m j=0
x∈Qjm, am(x−bj)
2x−bj∞+bj∈Dmj ifx=bjorx=bj otherwise
=Θm [α1, β1]
and ifx∈Qjm∩Θm([α1, β1])for somej ∈ {0, . . . , m}then hm(x)= |x3−sj| = x−bj∞ and r(x)=
x12+x22. (2.8)
Some classical computations (see [1] and [7]) lead to, forj=0, . . . , m,
(∂Qjm)\Dmj
|∇vm1|2
∂Qjm
|∇v|2+O(am2)
and therefore
Qjm\Θm([α1,β1])
∇vm1(x)2w(x)dxC1Λam
∂Qjm
|∇v|2+C2Λa3m.