2003 Éditions scientifiques et médicales Elsevier SAS. All rights reserved 10.1016/S0294-1449(02)00021-5/FLA
VORTEX PINNING WITH BOUNDED FIELDS FOR THE GINZBURG–LANDAU EQUATION
Nelly ANDREa, Patricia BAUMANb,1, Dan PHILLIPSb,∗,2
aUniversité François Rabelais, département de mathématiques, Parc Grandmont, 37200 Tours, France bDepartment of Mathematics, Purdue University, West Lafayette, IN 47907, USA
Received 31 January 2002
ABSTRACT. – We investigate vortex pinning in solutions to the Ginzburg–Landau equation.
The coefficient,a(x), in the Ginzburg–Landau free energy modeling non-uniform superconduc- tivity is nonnegative and is allowed to vanish at a finite number of points. For a sufficiently large applied magnetic field and for all sufficiently large values of the Ginzburg–Landau parameter κ=1/ε, we show that minimizers have nontrivial vortex structures. We also show the existence of local minimizers exhibiting arbitrary vortex patterns, pinned near the zeros ofa(x).
2003 Éditions scientifiques et médicales Elsevier SAS
RÉSUMÉ. – On étudie la localisation des vortex des solutions de l’équation de Ginzburg–
Landau. Dans l’énergie libre de Ginzburg–Landau, le coefficienta(x)modélise la supraconduc- tivité non uniforme. Ce coefficient est positif et s’annule en un nombre fini de points. On montre que, pour un champ magnétique assez grand et pour toutes les valeurs du paramètre de Ginzbug–
Landauκ=1/εassez grandes, les minimiseurs présentent des structures de vortex non triviales.
On montre aussi l’existence de minimiseurs locaux présentant une structure prescrite de vortex situés au voisinage des zéros dea(x).
2003 Éditions scientifiques et médicales Elsevier SAS
Introduction
In this paper we analyze several aspects of vortex pinning in superconductivity using the Ginzburg–Landau theory as our model. To describe these phenomena consider the energy
Jε(ψ, A)=
(∇ −iA)ψ2+ |∇ ×A−hee3|2+ 1 2ε2
a− |ψ|22
(1)
∗Corresponding author.
E-mail address: phillips@math.purdue.edu (D. Phillips).
1Partially supported by the National Science Foundation grant DMS-9971974.
2Partially supported by the National Science Foundation grant DMS-9971713.
for ε >0. Here is a bounded simply connected domain in R2with a smooth (C2,1) boundary and a:→R. The domain represents the cross-section of an infinite cylindrical body with e3as its generator. The body is subjected to an applied magnetic field,hee3wherehe0 is constant. The functionA:→R2is the magnetic potential and∇ ×A= ∇ ×(A1, A2,0)is the induced magnetic field in the cylinder. The function ψ is complex-valued where |ψ|2 =ψ∗ψ represents the density of superconducting election pairs and
j= −i
2(ψ∗∇ψ−ψ∇ψ∗)− |ψ|2A (2) denotes the superconducting current density circulating in the cross-section . The parameter ε =1/κ is a positive number where κ is the Ginzburg–Landau parameter associated to the material. We analyze the smallε(largeκ) regime. It is here that vortex dominated current patterns are expected in stable equilibria for Jε. The prototypical picture of this phenomenon is that of a finite number of non-superconducting points in (at whichψ =0, called vortices), each of which is surrounded by a ring of the super currentj.
If the material is homogeneous, the function a in Jε is taken to be a constant, proportional toTc−T. HereT is the body’s temperature andTc is the material’s critical temperature. For T Tc (a0), it is easy to show that the only equilibria for Jε are completely non-superconducting and have ψ≡0, ∇ ×A≡hee3. ForT < Tc (a >0), superconducting minimizers exist if the applied field strength he is not too large. There are a number of mathematical investigations of the relationship between he and the nature of stable superconducting states for this case. In [11] Sandier and Serfaty showed that there exists a constantHc1 proportional to|log(ε)|asε→0, such that ifheHc1, then minimizers forJεare purely superconducting, satisfying|ψ|>0 in. In [12] they showed that forhe slightly greater thanHc1 and such thatheε−2, minimizers are in a mixed state having a vortex-like structure. It was shown by Giorgi and Phillips in [5]
that forheCε−2for some constantC, superconductivity is completely suppressed, in that all equilibria forJεhaveψ≡0.
Inhomogeneous superconducting materials can arise naturally due to material defects or the presence of grain boundaries. Inhomogeneities can be inserted intentionally, as well, by adding non-superconducting (normal) impurities to the material. (See [3]
and [4].) A consequence of having material inhomogeneities is that they tend to pin or stabilize supercurrent patterns. The classical Ginzburg–Landau theory can be modified to take normal inclusions into account. This is done by having the critical temperature, Tc, depend on position which is equivalent to havinga=a(x). (See [10].) It is possible thata(x)may vanish or change sign within the domain.
A mathematical study for the Ginzburg–Landau equations corresponding to the energy (1) with variablea(x)was done by Aftalion, Sandier, and Serfaty in [1] where the case
1
2 a(x)1 was considered. They proved among other things, that Hc1 remains of order |log(ε)| asε→0. In this paper we consider the case where contains a finite number of point impurities,{x1, . . . , xn}, and thata(x)vanishes at these normal sites. In this instance, the strong pinning enables us to show that the transition threshold for he, denoted byHc1=Hc1(ε), separating the presence or absence of vortices, is of order 1 as ε→0. (See Corollary 4.4.) In addition, for eachheand allεsufficiently small, we show
that there are local minimizers forJε with prescribed vortex structure about each of the xi corresponding to the homotopy classes in\ {x1, . . . , xn}. (See Theorem 4.6.) In this way we are able to pin supercurrent patterns near the zeros ofa(x). (See Theorem 4.6.) Another way of introducing inhomogeneities is by making holes (voids) in the body.
In [8,9,13]Jε was studied witha=1,he=0 but withmultiply connected by Jimbo and Morita, Jimbo and Zhai, and Rubinstein and Sternberg, respectively. In that setting, local minimizers with prescribed vortex structures associated to the homotopy classes ofwere shown to exist.
We require thata(x)satisfy the following.
Assume:a∈C1(\{x1, . . . , xn})∩Cβ() for someβ >0, √
a∈H1(), a(x)0 for allxin, anda(x)=0 iffx∈ {x1, . . . , xn}wherex1, . . . , xnare distinct points in and n∈N. Moreover, assume that there are positive constants mi, Mi and αi so that mi|x−xi|αia(x)Mi|x−xi|αi in some neighborhoodUi ofxi for 1in.
DEFINITION. – Let ε >0 and let (ψε, Aε)∈H1(;C)×H1(;R2) ≡M. Then (ψε, Aε)is an equilibrium forJε if and only if(ψε, Aε)is a weak solution of the Euler–
Lagrange equations and natural boundary conditions for critical points of Jε in M, namely:
−(∇ −iAε)2ψε= 1 ε2
a− |ψε|2ψε in, (∇ −iAε)ψε·n=0 on∂,
(3) and
∇ × ∇ ×Aε= −i 2
ψε∗∇ψε−ψε∇ψε∗− |ψε|2Aε≡jε in,
∇ ×Aε=hee3 on∂.
(4) Forε=0 we set
J0(ψ, A)=
(∇ −iA)ψ2+ |∇ ×A−hee3|2. (5)
Denote
Ha1= ψ∈H1(;C)such that|ψ| =√
aalmost everywhere. Note that Ha1 is nonempty, since √
a ∈ Ha1 by our assumptions on a. We prove in Section 1 (see Theorem 1.4) that each ψ ∈ Ha1 can be written as ψ = √
aeiθ (x), where θ (x)=θ0(x)+ni=1diθi(x), θ0 is a measurable function determined up to an additive constant, 2π k fork∈Z, satisfyinga|∇θ0|2<∞,D=(d1, . . . , dn)∈Zn is uniquely determined, and θi(x) is the azimuthal angle about xi for 1i n(so that (cosθi(x),sinθi(x))=(x−xi)/|x−xi|for allx=xi inR2). Thusψcorresponds to a uniqueD∈Zndescribing a homotopy class forψ in\{x1, . . . , xn}. We write
Ha1=
D∈Zn
Ha,D1 .
We note that Ha,D1 is both open and closed in Ha1 and that if {un} ⊂Ha,D1 such that un% uinH1thenu∈Ha,D1 . (See Theorem 1.5.)
DEFINITION. – Let(ψ0, A0)∈Ha1×H1(;R2)≡M0. Then(ψ0, A0)is an equilib- rium forJ0if and only if(ψ0, A0)is a weak solution of the Euler–Lagrange equations and natural boundary conditions for critical points ofJ0inM0, namely:
div
−i 2
ψ0∗∇ψ0−ψ0∇ψ0∗− |ψ0|2A0
=0 in,
−i 2
ψ0∗∇ψ0−ψ0∇ψ0∗− |ψ0|2A0
·%n=0 on∂,
(6)
and
∇ × ∇ ×A0=
−i 2
ψ0∗∇ψ0−ψ0∇ψ0∗− |ψ0|2A0
≡j0 in,
∇ ×A0=hee3 on∂.
(7) The functionalsJε, forε0, are gauge invariant. By this we mean that if(ψ, A)∈ M (M0) and if φ ∈ H2(), then the gauge transformation, (ψ, A) =Gφ(ψ, A) defined by
ψ≡ψeiφ, A≡A+ ∇φ,
satisfies(ψ, A)∈M(M0),Jε(ψ, A)=Jε(ψ, A), and(ψ, A)is an equilibrium for Jε(J0)if (ψ, A)is one. In this paper we will fix a gauge by requiring (without loss of generality) thatAsatisfy
divA=0 in, A·n=0 on∂,
(8) since this can be accomplished by an appropriate gauge transformation. With this choice of gauge (the Coulomb gauge), A is determined from the value of ∇ ×A= (∂xA1−∂yA2)e3≡he3by first solving
)ξ=h in, ξ=0 on∂.
(9) From (8), (9), and the fact that is simply connected we have A= ∇⊥ξ where (∂x, ∂y)⊥≡(−∂y, ∂x). An important feature of the gauge choice (8) is that the boundary conditions in (3) and (6) can be replaced by
∇ψ· n=0 on∂
and, since∇ × ∇ ×A= −)A+ ∇(divA), the term∇ × ∇ ×Ain Eqs. (4) and (7) is equal to−)A.
We establish the following main results in this paper.
THEOREM 1. – Fixhe0. For eachD∈Zn,J0has an equilibrium (with our choice of gauge),(ψD, AD), inHa,D1 ×H1(;R2). Moreover,(ψD, AD)is unique up to uniform rotations ofψD in,ψD→ψDeicforc∈R. (See Theorem 3.2.)
We remark that(ψ, A)→(ψeic, A)is a gauge transformation inM(M0), and thus Jε(ψ, A)=Jε(ψeic, A)for allc∈Randε0.
THEOREM 2. – Fixhe0. Let(ψεk, Aεk)be an equilibrium forJεk fork=1,2, . . . such thatεk→0+ and
lim inf
k→∞ Jεk(ψεk, Aεk)c <∞. (10) There exists a finite subset D = D(c, he) of Zn, a subsequence {εk+}, and (ψ0, A0)∈ Ha,D1 ×H1(;R2)for someD∈Dsuch that
(ψεk+, Aεk+) %ψ0, A0 inM.
Moreover(ψ0, A0)is an equilibrium forJ0. (See Theorem 4.1.) Note that
Jε(√
a )= ||h2e+
|∇√
a|2 forε0. (11)
Thus, givenhe, it follows from Theorem 2 that a sequence of minimizers withεk→0+ will satisfy (10).
THEOREM 3. – Fixhe0. Let(ψεk, Aεk)be a minimizer ofJεk inMfork=1,2, . . . with εk→0+. Then a subsequence (ψεk+, Aεk+)→(ψD, AD) in M, where (ψD, AD) is a minimizer of J0 in M0 and (ψD, AD)∈Ha,D1 ×H1(;R2). Moreover, if R >0 and BR(xi) are disjoint subsets offor i=1, . . . , n, then for all +sufficiently large,
|ψεk+|>0 outsideni=1BR(xi)and the degree ofψεk+ inBR(xi)isdi for alli∈ {1, . . . , n} whereD=(d1, . . . , dn). (See Theorem 4.2.)
We prove in Corollary 3.6 that for he 0 fixed, the set of all D in Zn such that Ha,D1 ×H1(;R2)contains a minimizer ofJ0inM0is a nonempty finite set (depending only on,a(x), andhe), which we denote byD0=D0(he).
THEOREM 4. – Let(ψε, Aε) be a minimizer ofJε for each ε >0. Fix R >0 as in Theorem 3 and he0. There exists ε0=ε0(R, he) >0 such that for all 0< ε < ε0,
|ψε|>0 outsideni=1BR(xi)and the degree of ψε inBR(xi) fori=1, . . . , n, denoted by Dε=(d1,ε, . . . , dn,ε), is inD0. Moreover, there exists he>0 (depending only on anda(x)) such that ifhe> heand 0< ε < ε0(R, he), thenDε= 0. (See Theorem 4.3.)
We remark that Theorem 4 implies that{Hc1(ε)}is uniformly bounded inεasε→0+. (See Corollary 4.4.)
The equilibrium found in Theorem 1 is (by uniqueness) the minimizer for J0 in Ha,D1 ×H1(;R2). Since Ha,D1 is open in Ha1, it is also a local minimizer for J0 in M0. Given he0, let (ψD, AD)be such a solution. For local minimizers ofJε inM, we have (in contrast to Theorem 4) that all degrees inZnnearx1, . . . , xnare attainable:
THEOREM 5. – Fixhe0 and anyD inZn. For eachε >0 sufficiently small, there exists a local minimizer, (ψε, Aε), ofJεinMsuch that(ψε, Aε)→(ψD, AD)inMas ε→0. In addition, for anyR >0 as in Theorem 3, there existsε1(R, he) >0 such that
|ψε|>0 outsideni=1BR(xi), and the degree ofψε inBR(xi)isdi for allε < ε1, where D =(d1, . . . , dn). (See Theorem 4.6.)
1. Preliminaries
It is well known that if(ψ, A)∈Mandψ=ρeiθ, then∇θ is uniquely determined almost everywhere in{ρ >0},ρ∈W1,2(),ρ∇θ∈L2(;R2),
(∇ −iA)ψ2= |∇ρ|2+ρ(∇θ−A)2 and j≡ −i
2(ψ∗∇ψ−ψ∇ψ∗)− |ψ|2A=ρ2(∇θ−A)a.e. in.
(12) If(ψε, Aε)∈Mand(ψε, Aε)is an equilibrium forJε withε >0, then from (3) we can derive the equations
−div(ρε∇ρε)+ |∇ρε|2+|jε|2 ρε2 = 1
ε2
a−ρε2ρε2 in,
ρε∇ρε·%n=0 on∂, divjε=0 in, and
jε·%n=0 on∂,
(13)
where ψε =ρεeiθε and jε =ρε2(∇θε −Aε). These equations are obtained by using test functions of the form ϕ =ψε∗φ in the formulation (3) such that φ ∈L∞() and (1+ |ψε|)|∇φ| ∈L2(). Moreover, if we definehε by∇ ×Aε=hεe3 then (4) can be rewritten as
−∇⊥hε≡(∂y,−∂x)hε=jε in, hε−he on∂.
(14) Similarly, if(ψ0, A0)∈M0and(ψ0, A0)is an equilibrium forJ0then (6) and (7) can be rewritten as
divj0=0 in, j0·%n=0 on∂
(15) and
−∇⊥h0=j0 in, h0=he on∂,
(16) whereψ0=ρ0eiθ0=√
aeiθ0,h0is defined by∇ ×A0=h0e3, andj0=ρ02(∇θ0−A0).
The following three results concern maximum principles and regularity for equilibria ofJε. The proofs are only a slight variation of the proofs for the case in which a≡1 in. (See [5] and [6].)
LEMMA 1.1. – If (ψε, Aε) ∈ M(M0) and (ψε, Aε) is an equilibrium for Jε(J0) whereε0, then|ψε|sup√
a.
Proof. – For(ψ0, A0)∈M0, we have|ψ0| =√
ainand hence the result is trivial in this case. Ifε >0 and(ψε, Aε)is an equilibrium forJε, the result follows by using
φε≡max 0,ψε(x)−sup
√a/ψε(x)=ρε(x)−sup
√a+/ρε(x)
as a test function in the weak formulation of the first two equations in (13), which yields
0
E
|∇ρε|2=
E
(−φε)·|jε|2 ρ2ε + 1
ε2
E
a−ρε2ρε2φε0,
whereE= {x∈: φε(x) >0}. It follows thatEhas zero measure. Thusφε0 a.e. in which proves the lemma. ✷
LEMMA 1.2. – Forε >0 equilibria are of class,C2,β()for someβ >0.
Proof. – With our choice of gauge (8), we have∇ × ∇ ×Aε= −)Aε. The system (3) and (4) is thus uniformly elliptic and regularity follows from the classical theory. (See [6].) ✷
LEMMA 1.3. – Fixhe0. Assumeε0 and(ψε, Aε)is an equilibrium forJε. SetM=max(Jε(ψε, Aε),Jε(√
a,0), maxa). Then
Aε2,2C(M, ), (17)
ψε1,2C(M, ), (18)
and ifε >0
|∇ψε|C(M, )/ε in, (19)
where C(M, ) denotes a constant depending only on M, a(x), and , and the subscript k,2 denotes the norm inWk,2().
Proof. – We argue for ε >0. The proofs of (17) and (18) for the case ε =0 are identical.
We write (using (12)) Jε(ψ, A)=
|∇|ψ||2+ |ψ|2|∇θ−A|2+ 1 2ε2
a− |ψ|22+ |∇ ×A−hee3|2
.
Recall thatjε= |ψε|2(∇θε−Aε)andhε is defined by
hεe3= ∇ ×Aε. (20)
From this and (12), we have Jε(ψε, Ae)=
|∇|ψε||2+ |ψε|−2|jε|2+ 1 2ε2
a− |ψε|22+ |hε−he|2
. (21) Thusjε22supa·Jε(ψε, Aε)C(M, ), wherejε2denotes theL2norm ofjε
in. Then from (14), we have
∇hε2C(M.).
Using this estimate together with (9) we see that∇ξ2,2C(M, ). Thus Aε2,2= ∇ξ2,2C(M, ).
Note that this implies
AεCγ()C(M, , γ ) for eachγ ∈(0,1). (22) Now
∇ψε22C(∇ −iAε)ψε2
2+ Aεψε22
. So we see
∇ψε22C(M, ).
This proves (17) and (18) forε >0 (andε=0).
To prove (19) let y =x/ε, ε =/ε, ψε(y)=ψε(εy), andA˜ε =εAε. We have from the Ginzburg–Landau equation
)yψε−2iA˜ε· ∇yψε− | ˜Aε|2ψε=a(εy)− |ψε|22ψε inε,
∂nψε=0 on∂ε.
Here we have used the choice of gauge. From (22) we see that | ˜Aε(y)| = |εAε(y)| εC(M, ). It follows from local elliptic estimates and Lemma 1.1 that ψε∈W2,p(ε) forp <∞and
|∇yψε|C(M, ) inε for 0< ε1.
(Here we use that∂is of classC2,1.) Thus|∇ψε|C(M, )/εin. ✷ The remaining results in this section are facts about
Ha1= ψ∈H1(;C): |ψ| =√
aa.e. in which are used later in this paper.
THEOREM 1.4. – Eachu∈Ha1can be written as u(x)=a(x)·
n
j=1
z−zj
|z−zj| dj
·eiϕ(x)=a(x)·eiθ (x)
wherez=z(x)=x1+ix2forx=(x1, x2)in,zj =z(xj),ϕ∈Hloc1 (\ {x1, . . . , xn}), θ (x) =ϕ(x) +nj=1djθj(x), and θj(x) is the azimuthal angle of x about xj for 1j n. Moreover, for each u∈Ha1,D≡(d1, . . . , dn)∈Zn is unique,ϕ∈Hloc1 (\ {x1, . . . , xn}) is unique up to an additive constant 2π k for k ∈ Z, and ϕ satisfies
a|∇ϕ|2C(, a, D)+|∇u|2.
Proof. – Fixu∈Ha1 and set v(x)=u(x)/√
a(x)=u(x)/|u(x)|. Then v∈Hloc1 (\ {x1, . . . , xn};S1) whereS1= {z∈C: |z| =1}). It follows from Schoen and Uhlenbeck [14] that there exists a sequence{vm}such that
vm∈C2
\ n
j=1
B1
m(xj);S1
form=1,2,3, . . .and
vm→v inHloc1 \ {x1, . . . , xn}asm→ ∞.
(See also [2].) We compute the degree of eachvmnearxj; as follows:
We say that a radiusr is admissible for a givenxj andvmifBr(xj)∩ {x1, . . . , xn} = {xj}and ∂Br(xj)⊂\ni=1B1
m(xi). For any such r, since vmis smooth and |vm| =1 in\ni=1B1
m(x1, . . . , xn), the winding number ofvmon∂Br(xj)is defined by:
dj,m= − i 2π
∂Br(xj)
vm∗(vm)τ (23)
whereτ=ν⊥=(−ν2, ν1),νis the exterior unit normal on the boundary ofBr(xj), and (vm)τ is the derivative ofvmin the directionτ. It is well known from degree theory that dj,mis integer-valued and independent ofrfor all admissiblerwith respect toxj andm.
Thus if 0< r1< r2<∞andr2satisfiesBr2(xj)⊂andBr2(xj)∩ {x1, . . . , xm} = {xj}, then for allmsufficiently large, anyr∈ [r1, r2]is admissible forxj andvm, and we may integrate (23) to obtain
dj,m= − i 2π(r2−r1)
Bra(xj)\Br1(xj)
vm∗(vm)τdx. (24)
Sincedj,mis integer-valued andvm→vinHloc1 (\ {x1, . . . , xn})asm→ ∞, it follows from (24) that dj,m is independent ofm for all msufficiently large. Thus there exists dj ∈Z such thatdj=dj,mfor allmsufficiently large and, lettingm→ ∞, we have
dj= − i 2π(r2−r1)
Br
2(xj)\Br
1(xj)
v∗vτdx. (25)
We may use this to define the degree ofvnearxj, since (25) is independent ofr2> r1>0 provided that Br2(xj)⊂ and Br2(xj)∩ {x1, . . . , xn} = {xj} and it is clear (25) is
independent of the particular converging sequence {vm}. In particular, we can define the degree of u inBr(xj)by (25), forv=u/|u|andr1< r2as above. (See also [7].)
Now consider the real two-dimensional vector field Fm= −
n
j=1
dj∇θj−iv∗m∇vm (26)
inC1(\nj=1B1
m(xj)), whereθj(x)is the (multivalued) azimuthal angle ofxaboutxj
and thus∇θj(x)is well defined in\ {xj}for 1jn. Since∇ × ∇θj=0 in\ {xj} andvm isC2with|rm|2=vmvm∗ =1 in\nj=1B1
m(xj), it follows that∇ ×Fm=0 in \nj=1B1
m(xj). Thus ifmis sufficiently large so thatdj=dj,mfor 1jn, then by Stokes’ Theorem and (23),CFm·dr=0 for any closed curve,C, in\nj=1B1
m(xj).
Moreover, there existsϕm∈C2(\nj=1B1
m(xj))such that∇ϕm=Fmform=1,2, . . .. From this and (26) we obtain
vm∇ϕm= −vm
n
j=1
dj∇θj
−i∇vm
and hence
∇vme−iϕm·e−i n
j=1djθj
=e−iϕm·e−i n
j=1djθj
×
∇vm−i
vm∇ϕm+vm
n
j=1
dj∇θj
=0.
As a result (adding a constant toϕmif necessary), we have vm(x)=eiϕm(x)·ei
n j=1djθj(x)
=eiϕm(x)· n
j=1
z−zj
|z−zj| dj
.
By (26),∇ϕm= −nj=1dj∇θj−ivm∗∇vmand sincevm→vinHloc1 (\ {x1, . . . , xn}), we have∇ϕm→ −nj=1dj∇θj−iv∗∇vand
eiϕm≡vm· n
j=1
z−zj
|z−zj| −dj
→v· n
j=1
z−zj
|z−zj| −dj
inL2loc(\ {x1, . . . , xm}). It follows that{ϕm}(after possibly subtracting constants 2π km
wherekm∈Z) converges inHloc1 (\ {x1, . . . , xn}), to someϕ∈Hloc1 (\ {x1, . . . , xn}), andu=√
av=√ aei(ϕ+
n j=1djθj)
a.e. in. Settingθ (x)=ϕ(x)+nj=1djθj(x), we have
|∇u|2= |∇√
a|2+a|∇θ|2a|∇θ|2.
Since|∇θj(x)| =c(dj)|x−xj|−1anda(x)c|x−xj|αj whereαj>0 for 1jn,
a(x)∇θj(x)2C(, a, D) <∞ (27) whereD=(d1, . . . , dn). Thus
a|∇ϕ|2
|∇u|2+C· n
j=1
a|∇θj|2
|∇u|2+C(, a, D).
Finally, to show thatD∈Znis unique andϕ∈Hloc1 (\ {x1, . . . , xn})is unique (up to an additive constant 2π l wherel∈Z) for eachu∈Ha1, assume thatD=(d˜1, . . . ,d˜n)∈ Zn and ϕ˜ ∈Hloc1 (\ {x1, . . . , xn}) such that u=√
aei[ ˜ϕ+ n
k=1d˜kθk]. Then v≡u/√ a satisfies −iv∗∇v= ∇ ˜ϕ +nk=1d˜k∇θk in \ {x1, . . . , xn}. Fixing j ∈ {1, . . . , n} and integrating overBr2(xj)\Br1(xj)for 0< r1< r2as in (25), we have
dj= − i 2π(r2−r1)
Br
2(xj)\Br
1(xj)
v∗vτdx
= 1
2π(r2−r1)
Br2(xj)\Br1(xj)
˜ ϕτ+
n
k=1
d˜k(θk)τ
dx
= 1
2π(r2−r1)·0+ ˜dj·2π(r2−r1)
= ˜dj
where τ =τ (x) = (x −xj)⊥/|x −xj| for all j ∈ {1, . . . , n}. Thus ei(ϕ− ˜ϕ) =1 in \ {x1, . . . , xn}withϕ− ˜ϕinHloc1 (\ {x1, . . . , xn})and it follows thatϕ− ˜ϕ=2π lfor somel∈Z. ✷
For eachD∈Zn; we define Ha,D1 = u∈Ha1: u=√
aei[ϕ+ n
j=1djθj] whereϕ∈Hloc1 \ {x1, . . . , xn}. By Theorem 1.4, it follows that
Ha1=
D∈Zn
Ha,D1
andHa,D1 ∩Ha,D1 = ∅forD=DinZn. We will need the following additional properties ofHa,D1 :
THEOREM 1.5. – For each D∈Zn, Ha,D1 is a nonempty, open and closed subset of Ha1. In addition,Ha,D1 is sequentially weakly closed inH1(;C), i.e. if{uk} ⊂Ha,D1 and uk→uweakly inH1(;C), thenu∈Ha,D1 .
Proof. – Our hypotheses ona inensure that√
a∈H1()and√a∇θj∈L2()for each j ∈ {1, . . . , n}(see (27)); hence √
aei[ n
j=1djθj]∈Ha,D1 and Ha,D1 = ∅. To prove thatHa,D1 is open inHa1, assume thatu0=√
aei[ϕ+ n
j=1djθj]∈Ha,D1 and let BR(u0)= u∈Ha1:u−u0H1(;C)< R
where R >0. Since u∈Ha1, there existsϕ˜ ∈Hloc1 (\ {x1, . . . , xn}) and D ∈Zn such thatu=√
aei[ ˜ϕ+ n
j=1d˜jθj]
. Setv0=u0/|u0| =u0/√
a andv=u/|u| =u/√
a. By (25), there exist positive numbersr1< r2such that for eachj ∈ {1, . . . , n},
dj= − i 2π(r2−r1)
Sj
v0∗(v0)τdx (28) and
d˜j= − i 2π(r2−r1)
Sj
v∗(v)τdx
whereSj =Br2(xj)\Br1(xj). Sincea isC1and |a|>0 on Sj for eachj ∈ {1, . . . , n}, we have
v0∗∇v0−v∗∇vL1(Sj)v∗(∇v0− ∇v)L1(Sj)+(v0∗−v∗)∇v0
L1(Sj)
C(a, r1, r2, v0)·1+ u0H1(Sj)
· u−u0H1(Sj). From this and (28), it follows that if R is sufficiently small (depending onr1,r2,,a, and u0), we havedj = ˜dj and u∈Ha,D1 . ThusBR(u0)⊂Ha,D1 forR sufficiently small and we conclude that Ha,D1 is an open subset ofHa1. Now since Ha1=D∈ZnHa,D1 and Ha,D∩Ha,D1 = ∅forD=DinZn,Ha,D1 is also a closed subset ofHa1.
Finally, to prove that Ha,D1 is weakly sequentially closed inHa1, assume that{uk} ⊂ Ha,D1 and uk →u weakly in H1(;C). By compactness, a subsequence (which we relabel as{uk}) satisfiesuk→uinL2(). Thus|u| =√
aa.e. inand henceu∈H1
a,D for someD ∈Zn. It follows from (28) (withv0replaced by uk/√
a andv replaced by u/√
aand the weak convergence ofuk touthatD=D andu∈Ha,D1 . ✷ 2. A weighted Sobolev space
Set
V ≡
g∈H1():
a−1|∇g|2<∞
. ThenV is a Hilbert space with norm
gV =
a−1|∇g|2+g2
1/2
.