COMPLEMENTS TO THE PAPER

“W 1,1 -MAPS WITH VALUES INTO S 1 ”

07/05/04

Haïm Brezis(1),(2) , Petru Mironescu(3) and Augusto C. Ponce(1),(2)

The purpose of these notes is to complement some of our results in [BMP]. We

also establish some of the claims we stated there without proof.
A. Extending Theorem 10 to other seminorms in W 1,1 (Ω; S 1 ).
In view of Theorem 10, it is natural to introduce the following quantity

1 n o
ρ(P, N ) = Inf [g]W 1,1 ; g ∈ W 1,1 (Ω; S 1 ), T (g) = 2π(δP − δN ) .
2π

Here, [ ]W 1,1 is a general given semi-norm on W 1,1 (Ω; R2 ) equivalent to | |W 1,1 . We

require from [ ]W 1,1 some structural properties :
(P1) [αg]W 1,1 = [g]W 1,1 , ∀ g ∈ W 1,1 (Ω; R2 ), ∀ α ∈ S 1 ;
(P2) [ḡ]W 1,1 = [g]W 1,1 , ∀ g ∈ W 1,1 (Ω; R2 ) ;
(P3) [gh]W 1,1 ≤ kgkL∞ [h]W 1,1 + khkL∞ [g]W 1,1 , ∀ g, h ∈ W 1,1 (Ω; R2 ) ∩ L∞ .
It follows from (P3) that ρ is a distance.

[We refer the reader to [Mi] for the solution of the exercise given at the end of
Section 3.]

Alternatively, we may define ρ starting from maps in R :

Lemma A1. We have
( ¯ )
1 ¯ g ∈ C ∞ (Ω \ {P, N }; S 1 ) ∩ W 1,1 ,
¯
ρ(P, N ) = Inf [g]W 1,1 ¯ .
2π ¯ deg (g, P ) = +1, deg (g, N ) = −1

Proof. It suffices to prove that, for g ∈ W 1,1 (Ω; S 1 ) such that T (g) = 2π(δP − δN ),
we may find a sequence (gn ) ⊂ R such that T (gn ) = 2π(δP − δN ) and gn → g

Typeset by AMS-TEX
1
2 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

in W 1,1 . Note that the desired conclusion is invariant with respect to orientation-
preserving diffeomorphisms of Ω. Therefore, we may assume that Ω = S 2 and that
(x, y)
P, N are respectively the North and the South pole of S 2 . Let h(x, y, z) = .
|(x, y)|
Then h ∈ R, T (h) = 2π(δP − δN ). It follows that k = g h̄ ∈ Y . Thus, we may
find a sequence (kn ) ⊂ C ∞ (Ω; S 1 ) such that kn → k in W 1,1 . Set gn = hkn . Then
gn ∈ R, T (gn ) = 2π(δP − δN ) and gn → g in W 1,1 .

Another simple property of ρ is

(A1) C1 d(P, N ) ≤ ρ(P, N ) ≤ C2 d(P, N )

for some C1 , C2 > 0 independent of P, N . This follows from the equivalence of

| |W 1,1 and [ ]W 1,1 .
Part of Theorem 10 holds in this more general setting :
Theorem A1. Let Pi , Ni ∈ Ω be such that
X X
ρ(Pi , Ni ) < ∞ (or, equivalently, d(Pi , Ni ) < ∞).

Set nX X X o
e = Inf
L ρ(Pej , N
ej ) ; (δPej − δNej ) = (δPi − δNi ) .

Then
1 n X o
1,1 1
Inf [g]W ; g ∈ W (Ω; S ), T (g) = 2π
1,1 e
(δPi − δNi ) ≤ L.
2π

Proof. Let ε > 0 and let Pej , N

ej be such that
X X X
(δPej − δNej ) = (δPi − δNi ) and ρ(Pej , N
ej ) ≤ L
e + ε.

By definition, for each j we may find some gj ∈ W 1,1 (Ω; S 1 ) such that

T (gj ) = 2π(δPej − δNej )

and
ej ) + ε .
[gj ]W 1,1 ≤ 2πρ(Pej , N
2j
n
We claim that there is a sequence kn → ∞ such that Πkj=1 n
gj → g in W 1,1 for some
g ∈ W 1,1 (Ω; S 1 ). By Lemma 1, this implies that
X X
T (g) = 2π (δPej − δNej ) = 2π (δPi − δNi ).
j i
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 3

Using (P3), we will also have [g]W 1,1 ≤ 2π Le + ε. Therefore, the conclusion of
Theorem A1 follows if we prove the existence
P of the sequence (kn ). We adapt below
an argument used in [BBM2]. Set H = j≥1 |∇gj | ∈ L1 (this uses the equivalence
¡ ¢¯
of | |W 1,1 and [ ]W 1,1 ). Since |∇ Πkj=1 gj ¯ ≤ H, ∀k, we may find a sequence
kn → ∞ and a map g ∈ BV (Ω; S 1 ) such that hn = Πkj=1 n
gj → g a.e. Then, for
m > n, we have

|hm − hn |W 1,1 = |hn (hm h̄n − 1)|W 1,1

≤ |hm h̄n |W 1,1 + k(1 − hm h̄n )∇hn kL1
km
X
≤ k∇gj kL1 + k(1 − hm h̄n )HkL1 = Am,n + Bm,n .
j=kn +1

Let 0 < δ < 1. Then, clearly, Am,n < δ provided m, n are sufficiently large. On the
other hand, Z
Bm,n ≤ δkHkL1 + 2 |H|.
{x:|1−hm (x)h̄n (x)|≥δ}

Note that
© ª n δo n δo
x; |1 − hm (x)h̄n (x)| ≥ δ ⊂ x; |g(x) − hm (x)| ≥ ∪ x; |g(x) − hn (x)| ≥ .
3 3

Since hn → g a.e., we find that Bm,n ≤ δ(kHkL1 + 1), provided m, n are sufficiently
large. Therefore, (hn ) is a Cauchy sequence in W 1,1 and converges to the above g
in W 1,1 .

It is not clear whether the reverse inequality in Theorem A1 is valid in general :

P
Open Problem 8. Let Pi , Ni ∈ Ω be such that T (g) = 2π i (δPi − δNi ). Is it
true that
e?
[g]W 1,1 ≥ 2π L
Note that, by definition, the answer is yes if T (g) = 2π(δP − δN ).
B. Proof of Theorems 10 , 30 , and 50 .
Proof of Theorem 3 0 . Let us first assume that g ∈ C ∞ (Ω\{a1 , . . . , ak }; S 1 ) ∩ W 1,1 .
It is then easy to see that

k
X Z
hT (g), ζi = 2π dj ζ(aj ) + (g ∧ gτ )ζ, ∀ζ ∈ Lip (Ω; R),
j=1 ∂Ω
4 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

where dj denotes the topological degree of g with respect to any small circle centered
at aj . In particular,
k
X
(B1) hT (g), ζi = 2π dj ζ(aj ), ∀ζ ∈ W01,∞ (Ω).
j=1
P
Note that, in general, j dj 6= 0. This means that we do not have necessarily the
same number of positive and negative points as before. In order to compensate this,
we insert points from ∂Ω into (B1). Since ζ = 0 on ∂Ω, equality in (B1) remains
true. We can then relabel the points aj as P1 , . . . , P` , N1 , . . . , N` , according to
their multiplicity dj , so that (B1) becomes

X̀
T (g) = 2π (δPj − δNj ) in W01,∞ (Ω).
j=1

For a general g ∈ W 1,1 (Ω; S 1 ), we argue by density using Lemma 2 to conclude

that X £ ¤∗
T (g) = 2π (δPi − δNi ) in W01,∞ (Ω) .

Note that dΩ induces a metric on the space Ω/∂Ω, where ∂Ω is identified with
a single point. Moreover, Lipschitz functions ζ on Ω/∂Ω with |ζ|Lip ≤ 1 and
ζ(∂Ω) = 0 correspond to elements in W01,∞ (Ω) such that k∇ζkL∞ ≤ 1. Applying
Lemma 120 in [BBM2] to Ω/∂Ω, we obtain
X
L(g) = Inf dΩ (Pi , Ni ).
i

Remark B1. The main new feature when Ω is a bounded domain in R2 is that
a minimal connection is made of segments from a positive singularity Pi to some
negative Nj , but we can also have line segments joining the singularities Pi , Ni to
the boundary ∂Ω. This is the analog of Example 3 in [BCL].
Proof of Theorem 1 0 . The proof of E(g) = Erel (g) is exactly the same as in Propo-
sition 2 and we shall omit it.
We are left to show that
Z
(B2) E(g) = |∇g| + 2πL(g).
Ω

Let ϕ ∈ BV (Ω; R) be such that g = eiϕ . Using Vol’pert’s chain rule as in the
proof of Lemma 5, we have
(B3) |Dϕ|M(Ω) = |g|W 1,1 + |g ∧ ∇g − Dϕ|M(Ω) .
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 5

We claim that

(B4) |g ∧ ∇g − Dϕ|M(Ω) ≥ 2πL(g).

In fact, for every ζ ∈ C0∞ (Ω) such that k∇ζkL∞ ≤ 1,

Z Z
⊥
|g ∧ ∇g − Dϕ|M(Ω) ≥ (g ∧ ∇g) · ∇ ζ − Dϕ · ∇⊥ ζ = hT (g), ζi.
Ω Ω

Taking the supremum with respect to ζ, we conclude that (B4) holds.

Inequality ≥ in (B2) follows immediately from (B3) and (B4).
We now establish ≤ in (B2). Let us assume for the moment that g is smooth
outside finitely many points a1 , . . . , ak , and that g has topological degree ±1 at
each one of those points. Let C be a minimal connection between those points with
respect to the distance dΩ .
Note that on any closed curve contained in Ω\C, g has zero topological degree.
We conclude that g has a smooth lifting ϕ on Ω\C. Moreover, as we cross any one
of the line segments of C, ϕ jumps by 2π. Thus, ϕ ∈ BV (Ω; R) and
Z Z Z
|Dϕ| = |∇g| + 2π|C| = |∇g| + 2πL(g).
Ω Ω Ω

We can now argue by density, using Lemma 2, to conclude that for any g ∈
W 1,1 (Ω; S 1 ) there exists ϕ ∈ BV (Ω; R) such that g = eiϕ a.e. in Ω and
Z Z
(B5) |Dϕ| ≤ |∇g| + 2πL(g).
Ω Ω

This concludes the proof of the theorem.

Proof of Theorem 5 0 . Using exactly the same argument as in the proof of Proposi-
tion 3, we have

(B6) |hDet (∇g), ζi| ≤ T V (g)kζkL∞ , ∀ζ ∈ C0∞ (Ω).

Thus, if T V (g) < ∞, then Det (∇g) ∈ M(Ω). We now apply Proposition 3.2 in [S]
(see also [P]) to the quotient space Ω/∂Ω. We conclude that there exist distinct
points a1 , . . . , ak ∈ Ω and nonzero integers d1 , . . . , dk such that

k
X
(B7) Det (∇g) = π d j δ aj .
j=1
6 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

We now define
µ ¶−d1 µ ¶−dk
x − a1 x − ak
h(x) := ··· g(x) for a.e. x ∈ Ω.
|x − a1 | |x − ak |

Clearly, Det (∇h) = 0 in D0 (Ω). It follows from the analog of Theorem 7 for
domains in R2 (see also [D]) that h has a lifting in W 1,1 . In other words, we can
find ϕ ∈ W 1,1 (Ω; R) such that h = eiϕ a.e. in Ω. We then conclude that
µ ¶d1 µ ¶dk
x − a1 x − ak
g(x) = ··· eiϕ(x) for a.e. x ∈ Ω.
|x − a1 | |x − ak |

Arguing as in the proof of Theorem 5, this implies that

k
X
(B8) T V (g) ≤ π |dj |.
j=1

The reverse inequality already follows from (B6). We then conclude that (6.8)
holds.
Conversely, if Det(∇g) ∈ M(Ω), then (B7) holds. The above argument then
shows that T V (g) < ∞ and

k
X
T V (g) = | Det (∇g)|M = π |di |.
i=1

C. Proof of Theorem 3 0000 .

Theorem 30000 follows immediately from Theorem 3 and the next
Lemma C1. Given g ∈ W 1,1 (S 2 ; Γ), we define

g − aj
gj := ∈ W 1,1 (S 2 ; S 1 ).
|g − aj |

Then,

1X
(C1) Det (∇g) = |Aj | Det (∇gj ) in D0 (S 2 ).
π j

The proof of Lemma C1 relies on the following

 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 7

Lemma C2. For any u ∈ W 1,1 (S 1 ; Γ), we have

Z X
1 u − aj
(C2) u ∧ uτ = |Aj | deg .
2 S1 j
|u − aj |

Proof of Lemma C2.

Step 1. Γ is a simple curve.
It is well-known that (C2) holds if u ∈ C 1 (S 1 ; Γ) (see, e.g., [N]). By approximation,
we conclude that (C2) is also true for any u ∈ W 1,1 (S 1 ; Γ).
Step 2. Assume Γ has finitely many self-intersections, say q1 , . . . , qk .
Since u is continuous, the set

S 1 \u−1 ({q1 , . . . , qk })

is open and can be written as a countable union of open arcs in S 1 . Let α1 be

such an arc. It is easy to see that we can select disjoint arcs α2 , . . . , αj (oriented
anticlockwise) such that u at the positive endpoint of αi coincides with the value
of u at the negative endpoint of αi+1 for i = 1, . . . , j, with the convention that
αj+1 = α1 . By removing arcs from this list if necessary, we can assume that each
point qi appears only twice in the list
© ª
u(∂α1 ), . . . , u(∂αj ) .

This construction induces a function ũ ∈ W 1,1 (S 1 ; Γ) such that

(a) ũ = u on α1 ∪ · · · ∪ αj ;
(b) ũ is locally constant on S 1 \α1 ∪ · · · ∪ αj .
By construction, ũ(S 1 ) is a subset of a Jordan curve Γ̃ contained in Γ. Let
At1 , . . . , At` be the components of R2 \Γ enclosed by Γ̃. By our first step, we have
Z
1 ¯ ¯ ũ − a
(C3) ũ ∧ ũτ = ¯At1 ∪ · · · ∪ At` ¯ deg .
2 S1 |ũ − a|

for some point a inside Γ̃. Note, however, that

ũ − a ũ − ats
deg = deg , ∀s = 1, . . . , ` ;
|ũ − a| |ũ − ats |
moreover,
ũ − aj
deg = 0, if j 6∈ {t1 , . . . , t` }.
|ũ − aj |
8 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

We can thus rewrite (C3) as

Z X̀ X
1 ũ − ats ũ − aj
(C4) ũ ∧ ũτ = |Ats | deg = |Aj | deg .
2 S1 s=1
|ũ − ats | j
|ũ − aj |

We can proceed with the construction of ũ and “decompose” u ∈ W 1,1 (S 1 ; Γ) as

ũ1 , ũ2 , . . . so that
(a) ũi ∈ W 1,1 (S 1 ; Γ) and ũi (S 1 ) is contained in some Jordan curve for every i ;
(b) uτ = ũ1τ + ũ2τ + · · · in S 1 ;
(c) ũi coincides with u on finitely many arcs in S 1 \u−1 ({q1 , . . . , qk }) and ũi is
locally constant outside those arcs.
By (C4), we have
Z X
1 ũi − aj
ũi ∧ ũiτ = |Aj | deg , ∀i.
2 S1 j
|ũi − aj |

Note that, by (b) and (c),

Z XZ
u ∧ uτ = ũi ∧ ũiτ .
S1 i S1

For the same reason,

u − aj X ũi − aj
deg = deg , ∀j.
|u − aj | i
|ũi − aj |

We conclude that (C2) holds.

Proof of Lemma C1. Let g ∈ W 1,1 (S 2 ; Γ). By the coarea formula (see [BBM2]), we
have
Z µZ ¶
(C5) hDet (∇g), ζi = g ∧ gτ dλ,
R Σλ
© ª
where ζ ∈ C ∞ (S 2 ), and Σλ = x ∈ S 2 ; ζ(x) = λ is equipped with the appropriate
orientation, whenever λ is a regular value of ζ.
Recall that, for a.e. λ ∈ R, g|Σλ belongs to W 1,1 . Applying Lemma C2 to g|Σλ
for such λs we get
Z X Z
1X
g ∧ gτ = 2 |Aj | deg gj = |Aj | gj ∧ gjτ .
Σλ j
π j Σλ
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 9

Integrate both sides of the identity above with respect to λ. Using (C5), we conclude
that
1X
hDet (∇g), ζi = |Aj |hDet (∇gj ), ζi, ∀ζ ∈ C ∞ (S 2 ).
π j

This establishes (C1).

We also call the attention of the reader to the following analog of Lemma 120 in
[BBM2] :
Proposition C1. Let X be a metric space.
P Given two sequences (Pi ), (Ni ) in X
and nonnegative numbers αi such that i αi d(Pi , Ni ) < ∞, let
X £ ¤∗
(C6) T = αi (δPi − δNi ) in Lip(X) .
i

Define
L= Sup hT, ζi.
ζ∈Lip(X)
|ζ|Lip ≤1

Then,
X
(C7) L = Inf αi d(Pi , Ni ),
i

where the infimum is taken over all sequences (Pi ), (Ni ) in X and numbers αi ≥ 0
such that (C6) holds.
Proof. Let us denote by L̃ the infimum in (C7). Clearly, L ≤ L̃. We now establish
the reverse inequality.
Let ε > 0. We take k ≥ 1 sufficiently large so that
X
αi d(Pi , Ni ) < ε.
i>k

Without loss of generality, we can assume that each αi is rational for i = 1, . . . , k.

We then choose an integer J ≥ 1 sufficiently large so that Jαi is an integer for every
i = 1, . . . , k. Write the points Pi , Ni as p1 , p2 , . . . and n1 , n2 , . . . , with multiplicity
Jαi . It follows from Lemma 4.2 in [BCL] that we can find ζ0 ∈ Lip(X), with
|ζ0 |Lip ≤ 1, such that, after relabeling the points nj if necessary, we have :

k
X £ ¤ X£ ¤ X
Jαi ζ0 (Pi ) − ζ0 (Ni ) = ζ0 (pj ) − ζ0 (nj ) = d(pj , nj ).
i=1 j j
10 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

Thus,
∞
X £ ¤
L≥ αi ζ0 (Pi ) − ζ0 (Ni )
i=1
k
X £ ¤
≥ αi ζ0 (Pi ) − ζ0 (Ni ) − ε
i=1
1X X1 X
= d(pj , nj ) − ε ≥ d(pj , nj ) + αi d(Pi , Ni ) − 2ε.
J j j
J
i>k

Note that X1 X
T = (δpj − δnj ) + αi (δPi − δNi ).
j
J
i>k

We conclude that L ≥ L̃ − 2ε. Since ε > 0 was arbitrary, this implies L ≥ L̃. Thus,
L = L̃ as claimed.

D. Proof of Theorems 100 and 200 .

We begin with a few preliminary results
Lemma D1. Given ε > 0, let Φε : B̄ N → B̄ N be defined as

0 if |x| ≤ ε,
Φε (x) = |x| − ε x
 if ε < |x| ≤ 1.
1 − ε |x|

Then, for every f ∈ C ∞ (B̄ N ; RM ), we have

f ◦ Φε → f in W 1,N (B N ).

r−ε
Proof. Given x ∈ B N , |x| ≥ ε, let r = |x| and ψε (r) = (1−ε)r . Using this notation,
we have ³x x ´
i j 0
Φ0ε (x) = ψε (r) Id + ψε (r) .
r
ε
Since ψε0 (r) = (1−ε)r 2 ,
¯x x ¯
¯ i j 0 ¯ Cε
¯ ψε (r)¯ ≤ .
r r
Moreover,
Cε
|Φε (x) − x| ≤ Cε and |ψε (r) − 1| ≤ .
r
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 11

We then have
¯ ¯ ¯ ¯
¯∇fε (x) − ∇f (x)¯ = ¯ t Φ0ε (x) ∇f (Φε (x)) − ∇f (x)¯
¯ ¯ ¯ ¯¯ ¯
≤ ¯∇f (Φε (x)) − ∇f (x)¯ + ¯∇f (Φε (x))¯ ¯ Id −t Φ0ε (x)¯
Cε Cε
≤ C|Φε (x) − x| + C|ψε (r) − 1| + ≤
r r

for |x| ≥ ε. Therefore,

Z Z Z
N N dx
|∇fε − ∇f | ≤ Cε + |∇f | → 0 as ε → 0.
BN |x|N
ε≤|x|≤1 |x|<ε

Next, we establish the following

Lemma D2. Given f ∈ C ∞ (B̄ N × [0, 1]; RM ), let

fε (x, t) = f (Φε (x), t).

Then
fε → f in W 1,N (B N × [0, 1]).

Proof. Note that


 ∂f ∂f
∂fε ∂f  (0, t) − (x, t) if |x| < ε,
(x, t) − (x, t) = ∂t ∂t
∂t ∂t  ∂f (Φ (x), t) − ∂f (x, t)

if |x| ≥ ε.
ε
∂t ∂t

Thus, since |Φε (x) − x| ≤ Cε, we have

¯ ∂f ∂f ¯¯
¯ ε
¯ − ¯ ≤ Cε.
∂t ∂t

The result now immediately follows from Lemma D1.

As a consequence of Lemma D2 above, any map f ∈ C ∞ (B̄ N × [0, 1]; S k ) can be
approximated in W 1,N by maps fε such that

fε = f on S N −1 × [0, 1] and fε (x, t) = gε (t) if |x| ≤ ε.

12 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

Lemma D3. Given g ∈ C ∞ ([0, 1]; RM ), let

f (x, t) = g(t) ∀(x, t) ∈ B̄ N × [0, 1].

Then, there exists a sequence (fε ) in W 1,N such that

fε (x, t) = const if |x| ≤ ε, fε (x, t) = g(t) if |x| = 1,

and
fε → f in W 1,N (B N × [0, 1]).

Proof. Let ζε : B̄ N → R be given by

 |x|

 log ε if ε ≤ |x| ≤ 1,
ζε (x) = log 1ε


0 if |x| < ε.

It is easy to see that

Z Z
N
|∇ζε | →0 and |ζε − 1|N → 0.
BN BN

Since N ≥ 2, there exists Q ∈ S N such that Q 6∈ g([0, 1]). Let Ψ : S N \{Q} → RN

denote the stereographic projection. Set F = Ψ◦f . Clearly, in order to establish the
lemma, it suffices to approximate F in W 1,N by a sequence Fε such that |Fε | ≤ C,

Fε (x, t) = const if |x| ≤ ε and Fε (x, t) = F (x, t) if |x| = 1.

Set G(t) = Ψ ◦ g(t) and Fε (x, t) = ζε (x)G(t). We then have

ZZ Z Z
N N
|∇x Fε − ∇F | dx dt = |∇ζε | |G|N → 0

and ZZ Z Z
N N
|∂t Fε − ∂t F | dx dt = |1 − ζε | |G0 |N → 0.

The proof of Lemma D3 is complete.

We conclude from Lemma D3 that, given any u ∈ W 1,N −1 (S N ; S N −1 ), N ≥ 3,
there exists a sequence (un ) such that un → u in W 1,N , where each un satisfies the
following properties :
(i) un has a finite number of point singularities Pi , Ni ;
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 13

(ii) un is homogeneous of degree 0 in a neighborhood of each singularity ;

(iii) un ≡ Q in some conic neighborhood of a geodesic joining Pi and Ni .
In fact, (ii) holds since every topological singularity may be approximated by
homogeneous singularities (see Lemma E.5 below). We then apply Lemmas D.2
and D.3 to obtain property (iii).
We shall say that a map is good if it satisfies properties (i)–(iii).
Proof of Theorem 2 00 . Our goal is to show that, for every u ∈ W 1,N −1 (S N ; S N −1 ),
Z
(D1) Inf |D(u) − D(v)| = σN L(u).
v∈C ∞ (S N ;S N −1 ) S N

Proof of “≥”. For every ζ ∈ Lip (S N ) such that k∇ζkL∞ ≤ 1, we have

Z Z Z
£ ¤
|D(u) − D(v)| ≥ D(u) − D(v) · ∇ζ = D(u) · ∇ζ = hT (u), ζi.
SN SN SN

Taking the supremum with respect to ζ, we get

Z
|D(u) − D(v)| ≥ σN L(u) ∀v ∈ C ∞ (S N ; S N −1 ).
SN

Proof of “≤”. Assume u is a good map. We shall assume for simplicity that u has
a single dipole P, N . Given ε > 0, let Uε : S N −1 → S N −1 be such that
Z
ε N −1
Uε ≡ Q if |x − P | ≥ and |∇T Uε |N −1 ≤ σN (N − 1) 2 + ε.
2 S N −1

The existence of such map is established in [BCL, Section VIII].

Let W denote the ε-conic neighborhood of the geodesic segment joining P and
N . We decompose W as W = W1 ∪ W2 , where W1 is the cylindric part of W , and
W2 is the union of the two conic caps. We then define uε as

uε (x) = u(x) if x 6∈ W, uε (x) = Uε (x) if x ∈ W1

and uε is extended by homogeneity of degree 0 in W2 . We then have

Z Z
|D(u) − D(uε )| = |D(Uε )| + o(1).
SN W1

In [BCL, Section VIII], it is proved that

1
|D(Uε )| = |∇T Uε | ≤ N −1 |∇T Uε |N −1 .
(N − 1) 2
14 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

We conclude that
Z
(D2) lim |D(u) − D(uε )| ≤ σN L(u).
ε→0 SN

On the other hand, since uε has only singularities of degree 0, Hopf’s theorem
W 1,N −1
implies that uε ∈ C ∞ (S N ; S N −1 ) . It then follows from (D2) that inequality
“≤” in (D1) holds, at least when u is a good map.
We now establish ≤ in (D1) for any map u ∈ W 1,N −1 (S N ; S N −1 ). Let (un ) be
a sequence of good maps such that un → u in W 1,N −1 . For each n ≥ 1, we have
just shown that there exists vn : S N → S N −1 smooth such that
Z
1
|D(un ) − D(vn )| ≤ σN L(un ) + .
SN n

Thus
Z Z
1
|D(u) − D(vn )| ≤ |D(u) − D(un )| + σN L(un ) + = σN L(u) + o(1),
SN SN n

which gives the desired result.

Proof of Theorem 1 00 . We want to show that
Z
N −1
(D3) Erel (u) = |∇u|N −1 + (N − 1) 2 σN L(u).
SN

Proof of “≤”. It suffices to establish the result for good maps. In fact, if un → u
in W 1,N −1 , then
Erel (u) ≤ lim inf Erel (un ),
n→∞

while the right-hand side of (D3) is continuous with respect to the strong topology
in W 1,N −1 . Thus, we may assume that u is good and we can proceed exactly as in
the proof of Theorem 200 . We shall leave the details to the reader.
Proof of “≥”. As in [BCL, Section VIII], we have

³P ´ N2−1
2
¯ ¯ |w| j |vj |
(D4) ¯(w · v2 ∧ · · · ∧ vN , w · v1 ∧ v3 ∧ · · · ∧ vN , . . . )¯ ≤ N −1
(N − 1) 2

for every w, v1 , . . . , vN ∈ RN .
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 15

On the other hand, given a sequence (un ) ⊂ C ∞ (S N ; S N −1 ) such that (un ) is

bounded in W 1,N −1 and un → u a.e., we have
Z Z Z
N −1 N −1
(D5) |∇un | = |∇u| + |∇un − ∇u|N −1 + o(1).
SN SN SN

Let ζ ∈ Lip (S N ) be such that k∇ζkL∞ ≤ 1. Applying (D4) to w = un and

v = ∇un − ∇u, we get
Z
|∇un − ∇u|N −1 X Z
SN
N −1 ≥ εσ un · (unxσ2 − uxσ2 ) ∧ · · · ∧ (unxσN − uxσN ) ζxσ1 ,
(N − 1) 2 σ S N

where the sum is taken with respect to all permutations σ of N elements, and
εσ = ±1 denotes the sign of σ. Thus,
Z Z
N −1
N −1
(D6) |∇un − ∇u| ≥ −(N − 1) 2 D(u) · ∇ζ + Rn (ζ).
SN SN

Assume for the moment that Rn (ζ) → 0. It follows from (D5) and (D6) that
Z Z
N −1
N −1
lim inf |∇un | ≥ |∇u|N −1 + (N − 1) 2 hT (u), ζi.
n→∞ SN SN

Taking the supremum over ζ, we obtain “≤” in (D3).

Thus, in order to conclude the proof of Theorem 100 , we need to show that
Rn (ζ) → 0. Since each un is smooth,
Z
D(un ) · ∇ζ = 0.
SN

It then follows that

Rn (ζ) = Sn (ζ) + o(1),
where Sn (ζ) is a sum of integrals of the form
Z
In = ± un · vnxi1 ∧ · · · ∧ vnxik ∧ uxj1 ∧ · · · ∧ uxjl ζxt ,
SN

where k + l = N − 1 and vn = un − u. It is important to notice that k ≥ 1 and

l ≥ 1. Since
N −1
(un − u) uxj1 ∧ · · · ∧ uxjl → 0 strongly in L l ,
16 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

we may replace In by
Z
± u · vnxi1 ∧ · · · ∧ vnxik ∧ uxj1 ∧ · · · ∧ uxjl ζxt .
SN

We can now formally integrate by parts to write Sn (ζ) as a sum of integrals of the
form Z
± uxi1 · vn ∧ vnxi2 ∧ · · · ∧ vnxik ∧ uxj1 ∧ · · · ∧ uxjl ζxt .
SN

Such integration by parts can be easily justified by approximation ; note that all
the second derivatives are cancelled by symmetry.
Since
N −1
vn uxj1 ∧ · · · ∧ uxjl → 0 strongly in L l+1 ,

we conclude that
Rn (ζ) = Sn (ζ) + o(1) = o(1).

The proof of Theorem 10 is completed.

E. Proof of Theorem 500 .

Throughout this section, we assume that Ω = ∂G, where G is a domain in RN ;
more generally, Ω could be any smooth domain (with boundary) in ∂G. We start
with
Lemma E1. Let 0 < σ < ∞, 1 ≤ q < ∞. Then

W σ,q W σ,q
C ∞ (Ω; S N −1 ) ∩ W σ,q = ∞
C (Ω; S N −1 ) .

In other words, for each map u ∈ C ∞ (Ω; S N −1 ) ∩ W σ,q , there is a sequence (un ) ⊂
C ∞ (Ω; S N −1 ) such that un → u in W σ,q .
© ª
Proof. Let, for t > 0 sufficiently small, Ωt = x ∈ Ω ; d(x, ∂Ω) > t . Consider, for
any such t, a diffeomorphism Φt : Ω → Ωt such that

kDk Φ − Dk IdkL∞ ≤ Ck t, k = 0, 1, 2 . . .

and set, for u ∈ C ∞ (Ω; S N −1 ), ut = u ◦ Φt . Then

ut ∈ C ∞ (Ω; S N −1 ) and ut → u in W σ,q .

 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 17

Lemma E2. Let u ∈ W 1,N (Ω; S N −1 ). Then

W 1,N
L(u) = 0 ⇐⇒ u∈ C ∞ (Ω; S N −1 ) ∩ W 1,N .

Here, Z
hT (u), ζi = − D(u) · ∇ζ,
Ω

where ζ ∈ W 1,∞ (Ω) and ζ is constant on each connected component of ∂Ω ; L(u)

is computed accordingly. When N = 3, this result is due to Bethuel [B1] ; the same
proof yields Lemma E2.
µ ¶
1,p N −1 N −1 x
Lemma E3. Let 1 ≤ p < N . For g ∈ W (S ;S ), set g̃(x) = g for
|x|
x ∈ B N . Then g̃ ∈ W 1,p and the map

g ∈ W 1,p (S N −1 ; S N −1 ) 7−→ g̃ ∈ W 1,p (B N ; S N −1 )

is continuous and verifies |g̃|W 1,p ≤ C|g|W 1,p .

Proof. Trivial computation.
Lemma E4. In the definition of T V , we may replace C ∞ (Ω; RN )-maps by maps
in Lip (Ω; RN ).
Proof. Clear, by approximation.
Lemma E5. Let N ≥ 2, N − 1 ≤ p < N , and let u ∈ W 1,p (Ω; S N −1 ). Fix
a1 , . . . , ak ∈ Ω and define, for ρ > 0 sufficiently small,


 u(x), if d(x, {a1 , . . . , ak }) ≥ ρ

 u(y), if, for some j, d(x, a ) < ρ and
j
uρ (x) =

 x lies on the geodesic segment from


aj to y, where d(aj , y) = ρ

(here, d denotes the geodesic distance in S N .)

Then uρn → u for some sequence ρn → 0.
Proof. For simplicity, we may assume Ω is flat near each aj . Then the definition of
uρ becomes

 u(x)
µ ¶
if d(x, {a1 , . . . , ak }) ≥ ρ,
uρ (x) = x − aj
u ρ if |x − aj | < ρ.
|x − aj |
18 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

Alternatively, denoting by uρj (y) = u(aj + ρy), y ∈ S N −1 , then

µ ¶
x − aj
uρ (x) = uρj if |x − aj | < ρ.
|x − aj |

Assume, for simplicity, that there is only one singularity, say a1 = 0.

1 1
Let, for n ≥ 1, ρn be such that ≤ ρn ≤ and
2n n
Z Z 1 µZ ¶ Z
1 p
n
p
|∇u| ≤ |∇u| dρ = |∇u|p .
2n Sρ n 1
2n |x|=ρ 1 1
2n ≤|x|≤ n

Then
Z Z Z
p p n→∞
|∇uρn | ≤ CN,p ρn |∇u| ≤ C |∇u|p −→ 0.
1 1
Bρn Sρ n 2n ≤|x|≤ n

Thus
Z Z
p
|∇uρn − ∇u| = |∇uρn − ∇u|p
Ω Bρn
µZ Z ¶
p p n→∞
≤C |∇uρn | + |∇u| −→ 0.
Bρn Bρn

Proof of Theorem 5 00 . The equivalence

T V (u) < ∞ ⇐⇒ Det (∇u) is a measure

is established as in the proof of Theorem 5. As already noted, T V (u) < ∞ implies

σN X
Det (∇u) = (δPi − δNi ).
N
finite

Let a1 , . . . , ak be the collection of points Pi , Ni . Given n ≥ 1, let

½ ¾
n 1
Ω = x ∈ Ω ; d(x, {a1 , . . . , ak }) >
n

and ½ ¾
n 1 2
A = x∈Ω; ≤ d(x, {a1 , . . . , ak }) ≤ .
n n
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 19

Consider a sequence (um ∞ n

n ) ⊂ C (Ω ; S
N −1
) such that m
µ un ¶→ u in W
1,N −1
as
1 2
m → ∞ (clearly, L(u|Ωn ) = 0). There is some ρn ∈ , such that, up to a
n n
subsequence in m,
(i) umn |Σρn → u|Σρn in W
1,N −1
(Σρn ) as m → ∞,
Z Z
(ii) |∇u|N −1 ≤ Cn |∇u|N −1 .
Σ ρn An
© ª
Here, Σρ = x ; d(x, {a1 , . . . , ak }) = ρ .
Extend umn to Ω as in Lemma 5 (by homogeneity of degree 0) ; let (un )ρn = ũn
m m

be this extension. By (i) and Lemma E5, we have ũm n → uρn in W

1,N −1
(Ω) as
1,N −1
m → ∞. By (ii) and Lemma E3, we have uρn → u in W (Ω). Thus we may
find a sequence (vn ) ⊂ Liploc (Ω \ {a1 , . . . , ak }; S N −1 ) such that
(a) vn is homogeneous of degree 0 near each aj ;
(b) vn → u in W 1,N −1 (Ω) ;
(c) near each aj , the degree of vn is dj .
Assertion (c) follows from (i), the continuity of degree of maps from S N −1 into
N −1
S for W 1,N −1 convergence and the following clear fact
X X
T (vn ) = σN deg (vn , aj )δaj * T (u) = σN dj δaj .
j j

For the remaining part of the proof, assume for simplicity that there is only one
singularity a = 0 of degree d > 0 and that Ω is flat near a. Let ρn be such that vn
is homogenous of degree 0 in Bρn (0). Here, Bρ (0) is a ball in Ω centered at a = 0.
Fix d distinct points p1 , . . . , pd in B1 .
Let ε > 0 be sufficiently small. For S w ∈ Lip (∂BN1 ;−1 S N −1 ) (B1 is the unit ball
N
in R ), with deg w = d, let w̃ : B1 \ j Bε (pj ) → S be such that w̃|∂B1 = w,
x − pj
w̃(x) = if |x − pj | = ε, j = 1, . . . , d and w̃ is a Lipschitz function. (Such a
|x − pj |
x−p
map exists, since deg w = d.) We then extend w̃ to B1 by setting w̃(x) = ε j , if
|x − pj | ≤ ε. Thus w̃ is still Lipschitz.
Define, for 0 < ρ < ρn ,

 vn (x),
µ ¶
if d(x, 0) > ρ
vn,ρ (x) = x ;
 w̃n , if d(x, 0) ≤ ρ
ρ
20 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

here, wn (x) = vn (ρn x), if |x| = 1. Lemma E5 yields

vn,ρ → vn in W 1,N −1 as ρ → 0.

Clearly, by the definition of vn,ρ , we have

Z
| Jac vn,ρ | = ωN d.
Ω

Considering now the case of several singularities, we obtain by a diagonal procedure

a sequence (wn ) ⊂ Lip (Ω; RN ) such that
Z X
1,N −1
wn → u in W and | Jac wn | = ωN |dj |.
Ω j

F. Proof of Theorems 5000 .

We start with the following well-known
Lemma F1. Let 1/p < s < 1 and 1 < p < ∞. Given u ∈ W01,p (RN
+ ), let
½
u(x) if xN > 0,
ũ(x) =
0 if xN ≤ 0.

Then, u 7→ ũ is a continuous mapping from W01,p (RN

+ ) into W
1,p
(RN ).
Proof. By density, it suffices to deal with u ∈ C0∞ (RN + ). Using the Besov seminorm,
we have
Z ∞Z Z
p p p |ũ(x0 , t + τ ) − ũ(x0 , t)|p 0
(F1) |ũ|W s,p ∼ |u|W s,p + kukLp + dx dt dτ.
0 R RN −1 τ 1+sp
Denote by I the last term in the right-hand side of (F1). Clearly, it suffices to
estimate I :
(F2) Z Z ∞Z ∞ Z Z ∞
|u(x0 , t) − u(x0 , σ)|p 0 |u(x0 , t)|p
I∼ 1+sp
dσ dt dx + sp
dt dx0 .
R N −1 0 0 |t − σ| R N −1 0 t
It then suffices to estimate the last term in (F2). By Fubini, it suffices to consider
the 1-dimensional integral
Z ∞
|u(t)|p
sp
dt, u ∈ C0∞ (0, ∞).
0 t
This is a trivial consequence of the following
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 21

Lemma F2. Let 1/p < s < 1 and 1 < p < ∞. Given u ∈ C0∞ (0, ∞), we have
Z ∞ Z ∞Z ∞
|u(t)|p |u(x) − u(y)|p
(F3) dt ≤ C dx dy.
0 tsp 0 0 |x − y|1+sp

Proof. We first point out that both integrals are finite. Given 0 < α < β < 1, we
have
|u(t)| ≤ |u(x)| + |u(t) − u(x)| ∀x ∈ [αt, βt] =: It .
Thus,
|u(t)|p ≤ 2|u(x)|p + C|u(t) − u(x)|p ∀x ∈ It .
Integrating over x ∈ It , we get
Z ∞ Z
p 2 p C
(F4) |u(t)| ≤ |u(x)| dx + |u(t) − u(x)|p dx.
(β − α)t 0 (β − α)t It

Since α, β < 1, we have |t − x| ∼ t for every x ∈ It . Integrating (F4) with respect

to t we then have
Z ∞ Z ∞Z Z ∞Z ∞
|u(t)|p 2 |u(x)|p C |u(t) − u(x)|p
dt ≤ dx dt + dx dt
0 tsp β − α 0 It tsp β−α 0 0 |t − x|1+sp
Z Z ∞Z ∞
2(β sp − αsp ) ∞ |u(x)|p C |u(t) − u(x)|p
(F5) = dx + dx dt.
sp(β − α) 0 xsp β−α 0 0 |t − x|1+sp
sp sp
Note that βsp(β−α)
−α
= γ sp−1 for some γ ∈ [α, β]. Thus, by taking β ¿ 1/2 sufficiently
small, we have
2(β sp − αsp ) 1
≤ .
sp(β − α) 2
With such choice, (F3) trivially follows from (F5).
Lemma F3. Let ω1 , Ω be two smooth domains, ω1 ⊂⊂ Ω. Set ω2 = Ω\ω 1 . Assume
that un → u in W s,p (ω1 ) and vn → v in W s,p (ω2 ), with tr un = tr vn on ∂ω1 . Let
½ ½
un in ω1 u in ω1
wn = and w = .
vn in ω2 v in ω2

Then
wn → w in W s,p (Ω).

Proof. It suffices to show that

¡ ¢
kwkW s,p (Ω) ≤ C kukW s,p (ω1 ) + kvkW s,p (ω2 ) .
22 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

Let η = tr u = tr v. By the standard trace theory,

(i) kηkW s−1/p,p ≤ CkukW s,p (ω1 ) ;
(ii) kηkW s−1/p,p ≤ CkvkW s,p (ω2 ) ;
(iii) there exists an extension g ∈ W s,p (Ω) of η to ω1 and ω2 such that

kgkW s,p (Ω) ≤ CkηkW s−1/p,p .

Let ½
u−g in ω1 ,
w̃ =
v−g in ω2 .
By Lemma F1, w̃ ∈ W s,p (Ω) and
¡ ¢
kw̃kW s,p (Ω) ≤ C kukW s,p (ω1 ) + kvkW s,p (ω2 ) .

Since w = w̃ + g a.e. in Ω, Lemma F.3 follows.

In the sequel, we shall denote by C the cube (−1, 1)N . Let kxk∞ = Maxi {|xi |}.
Lemma F4 (Brezis-Mironescu [BM1]).¡ Let 0 <¢ s < 1 and 1 < p < ∞, with
sp < N . Given f ∈ W s,p (∂C), set f˜(x) = f x/kxk∞ , x ∈ C. Then, f˜ ∈ W s,p (C)
and the mapping
f 7→ f˜

is continuous from W s,p (∂C) into W s,p (C).

We refer the reader to [BM1, Lemma D.1] for a proof of Lemma F4.
We denote by Cε = (−ε, ε)N . The following lemma is a variant of the approxi-
mation procedure in [BM2]:
Lemma F5. Let 0 < s < 1 and 1 < p < ∞, with sp < N . Let f ∈ W s,p (RN ).
Given ε > 0 and Q ∈ Cε , set
½
f (x) if x 6∈ Cε (Q),
fε,Q (x) =
f (πε,Q (x)) if x ∈ Cε (Q),

x−Q
where πε,Q (x) = Q + ε kx−Qk ∞
is the projection of x to ∂Cε (Q), with respect to Q.
Then, there exist εn → 0 and Qn ∈ Cεn /2 such that

fεn ,Qn → f in W s,p (RN ).

 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 23

Proof. Set gε,Q = fε,Q − f . We have

Z Z
|fε,Q (x) − f (x)|p
|gε,Q |pW s,p =2 dy dx+
RN \Cε (Q) Cε (Q) |x − y|N +sp
Z Z
|gε,Q (x) − gε,Q (y)|p
+ dx dy
Cε (Q) Cε (Q) |x − y|N +sp
Z
|fε,Q (x) − f (x)|p
∼ dx+
Cε (Q) d(x, ∂Cε (Q))sp
Z Z
|fε,Q (x) − fε,Q (y)|p
+ dx dy + o(1)
Cε (Q) Cε (Q) |x − y|N +sp
=: Iε,Q + Jε,Q + o(1).

It suffices to show that

Z 1 ½Z ¾
dε
 (Iε,Q + Jε,Q ) dQ < ∞.
0 ε Cε

The proof of Z Z
1
dε
 Jε,Q dQ < ∞
0 ε Cε
may be found in [BM2, Appendix A]. Next,
Z 1 Z Z 1 Z Z
dε dε |fε,Q (x) − f (x)|p
 Iε,Q dQ ≤ dx  sp
dQ.
0 ε Cε 0 ε C2ε Cε (x) d(x, ∂Cε (Q))

We now make the change of variables Q = x − y, y ∈ Cε . We get

Z 1 Z Z 1 Z Z |f (x) − f (x − y + ε y )|p
dε dε kyk
 Iε,Q dQ ≤ dx  y sp dy.
0 ε Cε 0 ε C2ε Cε |y − ε kyk |

y
Let z = −y + ε kyk , whose Jacobian is O(( kzkεN −1 )) (see [BM2]). We then have
Z 1 Z Z 1 Z Z
dε dε |f (x) − f (x + z)|p N −1
 Iε,Q dQ ≤ C dx  ε dz
0 ε Cε 0 ε C2ε Cε |z|N +sp−1
Z Z Z 1
|f (x) − f (x + z)|p dε
≤C dx N +sp−1
dz 2
C2 C1 |z| |z| ε
ZZ
|f (x) − f (x + z)|p
≤ dx dz < ∞.
|z|N +sp
24 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

This concludes the proof of Lemma F5.

Given N − 1 < p ≤ N , any map u ∈ W (N −1)/p,p (S N ; S N −1 ) has a harmonic
extension U ∈ W N/p,p (B N +1 ; RN ) ⊂ W 1,N . We then define T (u) as
N Z
X
hT (u), ζi = − det (Ux1 , . . . , Uxj , u, Uxj+1 , . . . , UxN +1 ) ξxj ,
j=1 B N +1

where ζ ∈ Lip (S N ) and ξ is any extension of ζ to B N +1 . One can see that this
definition is independent of the extension ξ. Let
1
L(u) = Max hT (u), ζi.
σN k∇ζkL∞ ≤1
We have the following
Lemma F6. Assume N ≥ 2. Let 1 < p < ∞ and u ∈ W (N −1)/p,p (S N ; S N −1 ). If
W (N −1)/p,p
T (u) = 0, then u ∈ C ∞ (S N ; S N −1 ) .
Proof.
Case 1. Proof of the lemma if N ≥ 3.
Note that good maps are dense in W 1,N −1 (S N ; S N −1 ) and, by interpolation, in
W (N −1)/p,p (S N ; S N −1 ). Thus, it suffices to show that if u is a good map, then
there exists v ∈ C ∞ (S N ; S N +1 ) such that

ku − vkW 1,N −1 ≤ CL(u),

which can be done by a dipole construction. By interpolation, we obtain Lemma F6.

Case 2. Proof of the lemma if N = 2.
The interpolation argument does not work in this case. However, for any map
u ∈ R, the dipole construction in [BBM2] gives a sequence (vn ) such that

T (vn ) = T (u), |vn |pW 1/p,p ≤ CL(u), and vn → 1 a.e.

W 1/p,p
Clearly, uv̄n ∈ C ∞ and
ZZ
|u(x) − u(y)|p
|u − uv̄n |pW 1/p,p ≤ 2|1 − v̄n |pW 1/p,p +K |1 − v̄n (x)|p dx dy
|x − y|2
≤ 2CL(u) + o(1).

Using the density of R in W 1/p,p (S 2 ; S 1 ), we obtain the desired result.

We also point out the following extension of Lemma F6 whose proof is left to
the reader :
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 25

Lemma F7. Assume N − 1 < p ≤ N . Let Ω be a smooth subdomain of S N .

For any u ∈ W (N −1)/p,p (S N ; S N −1 ), T (u|Ω ) is well-defined when computed against
Lipschitz functions which are constant on each connected component of ∂Ω. If
W (N −1)/p,p
T (u|Ω ) = 0, then u|Ω ∈ ∞
C (Ω; S N −1 ) .
A key ingredient in the proof of Theorem 5000 is the following
Proposition F1. Assume N − 1 < p ≤ N . Let u ∈ W (N −1)/p,p (S N ; S N −1 ) be
such that X
T (u) = σN d i δM i .
finite

Then, there exist (un ) ⊂ W (N −1)/p,p (S N ; S N −1 ), εn → 0, and Min → Mi such that

S
(i) un ∈ C ∞ (S N \ i Cεn (Min ); S N −1 ) ;
(ii) un is homogeneous of degree 0 on each cube Cεn (Min ) ;
X
(iii) T (un ) = σN di δMin ;
finite

(iv) un → u in W (N −1)/p,p .
Proof. We first observe that the set of good pairs (ε, Q), in the sense of Lemma F5,
is “fat”. More precisely, there exists a sequence εn → 0 such that
¯©
¯ ª¯¯
¯ Q ∈ C εn /4 ; |f εn ,Q − f | W (N −1)/p,p < 1/n ¯ 1
≥ .
|Cεn /4 | 2
© ª
In particular, Q ∈ Cεn /4 ; |fεn ,Q − f |W (N −1)/p,p < 1/n intersects the complement
of any null set of Cεn /4 . For n ≥ 1 fixed, consider

vn = u|S N \ Si Cεn /4 (Mi ) .

Then, T (vn ) = 0, so that there exists a sequence (vnk )k ,

¡ [ ¢
vnk ∈ C ∞ S N \ Cεn /4 (Mi ); S N −1 ∀k ≥ 1,
i

such that
¡ [ ¢
vnk → vn in W (N −1)/p,p S N \ Cεn /4 (Mi ) as k → ∞.
i
26 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

By Fubini, for a.e. Q ∈ Cεn /4 , we have

vnk |Si ∂Cεn (Q+Mi ) → u|Si ∂Cεn (Q+Mi ) in W (N −1)/p,p as k → ∞.

By Lemmas F3 and F4, for any such Q we have

ṽnk → ũn in W (N −1)/p,p as k → ∞,

where ṽnk (resp., ũn ) is vnk (resp., u) extended by homogeneity of degree 0 on each
cube Cεn (Q + Mi ). By Lemma F5, we can choose Q = Qn such that ũn → u in
W (N −1)/p,p . Applying a diagonalization argument, un may be taken among (ũkn ).
We only need to show that (iii) holds. S
Note that ṽnk is locally Lipschitz on S N \ i Min , where Min = Qn + Mi . Thus,
X
T (ṽnk ) = σN d˜i δMin .
finite

Since T (un ) * T (u), for n large enough we have (iii). This concludes the proof of
Proposition F1.
We may now present the
Proof of Theorem 5 000 .
Step 1. If T V (u) < ∞, then Det (∇u) is a measure and

(F6) | Det (∇u)|M ≤ T V (u).

Clearly, we may replace un in the definition of T V (u) by

(
un if |un | ≤ 1,
ũn = un
|un | if |un | > 1.

We may thus assume that |un | ≤ 1. Since un → u in W (N −1)/p,p and |un | ≤ 1, we

have
hT (un ), ζi → hT (u), ζi
for every ζ ∈ Lip (S N ; S N −1 ). In addition,
Z
hT (un ), ζi = N Jac un ζ ≤ T V (u)kζkL∞ + o(1).
S N −1

Thus, T (u) is a measure and (F6) holds.

 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 27

Step 2. If Det (∇u) is a measure, then T V (u) < ∞ and

1
(F7) T V (u) = number of topological singularities of u.
ωn

By Proposition F1, it suffices to compute T V (u), where u is smooth outside

finitely many (disjoint) cubes and u is homogeneous of degree 0 inside each one of
these cubes. By (F6), we have “≥” in (F7). It then suffices to show the reverse
inequality.
Note that u ∈ W 1,q (S N ; S N −1 ) for every q < N . As in the proof of the case
W 1,N −1 (S N ; S N −1 ), we can find (un ) ⊂ C ∞ , un → u in W 1,q , with
Z
1
| Jac un | = number of topological singularities of u.
ωN S N −1

For N − 1 < q < N , we have W 1,q ∩ L∞ ⊂ W (N −1)/p,p , so that un → u in

W (N −1)/p,p . We conclude that (F7) holds.

G. Proofs of Theorems 14–17.

We start by establishing the precise value of E(g) :
Lemma G1. Let g ∈ BV (I; S 1 ) and let A be the set of jump points of g. Then
X
(G1) E(g) = |ġd | + dS 1 (g(a+), g(a−)).
a∈A

Proof. Let ϕ ∈ BV (I; R) be any lifting of g. We claim that

(G2) |ϕ̇d |M(I) = |ġd |M(I) .

Indeed, recall that, by the chain rule, we have

ϕ̇d = −iḡ ġd .

Set ν = ġd and µ = |ν|. Then there is some k ∈ L∞ ((I, dµ); S 1 ) such that ν = kµ.
Since ν is diffuse and −iḡ ∈ BV , we have −ig ∈ L∞ ((I, dµ); S 1 ), and thus ϕ̇d = `µ,
where ` = −iḡk ∈ L∞ ((I, dµ); S 1 ). It follows that

|ϕ̇d |M(I) = Sup hϕ̇d , ζi = Sup hµ, `ζi = hµ, |`|i = |µ|M(I) = |ġd |M(I) .
ζ∈C0 (I;C) ζ∈C0 (I;C)
|ζ|≤1 |ζ|≤1
28 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

Let now B denote the set of the jump points of ϕ; clearly, B ⊃ A. For each
a ∈ A, we have (
g(a+) = eiϕ(a+) ,
g(a−) = eiϕ(a−) ,
so that
|ϕ(a+) − ϕ(a−)| ≥ dS 1 (g(a+), g(a−)).
If a ∈ B \ A, then eiϕ(a+) = eiϕ(a−) ; thus

|ϕ(a+) − ϕ(a−)| ≥ 2π.

Consequently,
X X
|ϕ̇|M(I) = |ϕ̇d |M(I) + |ϕ(a+) − ϕ(a−)| + |ϕ(a+) − ϕ(a−)|
a∈A a∈B\A
X
(G3) ≥ |ġd | + dS 1 (g(a+), g(a−)) + 2π card (B \ A)
a∈A
X
≥ |ġd | + dS 1 (g(a+), g(a−)).
a∈A

Equality holds if and only if

|ϕ(a+) − ϕ(a−)| = dS 1 (g(a+), g(a−)) ∀a ∈ A and B = A.

This proves “≥” in (G1).

In order to prove “≤”, we split A as A = A1 ∪ A2 , where
© ª © ª
A1 = a ∈ A; |g(a+) − g(a−)| = 2 and A2 = a ∈ A; |g(a+) − g(a−)| < 2 .

If a ∈ A2 we may define a signed distance

µ ¶
g(a+)
δS 1 (g(a+), g(a−)) = arg .
g(a−)

Here, arg stands for the argument in (−π, π). Set

X X
(G4) µ = −iḡ ġd + π δa + δS 1 (g(a+), g(a−)) δa .
a∈A1 a∈A2

We claim that µ is a measure. Indeed, A1 is finite, since g ∈ BV . On the other

hand,
Sup |g(a+) − g(a−)| = d < 2
a∈A2
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 29

(again since g ∈ BV ). Thus there is some C > 0 such that

¯ ¯
¯δS 1 (g(a+), g(a−))¯ ≤ C|g(a+) − g(a−)|.

It follows that
X ¯ ¯ X
¯δS 1 (g(a+), g(a−))¯ ≤ C |g(a+) − g(a−)| < ∞.
a∈A2 a∈A2

Assume that I = (0, α) for some α > 0 and set ϕ0 (x) = µ((0, x)), x ∈ I. We claim
that, up to a constant, ϕ0 is a lifting of g and that |ϕ̇0 |M(I) = E(g).
Indeed, using the chain rule for a product we have

˙ X¡ ¢
(G5) ge−iϕ0 = e−iϕ0 ġd − ig e−iϕ0 (ϕ̇0 )d + ge−iϕ (a+) − ge−iϕ0 (a−) δa .
a∈A

Here, we have used the fact that ϕ is continuous outside A. For a ∈ A, we have

ϕ0 (a+) = ϕ0 (a−) + µ({a}),

so that
g(a−)
e−iϕ0 (a+) = e−iϕ0 (a−) ,
g(a+)
by our definition of µ. Thus the sum in (G5) vanishes. On the other hand,

(ϕ̇0 )d = −iḡ ġd ,

so that
e−iϕ0 ġd − ige−iϕ0 (ϕ̇0 )d = e−iϕ0 (ġd − ġd ) = 0.
Thus, there is some C ∈ C such that ϕ = ϕ0 + C is a lifting of g.
On the other hand,
X ¯ ¯
|ϕ̇|M(I) = |ϕ̇0 |M(I) = | − iḡ ġd |M(I) + π card (A1 ) + ¯δS 1 (g(a+), g(a−))¯
a∈A2
X
= |ġd |M(I) + dS 1 (g(a+), g(a−)).
a∈A

The proof of Lemma G1 is complete.

Proof of Theorem 14. We shall prove a slightly stronger assertion, which implies
all the properties stated in the theorem. Namely, a lifting ϕ ∈ BV (I; R) of g is a
canonical lifting if and only if
30 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

(i) ϕ and g have the same jump sets ;

(ii) for a ∈ A1 , we have ϕ(a+) − ϕ(a−) = ±π ;
(iii) for a ∈ A2 , we have ϕ(a+) − ϕ(a−) = δS 1 (g(a+), g(a−)).
Property (i) was seen to be necessary for optimality in the proof of Lemma G1.
Recall that, in addition to (i), equality in (G3) amounts to

(G6) |ϕ(a+) − ϕ(a−)| = dS 1 (g(a+), g(a−)) ∀a ∈ A.

If a ∈ A1 , then
|ϕ(a+) − ϕ(a−)| = π,
so that (ii) holds. Assume a ∈ A2 . Since eiϕ(a+) = g(a+) and eiϕ(a−) = g(a−),
then by (G6) we have
µ ¶
g(a+)
ϕ(a+) − ϕ(a−) = arg ,
g(a−)

which gives (iii). Conversely, it is easy to see that, if (i)–(iii) are fullfilled, then
equality holds in (G3).
Proof of Theorem 15. We identify S 1 \ {z} with an interval I. Let A, A1 , A2 be
defined as in the proof of Lemma G1. We claim that, for each choice of integers
εa ∈ {−1, 1}, a ∈ A1 , there is a canonical lifting ϕ of g on I such that

ϕ(a+) − ϕ(a−) = εa π ∀a ∈ A1 .

This ϕ is obtained as in the proof of Lemma G1, as ϕ = µ((0, x)) + C, x ∈ I. One

simply has to modify the definition of µ by taking
X X
µ = −iḡ ġd + εa πδa + δS 1 (g(a+), g(a−)) δa .
a∈A1 a∈A2

Moreover, the proof of Theorem 14 shows that, by this procedure, we obtain all
canonical liftings. We claim that if ϕ is the canonical lifting corresponding to the
choice εa , a ∈ A1 , and ϕ̃ the one corresponding to ε̃a , a ∈ A1 , then
ϕ̃(z−) − ϕ̃(z+) ϕ(z−) − ϕ(z+) 1 X
= + (ε̃a − εa ) ∈ Z.
2π 2π 2
a∈A1

If we identify S 1 \ {z} with I = (0, α), α > 0, this amounts to proving that

ϕ̃(α−) − ϕ̃(0+) ϕ(α−) − ϕ(0+) 1 X

= + (ε̃a − εa ).
2π 2π 2
a∈A1
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 31

We have ϕ̃(α−) − ϕ̃(0+) = µ̃(I) (where µ̃ is the corresponding measure); a similar

assertion holds for ϕ. Thus

ϕ̃(α−) − ϕ̃(0) ϕ(α−) − ϕ(0+) (µ − µ̃)(I) 1 X

− = = (ε̃a − εa ).
2π 2π 2π 2
a∈A1

Let ϕ be the canonical lifting corresponding to the choice εa = −1, a ∈ A1 . Then

it is clear that, with d = ϕ(z−)−ϕ(z+)
2π and k = card A1 , we have

Deg1 g = {d, d + d, . . . , d + k}.

We next prove that d is an integer. This follows easily from the fact that eiϕ(z−) =
iϕ(z+)
e , so that ϕ(z−) − ϕ(z+) ∈ 2πZ.
It remains to establish that Deg1 g does not depend on the choice of z. Let w
be any other continuity point of g. Let ψ be a canonical lifting of g on S 1 \ {w}.
Since g is continuous at w, there is some k ∈ Z such that ψ(w−) = ψ(w+) + 2kπ.
We set ½
ψ(ξ) if ξ ∈ (w+, z−),
ϕ(ξ) =
ψ(ξ) − 2kπ if ξ ∈ (z+, w−).
Clearly, ϕ ∈ BV and ϕ is continuous at w. It is obvious that

|ϕ̇|M(S 1 \{z}) = |ψ̇|M(S 1 \{z,w}) = |ψ̇|M(S 1 \{w}) .

It follows that ϕ is a canonical lifting of g on S 1 \ {z}. Indeed, by Lemma G1 we

have
E(g|S 1 \{z} ) = E(g|S 1 \{w} ) = |g|BV S 1
if z, w are continuity points of g. Since

ϕ(z−) − ϕ(z+) = ϕ(z−) − ϕ(w+) + ϕ(w+) − ϕ(w−) + ϕ(w−) − ϕ(z+)

= ψ(w−) − ψ(w+),

we find that the degrees obtained by cutting at z are among the ones obtained by
cutting at w. By reversing the roles, we obtained that Deg1 g is independent of z.
Proof of Theorem 16. Let z be a continuity point of g and let ϕ be a canonical
lifting of g in S 1 \ {z}. Assume, e.g., that z = 1; we identify S 1 \ {1} with (0, 2π).
Consider a sequence (ϕn ) ⊂ C ∞ ([0, 2π]) such that
Z 2π
|ϕ̇n | → |ϕ̇|M((0,2π)) and ϕn → ϕ a.e.
0
32 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

We may assume, in addition, that

ϕn (0) → ϕ(0+) and ϕn (2π) → ϕ(2π−).

(This is the case, e.g., if the functions ϕn are obtained from ϕ by mollification).
By replacing ϕn with ϕn + αn x + βn , for some appropriate αn → 0 and βn → 0,
we may further assume that

ϕn (0) = ϕ(0+) and ϕn (2π) = ϕ(2π−) ∀n ≥ 1.

Set gn = eiϕn . Then, clearly,

ϕ(1−) − ϕ(1+)
gn ∈ C ∞ (S 1 \ {z}) ∩ C 0 (S 1 ) and deg gn = .
2π

By further mollifying gn , we find a sequence (hn ) ⊂ C ∞ (S 1 ; S 1 ) such that

ϕ(1−) − ϕ(1+)
hn → g a.e., deg hn = ∀n ≥ 1,
2π
and Z Z
|ḣn | → |ϕ̇| = E(g|S 1 \{z} ).
S1 S 1 \{z}

It follows that
Deg2 g ⊃ Deg1 g.
Conversely, let d ∈ Deg2 g and let (gn ) ⊂ C ∞ (S 1 ; S 1 ) be such that
Z
gn → g a.e., |ġn | → |g|BV S 1 and deg gn = d ∀n ≥ 1.
S1

Let z ∈ S 1 be a continuity point of g. Write, in S 1 \ {z}, gn = eiϕn . Then

Z
|ϕ̇n | → |g|BV S 1 .
S1

Up to some subsequence and after subtracting a suitable multiple of 2π, we may

assume that ϕn → ϕ a.e., where ϕ ∈ BV is a lifting of g. Since |ϕ̇|M(S 1 \{z}) ≤
|g|BV S 1 , we find that ϕ has to be a canonical lifting of g. Let ε > 0. There is
some δ > 0 such that, if I is the interval of size δ centered at z, then we have
|g|BV S 1 (I) < ε. We may further assume that g is continuous at the endpoints of I.
Then
|g|BV S 1 (S 1 ) = |g|BV S 1 (I) + |g|BV S 1 (S 1 \I) .
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 33

Arguing as above, we find that

Z Z
|ϕ̇n | → |g|BV S 1 (S 1 \I) and |ϕ̇n | → |g|BV S 1 (I) .
S 1 \I I\{z}

In particular, for every n ≥ 1 sufficiently large,

¯ ¯
¯(ϕn (z − t) − ϕn (z + t)) − (ϕn (z−) − ϕn (z+))¯ < 2ε ∀t ∈ (0, δ).

We pick such t so that, in addition,

ϕn (z − t) → ϕ(z − t) and ϕn (z + t) → ϕ(z + t).

We then find ¯ ¯
¯ϕ(z − t) − ϕ(z + t) − 2πd¯ ≤ 2ε.

Letting ε → 0 and δ → 0, we obtain

ϕ(z−) − ϕ(z+) = 2πd,

i.e., d ∈ Deg1 g.
We complete the proof of Theorem 16 by proving that g 7→ Deg g is continuous
in the multivalued sense. Since Deg is Z-valued, this amounts to proving that, for
each d ∈ Z, the set © ª
g ∈ BV (S 1 ; S 1 ) ; d ∈ Deg g
is open.
To this purpose, we start with the following
Lemma G2. Let g ∈ BV (I; S 1 ). Let ϕ ∈ BV (I; R) be a lifting of g. If ϕ is not a
canonical lifting of g, then
|ϕ̇|M(I) ≥ E(g) + π.

Proof. Let A, B be the set of jump points of g, ϕ, respectively. Recall that

X X
|ϕ̇|M(I) ≥ |ġd |M(I) + |ϕ(a+) − ϕ(a−)| + [ϕ(a+) − ϕ(a−)|
a∈A a∈B\A
X
≥ E(g) + |ϕ(a+) − ϕ(a−)|.
a∈B\A

If B 6= A, then
|ϕ(a+) − ϕ(a−)| ≥ 2π ∀a ∈ B \ A,
34 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

and the conclusion is clear. If B = A, then there is some a ∈ A such that

|ϕ(a+) − ϕ(a−)| > dS 1 (g(a+), g(a−)),

for otherwise ϕ would be a canonical lifting. For any such a, we have

|ϕ(a+) − ϕ(a−)| ≡ dS 1 (g(a+), g(a−)) mod 2π.

Since dS 1 (g(a+), g(a−)) ≤ π, we find that

|ϕ(a+) − ϕ(a−)| ≥ dS 1 (g(a+), g(a−)) + π.

Since, for any b ∈ A \ {a}, we have

|ϕ(b+) − ϕ(b−)| ≥ dS 1 (g(b+), g(b−)),

and the conclusion follows.

Proof of Theorem 16 completed. If g, h ∈ BV (S 1 ; S 1 ), then

(G7) |g h̄|BV = |g h̄ − 1|BV = |g (h̄ − ḡ)|BV ≤ |g|BV kh − gkL∞ + |h − g|BV .

Let g ∈ BV (S 1 ; S 1 ) and let d ∈ Deg g. In view of (G7), there is some ε > 0 such
that if
h ∈ BV (S 1 ; S 1 ) and kg − hkBV < ε,
then
1
|g h̄|BV <.
10
We claim that d ∈ Deg h for any such h. Indeed, let z be a continuity point for
both g and h, and let ϕ be a canonical lifting of g in S 1 \ {z}. Set k = ḡh and let
1
ψ be a canonical lifting of k. Since |k|BV < 10 , each jump point a of k is such that
1
|k(a+) − k(a−)| < 10 . Thus

|ψ(a+) − ψ(a−)| ≤ 2|k(a+) − k(a−)|

for any such a. It follows that

X 1
|ψ|BV = |k̇d |M(I) + |ψ(a+) − ψ(a−)| ≤ 2|k|BV < .
5
jump points
of k

Set φ = ϕ + ψ. Then φ is a lifting of h and

¯ ¯
¯(φ(z−) − φ(z+)) − (ϕ(z−) − ϕ(z+))¯ < 2 ,
5
 COMPLEMENTS TO THE PAPER “W 1,1 -MAPS WITH VALUES INTO S 1 ” 07/05/04 35

so that
φ(z−) − φ(z+) = ϕ(z−) − ϕ(z+)
(since both quantities are multiple of 2π). In order to complete the proof of Theo-
rem 16, it suffices to prove that φ is a canonical lifting of h.
Indeed, on the one hand we have

1
E(h) ≤ |φ̇|M(S 1 \{z}) ≤ |ϕ̇|M(S 1 \{z}) + ,
5

so that E(h) ≤ E(g) + 15 . By reversing the roles, we obtain on the other hand that
E(g) ≤ E(h) + 51 ; thus

2
E(h) ≤ |φ̇|M(S 1 \{z}) ≤ E(h) + .
5
Lemma G2 implies that φ is a canonical lifting of h.
Proof of Theorem 17. With the notation we already used, we have
© ª
g ; Deg g is single-valued = {g ; A1 (g) = φ} =: U1 .

Thus, we have to prove that U1 is dense in BV (S 1 ; S 1 ).

Let g ∈ BV (S 1 ; S 1 ); then A1 (g) is finite. If A1 (g) = φ, then g ∈ U1 . Otherwise,
we may assume, for simplicity, that A1 consists of a single point, say A1 = {1}; the
general case can be treated along the same lines. We have |g(1−) − g(1+)| = 2.
Without loss of generality, we may assume that g(1−) = −1 and g(1+) = 1.
Given ε > 0, let hε : S 1 → S 1 be given by
 iθ
e
 if 1 ≤ θ ≤ 2π − 1,
iθ i(ε+(1−ε)θ)
hε (e ) = e if 0 ≤ θ ≤ 1,

 i((2π−1)(2π+ε)−(2π−1+ε)θ)
e if 2π − 1 ≤ θ < 2π.

It is immediate that hε (1+) = eiε , hε (1−) = e−iε ,

|hε |BV → 0 and hε → 1 uniformly.

Thus,
ghε → g in BV as ε → 0.
On the other hand, since hε ∈ C 0 (S 1 \{1}), we have A1 (ghε )\{1} = A1 (g)\{1}. In
particular, A1 (ghε ) ⊂ {1}. Since, by construction, 1 ∈
/ A1 (ghε ), we have ghε ∈ U1 .
The proof of Theorem 17 is complete.
36 HAÏM BREZIS(1),(2) , PETRU MIRONESCU(3) AND AUGUSTO C. PONCE(1),(2)

(1)LABORATOIRE J. -L. LIONS

UNIVERSITÉ P. ET M. CURIE, B.C. 187
4 PL. JUSSIEU
75252 PARIS CEDEX 05
E-mail address: brezis@ccr.jussieu.fr, ponce@ann.jussieu.fr

(2)RUTGERS UNIVERSITY
DEPT. OF MATH., HILL CENTER, BUSCH CAMPUS
110 FRELINGHUYSEN RD, PISCATAWAY, NJ 08854
E-mail address: brezis@math.rutgers.edu, augponce@math.rutgers.edu

(3) DÉPARTEMENT DE MATHÉMATIQUES

UNIVERSITÉ PARIS-SUD
91405 ORSAY
E-mail address: Petru.Mironescu@math.u-psud.fr