Removable Set GG

http://www.elsevier.
com/locate/aim
Advances in Mathematics 183 (2004) 155182
The GaussGreen theorem and removable sets
for PDEs in divergence form
Thierry De Pauw
a,1
and Washek F. Pfeffer
b
a
Equipe danalyse harmonique, Universite de Paris-Sud, Batiment 425, F-91405 Orsay Cedex, France
b
Department of Mathematics, University of California, Davis, CA 95616, USA
Received 21 June 2002
Communicated by R.D. Mauldin
Abstract
Applying a very general GaussGreen theorem established for the generalized Riemann
integral, we obtain simple proofs of new results about removable sets of singularities for the
Laplace and minimal surface equations. We treat simultaneously singularities with respect to
differentiability and continuity.
r 2003 Elsevier Science (USA). All rights reserved.
MSC: primary 26B20; 35D99
Keywords: Divergence theorem; Weak charges; Integral extensions; Laplace and minimal surface
equations
0. Introduction
In an open set UCR
m
we consider the second order partial differential equation
divh 3 rux f x; ux;
where h : R
m
-R
m
and f : U R
m
-R are given. Note that depending on the maps h
and f ; Eq. (
) need not be linear; e.g., the minimal surface equation (4.2) below. A set
ARTICLE IN PRESS
E-mail addresses: thierry.de-pauw@math.u-psud.fr (T. De Pauw), wfpfeffer@ucdavis.edu (W.F.
Pfeffer).
1
Supported in part by Marie Curie Fellowship, Human Potential European Community Program,
Contract HMPF-CT-2001-01235.
0001-8708/03/$ - see front matter r 2003 Elsevier Science (USA). All rights reserved.
doi:10.1016/S0001-8708(03)00085-9
ECU is called removable if u satises (
) in U whenever it satises (
) in U E: The
goal is to nd a useful class of removable sets.
While this is a classical problem studied by many mathematicians, e.g., [3,911,20]
to mention a few, it appears that our approach has not been used previously.
Applying a very general GaussGreen theorem established for the generalized
Riemann integral [17], we obtain simple proofs of new results about removable sets
for the Laplace and minimal surface equations. We deal simultaneously with
removable sets where the solution u of (
), or h 3 ru; or both, lack differentiability,

denoted by E
d
; as well as those where they lack continuity, denoted by E
c
: Typically,
we assume the Hausdorff measure H
m1
of E
d
is s-nite, and that of E
c
is zero.
Notwithstanding, we also study the case where the size of removable sets is measured
by the integral-geometric measure I
m1
1
: In the main result (Theorem 4.2 below), no
topological restrictions are imposed on removable sets.
We illustrate our idea by proving a known fact for the Laplace equation Du 0;
i.e., for Eq. (
) where h is the identity map and f 0: Let E be a relatively closed

subset of U whose Hausdorff measure H
m1
is s-nite. Assuming a function
uAC
1
U is harmonic in U E; we wish to prove that it is harmonic in U: Since
each weak solution (in the sense of distributions) of the Laplace equation is
harmonic [18, Corollary of Theorem 8.12], it sufces to show
_
U
uDj 0 for every
jAC
N
c
U: In view of our assumptions, this is true whenever the following
integrations by parts are valid:
_
U
uDj
_
U
ru rj
_
U
jDu:
While the rst equality is standard, the second depends on the GaussGreen theorem
for the vector eld jru that is continuous in U but differentiable only in U E:
Since such a theorem holds [17, Proposition 5.1.2, Corollary 5.1.13], the removability
of E follows.
In Theorem 4.1 below, we show that the same technique provides a simple proof
of a slightly improved classical result, due to Besicovitch [2], about removable sets
for holomorphic functions.
As all functions involved in the previous example, as well as in the proof of
Besicovitchs theorem, are Lebesgue integrable, the above mentioned generalized
Riemann integral, called the R-integral in [17, Chapter 5], is used only indirectly: we
merely apply the GaussGreen theorem established for the R-integral to the
Lebesgue integral. However, the Lebesgue integral cannot be used when f a0 and it
is not a priori clear, or actually not true, that the function x/f x; ux belongs to
L
1
loc
U; cf. Theorem 4.2 below.
To deal simultaneously with the sets E
d
and E
c
; we need a GaussGreen theorem
for discontinuous vector elds, which has not been available previously even in the
context of Lebesgue integrationcf. [1,19]. The necessary result is obtained in
Section 3 by extending the R-integral to a larger class of integrable functions; the
denition and basic properties of the R-integral are stated in this section without
ARTICLE IN PRESS
T. De Pauw, W.F. Pfeffer / Advances in Mathematics 183 (2004) 155182 156
proofs. The R-integral, its extension, and the associated GaussGreen theorem
depend on the concepts of charge and weak charge. These are linear functionals,
continuous with respect to suitable topologies, on the linear space of all bounded BV
functions with compact support. Under the name of continuous additive functions,
charges were dened in [16, Section 4]. Their properties are summarized in Section 2;
the details can be found in [17, Chapter 2]. In addition, Section 2 contains the main
results concerning weak charges, which are more restrictive than bounded additive
functions introduced in [16, Denition 10.1]. Applications to removable sets for
Eq. (
) are given in Section 4.

1. The setting
The ambient space of this paper is R
m
where mX1 is a xed integer. In R
m
we shall
use exclusively the Euclidean norm j j induced by the usual inner product x y: The
diameter of a set ECR
m
is denoted by dE: We denote by Bx; r and Bx; r;
respectively, the open and closed ball of radius r40 centered at xAR
m
: The origin of
R
m
is denoted by 0; and we write Br and Br instead of B0; r and B0; r;
respectively.
By a measure we always mean an outer measure. Lebesgue measure in R
m
is
denoted by L
m
; however, for ECR
m
; we usually write jEj instead of L
m
E: If
0pspm; we denote by H
s
the s-dimensional Hausdorff measure in R
m
: In addition,
we shall use the integral-geometric measure I
m1
1
dened in [7, Section 2.10.5].
Unless specied otherwise, the words measure, measurable, and negligible as
well as the expressions almost all and almost everywhere always refer to
Lebesgue measure L
m
; similarly, the symbols
_
E
f and L
p
E refer to L
m
:
Let ECR
m
: We denote by cl E; int E; and @E the closure, interior, and boundary
of E; respectively. If E is measurable, we denote by cl
E and @
E; the essential
closure and essential boundary of E; respectively. A measurable set E is called
essentially closed whenever cl
E is closed.
Let UCR
m
be an open set. The collections of all BV subsets of U and all locally
BV subsets of U are denoted by BVU and BV
loc
U; respectively. We denote by
BV
c
U the collection of all bounded BV subsets of U whose closure is also
contained in U: In the absence of additional attributes, a BV set or a locally BV set is
always a BV subset or a locally BV subset of R
m
; respectively. We write, respectively,
BV; BV
loc
; and BV
c
instead of BVR
m
; BV
loc
R
m
; and BV
c
R
m
: The perimeter
and unit exterior normal of a BV set A are denoted by jjAjj and n
A
; respectively. The
regularity of a BV set A is the number
rA :
jAj
dAjjAjj
if jAj40;
0 if jAj 0:
_
Throughout, by a function we mean a real-valued function. When considered
individually, functions are generally not identied with the equivalence classes they
ARTICLE IN PRESS
determine. On the other hand, by spaces of functions, we usually mean the spaces of
the equivalence classes determined by these functions. Often we denote by f both a
function dened on a set A and its restriction f pB to BCA:
Let UCR
m
be an open set. We denote by BVU the family of all BV functions in
U; and give the symbols BV
loc
U and BV
c
U the obvious meaning. We let
BV
N
U : BVU-L
N
U;
and dene BV
N
loc
U and BV
N
c
U similarly. We write BV instead of BVR
m
; and
use the same convention for the other spaces introduced in this paragraph. If
gABV
loc
U; we denote by Dg the distributional gradient of g; and by jjDgjj the
variational measure of g: We let jjgjj : jjDgjjU:
Observation 1.1. If gABV
c
R; then jgj
N
pjjgjj:
Proof. If fga0gCa; b and g
is the precise value of g; denote by Vg
the classical
variation of g
in the interval a; b: In view of [7, Theorem 4.5.9, (23)],

jgj
N
jg
j
N
p sup
tAa;b
jg
tjpVg
pjjgjj: &
Proposition 1.2. If E is a bounded measurable subset of R
m
; then jj jj is a Banach
norm in the linear space
BV
E
: fgABV: fga0gCEg:
Proof. Clearly jj jj is a norm in BV
E
: If fg
i
g is a Cauchy sequence in BV
E
; jj jj;
then it is a Cauchy sequence in L
1
R
m
: Indeed, this follows from Observation 1.1 if
m 1; and from the Ho lder and Sobolev inequalities if mX2: Thus fg
i
g converges to
a gAL
1
R
m
; and we may assume fga0gCE: As fjjg
i
jjg is a Cauchy sequence of real
numbers, jjgjjplim jjg
i
jjoN: Consequently gABV
E
:
Now given e40; there is an integer kX1 such that jjg
i
g
j
jjoe for all i; jXk: If
iXk is an integer, then the sequence fg
i
g
j
g
j
converges to g
i
g in L
1
R
m
:
Therefore
jjg
i
gjjplim inf
j
jjg
i
g
j
jjpe;
and the proposition follows. &
Let ECCCR
m
and let v : C-R
n
: We say v is pointwise Lipschitz in E if given
xAE; we can nd c
x
40 and d
x
40 so that
jvx vyjpc
x
jx yj
ARTICLE IN PRESS
for all yAC-Bx; d
x
: Recall v is called continuous in E if it is continuous at each
xAE: An easy variant of Whitneys extension theorem [21, Chapter 6, Section 2]
yields the following:
If C is a closed set and v is continuous or pointwise Lipschitz in E; then v has an
extension w : R
m
-R
n
that is C
N
in R
m
C and continuous or pointwise Lipschitz
in E; respectively.
In particular, if v is pointwise Lipschitz in E; then by Stepanoffs theorem, w is
differentiable at almost all xAE; moreover, for almost all xAE; the derivative Dwx
depends only on v and not on the extension w [17, Lemma 1.6.3].
2. Charges
For n 1; 2; y; we topologize the convex set
BV
n
: fgABV
N
c
: supp gCBn and jjgjj jgj
N
pn 1g
by two different metrics:
t : f ; g/jf gj
1
and $ : f ; g/jjf gjj:
The space BV
n
; t is compact by [5, Theorem 4, Section 5.2.3], and it follows from
Proposition 1.2 that the space BV
n
; $ is complete. In BV
N
c
we consider the largest
topology T for which all inclusion maps
BV
n
; t+BV
N
c
; T
are continuous, and the largest topology Wfor which all inclusion maps
BV
n
; $+BV
N
c
; W
are continuous. Both topologies T and W are Hausdorff, sequential and
sequentially complete, but not metrizable. Moreover, the topology T is locally
convex; whether the same is true about the larger topology Wis unclear. Identifying
each set B in BV
c
with its indicator w
A
ABV
N
c
; we view BV
c
as a closed subspace of
BV
N
c
; T and BV
N
c
; W:
Let fg
i
g be a sequence in BV
N
c
and gABV
N
c
: We write fg
i
g-g or fg
i
g4g
according to whether fg
i
g T-converges or W-converges to g; respectively.
Observe
*
fg
i
g-g if and only if each g
i
vanishes outside a xed compact set KCR
m
; and
supjjg
i
jj jg
i
j
N
oN and lim jg
i
gj
1
0;
ARTICLE IN PRESS
*
fg
i
g4g if and only if each g
i
vanishes outside a xed compact set KCR
m
; and
sup jg
i
j
N
oN and lim jjg
i
gjj 0:
Using Observation 1.1 if m 1; and the Ho lder and Sobolev inequalities if mX2; it
is easy to show that fg
i
g4g implies fg
i
g-g: For a sequence fB
i
g in BV
c
and a set
BABV
c
; the meaning of the symbols fB
i
g-B and fB
i
g4B is obvious.
Note fg
i
g-g and fB
i
g-B can be dened for gAL
1
R
m
and a bounded
measurable set BCR
m
: However, in this case [5, Section 5.2.1] implies gABV
N
c
and
BABV:
Denition 2.1. A linear functional F : BV
N
c
-R is called a charge or a weak charge
(abbreviated as w-charge) according to whether F is T-continuous of W-
continuous, respectively.
It is easy to see a linear functional F : BV
N
c
-R is, respectively, a charge or w-
charge whenever lim/F; g
i
S 0 for each sequence fg
i
g in BV
N
c
for which fg
i
g-0
or fg
i
g40:
Remark 2.2. It follows from [17, Section 4.1] that a charge F is uniquely determined
by the restriction FpBV
c
; and that each additive T-continuous function F on BV
c
denes a charge by the formula
/F; gS :
_
N
0
Ffg
4tg dt
_
N
0
Ffg
4tg dt
for each gABV
N
c
: In general, neither is true for w-charges.
Clearly, each charge is a w-charge, and the next example shows the converse is
false.
Example 2.3 (The ux of a vector eld). If v : R
m
-R
m
is a locally bounded Borel
vector eld, we dene a linear functional F
v
on BV
N
c
by the formula
/F
v
; gS :
_
R
m
v dDg
for each gABV
N
c
: If jvxjpy for every xAsupp g; then
j/F
v
; gSjpyjjDgjjR
m
yjjgjj;
and it follows F
v
is a w-charge. Since
/F
v
; w
B
S
_
@
B
v n
B
dH
m1
;
ARTICLE IN PRESS
for each bounded BV set B; we call F
v
the ux of v: Choosing a vector eld v with a
suitable discontinuity, it is easy to see that the ux of v need not be a charge [17,
Example 2.1.11].
Proposition 2.4. If F is a linear functional on BV
N
c
; then
*
F is a charge if and only if given e40; there is a y40 such that
j/F; gSjoyjgj
1
ejjgjj jgj
N
for each gABV

N
c
with fga0gCB1=e;
*
F is a w-charge if and only if given e40; there is a y40 such that
j/F; gSjoyjjgjj ejgj
N
for each gABV
N
c
with fga0gCB1=e:
Proof. The statement about charges follows immediately from Remark 2.2 and [17,
Proposition 2.2.6].
As the converse is obvious, suppose F is a w-charge, and choose an e40: Observe
there is an Z40 such that j/F; gSjoe=2 for each gABV
N
c
with jjgjjoZ; jgj
N
o1;
and fga0gCB1=e: Let y : e=2Z and select a gABV
N
c
for which
fga0gCB1=e: With no loss of generality, we may assume gX0:
Let p and q be the smallest positive integers for which jjgjj=poZ and jgj
N
=qo1:
Note ppjjgjj=Z 1 and qpjgj
N
1: Since
s/
_
s
0
jjfg4tgjj dt : 0; jgj
N
-0; jjgjj
is an increasing continuous function, the coarea theorem implies there are 0
a
0
o?oa
p
jgj
N
such that
_
a
i
a
i1
jjfg4tgjj dt
1
p
jjgjjoZ; i 1; y; p:
For i 0; y; q; let b
i
i=qjgj
N
; and order the set fa
0
; y; a
p
; b
0
; y; b
q
g
into a sequence 0 c
0
o?oc
r
jgj
N
: Clearly rpp q 1: Now it is easy
to verify
g
i
: maxfminfg; c
i
g; c
i1
g c
i1
; i 1; y; r;
ARTICLE IN PRESS
are BV functions with jg
i
j
N
o1; fg
i
a0gCB1=e; and
r
i1
g
i
g: As each c
i1
; c
i
is contained in some a
j1
; a
j
;
jjg
i
jj
_
1
0
jjfg
i
4tgjj dt
_
c
i
c
i1
jjfg4tgjj dt
p
_
a
j
a
j1
jjfg4tgjj dtoZ;
by the coarea theorem. We conclude
j/F; gSjp
r
i1
j/F; g
i
Sjo
er
2
p
e
2
p q 1
p
e
2
jjgjj
Z
jgj
N
1
_ _
yjjgjj
e
2
jgj
N
1;
from which the desired inequality follows whenever jgj
N
X1: If 0ojgj
N
o1; we apply
the previous result to h : g=jgj
N
:
/F; gS j j jgj
N
/F; hS j jojgj
N
yjjhjj ejhj
N
yjjgjj ejgj
N
:
As the case jgj
N
0 is trivial, the proposition is established. &
If F is a charge and f ABV
N
loc
; then it is easy to see we can dene a charge F L f by
the formula
/F L f ; gS : /F; fgS
for all gABV
N
c
: Showing that the same construction is possible for w-charges
requires some work.
Observation 2.5. The variational measure jjDf jj of a BV function f is absolutely
continuous with respect to the Hausdorff measure H
m1
:
Proof. By [17, Theorem 1.8.2, (3)], this is true if f is the indicator of a locally BV set.
For an arbitrary BV function, the observation follows from the coarea theorem [17,
Proposition 1.8.10]. &
Lemma 2.6. Let fg
i
g be a sequence in C
1
c
R
m
: If lim jjg
i
jj 0; then fg
i
g has a
subsequence that converges to zero H
m1
-almost everywhere.
Proof. For m 1; Observation 1.1 implies fg
i
g converges uniformly to zero
everywhere.
ARTICLE IN PRESS
If mX2; select a subsequence of fg
i
g; still denoted by fg
i
g; so that jjg
i
jjp2
i
for
i 1; 2; y; and let
B
i;k
: xAR
m
: g
i
x4
1
k
_ _
and B
k
:
N
j1
_
N
ij
B
i;k
:
It follows directly from the denition of capacity Cap
1
[5, Section 4.7.1] that
Cap
1
B
i;k
pjjkg
i
jjpk2
i
: Since Cap
1
is a measure in R
m
;
Cap
1
B
k
pCap
1
_
N
ij
B
i;k
_ _
pk2
1j
for j 1; 2; y: We infer Cap
1
B
k
0; and if B
N
k1
B
k
then Cap
1
B 0:
According to [5, Section 5.6.3, Theorem 3], we have H
m1
C 0 for each compact
subset of B: As B is a Borel set, it follows from [6, Theorems 1.6 and 5.6] that
H
m1
B 0: A direct verication reveals
B fxAR
m
: lim sup g
i
x40g;
which means lim sup g
i
p0 H
m1
-almost everywhere. Applying this result to the
sequence fg
i
g; we obtain lim inf g
i
X0 H
m1
-almost everywhere, and the lemma
follows. &
Lemma 2.7. Let fg
i
g be a sequence in BV
N
c
such that fg
i
g40; and let f ABV
N
loc
: Then
ffg
i
g40:
Proof. Select a compact set KCR
m
with supp g
i
Cint K; and let c sup jg
i
j
N
: We
may assume ff a0gCK; in which case
sup jfg
i
j
N
pcjf j
N
oN:
Thus we only need to prove lim jjfg
i
jj 0: Clearly, it sufces to show that this is true
for a subsequence of fg
i
g:
Assume rst that fg
i
g is a sequence in C
1
c
R
m
: Using Lemma 2.6, nd a
subsequence of fg
i
g; still denoted by fg
i
g; that converges to zero H
m1
-almost
everywhere, and in view of Observation 2.5, also jjDf jj-almost everywhere. Choose a
sequence ff
j
g in C
1
c
R
m
so that lim jf
j
f j
1
0 and lim jjf
j
jj jjf jj: If UCR
m
is an
open set, then
jjDf jjUplim inf jjDf
j
jjU
by [5, Section 5.2, Theorem 1]. As jjDf jj and jjDf
j
jj are nite measures and
lim jjDf
j
jjR
m
jjDf jjR
m
; the previous inequality implies
lim sup jjDf
j
jjCpjjDf jjC
ARTICLE IN PRESS
for each closed set CCR
m
: According to [5, Section 1.9, Theorem 1], the measures
jjDf
j
jj converge weakly to jjDf jj: Since
jjfg
i
jjp lim inf
j-N
jjf
j
g
i
jj lim inf
j-N
_
R
m
jD f
j
g
i
j dL
m
p lim
j-N
_
R
m
jDf
j
j jg
i
j dL
m
lim
j-N
_
R
m
jf
j
j jDg
i
j dL
m
_
R
m
jg
i
j djjDf jj
_
R
m
jf j jDg
i
j dL
m
p
_
R
m
jg
i
j djjDf jj jf j
N
jjg
i
jj;
the dominated convergence theorem implies lim jjfg
i
jj 0:
In the general case, nd functions g
i;k
; i; k 1; 2; y; in C
1
c
R
m
so that each
fg
i;k
a0g is contained in K; all jg
i;k
j
N
are bounded by a xed constant, and for
i 1; 2; y;
lim
k-N
jg
i;k
g
i
j
1
0 and lim
k-N
jjg
i;k
jj jjg
i
jj:
Since ffg
i;k
g converges to fg
i
in L
1
; we have jjfg
i
jjplim inf jjfg
i;k
jj: Thus for i
1; 2; y; there is an integer k
i
X1 such that
jjg
i;k
i
jjojjg
i
jj
1
i
and jjfg
i
jj
1
i
pjjfg
i;k
i
jj;
in particular, lim jjg
i;k
i
jj 0: Now lim jjfg
i;k
i
jj 0 by the rst part of the proof, and
consequently lim jjfg
i
jj 0: &
Corollary 2.8. If F is a w-charge and f ABV
N
loc
; then
F L f : g//F; fgS: BV
N
c
-R
is a w-charge.
Let F be a charge or a w-charge, and let AABV: We dene
FA : /F; w
A
S and F L A : F L w
A
;
and say that F is, respectively, a charge or w-charge in A whenever F F L A:
Note. Independently of the previous paragraph, when m is a measure in R
m
and
ECR
m
; we dene the measure m L E in the customary way:
m L EA : mA-E
for each set ACR
m
:
ARTICLE IN PRESS
3. Integrals
A set ECR
m
is called thin whenever H
m1
L E is a s-nite measure. A gage on a
set ECR
m
is a function d : E-0; N such that fda0g is a thin set. A partition is a
nite collection
P fA
1
; x
1
; y; A
p
; x
p
g;
where A
1
; y; A
p
are disjoint bounded BV sets and x
1
; y; x
p
are points of R
m
:
Denition 3.1. A function f dened almost everywhere in a bounded BV set A is
called R-integrable in A if there is a charge F in A and an extension of f to cl
A; still
denoted by f ; such that the following condition is satised: given e40; we can nd a
gage d on cl
A so that
p
i1
j f x
i
jA
i
j FA
i
joe
for every partition fA
1
; x
1
; y; A
p
; x
p
g with x
i
Acl
A; A
i
CA;
rA
i
,fx
i
g4e and dA
i
,fx
i
godx
i
for i 1; y; p:
The charge F of Denition 3.1, which is uniquely determined by f ; is called the R-
primitive of f ; denoted by R
_
f : For a BV set BCA; the charge F L B is the R-
primitive of f pB; in particular, f pB is R-integrable in B: The R-integral of f over A is
the number R
_
A
f : FA: By [17, Proposition 5.1.3], the R-integral is a
nonnegative linear functional on the linear space RA of all R-integrable functions
in A:
Without proofs, we summarize the main properties of the R-integral established in
[17, Chapter 5].
Theorem 3.2. Let A be bounded BV set.
(1)
L
1
ACRA and R
_
A
f
_
A
f for each f AL
1
A: The inclusion
L
1
ACRA is proper whenever int Aa|:
(2) Each function f ARA is measurable; moreover, if f ARA is nonnegative, then
f AL
1
A:
(3) If the R-primitive of f ARA equals zero, then f x 0 for almost all xAA:
(4) Suppose B is a bounded BV set disjoint from A: Let f be a function dened on
A,B that is R-integrable in A and B; and let F
A
and F
B
be the R-primitives of
f pA and f pB; respectively. If both A and B are essentially closed, then
f ARA,B and F
A
F
B
is the R-primitive of f :
ARTICLE IN PRESS
(5) Let F be the R-primitive of f ARA; and let gABV
N
loc
: Then fgARA and F L g
is the R-primitive of fg: In particular,
R
_
A
f xgx dx
_
N
0
R
_
fg4tg
f x dx
_ _
dt
whenever g is nonnegative.
(6) Let vACcl A; R
m
; let T be a thin set, and let v be pointwise Lipschitz at each
xAcl
A T: Then div v belongs to RA and the ux of v is the R-primitive of

div v:
(7) If f: A-R
m
is a lipeomorphism and f ARfA; then f 3 fJ
f
belongs to RA
and
R
_
A
f fxJ
f
x dx R
_
fA
f y dy;
here J
f
: jdet Dfj is the Jacobian of f:
Note. Four comments concerning Theorem 3.2 are in order.
(i) An easy consequence of part (1) is the following observation: if the pair f ; F
satises the conditions of Denition 3.1 for a particular extension of f to cl
A;
then it satises these conditions for an arbitrary extension of f to cl
A:
(ii) Part (4) is false without assuming the sets A and B are essentially closed [17,
Proposition 6.1.1, Remark 6.1.2, (4)]. We shall improve on this situation in
Theorem 3.5 below.
(iii) As part (6) implies that Fubinis theorem is generally false for the R-integral
[17, Example 5.1.14], part (5) asserts a nontrivial fact; cf. [17, Remark 5.2.3].
(iv) Part (7) can be generalized to a geometrically intuitive transformation formula
for local lipeomorphisms [17, Section 5.3].
If f is a function dened almost everywhere in AABV and F is a w-charge, we
denote by R f ; F; A the family of all BV sets BCA for which F L B is the R-
primitive of f pB:
Denition 3.3. A function f dened almost everywhere in AABV is called W-
integrable in A if there is a w-charge F in A and a sequence fA
i
g in R f ; F; A with
fA
i
g4A:
The family of all W-integrable functions in AABV is denoted by WA: Each
charge F associated with f AWA according to Denition 3.3 is called a W-primitive
of f :
Let F be a W-primitive of f AWA; and choose a sequence fA
i
g in R f ; F; A
with fA
i
g4A: If BCA is a BV set, then fB-A
i
g is a sequence in R f pB; F L B; B;
ARTICLE IN PRESS
and it follows from Lemma 2.7 that fB-A
i
g4B: Consequently, F L B is a W-
primitive of f pB:
Observation 3.4. Let A be a bounded BV set. Each f AWA has a unique W-
primitive, denoted by W
_
f :
Proof. Suppose F and G are W-primitives of f AWA; and nd sequences fA
i
g in
R f ; F; A and fB
i
g in R f ; G; A so that fA
i
g4A and fB
i
g4A: Then fA
i
-B
i
g is a
sequence in R f ; F; A-R f ; G; A and fA
i
-B
i
g4A: As each R-integrable
function has a unique R-primitive,
FA lim FA
i
-B
i
lim GA
i
-B
i
GA;
and the observation follows from the previous paragraph. &
Let A be a bounded BV set. If F is the W-primitive of f AWA; we call the
number W
_
A
f : FA the W-integral of f over A: Employing proofs similar to
that of Observation 3.4, it is easy to show that the W-integral is a linear functional
on WA; and that parts (1)(3) and (7) of Theorem 3.2 hold for the W-integral.
Theorem 3.5 (Additivity). Let A and B be bounded BV sets, and let f be a function
dened almost everywhere on A,B: If f is W-integrable in A and B; then it is W-
integrable in A,B; and
W
_
A,B
f W
_
A
f W
_
B
f
whenever A and B are disjoint.
Proof. It sufces to prove the theorem when A and B are disjoint. Denote by F
A
and
F
B
the W-primitives of f in A and B; respectively, and nd sequences fA
i
g in
R f ; F
A
; A and fB
i
g in R f ; F
B
; B so that fA
i
g4A and fB
i
g4B: According to
[22], for i; j 1; 2; y; there are essentially closed BV sets A
i;j
CA
i
and B
i;j
CB
i
such
that fA
i;j
g
j
4A
i
and fB
i;j
g
j
4B
i
: For each i; nd a j
i
so that
jjA
i
A
i;j
i
jjo1=i and jjB
i
B
i;j
i
jjo1=i;
and let C
i
: A
i;j
i
,B
i;j
i
: Part (4) of Theorem 3.2 shows that fC
i
g is a sequence in
R f ; F
A
F
B
; A,B: Since fC
i
g4A,B; the theorem follows. &
Theorem 3.6 (Multipliers). Let A be a bounded BV set, and let F be the W-primitive
of f AWA: If gABV
N
loc
; then fgAWA and F L g is the W-primitive of fg: If, in
addition, fg
i
g is a sequence in BV
N
loc
such that fg
i
w
A
g4gw
A
; then
lim W
_
A
fg
i
W
_
A
fg:
ARTICLE IN PRESS
Proof. There is a sequence fA
i
g in R f ; F; A with fA
i
g4A: As part (5) of Theorem
3.2 implies R f ; F; ACR fg; F L g; A; it follows from Corollary 2.8 that fgAWA
and F L g is the W-primitive of fg: If fg
i
g is a sequence in BV
N
loc
with fg
i
w
A
g4gw
A
;
then
W
_
A
fg F L gA /F; gw
A
S lim/F; g
i
w
A
S
lim F L g
i
A lim W
_
A
fg
i
: &
If UCR
m
is an open set, we denote by W
loc
U the linear space of all functions
f : U-R such that f pA belongs to WA for each AABV
c
U: The elements of
W
loc
U are called locally W-integrable functions in U: If f AW
loc
U and
gABV
N
c
U; nd an AABV
c
U with supp gCint A and, using the multipliers
theorem, let
W
_
U
fg : W
_
A
fg:
The W-integral W
_
U
fg is well dened, since by the additivity theorem, its value
does not depend on the choice of A: In particular, it is easy to see that for each
f AW
loc
U; the linear map
L
f
: j/W
_
U
f j: C
N
c
U-R
is a distribution in U [18, Denition 6.7]. Showing that L
f
0 implies f x 0 for
almost all xAU requires some work.
Lemma 3.7. Let F be a w-charge in AABV
c
; and suppose there is a sequence fA
i
g of
BV subsets of A such that fA
i
g4A and each F L A
i
is a charge. If C
x
: C x is the
translation of CABV
c
by xAR
m
; then the function x/FC
x
is uniformly continuous
on R
m
and has compact support.
Proof. Choose an e40; and using Proposition 2.4, nd a y40 so that
jFEjoyjjEjj e for each EABV
c
: Let a : e=y; and for i 1; 2; y; let F
i
:
F L A
i
and B
i
: A A
i
:
Given a positive integer jpm; denote by P
j
the m 1-dimensional subspace of
R
m
perpendicular to the jth coordinate axis. For each yAP
j
denote by l
y
the line
passing through y and perpendicular to P
j
: According to [5, Section 5.10.2, Theorem
2], for H
m1
-almost all yAP
j
; the intersection l
y
-C is a one-dimensional BV set,
and we denote by jjl
y
-Cjj its perimeter. We employ the usual relationships between
jjl
y
-Cjj and jjCjj; see [17, Section 1.9]. As
_
P
j
jjl
y
-Cjj dH
m1
ypjjCjjoN;
ARTICLE IN PRESS
there is a b
j
40 such that
_
E
jjl
y
-Cjj dH
m1
yoa for each H
m1
-measurable set
ECP
j
; with H
m1
Eob
j
: The set
E
i;j
: fyAP
j
: jjl
y
-B
i
jj40g fyAP
j
: jjl
y
-B
i
jjX2g 3:1
is H
m1
-measurable by [5, Section 5.10.2, Lemma 1]. Denote by x
j
the orthogonal
projection of xAR
m
to P
j
; and by E
i;j
x
j
the translation of E
i;j
by x
j
: Observe
_
P
j
jjl
y
-B
i
-C
x
jj dH
m1
y
_
E
i;j
jjl
y
-B
i
-C
x
jj dH
m1
y
p
_
E
i;j
jjl
y
-B
i
jj dH
m1
y
_
E
i;j
jjl
y
-C
x
jj dH
m1
y
pjjB
i
jj
_
E
i;j
x
j
jjl
y
-Cjj dH
m1
y:
Let b minfb
1
; y; b
m
g; and nd an integer kX1 with jjB
k
jjominfa; bg: In view of
(3.1),
H
m1
E
k;j
x
j
H
m1
E
k;j
p
_
P
j
jjl
y
-B
k
jj dH
m1
ypjjB
k
jjob:
Consequently,
jjB
k
-C
x
jjp
m
j1
_
P
j
jjl
y
-B
k
-C
x
jj dH
m1
yo2ma;
and we conclude that for each xAR
m
;
jFC
x
F
k
C
x
j jFB
k
-C
x
jo2may e e2m 1:
Since F
k
is a charge, Proposition 2.4 and [17, Lemma 4.2.1] imply there is a d40 such
that jF
k
C
x
F
k
C
z
joe for each x; zAR
m
with jx zjod: As FC
x
0 whenever
jxj is sufciently large, the lemma follows. &
Proposition 3.8. Let UCR
m
be an open set and let f AW
loc
U: If W
_
U
f j 0 for
each jAC
N
c
U; then f x 0 for almost all xAU:
Proof. As U is Lindelo f, it sufces to show f 0 almost everywhere in each
open ball VCU: Choose an open ball VCU and a diffeomorphism f from
R
m
onto V: Observe J
f
x40 for each xAR
m
; and g : f 3 fJ
f
belongs to
W
loc
R
m
: Hence W
_
R
m gf 0 for each fAC
N
c
R
m
; and g 0 almost everywhere
yields f x 0 for almost all xAV: It follows we may assume U R
m
from the
onset.
ARTICLE IN PRESS
Select bounded BV sets A and B: Let B
: fx: xABg and B

x
: B x for each
xAR
m
: Denote by F the W-primitive of f pA; and for xAR
m
; let
w
B
f x :W
_
A
w
B
x yf y dy
W
_
A-B
x
f y dy FB
x
: 3:2
By Lemma 3.7, the convolution w
B
f is a uniformly continuous function on R
m
with compact support. Select a jAC
N
c
R
m
; and observe that w
B
j is the usual
convolution of w
B
and j; in particular w
B
j belongs to C
N
c
R
m
:
Claim. W
_
A
f w
B
j
_
A
w
B
f j for each jAC

N
c
R
m
:
Proof. If H is a charge in A and nX1 is an integer, let
jjHjj
n
: supfjHCj: jjCjjpng:
According to [17, Proposition 2.2.4], there is an integer kX1; depending on A; such
that jj jj
k
; jj jj
k1
; yare equivalent Banach norms in the linear space of all charges
in A: If H is the R-primitive of hARA; then h/jjHjj
n
; n k; k 1; y are
equivalent norms in RA: The topology in RA induced by any of these norms is
denoted by S: Given jAC
N
c
R
m
; the linear functionals
R : h/R
_
A
hw
B
j and L : h/
_
A
w
B
hj
dened on RA are S-continuous. The S-continuity of R follows from [17,
Proposition 4.5.2]. The dominated convergence theorem implies the S-continuity of
L: since jjB
x
jj jjBjj for each xAR
m
; we infer from (3.2) the sequence fw
B
h
i
g
converges uniformly to zero for every sequence fh
i
g in RA that S-converges to
zero. The standard manipulation of convolutions by means of Fubinis theorem
shows that Rh Lh for each hAL
1
A: This equality extends to every hARA;
because L
1
A is a dense subspace of RA; S; see [17, Corollary 4.2.3]. There is a
sequence fA
i
g of BV subsets of A such that fA
i
g4A and f ARA
i
for i 1; 2; y:
Applying what we have already proved, and observing that lim jA A
i
j 0
establishes the claim:
W
_
A
f w
B
j lim R
_
A
i
f w
B
j
lim
_
A
i
w
B
f j
_
A
w
B
f j:
ARTICLE IN PRESS
Now choose a jAC
N
c
R
m
; and nd an AABV
c
whose interior contains the
supports of both w
B
j and w
B
f : In view of the claim

_
R
m
w
B
f j
_
A
w
B
f j
W
_
A
f w
B
j W
_
R
m
f w
B
j 0:
As w
B
f is continuous, it is equal to zero everywhere. In particular FB

w
B
f 0 0; and the proposition follows from Theorem 3.2, part (3), and the
arbitrariness of B: &
An H
m1
-negligible subset of R
m
is called slight. Clearly, each slight set is thin but
not vice versa.
Observation 3.9. Given a bounded slight set S and e40; there is an open set UABV
such that SCU and jjUjjoe:
Proof. Choose a bounded open set V containing S: Since S is negligible with respect to
the m 1-dimensional spherical measure [13, Section 5.1], we can cover S by open
balls B
i
CV so that
N
i1
jjB
i
jjoe: It follows that U
N
i1
B
i
is the desired set. &
Theorem 3.10 (GaussGreen). Let A; S; and T be, respectively, a bounded BV set, a
slight set, and a thin set. Suppose v : cl A-R
m
is a bounded vector eld that is
continuous in cl A S and pointwise Lipschitz in cl
A T: Then div v belongs to

WA and
W
_
A
div v dL
m
_
@
A
v n
A
dH
m1
:
Proof. Extend v to a bounded vector eld w : R
m
-R
m
that is continuous on R
m
S-cl A; and use Example 2.3 to dene the ux F of w: By Observation 3.9, there is
a sequences fU
i
g of open bounded BV sets such that S-cl ACU
i
for i 1; 2; y;
and lim jjU
i
jj 0: Let A
i
A U
i
and observe fA
i
g4A: By part (6) of Theorem
3.2, each A
i
belongs to Rdiv v; F; A and the theorem follows. &
Theorem 3.11 (Integration by parts). Let O; S; and T be, respectively, a Lipschitz
domain, a slight set, and a thin set. Suppose v : cl O-R
m
is a bounded vector eld that
is continuous in cl O S and pointwise Lipschitz in O T: Then
W
_
O
g div v dL
m
_
@O
Tr gv n
O
dH
m1
_
O
v dDg
for each gABV
N
O:
ARTICLE IN PRESS
Proof. It sufces to prove the theorem for a nonnegative gABV
N
O: Extend v to a
bounded vector eld w : R
m
-R
m
that is continuous on R
m
S-cl O: Using the
standard molliers, nd a sequence fv
k
g in C
N
R
m
; R
m
having the following
properties
(i) sup jv
k
j
N
pjwj
N
;
(ii) lim v
k
x vx for each xAcl O S;
(iii) fv
k
g converges to w uniformly on each compact set K contained in R
m
S-cl O:
According to [5, Section 5.3, Theorem 1],
_
O
g div v
k
dL
m
_
@O
Tr gv
k
n
O
dH
m1
_
O
v
k
dDg
for k 1; 2; y: Properties (i) and (ii), the dominated convergence theorem, and
Observation 2.5 yield
lim
_
@O
Tr gv
k
n
O
dH
m1
_
@O
Tr gv n
O
dH
m1
lim
_
O
v
k
dDg
_
O
v dDg;
and consequently
lim
_
O
g div v
k
dL
m
_
@O
Tr gv n
O
dH
m1
_
O
v dDg:
In view of the multipliers and GaussGreen theorems, g div v is W-integrable in O;
and we only need to show
W
_
O
g div v dL
m
lim
_
O
g div v
k
dL
m
:
By Observation 3.9, there is a sequences fU
i
g of open bounded BV sets such that
S-cl OCU
i
for i 1; 2; y; and lim jjU
i
jj 0: Letting A
i
: O U
i
; we see that
fA
i
g4O; and that cl A
i
is a compact subset of R
m
S-cl O: Part (6) of Theorem
3.2 implies
W
_
A
i
g div v dL
m
R
_
A
i
g div v dL
m
for i 1; 2; y; and as the W-primitive of g div v is a w-charge,
W
_
O
g div v dL
m
lim
i-N
R
_
A
i
g div v dL
m
:
ARTICLE IN PRESS
Choose an e40; and nd an integer pX1 so that
W
_
O
g div v dL
m
R
_
A
p
g div v dL
m
oe; 3:3
jjA
p
jjpjjOjj 1 and jjgw
OA
p
jjoe; the last inequality follows from Lemma 2.7. Let
B
p
: O A
p
and h : gw
B
p
: The Fubini, GaussGreen, and coarea theorems,
together with property (i), imply
_
B
p
g div v
k
dL
m
_
N
0
_
fh4tg
div v
k
dL
m
_ _
dt
_
N
0
_
@
fh4tg
v
k
n
fh4tg
dH
m1
_ _
dt
pjwj
N
_
N
0
H
m1
@
fh4tg dt
jwj
N
jjhjjoejwj
N
for k 1; 2; y; and hence
_
A
p
g div v
k
dL
m
_
O
g div v
k
dL
m
oejwj
N
: 3:4
By property (iii), there is a q such that jv
k
x vxjoe for each xAcl A
p
and each
kXq: Let u : gw
A
p
; and select kXq: Part (5) of Theorem 3.2 yields
R
_
A
p
g div v
_
A
p
g div v
k
dL
m
_
N
0
R
_
fu4tg
divv v
k
dL
m
_ _
dt
_
N
0
_
@
fu4tg
v v
k
n
fu4tg
dH
m1
_ _
dt
pe
_
N
0
H
m1
@
fu4tg dt ejjujj
pejjgjj jgj
N
jjA
p
jjoejjgjj jgj
N
jgj
N
jjOjj:
Combining the previous inequality with inequalities (3.3) and (3.4), we conclude
that for each kXq;
W
_
O
g div v dL
m
_
O
g div v
k
dL
m
obe;
where b 1 jwj
N
jjgjj jgj
N
jgj
N
jjOjj: &
ARTICLE IN PRESS
Remark 3.12. Given gABV
N
O; the denition of Dg implies
_
O
g div v dL
m

_
O
v dDg
for each vAC
1
c
O; R
m
: Employing the W-integral, we substantially generalized this
fact in Theorem 3.11. However, the W-integral is merely a tool, which can be
replaced by the Lebesgue integral whenever g div v belongs to L
1
O (part (1) of
Theorem 3.2).
In conclusion we show that in the GaussGreen theorem the exceptional sets can
be dened by means of the integral-geometric measure I
m1
1
provided jDvj belongs
to L
1
A: Note I
m1
1
EpH
m1
E for each set ECR
m
; and the equality holds
whenever E is H
m1
; m 1 rectiable [7, Section 2.10.6, Theorem 3.2.26].
Denote by G the Grassmanian Gm; m 1; and by g the probability measure
g
m;m1
on G dened in [13, Section 3.9]. For PAG; the unique orthogonal projection
of R
m
onto P is denoted by p: If BCR
m
is a Borel set, then
I
m1
1
B k
_
G
_
P
H
0
B-p
1
x dH
m1
x
_ _
dgP; 3:5
where k40 is a constant depending only on the dimension [13, Section 5.14].
Observation 3.13. If NCG is a g-negligible set, then R
m
has an orthonormal base
fe
1
; y; e
m
g such that each PAG perpendicular to some e
i
belongs to G N:
Proof. Choose any orthonormal base fu
1
; y; u
m
g in R
m
; and denote by P
i
the
elements of G orthogonal to u
i
: If y
m
is the Haar measure on the orthogonal group
Om and
O
_
m
i1
fgAOm: gP
i
ANg;
then y
m
OpmgN 0 by the denition of g in [13, Section 3.9]. Thus there is a
gAOn O; and fgu
1
; y; gu
m
g is the desired base in R
m
: &
Theorem 3.14. Let A be a bounded BV set, and let E
0
and E
s
be Borel subsets of R
m
such that I
m1
1
E
0
0 and the measure I
m1
1
L E
s
is s-nite. Suppose v : cl A-R
m
belongs to L
1
@
A; H
m1
; R
m
; is continuous in cl A E
0
; and pointwise Lipschitz in
cl
A E
s
: Then
_
A
div v dL
m
_
@
A
v n
A
dH
m1
3:6
whenever jDvjAL
1
A; L
m
:
ARTICLE IN PRESS
Proof. Denote by G
0
and G
s
the families consisting of all PAG such that the sets
E
0
-p
1
x and E
s
-p
1
x are, respectively, empty and countable for H
m1
-almost
all xAP: In view of (3.5), the set N G G
0
,G
s
is g-negligible. Observation 3.13
implies R
m
has an orthonormal base fe
1
; y; e
m
g such that each PAG perpendicular
to some e
i
belongs to G
0
,G
s
: Since neither our assumptions nor the equalities (3.5)
and (3.6) depend on the choice of an orthonormal base in R
m
; we may assume
fe
1
; y; e
m
g is the standard base of R
m
: We show that if v v
1
; y; v
m
and n
A

n
1
; y; n
m
; then
_
A
@v
i
@x
i
dL
m
_
@
A
v
i
n
i
dH
m1
3:7
for i 1; y; m: Our argument relies on @v
i
=@x
i
AL
1
Aa fact guaranteed by the
assumption jDvjAL
1
A for any choice of a base in R
m
:
In view of symmetry, it sufces to verify equality (3.7) only for i m: As P
fx; 0AR
m
: xAR
m1
g is perpendicular to e
m
; our choice of e
m
and [12, Section 2.2.1,
Theorem 2] imply there is an L
m1
-negligible set ECR
m1
such that for all x in
R
m1
E; the section
A
x
ftAR: x; tAAg
is a BV subset of R; and the function t/v
m
x; t is continuous in cl A
x
and Lipschitz
at all but countably many tAcl A
x
: According to Fubinis theorem, making E larger,
we may assume that the function t/@v
m
=@x
m
x; t belongs to L
1
A
x
; L
1
for
every x in R
m1
E: Parts (1) and (6) of Theorem 3.2 yield
_
A
x
@v
m
@x
m
x; t dL
1
t
_
@
A
x
v
m
x; t n
m
x; t dH
0
t
_
@
A
x
v
m
x; t n
m
x; t dH
0
t
whenever x belongs to R
m1
E: Integrating over R
m1
and using Fubinis theorem,
we obtain
_
A
@v
m
@x
m
dL
m
_
@
A
v
m
n
m
dH
m1
: &
Remark 3.15. Let A be a bounded BV set, and let E
0
CR
m
be I
m1
1
-negligible. If
v : cl A-R
m
is bounded and continuous in @
A E
0
; then vAL
1
@
A; H
m1
; R
m
:
Indeed, as E
0
-@
A is an H
m1
; m 1 rectiable set, it is H
m1
-negligible by [7,
Theorem 3.2.26].
ARTICLE IN PRESS
4. Removable singularities
To illustrate our technique, we present rst a simple proof of a slightly improved
classical result of Besicovitch [2] concerning the removable singularities of
holomorphic functions dened on an open subset of the complex plane C:
Theorem 4.1. Let UCC be an open set, and let f : U-C be locally bounded. Suppose
f is continuous outside a slight set E
c
CU and pointwise Lipschitz outside a thin set
E
d
CU: If f has a complex derivative almost everywhere in U; then it can be redened
on E
c
so that it is holomorphic in U:
Proof. Let Rf and If denote, respectively, the real and imaginary part of f : The
vector elds u Rf ; If and v If ; Rf are locally bounded in U and
continuous in U E
c
; in particular, they belong to L
1
loc
U; R
2
: The Cauchy
Riemann equations yield
div u div v 0
almost everywhere in U: Using the integration by parts theorem, we infer
_
U
u rj
_
U
j div u 0;
_
U
v rj
_
U
j div v 0
for each jAC
1
c
U: This means the vector eld Rf ; If is a distributional
solution of the CauchyRiemann equations. As these equations form an elliptic
system, an application of the regularity theorem completes the argument; cf. [18,
Example 8.14]. &
Throughout the remainder of this paper UCR
m
is a nonempty open set, in which
we consider the equation
divh 3 rux f x; ux; 4:1
where
u : U-R; h : R
m
-R
m
; and f : U R-R
are maps whose properties will be specied below. To avoid trivialities, we assume
mX2:
A classical solution of Eq. (4.1) is a function uAC
2
U such that (4.1) holds for all
xAU: A weak solution of Eq. (4.1) is an almost everywhere differentiable function u
such that
_
U
hrux rjx dx
_
U
jxf x; ux dx
ARTICLE IN PRESS
for each jAC
N
c
U: Under mild assumptions, we show that if u satises (4.1) at
almost all xAU; then u is a weak solution of (4.1).
Theorem 4.2. Let E
c
CE
d
be, respectively, a slight and a thin subset of U; and suppose
the following conditions are satised:
(1) u is differentiable almost everywhere in U;
(2) h 3 ru has a locally bounded extension to U that is continuous in U E
c
and
pointwise Lipschitz in U E
d
;
(3) u satises equation (4.1) for almost all xAU:
Then the equality
_
U
h 3 ru dDg W
_
U
gxf x; ux dx
holds for each gABV
N
c
U:
Proof. Select a gABV
N
c
U: Since supp g is a compact subset of U; there is a
Lipschitz domain O such that supp gCO and cl OCU: By our assumptions, h 3 ru
can be extended to a vector eld v : cl O-R
m
that satises the assumptions of
Theorem 3.11. Since Tr g 0 on @O; the theorem follows. &
Under an additional assumption, Theorems 3.2, part (1), and 4.2 imply the
aforementioned result.
Corollary 4.3. If, in addition to the assumptions of Theorem 4.2, we assume the
function x-f x; ux belongs to L
1
loc
U; then u is a weak solution of (4.1).
We apply Corollary 4.3 to the Laplace equation Du 0 and to the minimal surface
equation
div
ru
1 jruj
2
_ 0:
Proposition 4.4. Let E
c
CE
d
be, respectively, a slight and a thin subset of U; and
suppose the following conditions are satised:
(1) u is continuous in U E
c
; and pointwise Lipschitz in U E
d
;
(2) ru has a locally bounded extension to U that is continuous in U E
c
and
pointwise Lipschitz in U E
d
;
(3) Du 0 almost everywhere in U:
Then u can be redened on E
c
so that it is harmonic in U:
ARTICLE IN PRESS
Proof. By Stepanoffs theorem, ru and Du are dened almost everywhere in U: We
show rst that uAL
N
loc
U: To this end, choose a ball B : Bz; r whose closure is
contained in U; and a c40 so that jruxjpc for almost all xAB: An admissible
segment is the nonempty intersection l of B and a line in R
m
such that l-E
c
|; and
l-E
d
is countable. Since I
m1
1
pH
m1
; proceeding as in the proof of Theorem 3.14,
we can nd an orthonormal base e
1
; y; e
m
in R
m
so that the union S
i
of all
admissible segments parallel to e
i
differs from B by a negligible set. If l is an
admissible segment and x; yAl; then the one-dimensional version of the Gauss
Green theorem yields juy uxjpcjy xj: We infer u is Lipschitz on S
m
i1
S
i
with the Lipschitz constant not larger than mc: As B S is a negligible set, our
assertion is proved.
Select a jAC
N
c
U; and a bounded BV set A such that cl ACU and supp jCint A:
Corollary 4.3, applied to the function h : x/x; implies
_
U
ru rj 0: As the vector
urj satises the assumptions of the GaussGreen theorem and uDjAL
1
A; we
have
_
U
uDj
_
A
uDj
_
A
uDj ru rj
_
A
divurj
_
@
A
urj n
A
0;
indeed, since uDj belongs to L
1
A by the rst part of the proof, part (1) of Theorem
3.2 shows the W-integrals, which would normally occur in the previous equality, can
be replaced by the Lebesgue integrals. Thus u is a distributional solution of Du 0;
and the proposition follows from [18, Corollary of Theorem 8.12]. &
Corollary 4.5. Let E
c
CE
d
be relatively closed subsets of U that are, respectively,
slight and thin, and suppose uAC
1
U E
c
: If u is locally Lipschitz in U and harmonic
in U E
d
; then it is harmonic in U:
Proof. As E
c
is relatively closed in U; the bounded vector eld ru has a locally
bounded extension u
0
: U-R
m
that is continuous in U E
c
: By our assumption, u is
C
N
in U E
d
: Since E
d
is relatively closed in U; both u and u
0
are differentiable in
U E
d
: Now the corollary follows from Proposition 4.4. &
Remark 4.6. Let K be a compact subset of U; and suppose u is locally Lipschitz in U
and harmonic in U K: If H
m1
K 0; then letting E
c
E
d
K; Corollary 4.5
yields the classical result: u can be redened on K so that it is harmonic in U: In special
situations, the same conclusion holds under weaker assumptions. For m 2; David
and Mattila [3] have shown it sufces to assume H
1
KoN and I
1
1
K 0: For
mX3; it is not known whether the assumptions H
m1
KoNand I
m1
1
K 0 are
sufcient; however, Mattila and Paramonov [14] proved that they are sufcient when
K belongs to a class of self-similar sets.
ARTICLE IN PRESS
Since closed I
m1
1
-negligible sets are generally not removable for Lipschitz
harmonic functions [14], the next proposition is interesting.
Proposition 4.7. Let E
0
CE
s
be relatively closed subsets of U such that I
m1
1
E
0
0
and the measure I
m1
1
L E
s
is s-nite. Suppose uAC
1
U E
0
is harmonic in U E
s
:
If u belongs to the Sobolev space W
2;1
loc
U; then it can be redened on E
0
so that it is
harmonic in U:
Proof. Select a jAC
N
c
U; and a bounded BV set A such that cl ACU and
supp jCint A: As j rj 0 on @
A; we can apply Theorem 3.14 to the vector

elds urj and jru: As in the proof of Proposition 4.4, we obtain
_
U
uDj
_
A
jDu 0; and the proposition follows from [18, Corollary of Theorem 8.12]. &
Remark 4.8. In view of the Ho lder and Sobolev inequalities, in Proposition 4.7 the
assumption uAW
2;1
loc
U is equivalent to assuming that all second partial derivatives
of u belong to L
1
loc
U:
Prior to considering the minimal surface equation
div
ru
1 jruj
2
_ 0; 4:2
recall that according to the regularity result of De Giorgi [8], each weak solution of
(4.2) in U which is locally Lipschitz is a real analytic function in U that solves (4.2) in
the classical sense. In other words, among locally Lipschitz functions in U there is no
difference between weak and classical solutions of (4.2).
Corollary 4.9. Let E
c
CE
d
be relatively closed subsets of U that are, respectively,
slight and thin, and suppose uAC
1
U E
c
: If u solves equation (4.2) in U E
d
; then u
can be redened on E
c
so that it is locally Lipschitz in U and solves equation (4.2) in U:
Proof. Since the function u is locally Lipschitz in the open set U E
c
; it is real
analytic in the open set U E
d
: Letting
hx :
x
1 jxj
2
_
for each xAU; the map h3ru is bounded and continuous in U E
c
; and
differentiable in U E
d
: As E
c
is a closed subset of U; the map h 3 ru has a
bounded extension to U which is still continuous in U E
c
and differentiable in
U E
d
: Theorem 4.2 implies that u is a weak solution of (4.2) in U: Consequently u
is a classical solution of (4.2) in U E
c
; and the corollary follows from [20]; cf.
Remark 4.10, (ii) below. &
ARTICLE IN PRESS
Remark 4.10. Special cases of Corollary 4.9 were obtained previously by various
authors. Specically, the corollary was proved by
(i) De Giorgi and Stampacchia [9] if E
d
E
c
is a slight compact subset of U;
(ii) Simon [20] if E
d
E
c
is a slight relatively closed subset of U;
(iii)
Harvey and Lawson [10] if E
c
| and the measure H
m1
L E
d
is locally nite;
cf. [11].
5. Closing remarks
In U consider the equation
divh 3 ru 0; 5:1
and dene a distribution F by the formula /F; jS :
_
U
h 3 ru rj for each
jAC
N
c
U: Let E be a relatively closed subset of U: If (5.1) has a weak solution in
U E; then supp FCE: Moreover, u is a weak solution of (5.1) in U whenever
F 0: Stated differently, E is removable whenever supp FCE implies F 0: In
Section 4 the conclusion F 0 was inferred from the smallness of E (in the sense of
measures H
m1
or I
m1
1
) and the regularity results for weak solutions.
The previous paragraph suggest the following denition.
Denition 5.1. Let Fbe a family of distributions in an open set UCR
m
: A collection
E of subsets of U is called removable with respect to Fif F 0 for each FAFsuch
that supp FCE for some EAE:
If v : R
m
-R
m
is a locally bounded Borel vector eld, denote by F
v
the ux of v
(Example 2.3), viewed as a distribution.
Example 5.2. Let F
c
consist of all F
v
such that vAC
1
R
m
supp F
v
: For instance,
ux :
0 if x
1
x
2
0;
1
x
1
2
x
2
2
p
x
2
; x
1
; 0; y; 0 otherwise
_
for each x : x
1
; y; x
m
in R
m
; denes F
u
AF
c
: We claim the collection of all slight
sets is removable with respect to F
c
: Indeed, if F
v
AF
c
and S supp F
v
is a
negligible set, then by Theorem 3.11,
_
R
m
jdiv v
_
R
m
v rj /F
v
; jS 0
for every jAC
N
c
R
m
S: Thus div v 0 almost everywhere outside S; and
consequently almost everywhere in R
m
: Now if S is a slight set, then another
ARTICLE IN PRESS
application of Theorem 3.11 shows that /F
v
; jS 0 for each jAC
N
c
R
m
: In
particular F
u
0; since supp F
u
is a linear space of dimension m 2: On the other
hand, if h is the indicator of 0; NCR and wx : hx
1
; 0; y; 0 for each x
x
1
; y; x
m
in R
m
; then F
w
AF
c
; F
w
a0; and supp F
w
is not a slight set.
Example 5.3. If F
d
consists of all F
v
AF
c
for which v is continuous, then proceeding
as Example 5.2, it is easy to see that the collection of all thin sets is removable with
respect to F
d
: Let h be the Cantor-Vitali function extended to a continuous function
on R by 0 and 1, and let wx : hx
1
; 0; y; 0 for each x x
1
; y; x
m
in R
m
:
Then F
w
AF
d
; and as F
w
a0; we see that supp F is not a thin set.
It is legitimate to ask whether the removable collections for the families F
c
and
F
d
indicated in the previous examples are the largest possible. The next example,
based on the continuum hypothesis (CH), may bear on this question with regard to
the family F
c
:
Example 5.4. Let KCR
m
be a compact set with H
m1
K40: According to
Frostmans lemma [13, Theorem 8.8], there is a nite Radon measure m in R
m
such
that mK40; supp mCK; and mBx; rpr
m1
for each xAR
m
and r40: In view of
[15, Theorem 4.7], we can nd a positive constant c so that j
_
R
m g dmjpcjjgjj for each
gABV; in particular,
F : g/
_
R
m
g dm : BV
N
c
-R
is a nontrivial w-charge with supp FCK: Yet, following the argument of [4, Section
3], one can show that under CH there is an H
m1
-measurable vector eld
v : R
m
-R
m
such that Fg
_
R
m v dDg for every gABV
N
c
: In particular, v is
Dg-measurable for all gABV
N
c
; and F is the ux of v:
References
[1] T. Bagby, W.P. Ziemer, Pointwise differentiability and absolute continuity, Trans. Amer. Math. Soc.
191 (1974) 129148.
[2] A.S. Besicovitch, On sufcient conditions for a function to be analytic, and behaviour of analytic
functions in the neighbourhood of non-isolated singular points, Proc. London Math. Soc. 32 (1931)
19.
[3] G. David, P. Mattila, Removable sets for Lipschitz harmonic functions in the plane, Rev. Mat.
Iberoamericana 16 (1) (2000) 137216.
[4] T. De Pauw, On SBV dual, Indiana U. Math. J. 47 (1) (1998) 99121.
[5] L.C. Evans, R.F. Gariepy, Measure Theory and Fine Properties of Functions, CRP Press, Boca
Raton, FL, 1992.
[6] K.J. Falconer, The Geometry of Fractal Sets, Cambridge University Press, Cambridge, 1985.
[7] H. Federer, Geometric Measure Theory, Springer, New York, 1969.
[8] E. De Giorgi, Sulla differentiabilita` e lanaliticita` delle estremali degli integrali multipli regolari, Mem.
Acad. Sci. Torino s. III (parte I) (1957) 2543.
ARTICLE IN PRESS
[9] E. De Giorgi, G. Stampacchia, Sulle singolarita` eliminabili delle ipersuperci minimali, Atti Accad.
Naz. Lincei Rend. u. Sci. Fis. Mat. Natur. 38 (1965) 352357.
[10] R. Harvey, H.B. Lawson, Extending minimal varieties, Inv. Math. 28 (1975) 209226.
[11] R. Harvey, J.C. Polking, Removable singularities of solutions of linear partial differential equations,
Acta Math. 125 (1970) 3956.
[12] U. Massari, M. Miranda, Minimal Surfaces of Codimension One, North-Holland, Amsterdam, 1984.
[13] P. Mattila, Geometry of Sets and Measures in Euclidean Spaces, Cambridge University Press,
Cambridge, 1995.
[14] P. Mattila, P.V. Paramonov, On geometric properties of harmonic Lip
1
-capacity, Pacic J. Math. 171
(1995) 469491.
[15] N.G. Meyers, W.P. Ziemer, Integral inequalities of Poincare and Wirtinger type for BV functions,
Amer. J. Math. 99 (1977) 13451360.
[16] W.F. Pfeffer, The GaussGreen theorem, Adv. Math. 87 (1991) 93147.
[17] W.F. Pfeffer, Derivation and Integration, Cambridge University Press, New York, 2001.
[18] W. Rudin, Functional Analysis, McGrawHill, New York, 1991.
[19] V.L. Shapiro, The divergence theorem for discontinuous vector elds, Ann. Math. 68 (1958) 604624.
[20] L. Simon, On a theorem of De Giorgi and Stampacchia, Math. Z. 155 (1977) 199204.
[21] E.M. Stein, Singular Integrals and Differentiability Properties of Functions, Princeton University
Press, Princeton, 1970.
[22] I. Tamanini, C. Giacomelli, Un tipo di approssimazione dallinterno degli insiemi di perimetro
nito, Rend. Mat. Acc. Lincei 1 (9) (1990) 181187.
ARTICLE IN PRESS

Removable Set GG

Încărcat de

Informații document

Descriere originală:

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Removable Set GG

Încărcat de

Drepturi de autor:

Formate disponibile

http://www.elsevier.

), or h 3 ru; or both, lack differentiability,

) where h is the identity map and f 0: Let E be a relatively closed

) are given in Section 4.

is the precise value of g; denote by Vg

in the interval a; b: In view of [7, Theorem 4.5.9, (23)],

for each gABV

A T: Then div v belongs to RA and the ux of v is the R-primitive of

: fx: xABg and B

f j for each jAC

f : In view of the claim

f is continuous, it is equal to zero everywhere. In particular FB

A T: Then div v belongs to

A; we can apply Theorem 3.14 to the vector

S-ar putea să vă placă și