Professional Documents
Culture Documents
(Modern Birkhäuser Classics) Bethuel, Fabrice - Brézis, Haim - Hélein, Frédéric - Ginzburg-Landau Vortices-Birkhäuser (2017) PDF
(Modern Birkhäuser Classics) Bethuel, Fabrice - Brézis, Haim - Hélein, Frédéric - Ginzburg-Landau Vortices-Birkhäuser (2017) PDF
Fabrice Bethuel
Haïm Brezis
Frédéric Hélein
Ginzburg-
Landau
Vortices
Modern Birkhäuser Classics
Fabrice Bethuel
Haïm Brezis
Frédéric Hélein
Frédéric Hélein
Université Paris Diderot - Paris 7
Paris, France
Ginzburg-Landau Vortices
ISBN 978-0-8176-3723-1
9876.5432
TABLE OF CONTENTS
Introduction xi
I. Energy estimates for 8 1-valued maps
1. An auxiliary linear problem 1
2. Variants of Theorem 1.1 6
3. 8 1-valued harmonic maps with prescribed isolated
singularities. The canonical harmonic map 10
4. Shrinking holes. Renormalized energy 16
II. A lower bound for the energy of 8 1-valued maps
on perforated domains 31
III. Some basic estimates for u"
1. Estimates when G = BR and g(x) = x/lx! 42
2. An upper bound for Ee (Ue) 44
3. An upper bound for -bfa(lu"l 2 - 1)2 45
4. luel ~ 1/2 on "good discs" 46
IV. Towards locating the singularities:
bad discs and good discs
1. A covering argument 48
2. Modifying the bad discs 49
V. An upper bound for the energy of u" away from
the singularities
1. A lower bound for the energy of u" near ilj 53
2. Proof of Theorem V.l 54
CONTENTS
BffiLIOGRAPHY 154
INDEX 159
Acknowledgements
The original motivation of this study comes from the following questions
that were mentioned to one of us by H. Mata.no.
Let
G = B1 = {~ = (~1.~2) E R2 ; ~~ +~~ = /z/ 2 < 1}.
Consider the Ginzburg-Landau functional
g(x) =x on OG
H: = {u
and set
E H 1 (G;C}; u= g on OG}.
It is easy to see that
(2)
(i.e., the winding number of g considered as a map from CJG into 8 1 ) plays
a crucial role in the asymptotic analysis of u~.
Cased= 0. This case is easy because n;(G; 8 1 ) ':f: 0 and thus the mini-
mization problem
(6)
makes sense. In fact, problem (6) has a. unique solution uo that is a. smooth
harmonic map from G into S 1 , i.e.,
We have proved in F. Bethuel, H . Brezis and F. Helein [2) (see also Appendix
I at the end of the book} that u~ -.... Uo in C 1 •0 (G) and in Ct,c(G) Vk; in
particular,
We have also obtained rates of convergence for Uu~- uol! in various norms.
Introduction xiii
Case d¥0. Throughout the book we assume that d > 0 since the case
d < 0 reduces to the previous case by complex conjugation. Here, the main
difficulty stems from the fact that
(8)
Indeed, suppose not, and say that n; (G; S 1) ¥ 0, then we could consider,
as above,
{9)
Min E~.
HJ(G;C)
(naturally, with fc IY'u.,l 2 = oo). If this is indeed the case then u., can be
viewed as a "generalized solution" of problem (9).
Of course, many other "penalties" can be devised. They all seem to lead
to the same class of generalized solutions. For example, one other natural
penalty consists of drilling a few little holes B (ai, p) in G and consider-
ing the domain Gp = G \ UB(ai,p). In this case there is no topological
i
obstruction and
(we do not impose a Dirichlet condition on 8B(ai, p)). Then, one may
consider the problem
and analyze what happens as p -+ 0. Here, the points (ai) are free to
move and some configurations will turn out to be "better" than others (see
Section 1.4 and Chapter VIII).
Ut:,. -+ u., inCiocCG \ L,J{ai}) 'Vk and in C 1 •Q(G \ L,J{a,}) 'Vo: < 1.
• •
In addition, each singularity has degree +1 and, more precisely, there are
complex constants (O:i) with la:i I = 1 such that
(11)
: = 2~ ,t, 6,, in G,
{
011 = 9 X 9-r on 8G,
d
R(x) = ~(x)- Llogjx- bil·
i=l
(13)
xvi Introduction
One proves (see Theorem 1.2} that there exists a unique minimizer up for
the problem
{14)
(18)
(19)
~r.p=O inG
(20) {
r.p = tpo on8G
{21)
i'Po(z) _ ( ) lz- bd lz- ~~ Jz- bdj
e - g z (z- bl) (z - ~) · · · (z- bd) ·
Introduction xvii
(Note that the right-hand side in (21) is a map from 8G into 8 1 of degree
zero so that v>o is well defined as a single--valued smooth function.)
For a general configuration b estimate (18) cannot be improved.
However, for the special configuration as described in Theorem 0.1 we
have the better estimate (11). That property, which may be written as
(22)
or -Au = uP + cU in nc Rn
is u(x) = xflxl. More generally, F.H. Lin [1) has obtained the same con-
clusion for the problem
Next, we study the zeroes of u,.. Let us recall some earlier works on
that question. It has been proved by C. Elliott, H. Matano and T. Qi [1]
that (for every e > 0) the zeroes of any minimizer u" of (2) are isolated.
P. Bauman, N. Carlson and D. Phillips [1] have shown, in particular, that
if G = B 1 and deg(g, 8G) = 1 with g(O) strictly increasing then (for every
E > 0) there is a unique zero of any minimizer u" of (2).
Moreover, there exist integers d1 , d2 , ••• , d~: E Z \ {0} and a smooth har-
monic function cp : G -+ R such that
(z- a~:)d*
lz- a~:ld"'.
In addition, we have
which expresses that (a;, dj) is a critical point of some appropriate renor-
malized eneryy W.
e = 0) has no physical meaning. Note that ~(T) plays the role of a charac-
teristic length: the values of l'I/J(x)j may vary significantly at two points
x11 x2 whose distance !x1 - x2l is of the order of ~(T).
[Warning: Instead of equation (3), i.e.,
is the so-called mixed state (or vortex state) characterized by the coex-
istence of two phases: near the cores of the vortices 1¢1 ~ 0, i.e., normal
state; away from the cores of the vortices 1¢1 ~ 1, i.e., superconducting
state. Again, vortices have a core radius of order e.
As H increases the
vortices arrange themselves in a regular pattern, the Abrikosov lattice
{predicted on a theoretical basis by A. Abrikosov [1]).
In the cross-section picture '¢ seems to have basically the same behavior
as our uc fore small with vortices located around (and close to) the zeroes
of uc (or in theE - 0 limit at the singularities of u.). The conclusion of
Theorem 0.1 is consistent with the observation that all vortices have the
same circulation +1 (in our language degree +I). Moreover, our analysis
yields cores of radius e that correspond to the physical cores of size e.
Remark 0.3. The conclusion of Theorem 0.1 bears some resemblance
to earlier results motivated by the theory of nematic liquid crystals (see H.
Brezis [1], [2], H. Brezis, J. M. Coron and E. Lieb [1], R. Hardt and F. H. Lin
[1], D. Kinderlehrer [1]). Let G C R3 be a smooth bounded domain and let
g: EJG- S2 be a (smooth) boundary condition. We now consider maps
u: G- R 3 (not JR2 ); in the theory of liquid crystals u corresponds to the
"director", which is an order parameter describing the orientation of the
optical axis of the (rod-like) molecules. As above, set
and
since, now,
uo(x) ~ ±R ( lx
x-a)
_ al as x --+ a
Remark 0.4. The results presented here differ considerably from the
conclusions obtained by numerous authors when u is a scalar func-
tion,i.e., u: G- R; the motivation there comes from the Vander Waals
and Cahn-Hilliard theory of phase transition (seeM. Gurtin [1), L. Modica
[1], P. Sternberg {1], R. Kohn and P. Sternberg {1], E. DeGiorgi [2]). In
the scalar case u€ --+ u* which takes only the values + 1 and -1; the sets
[u... = +1] and [u* = -1] are separated by an interface S having mini-
mal area. The phase transition region [-1 + 8 < Ut: < 1 - 8] consists of
a thin layer enclosing S (see Figure 1). By contrast, in our situation the
phase transition region flue I < 1-6) occurs in small neighborhoods of point
singularities (see Figure 2).
The difference in the analysis stems from the fact that, in the scalar case,
the potential (lul 2 -1 ) 2 is a two-wells potential while in the complex case
the same potential has an S 1-well. In the scalar case it is easy to pass to the
limit a.e. once the natural estimate fa IVcp(ue)l S C, for some appropriate
Introduction xxiii
FIGURE 1.
lu I: I
~,,,, (. b
(/ (j)
FIGURE 2.
A final comment:
Remark 0.5. Most of the results presented here hold (with identical
xxiv Introduction
where <I> is a scalar function that satisfies .!lei> = 0 together with boundary
conditions. The Dirichlet condition for u transforms into a Neumann condi-
tion for <I>, while the degree condition on u transforms into a Dirichlet-type
condition for <P.
We also study in detail the behavior of the minimizer u and its energy as
the size of the holes shrinks to zero. More precisely, given any configuration
b = (bt. ~ •... , bd) of distinct points in G, set Gp = G \ U B(bi, p) and
consider the class C.p defined by (13). We prove that, asp - t 0,
In Chapter III we start the study of Problem (2) and we derive some
elementary but basic estimates. The main estimates are the following:
(23}
(24)
and
{26)
where the constant>.> 0 and the integer N depend only on G and g {they
are independent of e).
As e-+ 0 {along a subsequence) the points x~ converge to some points
denoted (aj}jeJ with aj E G and card J $ N. The points (aj) are the
natural candidates for being the singularities of lim Ue,. (assuming such a
limit exists- which we don't know yet!).
A central part of the proof consists in showing that, for any fixed f1 > 0,
if we set
then
(27)
xxvi Introduction
where C(17) depends only on 17, G and g (but not on en)· This is the content
of Theorem V .1.
For this purpose, we establish, in Chapter V, lower bounds for
Set
(28)
(In principle ~i = K'J depends on n, but we provide a bound for K'J inde-
pendent of n so that, by passing to a further subsequence, we may always
assume that K'J is independent of n. Another difficulty stems from the fact
that ai may lie on lJG; to get around this difficulty it is convenient to en-
large a little bit G, say by a domain G', and to extend U~: by a fixed map
G on G' \ G with 1~1 = 1 on G' \G).
Roughly speaking, we prove that, for every fJ > 0,
(30)
{31)
L:(ttJ- ltj) ~ 0.
jEJ
Introduction xxvii
Using the results of F. Bethuel, H. Brezis and F. Helein [2], we are able to
exclude the possibility that "'i = 0. We are thus left with
(32} Vj
and
If one of the points a;, say alt belongs to 8G we may improve (29).
Instead of (29) (with 11:1 = 1) we now have
(34)
Theorem 0.2 and estimate (11) in Theorem 0.1 are proved in Chapters
VII and VIII. In Chapter VII we use the stationarity of Ur; with respect to
defonnations of ue induced by the group of diffeomorphisms of G (in the
same spirit as Pohozaev-type identities). This leads to a precise description
of the Hopf differential
I&* l2 - I8x2
w = 8x1
au.. l2 2 . au. au.
- 1 8x1 . 8x2
of the limiting map u* near its singularities a;. (Warning: Here the dot
product refers to the scalar product, not complex multiplication). These
infonnations yield a good control of the behavior of u* near a;, which,
combined with the results of Chapter I, allows us to identify u. with the
canonical harmonic map defined above (see (19)).
In Chapter VIII we prove that the configuration (a;) minimizes the
renonnalized energy W using appropriate comparison functions and the
strong convergence of u,_., in Ck nonns away from the singularities. We
also discuss the relationship between the fact that a= (a1. a2, ... , ac~) is a
critical point of W and property (22) (or equivalently (11)). In Sections
VIII.4 and VIII.5 we study two specific examples: G = B 1 with g(O) = ei9
and g(8) = ed'9 • In particular, we prove Theorem 0.4.
Section IX.l is devoted to the proof of Theorem 0.5. In Section IX.2 we
prove that (see Theorem IX.3)
(35}
~<ll=O inn,
c) = Const. = C, on OWi, i = 1,2, . .. ,n,
(3) <ll=O on 8G,
f
law;
c:: = 21rdi
uJJ
i = 1,2, ... ,n,
where vis the outward normal to Wi and also the outward normal to G.
Here the Ci 's are not given but are unkown constants that are part of the
problem. It is well known (seeR. Temam [1], H. Berestycki and H. Brezis
[1], [2]) that problem (3) has a unique solution. Moreover, ~ is obtained
by minimizing
(4)
in the class
(5)
hence
{6)
Set
(7) div D = 0.
On the other hand we also have
Indeed,
8~
D·v=-(vxv.,.)+ 011 ,
where -r is the unit tangent vector to &wi such that (v, -r) is direct (see
Figure 3). Then (8) follows from the fact that
FIGURE 3.
div D = 0,
and
{ D·v=O
lr,
for each connected component ri of 00. Then there exists a function H on
n such that
4 I. Energy estimates for 5 1 -valued maps
Note that
(11)
k1Vul LIV~I2 •
2 =
1. An auxiliary linear problem 5
Proof of Step 2. In view of (12) we shall try to find some u in£ that
satisfies
(13)
in 0,
(14)
in 0.
(15)
and
(16) l, F · r E 211" Z
for each connected component ri of 00. Note that in our situation (15)
holds since c) is harmonic and (16) holds since F · r = ~~.
Indeed we may write locally u = ei.P, and then (14) becomes
which has a local solution by {15). Moreover if u 1 and u2 a.re two local
solutions of (14) then u1 = au2, where a is a complex constant with lal = 1.
The local solution 1/J is continued globally by integrating over paths. The
corresponding integrals may differ, but the difference belongs to 21rZ, and
thus u = ei1/J is well defined up to a constant phase. Moreover u belongs to
£,since
6 /. Energy estimates for 8 1 -valued maps
(17}
Inf { -21 {
~~ k
IVcpj 2 + 21T t
~1
di 'P!8w; - {
~
cp (g X ~g)
~
}
where}}= {cp E H 1 (f2;JR); cp = Const. = lf'l8w; on each {)wi}·
We have
Theorem 1.2. The infimum in {17) is achieved by a unique solution.
Moreover
f Hf»t =- { ~ 1 8H
lac 8T laa 8r
2. Variants of Theorem 1.1 7
and, by (9),
{)H =-
&r
(gx OT
89 ) + Mt
01/
=0 on8G.
inn,
inn.
V X V.,. = Mt
{),_, = g X g.,. on8G
n
deg(go, 8G) = L deg(g,, ow,).
i=l
Set
(19)
inn,
(20)
on aw,, i = 0,1, 2, ... ,n.
8 I. Energy estimates for 8 1 -valued maps
where
inn,
on8wh
on 00\8w;.
Proof.
Step 1: We prove that
inn
inn.
Ov 8"P2 &g;
v x {h = &v = g; x ar on Ow; for i = 0, 1, ... , n.
e. Variants of Theorem 1.1 9
{ ~::~j
'1/J=O
inn,
on Ow;, for j = 1,2, .. . ,n,
on8G.
Hence
JVul 2 = IV'I/JI 2+ IV~2I 2 - 2 ( 8<P 2 81/J - a<P 2 8'1/J)
OX! OX2 ax2 8xl
~2
= IV'I/JI 2 + IV~2I 2 + 2 dtv
. [
'1/J ox2 , -'1/J 8~2]
OXt
and therefore, as in (11), we obtain
LIV'I/11 2 ~ 471' 2
n
t
i=l
capwi.
and set
(21)
10 I. Energy estimates for 8 1 -valued maps
in G,
(22)
oniJG.
inn,
{23}
inn.
This follows from the fact that Aq,o = 0 inn, 8 0 = g !, X 9-r on IJG, and
~ E Z (see the arguments in the proof of Theorem 1.2).
This u0 will play an essential role. We shall call it the canonical har-
monic map associated to (g, a, d).
3. 8 1 -wlued harmonic map.s 11
(25)
in 0,
(26)
in 0.
In fact, fb is globally defined, since for every closed curve C inn enclosing
a domain w,
f
lc
F·-r= f
lc
uxu.,.-1c Mo 8v
where vis the outward unit normal tow, and (v,-r) is direct.
1
We have
u xu.,. = 21r deg(u, C) = 21r L di.
c ~~
and therefore
LF·T=O.
12 I. Energy estimates for 8 1 -valued maps
at/; a~Po
a, = g X 9T - ov = 0.
Thus t/J is constant on aG.
Set
~'1/J = -a
OX}
(u x -au) +a- (u x
ax. OX2
-au) = u
ax2
x Au= 0.
We claim that (25) holds. Indeed, we may write locally u = ei"' and
uo = ei'l'o.
Then (23) becomes
ar.po
--=---,
8-Po
OXt 0X2
8r.po &<Po
ax2 = OXt
and (26) becomes
o'ljJ acp a~o
- = -ax. -,
+ -OX2
OXt
81/; &cp a~o
= OX2 - OXt .
OX2
Thus 'V (cp 0 + 1/J- cp) = 0. Hence cpo + 'ljJ - cp is locally constant and vanishes
on 8G. It follows that I{Jo + 'lj;- cp ::: 0. This completes the proof of the
theorem.
Remark 1.1. Since I'Vuol = IY'~ol and I'V-Pol :::: lxl:i~il near ai then
Uo E W 1•1(G) and even tto E W 1•P(Q) for any p < 2, but u 0 fj. W 1•2 (G).
Moreover u 0 satisfies the equation
In fact, here, we have only a,p = 21f E CiOa, I because the left-hand side of
(28) is a sum of (first order} derivatives of L 1 functions. Hence
where xis a smooth harmonic function on G, and the Ci'S are some real
constants. This means that any u E W 1•1 (G) nC is of the form
F
(29) u= IFI in 0.
(33)
(34)
Definition. Let G be an open set and let a1. a2, ... , an E G be n points in
G. Let u: G\U{ai} ~ S 1 be a smooth harmonic map. Set di = deg(u,ai)·
i
We say that u is proper if, for every i,
.
1liD lz- ail~ 1J
(z ) exists.
z-+a, (z - Bi)d•
and
1/J = 0 on aa.
Note that deg(g, l:JG) = 0 and hence there is a well defined single--valued
(smooth) function r.p 0 : 8G-+ R such that
Min { 1Vul2 ,
uEEp lo.P
where t'p = {v E H 1 (0p; 8 1 ); deg(v,8B(a,,p)) = di and v = g on 8G}. We
already know (by Theorem 1.2) that there exists a unique minimizer, which
we denote up.
Theorem 1.6. Asp tends to 0, Up converges to uo (the canonical harmonic
map) uniformly on every compact subset of n UaG.
r ~p = o.
laa
4. Shrinking holes. Renormalized energy 17
Then
n
(35) Sup v- In£ v $ :L)Sup v- Inf v).
n n i=t Bw; Bw;
We claim that
n •
(36} U li 1s connected.
i=l
i = 1,2, ... ,k
and
i = k + 1, ... , n,
for some 6 > 0. Choose any smooth function (} : R -+ R such that
if t $ t 0 o
={ ~
-
O(t)
ift~t 0 +6
18 I. Energy estimates for 8 1 -valued maps
r ~ =0
law, ull
fori= l,2, ... ,n.
Thus we obtain
i O'(v)!Vvl 2 = 0
A= Max {3•.
l::;t$n
Multiplying the equation L\v = 0 by (v- A)+ and integrating we see that
dv=O in 0
(38) { av
!l)w1
--0
8v- for j = 1, 2, ... , n.
Then
In particular if v =0 on 8G then
{40)
$I:
n
and thus there is a point on 8G where ~P- ~o = 0. From this and (41),
we deduce that
(42)
Proof of Theorem 1.6 completed. It follows from Lemma 1.2 and elliptic
estimates that
and
(43}
and
inn
inn.
(44)
L
n
(45} Ro(x) = ~o(x)- d; log lx- a;!
j=l
where
(47)
= W(a,d,g).
Note that W is independent of p and depends only on G, (a.), (di) and
g. W will be called the "renormalized energy". Here O(p) stands for a
quantity X such that [X[~ Cp where C depends only on G, (ai), (di) and
g.
Proof. By Theorem 1.2 we know that
8cP
Recall that a: = g X gT on aa, «Pp =Canst. on 8B(a,, p) and that
Thus, we obtain
Theorem 1.8. Lettt0 be the canonical harmonic map associated to (g, a, d),
then, as p - 0,
(50) iI l'lp
Vuol 2 = 1I
Op
V4>ol 2 = 1 &G
lJ4>o
a4>o- ~
L..t
II i=l
l
&B(a;,p)
lJ4>o
a4>o.
II
84>o
011 = g x g.,. on 8G
and that n
Ro(x) = 4>o(x)- L d; log lx- a; I
j=l
Thus, we have
i 84>o 0
-4>
BB(a,,p) 811
= h &B(a,,p)
(lJS·
_t
811
+ -£4) (Si + dt.
P
log p)
4- Shrinking holes. Renormalized enefYY 23
and, using the fact that Si is harmonic in B(a,, p), we deduce that
(52)
Mip. { 1Vul2
uE£p lnP
where
(55)
a~p=O
M onaG,
(56) a:=gxg'T
8~p £4
()v = p on fJB(ai,p), i = 1, 2, ... , n.
24 I. EnertnJ estimates for 8 1 -valued maps
+
Since P is unique up to an additive constant we may normalize it by
assuming that
r •p = o.
loc
Set
n
(57) 'lllp(z) =+p(x)- Ld;logJx-ajl·
j=l
We shall use the following lemma., the proof of which was suggested to
us by L. Nirenberg:
Lemma 1.5. We have, asp- 0,
(58)
and
1 8\11*
_P=j
lJr
{}iJ!*
on 8G,
Hence, we have
in Op,
onaG,
(62) on8B(ai,p), i=1,2, ... ,n,
where F (resp. Gi) is a primitive with respect to arc length of I (resp. 9i)
on 80 (resp. 8B(ai, p)), i.e.,
8F =f
8r
(63) { !l.w* = o in G,
'II*=F on8G.
L
n
From (62), (63), (64) and standard elliptic estimates (see e.g., D. Gilbarg
and N. Trudinger [I]) we deduce that for every compact subset
K c G\U{ai}
i
{65)
in G
{66)
in G
26 I. Energy estimates for 8 1 -valued maps
so that W satisfies
in G
{67}
on 8G.
Recall that Ro also satisfies (67) and since the solution of {67) is unique up
to an additive constant we may as well choose
(68) 'II=~.
{69)
(70)
(71)
Fix any cr > 0 so that B( eli, cr) C G and B( ai, cr) does not contain any
other point aj, j i- i. From (60) we deduce that
and
,4. Shrinking holes. Renormalized energy 27
by {71). In particular,
Therefore we have
a.nd
28 1. Energy estimates for 8 1 -valued maps
Remark 1.5. One could also work with a still more restrictive class of
testing maps:
We have, as p _. 0,
where
v =g on 8G and }
ip = { v E H 1 (0p;S1 ) (
z -a· )d·• .
v(z) = pd,' on 8B(a,,p), Vi
This follows from Theorems 1.3 and 1.9 and the fact that cap(B(ai,p)) =
0(1/llog pi).
Remark 1.6. Suppose now that the integer nand the points (a,) are not
prescribed: they are free to move in G. Suppose that the degrees (di) are
n
not given; they are only constrained by the relation :L;£4 = d = deg(g, BG)
i=l
where g : 8G _. 8 1 is given. If we want to minimize Jn p 1Vupl 2 (for p
small) among all possible choices of n, (a1) and (d1) we are led to:
(i) Choose n = d and each dt = +1. This follows from the obvious fact
that
In other words, W - t +oo as two of the points (ai) coalesce or as one of the
points ai tends to 8G. Therefore, Min W is achieved and every minimizing
configuration consists of d distinct points in Gd (not (id).
f j(>ol ::; C.
laG
Proof. Let 1/J be the solution of
in G,
on8G.
Lemma I. 7. We have
Proof. For simplicity, we shall only consider where d = 1 (i.e., the con-
figuration (a;) consists of a. single point). We shall sketch the proof in the
case of a fiat boundary. Assume that 0 E &G and that, locally, near 0, G is
the half-plane, G = {(x1, x2)i Xt < 0}. Let (an) be a sequence in G such
that an -+ 0. Consider the function
(75)
where a: is the reftected point of an about the x2-a.xis and o:n is a constant
chosen such that
r
loc
Vn = 0.
Note that lanl :5 G since J00 pog lxll < oo. We have (calling ~n the
function 4>o associated to an),
in G,
{76)
on {}G.
I Ovnll <G.
011 L""(BG) -
and the desired conclusion follows since log !an- a:l- -oo.
CHAPTER II
Wi = B{zi,p),
n
O=G\ .u w;.
J=l
n
Let di E Z, fori= 1, 2, ... , n, be given and set d = Ed&.
i=l
We assume
and
The main purpose in this chapter is to provide a lower bound for the
energy of maps v: 0 - S 1 in terms of their degrees on 8wi· In view of the
results of Chapter I it is natural to introduce the solution ~ of the problem:
l!l~=O inO
~ = Const. = C; on 8w;, j = 1,2, ... ,n,
(4)
f M =211'd;, j = 1,2, .•. ,n
)&13 8v
~=0 on8G
Theorem 11.1. Under the o.ssumptions (1), (2}, (3) and also d > 0 we
have
(5)
where P = {j E {1,2, ... ,n}; d; > 0} and the minimum in (5) is taken
over all choices of integers 6; such that 0 :S 6; ~ d; and E 6; = d, and
jEP
C = 6w (t.ld;l) 2
( Wg (rua;G) +n log 2).
Before proving Theorem II. I we derive some easy consequences for the
energy of the S 1-valued maps. Consider, as in {1.1), the class
Corollary II.l. Assume {1), (2}, (3). Then, for every v E £, we have
where ~ is the solution of (4). We may then apply Theorem ILl and note
that
to infer that
z=c; ~ z=c; =d.
jEP jEP
A lower bound for the eneryy 33
n
and the conclusion follows since E dJ = E dJ.
jEP j=l
The proof of Theorem 11.1 relies on the following lemma which concerns
the solution 'Pi of the problem:
Arpi = 0 in 0
'Pi= Const. on each awj,j = 1, 2, ... ,n,
(6) { orpi = {211'
d, if j = i
law; lJv 0 if j =/d
'Pi= 0 on lJG.
(7)
diamG)
(8) I.Rt(x)l :5 log ( JJ + {n- I) log 2 =A.
Moreover
{9)
34 II. A lower bound for the energy
t
Set
X= (supv- Infv).
. &w; 8wJ
J=l
~ log ( lx; - Xj I +
lxi- xil- p
p) :-=::; log 2
by assumption (3}.
Also
Supv- Infv = 0 since v is constant on ow;.
8w; 8wo
Thus
X~ (n- 1} log 2.
Supv
n
< lnfv
- n
+ Supv- Infv +X lJG
80
< Infv + Supv- Infv +X
- 8G 8G IJG
= Supv +X.
lJG
But
Supv = - Inf log (lx -x;l) ~ 0 by (1).
8G :z:E8G Jl
Thus
Supv ~X.
Sl
Similarly
Inf v > Inf v - X > - log ( diamG) - X.
Sl -IJG - J.&
A lower bound for the energy 35
:i
We now prove (9). We have
rJVtpiJ2 = - t lewj
ln
r tp;,
J=t
= -271' 'Pi(awi>·
Applying (7) and (8) we have
I'Pi(awi) -log(p/p,)j:::; A
and thus
kiV<.Oil 2 ~ 271' log(p,/p)- 211'A.
Proof. We have
o = - { (atpi)tpt = f IV'tptl2 + { otpi tpt
ln ln lew, ov
= fn1Vtptl 2 +21rdi'Pt(owi) ~ fo!V''Pti 2 -
Thus tpt = 0 in 0 and hence tpi :::; 0 in 0.
We now turn to the proof of (12). Set Ci = tpi(8wi)· We claim that
{13) (J'i. ~ ci on n.
This clearly implies (12).
=-In
We have
0 A(Ci- tpi)(Ci- tpi)+ = fo!V(Ci- tpi)+j 2 +
nla . 8{})Ci- (J'i)(ci- (J'i)+- Ia 8{})Ci- (J'i.)(ci- (J'i)+.
I:
j"'l 8w3 8G
Note that all boundary integrals vanish. Indeed if j -::/= i, then (C, - <,Oi)+
is constant on ow; and few.
3
8!i
vV
= 0. On 8w; (Ci - (J's) = 0, while on
00, (Ci- 'Pi)+= ct = 0 by (11). Therefore (Ci- tpi)+ = 0 in 0 and this
proves (13).
36 II. A lower bound for the energy
such that
d(P;, ni) 1 .
{14) d(P;,7r&) ~ 2 VJ = 1,2, ... ,d, Vi= 1, 2, ... ,k
(15)
Set
We have
d(p;,nl) 1 . k d
(16) . ) ~ -2 VJ = 1, 2, ... I + .
d(PJ,7rl
Thus
d(p;, ni) < d(n1, 1r1)
which contradicts (15). Thus we have proved (16).
Next, we eliminate the pair {n 1 , 1r1 } and we reiterate the same procedure
with n 2. This yields some +1 point denoted 1r2 E {Pt.P2 •... ·Pk+d}\{p,},
such that
Vj f. i.
A lower bound for the enerm~ 37
Then we start with n 3 and so on. When we have exhausted all -1 points
we are left with +1 points denoted P1, P2, ... , Pd. By construction, we see
immediately that (14) holds.
Proof of Theorem 11.1. Consider the points (xi) with their associated
integers (d1). We say that a point x; is a negative point if di < 0; the
negative points are repeated according to their multiplicity ld, I and we
denote them by
We say that Xi is a positive point if dt > 0; the positive points are repeated
according to their multiplicity dt and we denote them by
(note that d = Ed,>O d;- Ed;<O ldtl = :E7=l di)· The remaining points Xi,
associated to di = 0, are denoted
We say that
(7rt. nl), (1r2, n2), ... , (7rk, nk)
are "positive ions". We regroup the positive ions according to their multi-
plicity and write them as
38 II. A lower bound for the energy
6.1/J, = 0 inn
1/Ji = Const. on each 8w;, j = 1,2, ... ,n
1/Ji = 0 on8G
and
(ii) The function t]i associated to the ion fPt, 6t] is the solution of
Ll.1]i =0 inn,
TJi = Const. on each 8w;, j = 1, 2, ... , n,
1Ji =0 on8G,
and
if w is the disc centered at Pi,
otherwise.
A.O, =0 inn
Oi = Const. on each 8wj, j = 1, 2, ... , n,
ei = o on ac
and
r
Jawj av
ao, = 0 Vj = 1,2, ... ,n.
So that, in fact, 9, = 0.
A lower bound for the ene'!IY 39
By construction, we obtain
I; t
c)= E""i + L'1i·
i=l j=l
Hence
I; t
(17)
{ IV~I' = IE V'f/Ji 12
i=l
+ IE
i=l
VTJ; 12 + 2EV.,P, VTJ;
i,j
t
~IE V'7;1 2 + 2EV¢, V'l;·
i=l i,j
Step 1: We have
But
where w is the disc centered at P.. By Lemma 11.2 we know that '7; :5 0 in
In
n and thus V1J; v'l. ~ 0.
On the other hand, we deduce from Lemma 11.1 that
{19}
40 II. A lower bound for the energy
(21)
and therefore
l
- 21rk(2A +log 4) 2: Cj·
j=l
Note that
and n
k ~ Eldil·
i=l
Finally we obtain
(2)
(3) {
-~Ue: = : 2 u£(1 - luel 2 ) in G,
Ue =g on8G.
(4)
(5)
{ 1Vun[ 2 $ C,
lnl
{ (lunl 2 - 1) 2 $ C t~
ln1
and
un(x) =x on 8B1.
Passing to a subsequence we would find some u E H 1 (B1 ; 8 1 ) such that
u(x) = x on 8Bt. This is impossible (see the Introduction).
Lemma 111.1. We have, 'Vtt $ t2,
I(tt) :$ 1r log(t2/tt) + /(t2)
i.e., the function t ~--+ (I(t) + 1r log t) is nondecreasing.
In particular
/(t) :5 1t log(l/t) + 1(1) Vt E (0, 1}.
= ! 1
2 Btte 2
[Vu2l 2 + ! f
4 j Bl/t~
(lu2l 2 - 1) 2 + ! 1
2 Btto 1 \B. 1, 2
lv (.!...)
lxl
2
1
= l(h) + 1r log(t2/tt).
44 III. Some basic estimates for~
Remark III.2 One may also prove that there is a universal constant C
such that
(9)
Proof. Fix d distinct points a1o a2, ... , ad in G and fix R > 0 so small that
Let S1 = G\ ( Uf= 1 B(ai, R)) and consider the map g: 00- 8 1 defined by
if x E ac,
9-( X ) -- { g(x)
et·(J if x =a;+ Re* 9 E 8B(a;, R).
Since
deg(g, 00) = 0
(10)
(11)
But
(12) 8u · Vu) = (x · v)
-(x
av
(8u)
-
ov
2
&u ou
+ (x ·T)--
{h- oz,•
and
(13)
Theorem 111.3. There exist positive constants .Ao and p.0 (depending only
on G and g) such that if u<!' is a solution of {9) satisfying
(14) 12 r
e lanB3L
(lu~l2 - 1)2 :5 J.I.O
(16)
Proof. It follows from Lemma A.2 in the Appendix ofF. Bethuel, H. Brezis
and F. Helein [2] that
I!Vu<!'IIL""(G) :5 C/e,
where C depends only on G and g. Therefore, we have
(17) lu~(x)-
c
u!"(y)j :5 -jx- Yl "''x, y E G.
e
We argue by contradiction and assume that lu~(x 0 )1 < 1/2 for some
x0 E G n Bt. Then we have
and thus
Consequently
and consequently
47
Hence, we have
1GnB~t4c(xo) u 2
(I ~I - ) -
1 2 > o:e
{16C}2
2
If we choose >.o = l/4C and ~Jo < o:/(160)2 we are led to a contradiction.
CHAPTER IV
g=g on 8G.
(1) Xi E G, ViE/,
For this purpose, it suffices to consider a maximal family satisfying (1) and
(2).
We say that the disc B(x., .Xoe) is a good disc if
12 r
e 1B(xi,2Aoc)
(lu£12 - 1)2 < Jlo
(4) cardJ :5 N.
Remark IV .1. Strictly speaking we should have denoted the points (Xi)ie J
by (xDieJ.· The main content of Lemma IV.1 is that card J~ remains
bounded independently of e.
Proof. There is a. universal constant C such that
Proof. Let x E G\ UiEJ B(x,, Aoe: ). By {3), there is some j E 1\J such
that
x E B(x;, .Aoe:), which is a good disc.
In this section we shall replace the bad discs B(xi, Aoe)ieJ by slightly
larger discs (deleting if necessary some of the points (Xi)) in such a way
that the points Xi are far apart (relative toe:).
so IV. Towards locating the singularities
(5)
and
(6)
(7)
We take>.= 9.>.0 and J' = J\{1}. We are reduced to the previous case.
After a finite number of steps (at most N) we are led to the conclusion of
the theorem with Ao :::; >. :::; Ao gcard J.
Roughly speaking the points (xi)iEJ' = (xDieJ' correspond to points
where u~ may have a singular behavior. To simplify the notation we shall
write J = J~ instead of J'. We summarize the main properties of the family
(xi):
(9) lx·-
• x·lJ >- 8.Ae 'Vi, j E J, i ::f j
(10) cardJ :5 N
and
(11)
{12)
and
We cannot exclude the possibility that some of the points li are the
same, i.e., it may happen that x~" and xj" converge to the same limit.
We denote by
and this, in turn, will imply that, on K, ue:,. --+ u. uniformly. Then we
shall prove in Chapter VI that aj ¢. oG.
CHAPTER V
The conclusion follows from the fact that lu.rl ~ 1/2 on 8B(xf, >..e) and
JIVutllL<><> :5 C/e (see Lemma A.2 in the Appendix of F. Bethuel, H. Brezis
and F. Helein {2]).
Passing to a subsequence we may assume that
di = deg (Ue.,' oB(x~", Aen)) is independent of n.
For the same reason
K-j = deg (ue,., oB(ai> 17/2)) is also independent of n.
(5)
The proof of this theorem is indirect; it relies on a lower bound for the
energy near ai which is presented in Section V.l. We will return to the
proof of Theorem V.1 in Section V.2.
and consequently
Set
Proof. We write on ni
(7)
54 V. An upper bound for the energy of Uc
and therefore
(9)
(ll)
(12)
Combining (12) and the estimate lluelloo $ 1 we derive, with the help of
the Gagliarda.Nirenberg inequality, that
{13)
Going back to (10) and using (11) with (13) we are led to (9).
Finally, (7), (8) and (9) yield (6).
(14) L. ~
1Vut:nl 2 2= 211"1Kjlllog cnl- C(1J).
Hence
(16)
(17}
Going back to Theorem V.2, in the light of Lemma V.2, we may now
state
56 V. An upper bound for the eneryy of 'Ue
(18)
Proof. Suppose, by contradiction, that (18) does not hold. Then there
exists a sequence en ~ 1 such that
(19)
(20)
(1)
and
(2)
Since
faout,.l 2 -1)2 :5 Ce!
we deduce that lu.l = 1 a.e. We also have
(5)
(7) u* = g on {)G,
(8)
(9) :Ldeg(u*,aj) = d,
j
(10}
(11)
(13}
and
(14)
1. Proof of Theorem VI.1 59
Since
deg ( ue.. , 8B(xo, R')) = 0,
{16)
and thus
for n large.
Proof of (12}.
Step 1: uE,.- u* in H 1!x(G' \ U;{a;}) and in Gtc(G' \ U;{a;}).
It suffices, in view of (10), to consider a point xo E 8G that is not
a singularity (a;) and to show that for some /(, ue:,. converges to u. in
H 1 (B(x 0 , R') n G) and in CO(B(x 0 , R') n G).
Fix R < ~ dist (G, 8G'} such that B{xo, 2R) does not contain any singu-
larity. By Fubini we may find.« E {R, 2R) such that (up to a subsequence)
f 1Vue:,.l 2 ~C
i&B(:~:o,R')
60 VI. ~ .. converges: u... is born!
and
and we have
l 1/Jq
u
5: 4flvu~l'I/Jq-1 + 2e2
u
r
lou av
&'lj; .,pq-1_
(17)
(18)
a.nd moreover B(aj, R) does not contain any other singularity, so that
deg(ue:,.,oB(aj,R)) = K.j = 0. We may now apply Theorem 2, Step 1,
of F. Bethuel, H. Brezis and F. Helein [2] in B(ai,R) n G to conclude that
(19)
On the other hand, by the definition of ai, there exists at least one bad
disc B(xi,2.\cn) contained in B(a;,R). Recall that (see the beginning of
Section IV.1)
(20) 21
en
1
B(z,,2.>.e:,.}
( IUe,.l 2 - 1) 2 2:: /to > 0, for every i.
(22}
On the other hand, if we pass to the limit in Theorem V.l as n--+ oo, we
are led to
(23)
where C in (22) and (23) is independent of TJ. Note that the passage to the
limit is justified by Step 1 in the proof of Theorem VI.l.
Combining (22) and {23), we find
L(KJ- K.j) ~ 0.
j
In what follows we choose '1 E (0, R), which will tend to zero, as in Step 2.
Our next lemma plays an important role.
B. FUrther properties of u* 63
(24) u = g in (G'\G)nB(a,R)
and
(25}
we have
~ r
j B(a,R)\B(a,f1)
IV'ul 2 ~ 27rllog TJI- C, 'VTJ E (0, R}
(26)
{27}
where 0 = G' \ Ut= 1 B(a;, TJ), and C depends only on R,d, G, G' and R.
We now proceed as in the proof of Step 2, and conclude that
where St = St n {x2 > 0} and S't = St n {x2 < 0}. Note that
Therefore
471"- -a<
-
t .;t- s,
Ll
2
Vul.
u...
coincides with THE canonical harmonic
map having singularities (a;)
(z-a·) . (
1 e'"'j ~>, near a;
(2} uo(z) =
lz-a;l
where t/J; is a real valued harmonic function, smooth in some neighborhood
of a; (including a;). In other words, uo(z) behaves like e•<B+Bj) for some
constant phase 0; = tP;(a;).
The main result in this Chapter is the following:
Theorem VII.l. Let a; and"* be as in Chapter VI, then u. = u 0 = the
canonical harmonic map associated to the singularities (a;).
Proof.
The strategy is the following: we first prove that u. is in W 1 •1 (G). By
Remark 1.1, we then know that
which is the so-called Hop£ differential. (Warning: the dot product refers to
the scalar product of vectors, not the multiplication of complex numbers.)
This quantity was first introduced in the framework of minimal surfaces,
and plays a crucial role in many two-dimensional problems that are invari-
ant under conformal transformations (see e.g., J. Sacks and K. Uhlenbeck
[1], M. Griiter [1], R. Schoen [1]). Set
Wn = 4en
\ (Jue,.J 2 - 1) 2 •
1 2
-Aue,. = 2'Ue,. (1- Jup;,. I ),
en
satisfies
(5) az = .!!...
awn
{)z
(2W. )
n
where, as usual,
and
a= 21(a ~.a)
{)z &xt - fJx2 ·
In other words,
l Wn(x)((x)-+ ~ m;((a;),
J
V( E C(G).
VII. u.. coincides with THE canonical harmonic map 67
{1-lue,.l 2 ) ~ CK e! inK,
On the other hand, we know (see Step 1 in the proof of Theorem VI.l)
that for every point Xo on aa,
there exists a constant R' > 0, such that
f Wn-+ 0;
JB(xo,R')
this implies that supp(Jl) C U{a;}, and f-l has the required form.
j
T = .!!_
8z
(...!._)
1rz
=~(~log
8z 1r 2
r).
Set
in the sense of distributions,
where W n = Wn in G, and W n = 0 outside G.
We have
(7)
(8)
a
az(wn- 2an) =0 in V'(G).
Set
68 VII. u.., coincides with THE canoniedl harmonic map
Nn(K) = {x E G; dist(x,K) $ R}
(9)
{10)
Thus,
(12)
VII. ~ coincides with THE canonical harmonic map 69
(13)
(14)
(15)
Indeed, by (12),
a
{17) IVu.l ~ lz _a; I near each aj;
in particular,
Cj = 0, Vj.
70 VII. U;. coincides with THE canonical harmonic map
Fix one of the points a;; for simplicity take a1 = 0. Recall that, in polar
coordinates, (see (34) in Chapter I),
where Sis a harmonic function. From (3) and {19) it follows that
where x' is some harmonic function near zero (and including zero). We
now compute w* using (20). Note that if, locally, u = ei'P then
(25) \
4e:n (lut:,.l 2 - 1) 2 -+ i LDar j
(26)
(27)
1 2 1
llog ~n
JJVut:n I -o 0 in L (K)
(28)
with
From Theorem V.2 and Theorem VI.2 we deduce that, for every j and
every '17 > 0,
(31) 'Vj.
Combining (31) with (29) we conclude that mj = 21r, 'Vj and the desired
result follows.
The last result of this chapter is
72 VII. ~ coincides with THE canonical harmonic map
In other words,
(33) VH;(ai) = 0.
Remark VII.l. Note that property (33) does not hold for the canonical
harmonic map uo associated to an arbitrary configuration of points (a;).
This additional property is related to the fact that "*
is the limit of a
sequence of minimizers for Ee, and that the configuration (a;) minimizes
the renormalized energy W (see Section VIII.2}.
Proof of Theorem VII.4. In view of {22) and (23) we must have, as-
suming ai = 0,
x" (O) = 2 ox'
{jz
(o) = o
Recall that, near zero, u* = e•"' with cp = 9 + x' a.nd the desired conclusion
follows.
Remark VII.2. The method used in the proof of Theorem VII.l is in
the same spirit as the argument of H. Brezis and L. Peletier [1) concerning
the equation -Au= uP where p is the critical exponent (N + 2)/(N- 2).
There, precise information about the blow-up behavior-in particular the
location of the blow-up points-is obtained by multiplying the equation by
{j{Ju and using various Pohozaev-type identities.
Xi
We now present alternative proofs of Theorems VI1.2 and V11.4 assuming
Theorems VI.l, Vl.2 and VII.l. Recall that Ue satisfies
1 2
(34) -Aue = 2ue{l-luel ) in G.
c
VII. u.. coincides with THE canonical harnwnic map 73
(35)
(36)
{37)
(38) au.
8v
1=lao+
&v
2
&HI
&v
lonl
&v
2
=
2
and
'
(40) Rk
-2 I&HI
-
8Ba &v
2
+ 2ffii -- 71" + -2 Rk lanl
-
&BR &r
2
(41) !1H=0
74 VII. u.. coincides with THE canonical harmonic map
by E
k=l
XJc ~H
VXk
and integrate on BR we obtain
2RlaBR
f '8H' RlaBn
{ '8H'
(42) 2 2
8v = 2 8r
Comparing (40) and (42) yields mi = 1rj2 which is the desired conclusion.
2. Alternative proof of Theorem VII.4. Multiplying (34) by 0Bue: and
Xk
integrating we find
(43)
Passing to the limit in (43) (using (10) and (11) of Theorem Vl.l) yields
(44)
But we have
(45)
and
(46)
(47)- 1 8Bn
(T ·ek 8H)
ov -R- +8xk
-{)H - + 1 (--+-IVHI2)
8Bn
1 {)H
R 8T
1
2
(v·ek)=O.
(48)
VII. "* coincides with THE canonical harmonic map 75
(49)
But
(50)
aH
From (49), (50) and the mean-value theorem (recall that~ and -8 are
aH
v'XI X2
harmonic functions) we deduce that
VH{O) =0
which is the desired conclusion.
CHAPTER VIII
where O(p 2 ) stands for a quantity Y such that IYI ::5: Cp2 and C depends
only on G, a and g.
Recall that I(e,p) has been defined in Section III.l.
We postpone the proof of the lemmas and present the proof of the
theorem.
Proof of Theorem VIII. I. Fix p < min{J>o, Pt}· Combining (1) a.nd (2)
we obtain
W(a) ~ W(a)
Set
fi.p in Op
{
w= v; inB(ii;,p}, j=1,2, ... ,d.
In particular, there is an integer N 1 = N1(p), such that, for every n;::: N1o
(7)
On the other hand, recall (see Theorems 1.8, VI.2 and Vll.l) that
(8)
Combining (7) and (8) we see that, for n 2::: N 1 (p), we have
Combining (9) and (10) we are led to the conclusion of Lemma VIII.2.
Therefore, it remains only to prove the Claim.
Proof of Claim (10). Recall (see Theorem VII.2) that near each a; we
have
{11)
where ei 8 = t- I
a; and H; is a real~ valued smooth harmonic function in
z-a;
a neighborhood of a1 , including aj, with
(12)
By Theorem Vl.l we know that, given any p < Pl, we may find some
integer N3 = Na(p) such that, for every n;::: N3, we have
(13)
1. The general case 79
and
(14)
Using the same theorem we may also 8SSume that, for n ;;::: N 3 , we have
(15)
2
(16) IVu*l :5- + O(p) in B(a;,p) \ B(a;,p/2).
p
1 - lue:,. 12 4 - .
(17) 2
Cn
:5 n2
lr
+C = K(p) m B(a;,p) \ B(a;,p/2)
{19)
lw n - I
u e:,. <
- lei(B+Hj(ai)) - u* I+ lu* - ue:.,. I·
80 VIII. The configuration (a;) minimizes
Hence, we deduce {19) from (11), {12) and (13), for every n ~ N 3 . Differ-
entiating (18) we easily see that
IVw n
- Vu~;,. I < ~p !ei(8+H(ai))- u I+ !Vei(9+H(aj))- Vu I
- lln en •
Applying (11), (12), (13) and {14) we are led to (20) for every n ~ N3 .
The proof of (21) relies on the following variant of the parallelogram
identity
(24)
and then (21) follows from (23), (24), (ll), (12) and (13}.
We may now return to the proof of Claim (10).
Proof of Claim (10) completed. Set
if 1e1 ~ n,
if tel< R.
( ) = { Ut;,.(Z) if z E B(aj,p/2),
(27) z Pwn(z)
Vn if z E B(a1 ,p) \ B(ai,p/2).
(28)
(29)
(30)
and
(31)
(32)
c
{36) 11Vu*IIL 00 (B(ai,p)\B(a;,p/2)) ~ p·
c
(37) IIVu!,. IIL 00 (B(a;,p)\B(a;,p/2)) ~ p"
Going back to (35) we obtain
(38)
The desired estimate (10), with N2(p) = N4(p), follows from (29}, {32) and
(38).
We shall now prove that the vanishing gradient property (39) ma.y also
be derived as a. consequence of the fact that the configuration a = (a;)
minimizes the renormalized energy W (see Theorem VIII.l). Indeed, (39)
is exactly equivalent, a.s we are going to see, to the assertion that a = (ai)
is a critica.l point of W, i.e.,
(40) V'W(a) = 0.
2. The vanishing gradient property 83
inn= G \ U;{b;},
(42)
inn.
(45)
i.e.,
uo = (cos(H; + diO), sin(H; + d;O)).
Hence we find
(46)
84 VIII. The configuration (a;) minimizes
&H·
-
ae
3 +d;- =
&S·
- -3 -d;-loglxl
a
{ &xl axl &x2 8x2
(47)
&H· 88 88· &
-8 3 + d;~ = !:>._
3 + d;~ log lx!.
X2 uX2 U",f;J uXt
Finally, we note that the functions () and log lxl are harmonic conjugates,
i.e.,
(48)
Recall that the renormalized energy W(b) was defined in Theorem 1.7
where we found
n
-1r L: ~ Ro(b,)
i=l
and
n
{50) ~(x) = cl»o(x)- L ~log jx- b,j.
i=l
In the next result we present a simple formula for computing the differential
of W considered as a function of b = (bt.~ •... ,bn) E G".
Theorem VIII.3. We have
(51) DW(b) =
Corollary VIII.l. The property that b = (b1, ~ •... , bn) is a critical point
of W has several equivalent forms:
or
or
(54)
~ (b·- b-)
V.Ro(b;)+ L- d'lb~
i-#j
-b·i2
=0 'Vj.
J I
Note that the last condition comes from the fact (see (44) and (50)) that,
for each j,
and thus
V.Ro(x} = VSj{x)- " ' (x- b1}
L- d.; I _ b·l 2
#j X 1
a~ = 21r 6y in G,
a~
()v = g X 9-r on 8Q
f ib(a, y)da = 0.
laa
Note that ~(x, y) is well defined for x -# y. By analogy with (50) we
introduce
= f tP(a,b)(gxg.,.)da- fV'tP(x,b)V<t>(x,b)
JaG jG
= { (<i>(u, b)- tP(a, b)) (g X g,. )da + 21r<i>(b, b)
ji)G
and therefore
(59) 211" ( R:£(b, b)- Ry(b, b)) = - LG tP (u, b)(g x g.,. )da.
11
(60) 211' (Rx(b, b)- Ry(b, b)) = - JlJG tP 11 (u, b)(g X g.,. )da.
and thus
(61) DW(b) = ~ JlJG tP11 (a,b)(g x g,.)da -1rRx(b, b)- 1rRy(b, b).
Combining (60) and (61) we are led to
DW(b) = -21l"Rz(b,b).
it follows that
VS1(b) = R:,(b,b),
and therefore
DW(b) = -21rVS1(b),
and
R(x, y) = 11t(x, y)- d; log lx- yl,
so that R(x, y) is well defined, even for x = y, and it is smooth on G x G.
Note that
in G,
on&G.
and consequently
88 VIII. The configuration ( Uj) minimizes
- -
((x) = c)(x, b) - 4)(x, b) = R(x, b) - R(x, b)+ d; log lx _ bj
~-~
satisfies
inG,
(64)
onOG.
k
lc#j,l.f<j
Hence, we obtain
2. The vanishing gradient property 89
S;(x) = ci>(x, b)- d; log lx- bl = R(x, b)+ L ~log lx- btl
i~j
and consequently
w =
lJuol2 - l&uol2
l&xl ax2 -
.8uo au0
2' &xl . &x2 '
(69) DW(b)(B) = lm Ln !a 1
2w(z)(a; + i,B;) dz.
j=l 8B(I>;,a)
2
( )
wz = ( - .tz-
-d;- + &H;
b;
- - - t.&H;)
&xl
--
ax2
Hence
bt = b+tB
and
(72)
and
. { B)z-b
'Ub t(x) = e"P•a :z:+t - - in B(b, a)
' lz-bl
and hence
(75)
!!. The vanishing gradient property 91
(77)
(78)
(79)
_ 2t { 1tJvb,t
OX}
f\x) ax1
8yl
(y) + I avb,t 12 (x) ax2 (y)
OX2 Om
+ tJvb,t.
8x1
avb,t(x) (&X1
&x2 8y2
+ 8X2) (y}} +o(t).
&yt
(81)
-
t (
Jc\B(b,p)
{(18vb,tl2 18vb,tl2)
&x1 - &x2
(&Xt &X2)
&x1 - &x2
+ 2 &vb,t Ovb,t (&X1 + &X2)} dx + o(t).
8x1 8x2 8x2 &x1
We claim that
and
(83)
+ O(t2 ) + O(p).
Thus
(85)
and
or
(87) DW(b}(B) = lm [ f
laB(b,u)
~wb(a + i,B} dz].
Thus, to complete the proof, it remains to establish (82) a.nd (83).
Proof of (82). From {75) we deduce
which implies
3. Construction of critical points 93
and
1
laB(b,p) ull jG\B(b,p)
Oub
= -2 StUb X -8 = O(p).
lJB(b,p) V
The last equality holds because ';;: is bounded on 8B(b,p) by (74) and
because of (76}. Moreover (78) implies
(90)
(91)
We now set
(92) F(z) = 1z
zo
f(()d(;
note that this function may be multivalued since near each singularity b;,
f has a pole
ie ·
f(z) = ± z -Jb· + R;.
J
Thus, the real part of F is defined up to 211' Z. Hence
(93) u(z) = ei Re(F(z))
is a single-valued map, that takes its values into 8 1 . Moreover, from the
fact that w is holomorphic, it follows that
(94) (.!!_-
8x1
i~) (Re(F)) =
8x2 8z
= J, {)F
and thus
( 8Re(F) _ i8R.e(F)) = w.
2
(95)
8x1 8x2
Since F is meromorphic, u is a harmonic map from G \ {bt, ... , bn} into
8 1 . From (95) we deduce that the Hopf differential of u is w. The degree
of u around each singularity b; is d; = ±e;.
In conclusion, if d = (d 1 , ... ,dn), (b,d) is, by construction, a critical
point of the renormalized energy with boundary condition u1 8 a .
Example VIII.l. Let
1 1 2(z- i)(z + i)
(96) w = - (z- 1)2 - (z + 1)2 =- (z2 - 1)2
Choose any simply connnected domain G that does not contain i and -i
(the zeroes of w). Then w belongs to 1t {(1, -1), (1, 1)). It is easy to verify
that the associated harmonic map u has two singularities of degree ±1: one
at the point 1, the other at the point -1, with opposite degrees.
(100} a=O.
inG
(102)
on &G.
{104) O<a<l.
and
i'" (e) z(z- a) (1- a cosO)- aisinO
e ... o - -
- lz- al - (1- 2acos0 + a2)1/2 ·
~-----'-:.--------::-:-:-=
5. The case G = B1 and g(O) = ediB with d ~ 2 97
Thus
1- acos6
cos cpo(O) = (1- 2a cos 9 + a2)1/2 > 0 VO E [-71', +?r]
and
a sinO
sinrpo(O) =- (1- 2acos9 + a2)1/2 ·
It follows that
tpo(fJ) > 0 VO E ( -1r, 0),
{
( 105 ) cp0 {8) < 0 VB E {0, +1r).
We solve {102) explicitly using Poisson's formula (see e.g., F. Treves [1])
and we find, in polar coordinates,
0) - (1 - r2) 1+11" tpo(u) dn
cp(r, - 21!" -1r l-2rcos(O-u)+r2 •
In particular, we deduce that
otp (r O) = _ r(1- r 2) 1+.,.. cp 0 (u) sin{O- u)dn
{}(J ' 1r _.,.. (1 - 2r cos(O- u) + r 2 ) 2
and thus
ocp (a 0) = a(1- a 1+11" 2) rpo(u) sinu dn .
{}(J ' 1r _.,.. (1- 2acosu + a 2 ) 2
In view of (105) we see that
rp0 (u)sinu<0 'rluE (-1r,+1r),u#O.
and therefore
{)cp
(106) {)(}(a, 0) < 0.
On the other hand we know, by Theorem VII.4 (see also Section VIII.2)
that
Vcp(a) =0
This contradicts (106) and hence (100) is proved.
Proof of Theorem VIII.6. Going back to (103) with a= 0 we see that
tpo = 0 on {)G and thus cp = 0 on G. It follows from (101) that
z
u*(z) = j;T in G \ {0}.
1
is convex and therefore Ee is strictly convex for e ;::: JX!' In fact, a
monotonicity argument shows that the corresponding Euler equation
in G
on 8G
For e > 0 sufficiently small there are infinitely many distinct minimizers
for problem {98}.
Proof. First we observe that there is an 8 1 -group action on the minimizers
of (98). For every a E R and every function u(z) set
Note that
E€(Rc.u) = E€(u) 'Va., 'Vu, 'Vc.
and if g = u1aa satisfies (108) then
on&G.
5. The case G = B1 and g(O) = ediO with d ~ 2 99
In order to show that problem (98) has infinitely many distinct minimizers
it suffices to find some minimizer that is not invariant under this group
action.
We argue by contradiction and assume that there is a sequence en --+ 0
such that every minimizer of E~n is invariant under Ra. By Theorem VI.l
we may assume (for a further subsequence) that Ue,. --+ u*. Hence u* is
also invariant under Ro. Recall that u* has precisely d singularities; since
d ~ 2, one of the singularities is not at the origin. This contradicts the
invariance of u* under Ra.
CHAPTER IX
The main ingredient in the proof of Theorem IX.l is the following result,
which is a particular case of a theorem due to P. Bauman, N. Carlson and
D. Phillips [1].
Theorem IX.2. Let G = B., and let g(O) = r(O)ei<Po(O), with <po increasing
from [0, 21r] onto [0, 21r] and r(O) > 0. Then, for every c there is a unique
zero of any minimizer u£ of EE on H~.
Proof of Theorem IX.l. We argue by contradiction; we assume that
there is a sequence u£n with en - 0, such that, for every n, u£" does not
have d zeroes of degree +1. By passing to a subsequence we may assume
that
(2)
(3) 'r/j.
(4)
with
cpj(fJ} = (} + Hj(aj + peiU ).
Hence
1
dcp· >-
_)
(5) d(} - 2"
(6)
Combining (5) and (7) we may assume that, for n sufficiently large,
dcpj,n > ~
(8)
dO - 4"
Applying Theorem IX.2 we deduce that U~:n has exactly one zero of degree
+1 in each disc Bp(aj). On the other hand we also know from {1) that, for
n sufficiently large,
(9)
Ee:(uor) = J(e:)
and that t ~---+ I (t) + 1r log t is nondecreasing (see Lemma 111.1). Hence
Proof of Theorem IX.3. We shall use the estimates of Lemma VIII. I and
Lemma VIII.2 in conjunction with Lemma IX.l. First, by Lemma VIII.l,
we know that there is some Po such that for every p < Po and every e: > 0,
we have
Hence
(12)
+0(p2).
Hence
(16)
As in the proof of Theorem V.l we deduce with the help of (V.7) that
(18)
(19)
(20)
104 Some additional properties of u.:
(23)
(24)
Recall that
ViE J
and that the collection of distinct points in (l;.);.eJ coincides with { a 1 , a 2 , ..• , ad}·
We have introduced, for every j = 1, 2, ... , d,
Aj = {i E J; xf - a;}
and we have proved that, for every j,
d;.E{O,+l} ViE J.
limsup
lx.e:"- xt:"l < oo
{26) k t
n--->oo en
Proof. We argue by contradiction. Assume that, for some j, there exist
k E A; and lEA; such that (passing to a further subsequence)
(27) lim
lxe-"
k
-x.e:"l
t = +oo.
n-oo C:n
lxk,n - Xt,nl
Pn = en
•
Vk,n(Y) = Uen (xk,n + C:ny),
Vt,n(Y) = Ue,.(Xt,n + C:ny).
These functions are defined for y E B(O, Pn/3) and they satisfy
As n ___, oo we have
{31)
Applying a result of H. Brezis, F. Merle and T. Riviere [1] (see also Ap-
pendix III at the end of the book) we know that every solution of (31)
satisfies
(32) r (lvl
JR2
2 - 1) 2 = 27rq2 with q = 0, 1, 2, ... '00.
106 Some additional properties of Ue
(33)
Combining (32) and (33) we see that one of the integrals (at least)
va.nishes-say, for example, the integral corresponding to vk.
On the other hand, going back to (23) we have
(34)
Impossible.
deg(u£,.,8B(xf",o:en)) = +1 Vi, Vn
and
x~"---. ai Vi= 1, 2, ... ,d.
CHAPTER X
Non-minimizing solutions
of the Ginzburg-Landau equation
(4)
The proof is exactly the same as the proof of Theorem 111.2. Estimate (4)
plays a crucial role in our analysis. Therefore we assume throughout
the rest of this chapter that G is starshaped.
We also recall
Lemma X.2. Any solution Ve: of (1)-{2} satisfies
(5} in G,
(6) in G,
(7)
(8)
(9) 21
c
l
B(z;,2.\ 0 e)nG
(1-lvel 2 )2 ?. Jl.o,
where N, .>.0 and Jl.o are positive constants {depending only on g and G}.
The bad discs B(x;:, >..0 c), with i E J may intersect. To avoid this un-
pleasant situation, we replace (as in Section IV.2) the bad discs B(xt:, .>.oc)
by slightly larger discs B(x;:, .>.e), with i E J' C J and.>.> >.o is a constant
depending on g and G, such that
109
We set
Wi = B(xi, A.e)
nl!: = a\ uiEJ' wi
Ge: = G \ UieK wi,
where
K = {i E J'; oGnwi =F 0}
and
L = J' \ K = {i E J'; Wi c G} .
Clearly, we have, by (6),
(10)
(11)
(13)
{14)
Note, however, that we cannot write (12) globally since P~r vanishes at
some points; the corresponding '1/J~r need not be well defined as a single-
valued function. To overcome this difficulty we proceed as follows.
110 X. Non-minimizing solutions of the Ginzbury-Landau equation
(18) l p;1
Ow,
~c
--=2·m5i
fJv
fori E L,
where
(19)
We recall that
(20)
by {8), and hence (15) is elliptic and (19) is well defined. Therefore tit~~"
exists and is unique. Moreover, ~~~" is obtained by minimizing
in the class
(21)
This yields
(22)
3. Study of the associated linear problems 111
Note that
and
since
that is,
(25) V X Vz 1 + ~z2 = p2 Hz 1 ,
(26) V X Vz 2 - ~z 1 = p2 Hz 2•
We claim that
(27) inn.
Indeed
(28)
by (1).
By (25) and (26), we have
(31)
and that
Combining (31) and (32) we are led to (30). Putting together (29) and (30)
we obtain
(33)
This estimate will play a crucial role. In what follows, we shall estimate
successively IVIfl.e!. ]V'H.el and jVp.!J in various norms.
(35)
The proof is exactly the same as the proof of Lemma 1.4 (see also Lemma
1.3}, and we shall omit it.
In what follows we shall present some properties of the solution cp of the
equation
(38} div(aVI)) =0 in n,
(39) ci> = Const. = ci on 8wi' i = 1, 2, ... ' n
(40) cl>=O on8G
{44)
and thus
(45)
and thus
J&wt
{ al~l
uv
= { a M8
lor-~t v
= 211".
(49) in 0
3. Study of the associated linear problems 115
(56) 1 Ow,
a 0 (=0
011
for i =F 2 and i :,#: 0
(58)
loa a:~= J&J2 a:~.
We denote by "! the solution of (52}-(56) corresponding to h + 1. By the =
maximum principle (as in the proof of Lemma 11.2) we easily see that
and
Thus
{62}
(63)
1
Note that
(64) 1
80
a -8~
8v ( = 21r((8wt) +
8w,
a -8~
8v h
1
and
(65) 1 80
a -88( ~
v
=
8w2
a -8&c ~(8w2).
v
Combining (62), (63), (64) and (65) we find
(66) lJfJIAI, a~: hi::; 211' (I(( awl) I+ 1-r(&wt}l) ::; 411'
by (59) and (60). Since (66) holds for every h satisfying lhl $ 1 we obtain
f
JfJIAI2
a I v I$
aq,& 411'
(68) 'Vt.p E V
where
(69)
(70)
(71)
(72)
Set
0( k) = {X E 0; ( > k}
118 X. Non-minimizing solutions of the Ginzburg-Landau equation
1/2
(73) aiJV((- k)+ I2~ ( f ihlz ) :c::; llhll 9 /l(k)(l/2)-(1/q)
Jn(k)
and consequently
(74)
Next, we use the Sobolev embedding wt· (G) c L (G) to deduce that
1 2
(75)
(76)
where
3 1
"'(=---.
2 q
Finally, we write
jj((- k)+jj 1 = - jk
r+oc (t- k) dp.(t) =H(k)
and we are led to the differential inequality
(78) H(k) =0
But
H{O) = IJ(+II 1 :c::; f31GI.., by (76).
Therefore
H(k) = 0 for k ;;::: {3 ('y ~ l} IGI"'-1 ,
4. The basic estimates 119
i.e.,
11<:+11
L
00 :5 .2...11hll
aS
'Y
q ('Y- 1)
IGI(l/2 )-(l/q)
{79)
(80)
(81)
inn£.
We will estimate each of the terms on the right-hand side of (33) separately.
120 X. Non-minimizing solutions of the Ginzburg-Landau equation
(82)
div(p~V'H£) = 0.
We claim that for i E L, we have
(83) Ia
8w;
p2
£
oH£-
av - 0.
For simplicity we now drop the subscript e. Recall (see (28)) that
f v x aav =O.
law, v
On the other hand, by (25} and {26) we have
av 2 aH
vx -
011
= p -
ov on Owi
Here we have used the fact that p ~ ! in 0 together with (25)-(26). Simi-
larly
(86)
4. The basic estimates 121
and
(87)
(88)
(89) 'ViE L,
and hence
(90)
Combining all the above estimates and the fact that eard L $ card J $
N (independent of e) we obtain
(92)
122 X. Non-minimizing solutions of the Ginzburg-Landau equation
Hence
(93)
by (10) and (11). Combining (91), (92) and {93) we are led to (82).
(95)
where d = LiEL l6il and A is some universal constant. Note that, by (90),
we have
(96) d5,C
and therefore
(97)
by (97).
4. The basic estimates 123
(98)
(101)
and
124 X. Non-minimizing solutions of the Ginzburg-Landau equation
(102)
= f P (1 - P2)p - p) - 2: f op {1 - p) + f op (t - p).
ln e iEL law, 8v lao 8v
Note that
(here we have used (10) a.nd (11)). As in the proof of Lemma X.12 we have
and hence
1
Since
-1
e2 G
(1- p) 2 :S c
we deduce that
{105)
(106)
(107)
We cannot exclude the possibility that some of the points (x~") converge
to the same limit. We denote by
with No~ Nt
(109)
Note that 4>e and He are only defined in f'Ze. Therefore, we extend 41e
in G by setting
41e = ci in Wi, Vi E L,
41e = 0 in G \ Gl!.
We extend HI! by its harmonic extension in Wi if i E L, and by the solution
He of
t!J.He- =o in winG,
He= He on 8w;nG,
8v =0
8He on 8G nwi,
cJ:le=O on ac.
and by Lemma X.ll
(110)
'Vi E J',
(111)
(112)
(114)
{115) Ia H. =0.
{117)
(118} ll.H. = 0 in G.
(119)
(120)
Hence
{121)
(122)
1 -lve
(124) 2 ..
12
-+IVv.l 2 L
in C'"(K) as n-+ +oo.
en
Proof. The proof is divided into three steps.
Step 1: We have, for any compact subset K o£ G \ U{ai},
(125)
(126)
and
For n sufficiently large, the support of (is inn~" and therefore we may
multiply (15) by ( (~~ .. - ci-*) and integrate on G. We obtain
and hence
.
-2 LP~ (VHt: .. VH. + LP~. (IVH.I 2 •
Note that
(134)
and
Using (125), (126), the fact that Pe _. 1 a.e. and Lebesgue's dominated
convergence theorem, we see that the right-hand side of (137) tends to zero
as n ---+ +oo. Hence
L (IVPe.,.l 2 _. 0 as n _. +oo.
(138)
(139)
130 X. Non-minimizing solutions of the Ginzburg-Landau equation
{140)
so that
(141)
{142)
(143)
{144)
{145) ( =1 in B(xo,R)
Multiplying (15) by (.P 10n and (116) by (.P* we obtain (using the fact that
.PE,. = .P* = 0 on 8G n B(xo, 2R)) as in Step 1 of the proof of Theorem X.2
5. tip;,. converges to v. 131
and hence
lJHe:., 8ve:,.
--=ve: x - - on aG~
8v " &v
(149)
+ o(1).
Since He:,. --+ H* weakly in H 1 (G), He:,. --+ H. weakly in H 112 (8G) and
therefore by the Sobolev embedding
(150)
(151)
{153)
132 X. Non-minimizing solutions of the Ginzburg-Landau equation
Combining (147), (152) and (153) we prove (144) as in Step 3 of the proof
of Theorem X.2. Finally (142) and (143) follow from boundary estimate
techniques developed in F. Bethuel, H. Brezis and F. Helein [2J.
Remark X.l. We emphasize that Theorems X.l, X.2 and X.3 are proved
under the assumption that G is starshaped. We do not know whether
the conclusion still holds for a general simply connected domain. Note,
however, that the conclusion may fail when 0 is not simply connected.
Here is a simple example. Take G = B 2 \ n where 0 is simply connected
but not convex and f2 c Bt. Let
9 ={ +1 on 8B2
-1 on an.
Let vE be a real-valued solution of
X.6. Properties of v.
We have already proved in Section X.5 that v* is a. smooth harmonic
map from G \ U{ai} into 8 1 • We are first going to show that all the points
{ai) lie in G and not on the boundary oG.
{157)
6. Properties of v* 133
(158)
(159) g = e*'(>o
where <po is some smooth real-valued function on 8GnB(O, R).
We ma.y assume that the tangent vector to ac at 0 is along the Xt-axiS and
we set
r+ = {x E 8GnB(O,R); Xt > 0}
r- = {x E 8GnB(O,R); Xt < 0}.
Choosing R still smaller if necessary we ma.y assume that r + and r _ are
connected. Combining (156), (158), (159) and the fact that vis smooth on
r + U r _ we deduce that there are two integers k+ and k._ such that, near
0,
We claim that
(161)
(162)
kanB(O,R) I::1< 2
oo.
134 X. Non-minimizing solutions of the GinzbufY-Landau equation
acp1 =0 in GnB(O,R),
Bcp 1 =h on 8GnB(O,R).
av
Using a Pohozaev-type identity one deduces that 'PI E H 1 (8G n B(O, R} ).
Combining this fact with (160) we see that Ca = 0, Va and that k+ = k_.
Proof of Lemma X.14 completed. By adding a constant to cpo and
using {161) we may now assume that
cp ='Po on 8GnB(O,R).
(163)
laG
r Iav£OV.. 12 s c.
Passing to the limit in (163) with the help of Theorem X.3 we deduce that
All the conditions of Lemma X.14 are satisfied and we may thus assert that
v* is smooth in some neighborhood of the boundary aa.
6. Properties of v. 135
(165)
where C is independent of r.
Here we have used the fact that
combined with Theorem X.3 and the estimate (163). We are led to a
contradiction since r in {166) is arbitrarily small.
Theorem X.5. The map v. is the canonical harmonic map associated to
(a;,~), that is,
(167)
{168)
136 X. Non-minimizing solutions of the Ginzbury-Landau equation
Proof. Since v* is in W 1·P( G), for 1 ~ p < 2 and satisfies (109), the analysis
of Chapter I shows that v* is the canonical harmonic map associated to the
singularities (ai, di), and thus (167) holds (see Corollary 1.2).
Next, we note that the results in Chapter VII can be carried over to our
situation, namely with u£,. replaced by v£ ... In particular, this yields (168)
and the fact that
From Chapter VIII, we know that (169) is equivalent to the fact that
(ai, di) is critical for W.
CHAPTER XI
Open problems
Er:(u) = ~ L 1Vul 2 + 4 ! fa
2 (lul 2 - 1)2 w(x)
(2)
(3)
Note that W.J·"( Gi S 1 ) :j: 0 even if deg(g, 8G) -::/:- 0. One knows (see B. Chen
and R. Hardt [1]) that every minimizer up of (3) has only a finite number
of singularities having degree +1 or -1. Does limp" . . . . 2 up., exist? Does it
2 After our work was completed, L. Almeida. [1), {2] has constructed a solution to (2)
everywhere discontinuous.
XI. Open Problems 139
have the same properties as u. =lim Uc., stated in Theorems 0.1, 0.2, 0.3,
0.4 and 0.5?3
Problem 10. Assume G = B 1 is the unit disc and let g(O) = e'8 on {}G.
Let Uc be a minimizer for Ec. Can one say that, for every e > 0, U~: has
the form
(4)
[We already know that {4) holds fore large; see Theorem VIII.7 in Section
VIII.5 .J
Same question if we assume only that U~: is a solution of the Ginzburg-
Landau equation {3) in the Introduction.
Problem 11. Assume G = B 1 and g(O) = e2i 8 • Let Uc be a minimizer for
{0.2). We know that for e large (see Section VIII.3) 'Ue has only one zero
(namely, x = 0) and fore small Ue has precisely two zeroes (see Section
IX.l). Is there some critical value e2 such that for e ~ £2, ue: has one
zero, and fore< e- 2 , Ue has two zeroes? Can one study this problem via a
bifurcation analysis? Same question for g(O) = ediB, with Ue having one
zero when e is large and d zeroes when e is small. What is the dividing
line ed? How does ed depend on at Is this phenomenon related to the
dividing line K. = 1/v"i between type II-superconductors (K. > 1/./2) and
type !-superconductors (K. < 1/...;2)?
Problem 12. Assume G = B1 and g(O) = ediJJ. We know that Ucn -+ u.
having exactly d singularities. Do the singularities of u. form a lattice
as d --+ +oo? Here, we may play with the two parameters: e --+ 0 and
d --+ +oo. We could first let d--+ +oo (for fixed e) and then let e --+ 0;
alternatively, we could let e--+ 0 and d--+ +oo simultaneously (with some
relation between e and d).
Problem 13. In the framework of Theorem 0.1, let Ue be_A minimizer for
Ee such that tte., --+ u •. Let Xn be a zero of Ue.,, i.e., ue.. (xn) = 0. Assume
Xn --+ a where a is a singularity of u.. Estimate the rate of convergence
lxn - al (as n--+ oo). Study the blow-up limit of ue,., i.e., set
and study the behavior of Vn as n --+ oo. This is related to our next problem:
3 After our work was completed, Problem 9 was answered positively by R. Hardt and
F. H. Lin (2)
140 XI. Open Problems
(5)
(6)
Note that for each integer d there exists a solution of {5)-(6} having the
form
ud(r, 0) = ediB !d(r).
Are they the only solutions of (5)-(6), modulo translation and rotation?
[Some results and further open problems concerning (5)-(6) are presented
in H. Brezis, F. Merle and T. Riviere [1]; see also R. M. Herve and M. Herve
[1] and I. Shafrir [I].J
Problem 15. Replace the energy Ee and the Dirichlet boundary condition
by the appropriate physical expressions arising in type II-superconductors
under an applied magnetic field H or in a bucket of superftuid rotated with
an angular velocity n. Study the asymptotics of minimizers as c -+ 0.
[Some results have been obtained in that direction for superconductors by
F. Bethuel and T. Riviere [1J.J
Problem 16. Study the minimization problem (2) of the Introduction in
the framework of DeGiorgi's r-convergence theory.
Problem 17. Asswne G c Rn, n ~ 3, is a smooth bounded domain and fix
a (smooth} boundary condition g : 8G - sn- 1 such that deg(g, 8G) =I 0.
Consider the "energy"
We recall here, for the convenience of the reader, the main results of
F. Bethuel, H. Brezis and F. Helein (2].
(1) deg(g,OO) = 0.
eir;:>o =g on an.
{2)
(3)
(4)
(5)
(6)
{7) Il l -l:e-12
e
-1Vuol211
Ck(K)
::; CK,k £2.
Next we consider the case where the boundary condition g also depends
on e. More precisely, we have a family of boundary conditions 9e : 00 -+ C
(not necessarily into 8 1 ), and we make the following assumptions
and
{12} deg(g,OS1) =0
(note that, by {10) and (11), g takes its values into S 1 ).
Theorem A.2. Let ue- be a minimizer of Ee in H;Jn,C). Under the
assumptions {8}-{12}, we have
{16) 1 -lue-1 2
-....!:2-=-- -+ IVuol2 m
. k
Gtoc(fl), 'Vk.
e
{
-.6-UE ::= c12UE(1 -lu£1 2) in fl,
UE = gf: on on.
If, for some reason, we know that uE converges strongly to some limit u 0 in
H 1 , then all the conclusions of Theorems A.1, A.2 and A.3 remain valid.
Again, the proofs follow essentially the same arguments as in the proofs of
Theorems A.l, A.2 and A.3.
APPENDIX II
Radial solutions
(1)
where J(r) = J~,d(r) is a function from (0, 1) into itself such that /(0} = 0.
Proof. Inserting (3) into (1) we are led to
(the first condition comes from the smoothness of v at 0 while the second
condition comes from (2)). The solutions of (4)-(5) arise as critical points
of the functional
Clearly,
Min <I> is achieved
v
and this yields a solution of (4)-(5). We may always assume that the
minimizer J ~ 0 (otherwise replace f by ffl). Similarly, we have f ~ 1;
otherwise replace J by min{!, 1} without increasing q,,
Remark A.2. The uniqueness of a solution of (4)-(5) is proved in
R. M. Herve and M. Herve [1). Alternatively one can also use-as was
pointed out by I. Shafrir-the method of H. Brezis and L. Oswald {1).
Namely, let h and h be two positive solutions of (4)-(5). Dividing (4) by
f and subtracting the corresponding equations we see that
(6)
1 (!f- ~:12)
1 2
rdr+ 1 (12- ~~~~)
1 2
rdr= -1 1
(fr-Ji) 2 rdr
We state here, for the convenience of the reader, the main quantization
result of H. Brezis, F. Merle and T. Riviere (1}.
Theorem A.5. Assume v : R.2 -+ C is a smooth function satisfying
(1)
Then,
(2)
(3) -I"-!!'+~
r r2
I= /{1- P> on (O,oCl)
(1) Vj
and
(2) 'Vj, k, j 1= k,
so that, in particular, Ro :$ R/4.
Set
0 = BR \ UJ=lB(p;,Ro)
and let u be a (smooth) map from 0 into C.
We make the following assumptions:
d; = deg(u, 8B(p;,Ro})
is well defined and we consider the "reference map"
and we have
We may then extend t/J inside Bt by a standard reflection and cut-off tech-
nique.
Proof of Theorem A.6. Set p =lui. We may write, locally inn (but
not globally in !1),
150 Appendix IV
and then
(8)
where V;(z) is the unit vector tangent to the circle ofradius lz-pil, centered
at p;,
y-p· x-p·
(10) V;(z) = (-1 z- Pi31' Iz- Pi31).
It is convenient to introduce the function 1/J globally defined on f2 by
(11)
Thus, we have
(12)
and consequently
with
ld;l 1
(14)
jVuol $ ~ lz -p;l $lldll~ lz- P;l'
3 3
so that
Appendix IV 151
(17)
1n
V<po · Vt/J = E. d; loz-p;
Y;·V'¢
I I"
J
We extend .,P inside each disc B(p;, Ro) using Lemma A.l and we write, for
each j,
{19)
f Vj . v.,p I Vj . vtjj 2: f Vj . vtjj
Jo [z- Pil = lBs\B(p;,Ro) lz- P;l- k:f'j Js(p,.,Ro) [z- P;l"
Note that, for k :F j
(21)
1s.. (P:~>
V; - 1s..
·V¢=
(p,;)
-81{; =0
aT
Ils"\B(p;~,Ro)
r Vj . v~ I I r
[z- P;l
=
JBR\B(p;.P;)
Vj . v¢ $ _!_
[z- Pil
I r
P; lsR\B(p;,P;)
IV.,PI
~ .!_JIV:;jjii2('~~'R2 -11'pJ~)l/2.
P;
152 Appendix IV
Hence we obtain
(22) I r
}B11 \B(p3 ,R.J)
Vj · v"iii 1 ::;
lz- Pi!
cnv¢112·
(23)
lo
r1Vul ~ lorIVPI + Jnr1Vuol + a2 }rIV'I/11
2 2 2 2
0
2 - lldll 22m2 (4K +C)
a
where Cis some Wliversal constant. This is the desired conclusion (6}.
An application. We will show how to prove that if Ue is a minimizer of
Ee then
{24)
(26)
Combining (27) and {28) with the upper bound of Theorem III.l we see
that
(29)
(30)
Recall that
{31}
We easily deduce (24) and (25) from (29), (30) and (31).
Bibliography
lattice, 139
liquid crystals, xix
singularities
location of the, xii
prescribed isolated, 10
penalization, xi
perforated domains, 31, 148
phase transition, xvii
Poincare's lemma, 4