Local currents in metric spaces

Urs Lang∗
November 7, 2008; revised May 5, 2010

In memoriam Juha Heinonen, 1960–2007

Abstract
Ambrosio and Kirchheim presented a theory of currents with finite
mass in complete metric spaces. We develop a variant of the theory that
does not rely on a finite mass condition, closely paralleling the classi-
cal Federer–Fleming theory. If the underlying metric space is an open
subset of a Euclidean space, we obtain a natural chain monomorphism
from general metric currents to general classical currents whose image
contains the locally flat chains and which restricts to an isomorphism
for locally normal currents. We give a detailed exposition of the slic-
ing theory for locally normal currents with respect to locally Lipschitz
maps, including the rectifiable slices theorem, and of the compactness
theorem for locally integral currents in locally compact metric spaces,
assuming only standard results from analysis and measure theory.

Contents
Introduction 2

1 Preliminaries 7

2 Metric currents 10

3 Support, boundary, and push-forward 17

4 Mass 22

5 Normal currents 31

6 Slicing 38

7 Projections to Euclidean spaces 48

8 Integer rectifiable currents 54

References 61

∗
Supported by the Swiss National Science Foundation.

1
2 U. Lang

Introduction
Currents in the sense of geometric measure theory, linear functionals on spaces
of differential forms, were introduced by G. de Rham in 1955 for use in the
theory of harmonic forms [12]. A few years later, H. Federer and W. H. Flem-
ing devised the class of rectifiable currents, generalized oriented surfaces with
integer multiplicities, and the subclass of integral currents, whose boundary
is of the same type. Their fundamental paper [15] from 1960 also furnished
the compactness theorem for integral currents and thereby a solution to the
Plateau problem for surfaces of arbitrary dimension and codimension in Eu-
clidean spaces. The theory of currents then rapidly developed into a powerful
apparatus in the calculus of variations. Federer’s monograph [14] gives a com-
prehensive account of the state of the subject prior to 1970. Since then, the
theory has been extended in various directions and has found numerous ap-
plications in geometric analysis and Riemannian geometry, far beyond pure
area minimization problems.
A breakthrough was achieved in 2000, when L. Ambrosio and B. Kirch-
heim [3], following ideas of E. De Giorgi [7], presented a theory of currents in
complete metric spaces. This elegant approach employs (m+1)-tuples of real-
valued Lipschitz functions in place of differential m-forms and provides some
new insights even if the ambient space is Euclidean. Ambrosio and Kirchheim
discovered new proofs of the boundary rectifiability theorem and the closure
theorem for rectifiable currents, valid in any complete metric space. As an
application, they obtained existence results for generalized Plateau problems
in compact metric spaces and certain Banach spaces. The theory of metric
currents has been further developed in [8], [32], [34], and some geometric ap-
plications have been found [33], [35]. Various interactions with other areas
have emerged and should be further explored. In this context we just mention
the functions of bounded higher variation [21], [8], [9], [28], [25], the recent
progress on (generalized) flat chains [37], [38], [18], [1], the scans from [16],
[10], [11], the differentiation theory on metric measure spaces [4], [22], [5],
and the derivations from [30], [31]. We refer to [19] for an excellent survey of
some of these and further related topics.
The metric currents considered by Ambrosio and Kirchheim have finite
mass by definition. This a priori assumption plays a crucial role in their
development of the theory, in particular it is used to derive the properties that
qualify the functionals under consideration as analogues of classical currents.
Here we present a somewhat different approach, strongly inspired by the work
of Ambrosio and Kirchheim, but not relying on a finite mass condition. In
addition, we switch to local formulations. At a later stage, this enables us to
discuss metric currents with locally finite mass and locally rectifiable currents.
For this purpose it is appropriate to assume the underlying metric space to
be locally compact. However, once some basic properties are established, the
theory readily extends to a theory of currents with locally compact support
in arbitrary metric spaces. Furthermore, in the case of finite mass, it is also
Local currents in metric spaces 3

possible to dispense with the restriction on the support and to incorporate

the class of Ambrosio–Kirchheim currents. In the second half of the paper,
which discusses currents with locally finite mass, a number of results are
local versions of those in [3]. However, in some instances, we give different
arguments, using less prerequisites. A central aim was to provide a detailed
account of the fundamentals of the theory, including a complete proof of the
compactness theorem for locally integral currents in locally compact metric
spaces, in a style readily accessible to non-specialists.
We now describe the contents of the paper in a more systematic way.
Detailed references will be given later in the individual sections.
In Sect. 1 we fix the notation and gather a few basic facts from analysis
and measure theory.
In Sect. 2 we turn to currents. A classical m-dimensional current in an
open set U ⊂ Rn is a real-valued linear function on the space of compactly
supported differential m-forms on U , continuous with respect to convergence
of forms in a suitable C ∞ -topology. Given a locally compact metric space X,
we substitute differential m-forms by (m + 1)-tuples (f, π1 , . . . , πm ) of real-
valued functions on X, where f is Lipschitz with compact support spt(f )
and π1 , . . . , πm are locally Lipschitz. We denote by D m (X) the set of all such
tuples. The guiding principle is that if X = U is an open subset of Rn and
if (f, π1 , . . . , πm ) ∈ Cc∞ (U ) × [C ∞ (U )]m , then this tuple represents the form
f dπ1 ∧ . . . ∧ dπm . An m-dimensional metric current T in X is defined as
an (m + 1)-linear real-valued function on D m (X), continuous with respect to
convergence of tuples in a suitable topology involving locally uniform bounds
on Lipschitz constants, and satisfying T (f, π1 , . . . , πm ) = 0 whenever some πi
is constant on a neighborhood of spt(f ). The vector space of m-dimensional
metric currents in X is denoted by Dm (X). The terminology is justified by
the fact that the defining conditions of a metric current give rise to a set of
further properties, corresponding to the usual rules of calculus for differential
forms. Namely, every T ∈ Dm (X) is alternating in the m last arguments,
and the following product rule holds: If (f, π1 , . . . , πm ) ∈ D m (X), and if
g : X → R is locally Lipschitz, then

T (f, gπ1 , π2 , . . . , πm ) = T (f g, π1 , . . . , πm ) + T (f π1 , g, π2 , . . . , πm ).

We also obtain a chain rule, a special case of which states that if (f, π) =
(f, π1 , . . . , πm ) ∈ D m (X) and g = (g1 , . . . , gm ) ∈ [C 1,1 (Rm )]m , i.e., the partial
derivatives of gi are locally Lipschitz, then

T (f, g1 ◦ π, . . . , gm ◦ π) = T (f det((Dg) ◦ π), π1 , . . . , πm ).

Every function u ∈ L1loc (U ) on an open set U ⊂ Rm induces a metric current

[u] ∈ Dm (U ) satisfying
Z
[u](f, g) = uf det(Dg) dx
U
4 U. Lang

for all (f, g) = (f, g1 , . . . , gm ) ∈ D m (U ). This corresponds to the integration

of a simple m-form over U . The chain rule plays a crucial role in the devel-
opment of the theory. In particular, it is used to show (in Sect. 5) that for
every open set U ⊂ Rn and every m, there is an injective linear map Cm from
Dm (U ) into the space of general classical m-currents in U such that

Cm (T )(f dg1 ∧ . . . ∧ dgm ) = T (f, g1 , . . . , gm )

for all (f, g1 , . . . , gm ) ∈ Cc∞ (U )×[C ∞ (U )]m . This makes the aforesaid guiding
principle rigorous. Some more properties of these comparison maps Cm are
mentioned further below.
In Sect. 3 we define the support spt(T ) ⊂ X of a metric current T and
discuss the boundary and push-forward operators. For a classical m-current
T̄ , the boundary ∂ T̄ is the (m−1)-current satisfying ∂ T̄ (φ) = T̄ (dφ) for every
(m − 1)-form φ. Correspondingly, the boundary ∂T ∈ Dm−1 (X) of a metric
current T ∈ Dm (X) verifies

∂T (f, π1 , . . . , πm−1 ) = T (σ, f, π1 , . . . , πm−1 )

for all (f, π1 , . . . , πm−1 ) ∈ D m−1 (X) and for all σ such that σ = 1 on some
neighborhood of spt(f ). We have ∂ ◦ ∂ = 0, and, in case X = U is an open
set in Rn , ∂ ◦ Cm = Cm−1 ◦ ∂, so that the Cm form a chain map. The push-
forward F# T̄ of a general classical current T̄ is defined for every smooth map
F whose restriction to the support of T̄ is proper, and considerable efforts
are required to extend the definition, for particular classes of currents, to
locally Lipschitz maps. Given a metric current T ∈ Dm (X) and another
locally compact metric space Y , the push-forward F# T ∈ Dm (Y ) is defined
for every locally Lipschitz map F : D → Y such that spt(T ) ⊂ D ⊂ X and
F |spt(T ) is proper. In case D = X, we have

F# T (f, π1 , . . . , πm ) = T (f˜, π1 ◦ F, . . . , πm ◦ F )

whenever (f, π1 , . . . , πm ) ∈ D m (Y ) and f˜: X → R is a compactly supported

Lipschitz function that agrees with f ◦ F on spt(T ).
In Sect. 4 we discuss the notion of mass. Given a metric current T ∈
Dm (X), we define its mass MV (T ) in an open set V ⊂ X as the least number
M ∈ [0, ∞] such that X
T (fλ , π λ ) ≤ M
λ∈Λ

P set, (fλ , π ) ∈ D (X), π1 , . . . , πm are 1-Lipschitz,

whenever Λ is a finite λ m λ λ

spt(fλ ) ⊂ V , and λ∈Λ |fλ | ≤ 1. If a metric current T ∈ Dm (X) has

locally finite mass, then there is an associated Radon measure kT k on X,
characterized by kT k(V ) = MV (T ) for all open sets V ⊂ X, and
Z
T (f, π) ≤ |f | dkT k
X
Local currents in metric spaces 5

whenever (f, π) ∈ D m (X) and the restrictions of π1 , . . . , πm to spt(f ) are

1-Lipschitz. This last inequality allows to extend T to all tuples (f, π) such
that f is a bounded Borel function with compact support and π1 , . . . , πm are
still locally Lipschitz. For a Borel set B ⊂ X, the restriction T b B is then
defined as the m-current satisfying

(T b B)(f, π) = T (χB f, π)

for all (f, π) ∈ D m (X), where χB is the characteristic function of B. For a

locally bounded Borel function g : X → R, T b g is defined similarly.
A metric m-current T is called (locally) normal if both the mass of T
and the mass of ∂T are (locally) finite. A fundamental result about locally
normal metric currents, proved in Sect. 5, is the compactness theorem: If
T1 , T2 , . . . ∈ Dm (X) are currents with separable support, and if

sup(MV (Tn ) + MV (∂Tn )) < ∞

n

for every open set V ⊂ X with compact closure, then there is a subse-
quence Tn(1) , Tn(2) , . . . that converges weakly to some T ∈ Dm (X), i.e.,
limi→∞ Tn(i) (f, π) = T (f, π) for every (f, π) ∈ D m (X). Since MV is lower
semicontinuous with respect to weak convergence, and also ∂Tn(i) → ∂T
weakly, the limit T is locally normal. In case X = U is an open set in Rn , the
restriction of the comparison map Cm to the vector space of locally normal
currents is an isomorphism onto the space of classical locally normal currents.
An important technique in the theory of currents consists in relating in-
formation on the structure of a current T to properties of lower dimensional
slices of T in the level sets of a function. In Sect. 6 we discuss slicing of a
locally normal current T ∈ Dm (X) with respect to a locally Lipschitz map
π : X → Rk , where 1 ≤ k ≤ m. Let Tπ ∈ Dm−k (X) be the locally normal
current satisfying

Tπ (f, g1 , . . . , gm−k ) = T (f, π1 , . . . , πk , g1 , . . . , gm−k )

for all (f, g) ∈ D m−k (X). If spt(T ) is separable, then for almost every y ∈ Rk
there is a locally normal current in Dm−k (X) with support in π −1 {y}∩spt(T ),
denoted by hT, π, yi, such that
Z
hT, π, yi(f, g) dy = Tπ (f, g)
Rk

for all (f, g) ∈ D m−k (X). Moreover, for every Borel set B ⊂ X,
Z
khT, π, yik(B) dy = kTπ k(B).
Rk

The first identity also holds more generally if f is a bounded Borel function
with compact support.
6 U. Lang

Slicing is particulary important and useful when k = 1, for geometric ap-

plications, or when k = m. In the latter case, the slices are 0-dimensional,
and π maps any compactly supported portion of T to a current of maximal
dimension in Rm . This situation is closely inspected in Sect. 7. The slic-
ing theorem leads to a fundamental identity: If f : X → R is any bounded
Borel function with compact support, then π# (T b f ) ∈ Dm (Rm ) is a standard
current [uf ], for some uf ∈ L1 (Rm ), and
hT, π, yi(f ) = uf (y)
for almost every y ∈ Rm . Moreover, if f is Lipschitz, then [uf ] is normal,
and uf is a function of bounded variation. Such functions satisfy a Lipschitz
condition outside a set of small Lebesgue measure. Exploiting the resulting
Lipschitz property of the function y 7→ hT, π, yi(f ), we obtain a partial recti-
fiability result for every locally normal current T with separable support: Let
A ⊂ spt(T ) be the set of all x such that {x} is an atom of the corresponding
slice hT, π, π(x)i. Up to a set of kTπ k measure zero, A can be represented as
the union of countable many pairwise disjoint compact sets Bi ⊂ A such that
π|Bi is a bi-Lipschitz map into Rm .
In the final Sect. 8 we turn to rectifiable currents. For a general current
T ∈ Dm (X) with locally finite mass, the measure kT k may be diffused, so
that T does not correspond to a generalized m-dimensional surface in any
sense. We call T a locally integer rectifiable current if kT k is concentrated
on some countably m-rectifiable set E, i.e., the union of countably many
Lipschitz images of subsets of Rm , and if T satisfies the following integral-
ity condition: Whenever B ⊂ X is a Borel set with compact closure and
π : X → Rm is Lipschitz, then π# (T b B) = [uB,π ] for some uB,π ∈ L1 (Rm , Z).
From these conditions it follows that the support of T is separable and that
kT k is absolutely continuous with respect to m-dimensional Hausdorff mea-
sure H m . The slicing theory is then supplemented with the rectifiable slices
theorem, relying on the above partial rectifiability result and the identity
hT, π, yi(χB ) = uB,π (y): Given a locally normal current T ∈ Dm (X) with sep-
arable support and k ∈ {1, . . . , m}, T is locally integer rectifiable if and only
if for each Lipschitz map π : X → Rk , the (m − k)-dimensional slice hT, π, yi
is locally integer rectifiable for almost every y ∈ Rk . We call T ∈ Dm (X) a
locally integral current if both T and ∂T are locally integer rectifiable. In
particular, every such T is locally normal. By the boundary rectifiability
theorem, every locally integer rectifiable and locally normal current is locally
integral. This follows easily from the rectifiable slices theorem, by induc-
tion on m. The closure theorem for locally integral currents states that
if T1 , T2 , . . . ∈ Dm (X) is a sequence of locally integral currents converging
weakly to T ∈ Dm (X), and if
sup(MV (Tn ) + MV (∂Tn )) < ∞
n
for every open set V ⊂ X with compact closure, then T is locally integral. The
proof is another simple inductive application of the rectifiable slices theorem.
Local currents in metric spaces 7

By combining this result with the compactness theorem for locally normal
currents mentioned earlier, we obtain the compactness theorem for locally
integral currents.
Further results and applications will be discussed elsewhere.

Acknowledgements. My sincere thanks go to Luigi Ambrosio and Bernd

Kirchheim for their ground-breaking work [3] and for the hospitality I enjoyed
at the Scuola Normale Superiore in Pisa during a visit in January 2005, when
the first results of this paper were obtained. The many discussions with Mario
Bonk and Juha Heinonen at the University of Michigan in February 2007 led
to several improvements of the text and prepared the way for future work. I
gratefully remember those three weeks. I’m indebted to Pekka Koskela and
Jouni Parkkonen for their support during the 17th Jyväskylä Summer School,
where this theory was presented in August 2007. I also thank Robert Hardt
and Roger Züst for some useful comments and corrections.

1 Preliminaries
We now fix the notation and collect some basic facts from analysis and mea-
sure theory. A few more prerequisites will be discussed in individual sections.
Given a point x in a metric space X = (X, d), B(x, r) := {y : d(x, y) ≤ r}
and U(x, r) := {y : d(x, y) < r} denote the closed and open ball, respectively,
with center x and radius r.

1.1 Lipschitz maps

Let X = (X, d) and Y = (Y, d) be two metric spaces. For l ∈ [0, ∞), a map
f : X → Y is l-Lipschitz if

d(f (x), f (x0 )) ≤ l d(x, x0 )

for all x, x0 ∈ X. The map f : X → Y is Lipschitz if the Lipschitz constant

Lip(f ) := inf{l ∈ [0, ∞) : f is l-Lipschitz}

is finite, and f is locally Lipschitz if every point in X has a neighborhood such

that the restriction of f to this neighborhood is Lipschitz. Note that then
f |K is Lipschitz for every compact set K ⊂ X. We let Liploc (X, Y ) be the
set of locally Lipschitz maps from X into Y , Lip(X, Y ) and Lipl (X, Y ) the
subsets of Lipschitz maps and l-Lipschitz maps, respectively. In this notation,
we abbreviate (X, R) by (X). Note that Liploc (X) forms an algebra, while
Lip(X) is an algebra if and only if X is bounded. In fact, if f, g ∈ Lip(X)
are two bounded Lipschitz functions on a metric space X, then the product
f g is Lipschitz with Lipschitz constant

Lip(f g) ≤ kf k∞ Lip(g) + kgk∞ Lip(f ), (1.1)

8 U. Lang

where k · k∞ is the supremum norm.

A map f : X → Y is bi-Lipschitz if there is a constant b ∈ [1, ∞) such
that
b−1 d(x, x0 ) ≤ d(f (x), f (x0 )) ≤ b d(x, x0 )
for all x, x0 ∈ X.
If A ⊂ X and f ∈ Lipl (A), then there exists an extension f¯ ∈ Lipl (X),
i.e., f¯|A = f . In fact, one may simply define

f¯(x) := inf (f (a) + l d(a, x)) (1.2)

a∈A

for x ∈ X. By applying this result to each component of a map f =

(f1 , . . . , fm ) ∈ Lipl (A, Rm ), one obtains an extension f¯ ∈ Lip√ml (X, Rm )
of f .
Every uniformly continuous and bounded function f : X → R is a uniform
limit of a sequence of Lipschitz functions (see e.g. [17, Theorem 6.8]).
By Rademacher’s theorem, every f ∈ Liploc (Rm , Rn ) is differentiable at
L -almost all points of Rm , where L m denotes (outer) Lebesgue measure.
m

1.2 Borel functions and Baire functions

Given a topological space X, Bloc ∞ (X) denotes the algebra of real-valued,

locally bounded Borel functions on X, B ∞ (X) the subalgebra of bounded

functions, and Bc∞ (X) the subalgebra of bounded and compactly supported
functions.
A class Φ of real-valued functions on a set X is called a Baire class if
the following holds: Whenever f1 , f2 , . . . ∈ Φ and fi (x) → g(x) ∈ R for each
x ∈ X, then g ∈ Φ. In case X is a topological space, f : X → R is a Baire
function if it belongs to the smallest Baire class containing all continuous
functions. Since the Borel functions form a Baire class, every Baire function
is a Borel function. Conversely, if X is a metric space, then it is not difficult to
see that characteristic functions of Borel sets are Baire functions and, hence,
every Borel function is also a Baire function (cf. [14, 2.2.15]).

1.3 Outer measures and Radon measures

An outer measure ν on ν : 2X → [0, ∞] such that
Pa∞ set X is a set function S
ν(∅) = 0 and ν(A) ≤ k=1 ν(Ak ) whenever A ⊂ ∞ k=1 Ak . A set A ⊂ X is
called ν-measurable if ν(B) = ν(B ∩ A) + ν(B \ A) for all B ⊂ X. The family
of all ν-measurable sets forms a σ-algebra, and the restriction of ν to this
σ-algebra is a measure. If E ⊂ X is any set, then νb E is the outer measure
satisfying
(νb E)(A) = ν(E ∩ A)
for all A ⊂ X. Every ν-measurable set is also (νb E)-measurable. We say
that ν is concentrated on E if ν(X \ E) = 0 or, equivalently, ν = νb E.
Local currents in metric spaces 9

Let ν be an outer measure on a topological space X. The support spt(ν) of

ν in X is the closed set of all x ∈ X such that ν(U ) > 0 for every neighborhood
U of x. We call ν Borel regular if all Borel sets are ν-measurable and if every
set A ⊂ X is contained in a Borel set B with ν(B) = ν(A). If ν is Borel
regular and E is a Borel set, then νb E is Borel regular.
Now let ν be an outer measure on a metric space X. Carathéodory’s
criterion says that if ν(A)+ν(B) = ν(A∪B) for all A, B ⊂ X with inf{d(x, y) :
x ∈ A, y ∈ B} > 0, then all Borel sets are ν-measurable. If ν is Borel regular
and B is a ν-measurable set contained in the union of countably many open
sets Ui with ν(Ui ) < ∞, and if  > 0, then there is an open set V such that
B ⊂ V and ν(V \ B) < .
An outer measure ν on a locally compact Hausdorff space X is called a
Radon measure if Borel sets are ν-measurable, ν is finite on compact sets,

ν(V ) = sup{ν(K) : K ⊂ X is compact, K ⊂ V }

for every open set V ⊂ X, and

ν(A) = inf{ν(V ) : V ⊂ X is open, A ⊂ V }

for every set A ⊂ X. Then it is also true that if B is a ν-measurable set with
ν(B) < ∞, and if  > 0, then ν(B \ K) <  for some compact set K ⊂ B
(cf. [14, 2.2.5]).
For m ∈ N, we denote by

αm := L m (B(0, 1))

the Lebesgue measure of the unit ball in Rm , and we put α0 := 1. Given

a metric space X, the m-dimensional Hausdorff measure of a set A ⊂ X
is defined by H m (A) := limδ→0+ Hδm (A), where Hδm (A) is the infimum of
C∈C αm (diam(C)/2) over all countable coverings C of A with diam(C) :=
m
P
sup{d(x, y) : x, y ∈ C} ≤ δ for all C ∈ C . For every m, H m is a Borel regular
outer measure on X. With the chosen normalization, H m = L m on Rm .

1.4 Maximal functions and Lebesgue points

Suppose that µ is a finite Borel measure on Rm , i.e., a σ-additive function
µ : BX → [0, ∞), where BX is the σ-algebra of Borel sets in X. We denote
by Mµ : Rm → [0, ∞] the maximal function of µ, i.e.,

µ(B(x, r))
Mµ (x) := sup .
r>0 αm r m

A simple covering argument shows that

L m ({x ∈ Rm : Mµ (x) > s}) ≤ 3m s−1 µ(Rm ) (1.3)

10 U. Lang

for all s > 0. In particular, Mµ (x) < ∞ for L m -almost every x ∈ Rm . Note
also that Mµ is lower semicontinuous on Rm .
We further recall that, given a function u ∈ L1 (Rm ), L m -almost every
x ∈ Rm is a Lebesgue point of u, i.e.,
Z
1
lim |u(y) − u(x)| dy = 0.
r→0 αm r m B(x,r)

Moreover, if we associate to each x ∈ Rm a sequence of Borel sets Ei (x) with

the property that Ei (x) ⊂ B(x, ri (x)) and L m (Ei (X)) ≥ β(x)αm ri (x)m for
some ri (x) → 0 and β(x) > 0, then
Z
1
u(x) = lim u(y) dy (1.4)
i→∞ L m (Ei (x)) E (x)
i

at every Lebesgue point x of u, hence for L m -almost every x ∈ Rm . (See

e.g. the first section of [26, Ch. 7].)

1.5 Smoothing
We shall use the following basic facts regarding smoothing. Let η ∈ Cc∞ (Rm )
R a mollifier, so that spt(η) ⊂ U(0, 1),1 η(−z)
be = η(z) ≥ 0 for all z ∈ Rm , and
m
Rm η(z) dz = 1. Recall that for g ∈ Lloc (R ), the convolution defined by
Z Z
(η ∗ g)(x) := η(z)g(x − z) dz = η(x − z)g(z) dz
Rm Rm

for x ∈ Rm satisfies η ∗ g ∈ C ∞ (Rm ), and spt(η ∗ g) ⊂ spt(η) + spt(g). If

g ∈ Liploc (Rm ), then the partial derivatives of η ∗ g are given by

Dk (η ∗ g) = η ∗ Dk g, (1.5)

k = 1, . . . , m. If g is bounded, then kη ∗ gk∞ ≤ kgk∞ . If g is Lipschitz, then

η ∗ g is Lipschitz, with constant

Lip(η ∗ g) ≤ Lip(g). (1.6)

NowRput ηj (z) := j m η(jz) for z ∈ Rm and j ∈ N, so that spt(ηj ) ⊂ U(0, 1/j)

and Rm ηj (z) dz = 1. As j → ∞, (ηj ∗g)(x) → g(x) whenever x is a Lebesgue
point of g, moreover, the convergence is locally uniform if g is continuous.
(See e.g. [14, 4.1.2].)

2 Metric currents
From now on, unless otherwise stated, X will always denote a locally compact
metric space. We write A b X if A ⊂ X and the closure of A is compact.
We let D(X) be the algebra of all f ∈ Lip(X) whose support spt(f ) is
compact; these will serve as test functions. For every compact set K ⊂ X
Local currents in metric spaces 11

and every constant l ≥ 0 we put LipK,l (X) := {f ∈ Lipl (X) : spt(f ) ⊂ K},
so that D(X) is the union of all LipK,l (X). Then we equip D(X) with a
locally convex vector space topology τ with respect to which

fj → f in D(X) (2.1)

if and only if all fj belong to some fixed LipK,l (X) and fj → f pointwise
on X for j → ∞, which implies that fj → f uniformly on X. Explicitly,
this topology τ is given as follows. Let β be the collection of all absolutely
convex sets W ⊂ D(X) with the following property: For every pair (K, l)
and every f ∈ W ∩ LipK,l (X), there is an  > 0 such that g ∈ W whenever
g ∈ LipK,l (X) and kf − gk∞ < . (Recall that W is absolutely convex if and
only if sf + tg ∈ W for all f, g ∈ W and s, t ∈ R with |s| + |t| ≤ 1.) Then
β is a local base of τ at 0, thus τ is the collection of all unions of sets of
the form f + W , where f ∈ D(X) and W ∈ β. (See e.g. [27, p. 152] for the
corresponding construction in classical distribution theory.)
Similarly, we equip Liploc (X) with a locally convex vector space topology
with respect to which
πj → π in Liploc (X) (2.2)
if and only if for every compact set K ⊂ X there is a constant lK such that
Lip(πj |K ) ≤ lK for all j and πj → π pointwise, hence uniformly, on K for
j → ∞. For a fixed compact set K ⊂ X, let βK be the collection of all
absolutely convex sets W ⊂ Liploc (X) with the following property: For every
l ≥ 0 and every π ∈ W with Lip(π|K ) ≤ l, there is an  > 0 such that ρ ∈ W
whenever ρ ∈ Liploc (X), Lip(ρ|K ) ≤ l, and k(π − ρ)|K k∞ < . The union
of all βK forms a local base of the topology of Liploc (X). The topologies of
D(X) and Liploc (X) will only be used through (2.1) and (2.2).
We define the spaces

D 0 (X) := D(X), D m (X) := D(X) × [Liploc (X)]m (m ∈ N),

which will serve as substitutes for the spaces of compactly supported m-forms.
The guiding principle is that

(f, π1 , . . . , πm ) ∈ D m (X) represents f dπ1 ∧ . . . ∧ dπm (2.3)

if X is an open subset of Rn and the f, π1 , . . . , πm are smooth. This correspon-

dence will be made precise in Theorem 5.5. The space D m (X) is equipped
with the product topology. Thus,

(f j , π1j , . . . , πm
j
) → (f, π1 , . . . , πm ) in D m (X) (2.4)

if and only if f j → f in D(X) and πij → πi in Liploc (X) for i = 1, . . . , m.

Definition 2.1 (metric current). For m ∈ {0}∪N, an m-dimensional metric
current T in X is a function T : D m (X) → R satisfying the following three
conditions:
12 U. Lang

(1) (multilinearity) T is (m + 1)-linear;

(2) (continuity) T (f j , π1j , . . . , πm

j
) → T (f, π1 , . . . , πm ) if (f j , π1j , . . . , πm
j
)→
(f, π1 , . . . , πm ) in D m (X);

(3) (locality) in case m ≥ 1, T (f, π1 , . . . , πm ) = 0 whenever some πi is

constant on a neighborhood of spt(f ).

The vector space of m-dimensional metric currents in X is denoted by

Dm (X).

We endow Dm (X) with the locally convex weak topology with respect to which
Tn → T if and only if

Tn (f, π1 , . . . , πm ) → T (f, π1 , . . . , πm )

for every (f, π1 , . . . , πm ) ∈ D m (X).

As a first consequence of the defining conditions, we note the following

strict form of the locality property:

T (f, π1 , . . . , πm ) = 0 whenever some πi is constant on spt(f ). (2.5)

To see this, let βj : R → R (j ∈ N) be the 1-Lipschitz function satisfying

−βj (−s) = βj (s) = max{s − (1/j), 0} for s ≥ 0. Then βj ◦ f → f in D(X),
and spt(f ) contains a neighborhood of spt(βj ◦ f ) for every j, so that (2.5)
follows. An alternative proof uses the continuity in the respective argument.
Suppose that (πi − c)|spt(f ) = 0 for some c ∈ R. Then βj ◦ (πi − c) vanishes
on some neighborhood of spt(f ), and βj ◦ (πi − c) → πi − c in Liploc (X). This
implies that T (f, π1 , . . . , πm ) = 0.
The following simple lemma shows that the space [D(X)]m+1 would serve
the same purpose as D m (X). However, the guiding principle (2.3) suggests
to think of π1 , . . . , πm as coordinate functions, so that the choice of Liploc (X)
instead of D(X) seems more appropriate. In the proof of Theorem 2.5 (chain
rule) we shall also benefit from the fact that Liploc (Rn ) comprises the real
polynomials in n variables.

Lemma 2.2. Suppose T : [D(X)]m+1 → R is a function satisfying the three

conditions of Definition 2.1 with [D(X)]m+1 in place of D m (X). Then T
extends uniquely to a current T ∈ Dm (X).

In particular, every metric current T ∈ Dm (X) is determined by its values

on [D(X)]m+1 .

Proof. In view of conditions (1) and (3), the function T can be extended to
D m (X) so that

T (f, π1 , . . . , πm ) = T (f, σπ1 , . . . , σπm )

Local currents in metric spaces 13

whenever (f, π1 , . . . , πm ) ∈ D m (X), σ ∈ D(X), and σ = 1 on some neighbor-

hood of spt(f ). The right side is clearly independent of the choice of σ, for
if τ ∈ D(X) is another such function, then (σ − τ )πi vanishes on a neighbor-
hood of spt(f ) for i = 1, . . . , m. Note also that whenever K ⊂ X is compact,
there is a σ ∈ D(X) such that σ = 1 on some neighborhood of K, since X
is locally compact. Now each of the three properties of the given function T
implies the respective property of the extended function via an appropriate
choice of σ.

By inserting a number of locally Lipschitz functions into T and keeping

them fixed, we obtain a current of smaller dimension. This will often be used
to simplify notation.

Definition 2.3. For T ∈ Dm (X) and (u, v) ∈ Liploc (X)×[Liploc (X)]k , where
m ≥ k ≥ 0, we define the current T b (u, v) ∈ Dm−k (X) by

(T b (u, v))(f, g) := T (uf, v, g)

= T (uf, v1 , . . . , vk , g1 , . . . , gm−k )

for (f, g) ∈ D m−k (X).

In case k = 0, the definition simply reads

(T b u)(f, g) := T (uf, g).

In case k ≥ 1, the placement of the functions v1 , . . . , vk on the right side is in

accordance with the corresponding definition in the smooth case; it has the
property that
(T b (1, v))b (1, w) = T b (1, v, w). (2.6)
The tuple (u, v) ∈ Liploc (X) × [Liploc (X)]k corresponds to the k-form u dv1 ∧
. . . ∧ dvk . It is clear that T b (u, v) is indeed an element of Dm−k (X), as it
would be with any other placement of v1 , . . . , vk .
We now show that the defining properties of a general metric current give
rise to a set of further properties, corresponding to the usual rules of calculus
for differential forms.

Proposition 2.4 (alternating property and product rule). Suppose T ∈

Dm (X), m ≥ 1, and (f, π1 , . . . , πm ) ∈ D m (X).

(1) In case m ≥ 2, if πi = πj for some pair of distinct indices i, j, then

T (f, π1 , . . . , πm ) = 0.

(2) For all g, h ∈ Liploc (X),

T (f, gh, π2 , . . . , πm ) = T (f g, h, π2 , . . . , πm ) + T (f h, g, π2 , . . . , πm ).
14 U. Lang

Proof. To prove (1), it suffices to show that if T ∈ D2 (X) and (f, π) ∈ D 1 (X),
then T (f, π, π) = 0. For k ∈ Z, let ρk : R → R be the piecewise affine 1-
P function with ρk |[2k,2k+1] = 1 and spt(ρk ) = [2k − 1, 2k + 2]. Note
Lipschitz
that k∈Z ρk = 1. Let σ, σ̄ : R → R denote the piecewise affine 4-Lipschitz
functions such that σ|spt(ρk ) = 2k for k even and σ̄|spt(ρk ) = 2k for k odd.
Then X
T (f, σ ◦ π, σ̄ ◦ π) = T ((ρk ◦ π)f, σ ◦ π, σ̄ ◦ π);
k∈Z
note that (ρk ◦ π)f = 0 for almost all k since π|spt(f ) is bounded. By the strict
locality property (2.5), each summand is zero because σ ◦π or σ̄ ◦π is constant
on spt(ρk ◦ π) for k even or odd, respectively. Hence T (f, σ ◦ π, σ̄ ◦ π) = 0.
In the above definitions of the functions ρk , σ, σ̄ we may equally well replace
the unit by 1/j, for j ∈ N. The argument then shows that
T (f, σj ◦ π, σ̄j ◦ π) = 0,
where σj (s) = σ(js)/j and σ̄j (s) = σ̄(js)/j for s ∈ R. Letting j tend to
∞, we obtain T (f, π, π) = 0 by the continuity of T , since σj ◦ π → π and
σ̄j ◦ π → π in Liploc (X).
For the proof of (2), it suffices to show that if T ∈ D1 (X) and (f, g) ∈
D (X), then T (f, g 2 ) = 2T (f g, g). Let ρk , σ, σ̄ be defined as above. Since
1

σ ◦ g|spt(ρk ◦g) = 2k for k even and σ̄ ◦ g|spt(ρk ◦g) = 2k for k odd, and since
(ρk ◦ g)f = 0 for almost all k, the multilinearity of T and (2.5) give
T (f, (σ ◦ g)(σ̄ ◦ g))
X
= T ((ρk ◦ g)f, (σ ◦ g)(σ̄ ◦ g))
k∈Z
X X
= 2kT ((ρk ◦ g)f, σ̄ ◦ g) + 2kT ((ρk ◦ g)f, σ ◦ g)
k even k odd
X
= 2kT ((ρk ◦ g)f, σ ◦ g + σ̄ ◦ g)
k∈Z
= T ((τ ◦ g)f, (σ + σ̄) ◦ g)
P
for the piecewise affine 2-Lipschitz function τ := k∈Z 2kρk , which satisfies
τ |[2k,2k+1] = 2k for k ∈ Z. Rescaling by the factor 1/j, as in the proof of (1),
we obtain the identity
T (f, (σj ◦ g)(σ̄j ◦ g)) = T ((τj ◦ g)f, (σj + σ̄j ) ◦ g),
where τj (s) = τ (js)/j for s ∈ R. Taking the limit for j → ∞ we conclude
that T (f, g 2 ) = T (gf, 2g).

We now deduce a chain rule, which subsumes both the alternating prop-
erty and the case g = h of Proposition 2.4(2). For an open set U ⊂
Rn , C 1,1 (U ) denotes the space of all g ∈ C 1 (U ) with partial derivatives
D1 g, . . . , Dn g ∈ Liploc (U ). For n ≥ m ≥ 1, we let Λ(n, m) be the set of
all strictly increasing maps λ : {1, . . . , m} → {1, . . . , n}.
Local currents in metric spaces 15

Theorem 2.5 (chain rule). Suppose m, n ≥ 1, T ∈ Dm (X), U ⊂ Rn is an

open set, f ∈ D(X), π = (π1 , . . . , πn ) ∈ Liploc (X, U ), and g = (g1 , . . . , gm ) ∈
[C 1,1 (U )]m . If n ≥ m, then
X  m

T (f, g ◦ π) = T f det (Dλ(k) gi ) ◦ π i,k=1 , πλ(1) , . . . , πλ(m) .
λ∈Λ(n,m)

If n < m, then T (f, g ◦ π) = 0.

Proof. For illustration, suppose first that n = m = 1. In this case, the result
says that
T (f, g ◦ π) = T (f (g 0 ◦ π), π) (2.7)
whenever T ∈ D1 (X), U ⊂ R is an open set, f ∈ D(X), π ∈ Liploc (X, U ),
and g ∈ C 1,1 (U ). From the product rule, Proposition 2.4(2), we obtain the
power rule

T (f, π r ) = T (f rπ r−1 , π) ((f, π) ∈ D 1 (X), r ∈ N) (2.8)

by induction on r. Hence, (2.7) holds if g is (the restriction of) a polynomial.

Next, if g ∈ C 2 (U ), there is a sequence of polynomials pj such that pj → g,
p0j → g 0 , and p00j → g 00 locally uniformly on U as j → ∞. Then pj ◦ π → g ◦ π
and p0j ◦ π → g 0 ◦ π in Liploc (X), thus (2.7) follows by continuity. Finally, a
smoothing argument extends the result to all g ∈ C 1,1 (U ). For this, note that
there is no loss of generality in assuming that spt(g) is compact; by (2.5), we
may replace g by σg for any σ ∈ C 1,1 (U ) such that σ = 1 on π(spt(f )) and
spt(σ) is compact.
Now let n ≥ m = 1. We must show that
n
X

T (f, g ◦ π) = T f ((Dk g) ◦ π), πk (2.9)
k=1

whenever T ∈ D1 (X), U ⊂ Rn is an open set, f ∈ D(X), π = (π1 , . . . , πn ) ∈

Liploc (X, U ), and g ∈ C 1,1 (U ). As above, the product rule implies that this
identity holds if g is a polynomial in the variables x1 , . . . , xn . Furthermore, if
g ∈ C 2 (U ), there is a sequence of polynomials pj = pj (x1 , . . . , xn ) such that
pj → g, Dk pj → Dk g, and Dl Dk pj → Dl Dk g locally uniformly on U for all
k, l ∈ {1, . . . , n} (see e.g. [6, p. 57] for a direct proof). The continuity of T
then yields (2.9), and a smoothing argument shows the same identity for all
g ∈ C 1,1 (U ).
Finally, the general result for m, n ≥ 1 follows from (2.9), applied to each
gi , and the alternating property.

Suppose U is an open subset of Rn , T ∈ Dm (U ), and π is the identity map

on U , thus πi (x) = xi . In case m > n, the chain rule says that T (f, g) = 0 for
all (f, g) ∈ D(U ) × [C 1,1 (U )]m , and a smoothing argument then shows that
T = 0. Hence
Dm (U ) = {0} for m > n. (2.10)
16 U. Lang

In case m = n, we obtain

T (f, g) = T (f det(Dg), π) (2.11)

for all (f, g) ∈ D(U ) × [C 1,1 (U )]m . We now arrive at the first family of
examples of metric currents, corresponding to the integration of a simple
m-form over U ⊂ Rm .

Proposition 2.6 (standard example). Suppose U ⊂ Rm is open, m ≥ 1.

Every function u ∈ L1loc (U ) induces a current [u] ∈ Dm (U ) satisfying
Z
[u](f, g) = uf det(Dg) dx
U

for all (f, g) ∈ D m (U ).

Clearly [u] is (m + 1)-linear and satisfies the locality condition. The

1,∞
continuity follows from a well-known property of mappings in [Wloc (U )]m
(see e.g. [2, Theorem 2.16]). Since these examples will play a crucial role,
we include a proof below. For reasons of consistency (cf. Definition 8.1), we
also extend the notation to the case m = 0: Then R0 = {0}, u ∈ L1loc (R0 )
assigns the constant u(0) ∈ R, and [u] ∈ D0 (R0 ) is the current satisfying
[u](f ) = u(0)f (0) for all f ∈ D(R0 ).

Proof. We verify the continuity of [u]. Suppose (f j , g j ) → (f, g) in D m (U ).

Then there exist an open set V b U and a constant l such that spt(f j ) ⊂ V
and Lip(f j ) ≤ l for all j, and f j → f uniformly; moreover, for i = 1, . . . , m,
Lip(gij |V ) ≤ l for all j, and gij |V → gi |V uniformly. Put hji := gij − gi . Then
we have

[u](f j , g j ) − [u](f, g)
m
[u](f, g1 , . . . , gi−1 , hji , gi+1
j
X
j j j
= [u](f − f, g ) + , . . . , gm ).
i=1

Since the sequence (det(Dg j ))j∈N is bounded in L∞ (V ), the first term on the
right side clearly tends to 0 for j → ∞. Now consider the summand for i = 1;
the other summands are treated similarly. Since uf ∈ L1 (V ), we want to
show that Z
lim v det(D(hj1 , g2j , . . . , gm
j
)) dx = 0 (2.12)
j→∞ V

for all v ∈ L1 (V ). As Cc1 (V ) is dense in L1 (V ) and the sequence of deter-

minants is bounded in L∞ (V ), it suffices to prove (2.12) with v replaced by
w ∈ Cc1 (V ). We claim that
Z Z
w det(D(hj1 , g2j , . . . , gm
j
)) dx = − hj1 det(D(w, g2j , . . . , gm
j
)) dx. (2.13)
V V
Local currents in metric spaces 17

If hj1 , g2j , . . . , gm
j
∈ C 2 (V ), then d(whj1 dg2j ∧ . . . ∧ dgm
j
R
V ) = 0 and hence
Z Z
w dhj1 ∧ dg2j ∧ ... ∧ j
dgm =− hj1 dw ∧ dg2j ∧ . . . ∧ dgm
j
,
V V

which is just a restatement of (2.13). Now a smoothing argument relying on

the bounded convergence theorem shows (2.13) in the general case. Since
hj1 |V → 0 uniformly, the right side of (2.13) tends to zero for j → ∞.
Thus (2.12) holds with v replaced by w ∈ Cc1 (V ).

We conclude this section with some comments regarding Proposition 2.4

and Theorem 2.5. The proof of the alternating property does not require
the continuity of T in the first argument and corresponds essentially to the
last paragraph on p. 17 in [3]. In contrast, both the product and the chain
rule depend on the joint continuity condition, Definition 2.1(2). To exemplify
this, consider the functional T : D 1 (R) → R defined by
Z
T (f, π) := f 0 π 0 dx.
R

Clearly T is 2-linear and satisfies the locality condition. Moreover, an ap-

proximation and integration by parts argument as in the proof of Proposi-
tion 2.6 shows that if (f j , π j ) → (f, π) in D 1 (R), then T (f j , π) → T (f, π)
and T (f, π j ) → T (f, π). Yet, T fails to be a 1-dimensional metric current.
For instance, if f j (x) = χ[0,2π] (x) sin(jx)/j, then f j → 0 in D(R), while
T (f j , f j ) = π for all j ∈ N. Neither the product rule nor the chain rule holds
for T . In [3, Theorem 3.5], the proofs of the corresponding identities rely
on the finite mass axiom and use piecewise affine approximation rather than
polynomial approximation.

3 Support, boundary, and push-forward

For classical currents, these notions are defined in duality with support, ex-
terior derivative, and pull-back of forms. Similar constructions apply in the
metric context.

Definition 3.1 (support). Given a current T ∈ Dm (X), m ≥ 0, its support

spt(T ) in X is the intersection of all closed sets C ⊂ X with the property
that T (f, π) = 0 whenever (f, π) ∈ D m (X) with spt(f ) ∩ C = ∅.

The definition is justified by the next lemma, whose proof employs Lip-
schitz partitions of unity. This is made possible by the fact that the functions
in D(X) have compact support. In [3], the support of a current T is defined
as the support of the associated finite Borel measure kT k.

Lemma 3.2 (support). Suppose that T ∈ Dm (X), m ≥ 0.

18 U. Lang

(1) The support spt(T ) equals the set of all x ∈ X such that for every  > 0
there exists an (f, π) ∈ D m (X) with spt(f ) ⊂ B(x, ) and T (f, π) 6= 0.

(2) If f |spt(T ) = 0, then T (f, π1 , . . . , πm ) = 0.

(3) In case m ≥ 1, T (f, π1 , . . . , πm ) = 0 whenever some πi is constant on

{f 6= 0} ∩ spt(T ).

This shows in particular that T (f, π1 , . . . , πm ) depends only on the re-

strictions of f, π1 , . . . , πm to spt(T ).

Proof. Let Σ be the set described in (1). Suppose that x 6∈ spt(T ). There
is a closed set C with the property stated in Definition 3.1 such that x 6∈ C.
Then there is an  > 0 such that T (f, π) = 0 whenever spt(f ) ⊂ B(x, ). This
shows that x 6∈ Σ, so Σ ⊂ spt(T ).
Next we prove that T (f, π) = 0 whenever spt(f ) ∩ Σ = ∅. Since spt(f ) is
a compact subset of S X \ Σ, there exist finitely many open balls U1 , . . . , UN
such that spt(f ) ⊂ N k=1 Uk and T (g, π) = 0 whenever {g 6= 0} ⊂ Uk for
some k. Decomposing f by means of a Lipschitz partition of unity (ρk )N k=1
on spt(f ) with {ρk 6= 0} ⊂ Uk we see that T (f, π) = 0. As Σ is closed, this
shows in particular that spt(T ) ⊂ Σ.
For (2), let βj be the function defined after (2.5), j ∈ N. If f |spt(T ) = 0,
then spt(βj ◦ f ) ∩ spt(T ) = ∅. The argument of the previous paragraph then
shows that T (βj ◦ f, π) = 0 for all j, thus T (f, π) = 0 by the continuity of T .
To prove (3), by the linearity, locality, and the alternating property of
T
it suffices to show that T (f, π1 , . . . , πm ) = 0 if π1 = 0 on K := spt f |spt(T ) .
Then spt(βj ◦ π1 ) ∩ K = ∅ for βj as above, j ∈ N. For fixed j, since K is
compact, there is a function σ ∈ D(X) such that σ|K = 1 and βj ◦ π1 = 0 on
some neighborhood of spt(σ). Then (1 − σ)f |spt(T ) = 0, hence

T (f, βj ◦ π1 , π2 , . . . , πm ) = T (σf, βj ◦ π1 , π2 , . . . , πm ) = 0

by (2) and the locality of T . Since βj ◦ π1 → π1 in Liploc (X) for j → ∞, we

have T (f, π1 , . . . , πm ) = 0.

Suppose A is a closed subset of X, and TA ∈ Dm (A). For each f ∈ D(X),

the support of f |A in A is compact. Hence one obtains a current T ∈ Dm (X)
by defining
T (f, π1 , . . . , πm ) := TA (f |A , π1 |A , . . . , πm |A ) (3.1)
for all (f, π1 , . . . , πm ) ∈ D m (X). Clearly spt(T ) = spt(TA ). Conversely, the
following holds.

Proposition 3.3. Let T ∈ Dm (X), m ≥ 0, and let A ⊂ X be a locally

compact set containing spt(T ). Then there is a unique current TA ∈ Dm (A)
with the property that

TA (f, π1 , . . . , πm ) = T (f¯, π̄1 , . . . , π̄m )

Local currents in metric spaces 19

whenever (f, π1 , . . . , πm ) ∈ D m (A), (f¯, π̄1 , . . . , π̄m ) ∈ D m (X), f¯|A = f , and

π̄i |A = πi for i = 1, . . . , m. Moreover, spt(TA ) = spt(T ).

In particular, every current T ∈ Dm (X) may be viewed as a current in its

own support, i.e., as an element of Dm (spt(T )).

Proof. For every compact set K ⊂ A, every l ≥ 0, and every c > 0 there exist
a compact set K 0 ⊂ X containing K, an l0 ≥ l, and an operator

E : f ∈ LipK,l (A) : kf k∞ ≤ c → LipK 0 ,l0 (X)

such that (Ef )|A = f , kEf k∞ = kf k∞ , and kEf − Egk∞ = kf − gk∞ . In

fact, one may choose σ ∈ D(X) such that 0 ≤ σ ≤ 1 and σ|K = 1 and define

(Ef )(x) := σ(x) min kf k∞ , inf a∈A (f (a) + l d(a, x)) ,

cf. Sect. 1.1. This has the required properties, with K 0 = spt(σ) and l0 =
l+c Lip(σ). Now the result follows easily from Lemma 2.2 and Lemma 3.2.

Suppose for the moment that X is an arbitrary metric space. In view

of (3.1) and Proposition 3.3, it is possible to define the vector space Dm (X)
of general metric m-currents in X as follows: An element

T ∈ Dm (X) (3.2)

is a pair T = (XT , T ) consisting of a closed and locally compact set XT ⊂

X and a current T ∈ Dm (XT ) (Definition 2.1) with spt(T ) = XT . For
α ∈ R \ {0}, αT := (XT , αT ), and 0T := (∅, 0). To form the sum of two
elements T, T0 ∈ Dm (X), regard both T and T 0 as currents in XT ∪ XT0 , put
XT+T0 := spt(T + T 0 ) ⊂ XT ∪ XT0 , and then interpret T + T 0 as a current in
XT+T0 . With this understood, we may write again T instead of T and XT
or spt(T ) instead of XT . We shall briefly return to this discussion at the end
of Sect. 4, but otherwise we shall not pursue it in the present paper.
We now proceed to the definition of the boundary of a metric m-current
in the locally compact space X, which is easily seen to be an (m − 1)-current.

Definition 3.4 (boundary). The boundary of a current T ∈ Dm (X), m ≥ 1,

is the current ∂T ∈ Dm−1 (X) defined by

∂T (f, π1 , . . . , πm−1 ) := T (σ, f, π1 , . . . , πm−1 )

for (f, π1 , . . . , πm−1 ) ∈ D m−1 (X), where σ ∈ D(X) is any function such that
σ = 1 on {f 6= 0} ∩ spt(T ).

If τ ∈ D(X) is another such function, then f vanishes on {σ − τ 6=

0} ∩ spt(T ), thus T (σ − τ, f, π1 , . . . , πm−1 ) = 0 by Lemma 3.2(3). This shows
that ∂T is well-defined. Clearly ∂T is multilinear and continuous. To verify
the locality of ∂T , suppose that some πi is constant on a neighborhood of
20 U. Lang

spt(f ), choose σ such that πi is constant on a neighborhood of spt(σ), and

use the locality of T . We have
spt(∂T ) ⊂ spt(T ), (3.3)
and if A ⊂ X and TA are as in Proposition 3.3, then
∂(TA ) = (∂T )A . (3.4)
If m ≥ 2, then
∂(∂T ) = 0. (3.5)
To see this, let (f, π) ∈ D m−2 (X), and choose ρ, σ, τ ∈ D(X) such that
ρ|spt(f ) = 1, σ|spt(ρ) = 1, and τ |spt(σ) = 1. Then
∂(∂T )(f, π) = ∂T (σ, f, π) = T (τ, σ, f, π) = T (ρ, σ, f, π) = 0;
the third equality holds since f |spt(τ −ρ) = 0, the last since σ|spt(ρ) = 1. The
operator ∂ : Dm (X) → Dm−1 (X) is linear, and if Tn → T weakly in Dm (X),
then ∂Tn → ∂T weakly in Dm−1 (X).
The following lemma corresponds to the identity
(∂T )b φ = T b dφ + (−1)k ∂(T b φ)
for a classical m-current T and k-form φ, cf. [14, p. 356].
Lemma 3.5. For T ∈ Dm (X) and (u, v) ∈ Liploc (X) × [Liploc (X)]k , where
m > k ≥ 0, the identity
(∂T )b (u, v) = T b (1, u, v) + (−1)k ∂(T b (u, v))
holds.
In case k = 0, the identity simply reads
(∂T )b u = T b (1, u) + ∂(T b u). (3.6)
Note also that
(∂T )b (1, v) = (−1)k ∂(T b (1, v)) (3.7)
since T b (1, 1, v) = 0 by the locality of T .

Proof. Let (f, g) ∈ D m−k−1 (X), and choose σ ∈ D(X) with σ|spt(f ) = 1.
Then
((∂T )b (u, v))(f, g) = ∂T (uf, v, g)
= T (σ, uf, v, g)
= T (σf, u, v, g) + T (σu, f, v, g)
= T (f, u, v, g) + (−1)k T (σu, v, f, g)
= (T b (1, u, v))(f, g) + (−1)k (T b (u, v))(σ, f, g)
= (T b (1, u, v))(f, g) + (−1)k ∂(T b (u, v))(f, g);
the third step uses Proposition 2.4(2), the fourth 2.4(1).
Local currents in metric spaces 21

Next we define push-forwards of metric currents under locally Lipschitz

maps. Since the test functions have compact support, we need to restrict to
proper maps, as in the classical case (cf. [14, p. 359 and 4.1.14]).

Definition 3.6 (push-forward). Suppose that T ∈ Dm (X), m ≥ 0, A ⊂ X is

a locally compact set containing spt(T ), Y is another locally compact metric
space, and F ∈ Liploc (A, Y ) is proper, i.e., F −1 (K) is compact whenever
K ⊂ Y is compact. The push-forward of T via F is the current F# T ∈ Dm (Y )
defined by

F# T (f, π1 , . . . , πm ) := TA (f ◦ F, π1 ◦ F, . . . , πm ◦ F )

for (f, π1 , . . . , πm ) ∈ D m (Y ), where TA ∈ Dm (A) is as in Proposition 3.3.

Note that f ◦ F ∈ D(A) since F is locally Lipschitz and proper. One

readily verifies that F# T is a metric current. Since F is a proper continuous
map, F (spt(T )) is closed in Y , and

spt(F# T ) ⊂ F (spt(T )). (3.8)

If m ≥ 1, then
∂(F# T ) = F# (∂T ). (3.9)
To see this, let (f, π) ∈ D m−1 (Y ), and choose σ ∈ D(Y ) such that σ|spt(f ) = 1;
then

∂(F# T )(f, π) = (F# T )(σ, f, π) = TA (σ ◦ F, f ◦ F, π ◦ F )

= ∂(TA )(f ◦ F, π ◦ F ) = (∂T )A (f ◦ F, π ◦ F ) = F# (∂T )(f, π).

If Z is another locally compact metric space and G ∈ Liploc (Y, Z) is proper,

then
G# (F# T ) = (G ◦ F )# T. (3.10)
If F ∈ Liploc (X, Y ) is proper, then the operator F# : Dm (X) → Dm (Y ) is
linear, and F# Tn → F# T weakly in Dm (Y ) whenever Tn → T weakly in
Dm (X). Finally, suppose F ∈ Liploc (D, Y ) for some set D ⊂ X containing
spt(T ), and F |spt(T ) is proper. In this situation, we put


F# T := F |spt(T ) # T ; (3.11)

this is consistent with the above definition in case D is locally compact and
F is proper. When F ∈ Liploc (X, Y ) and F |spt(T ) is proper, it follows that

F# T (f, π1 , . . . , πm ) = T (σ(f ◦ F ), π1 ◦ F, . . . , πm ◦ F ) (3.12)

for (f, π1 , . . . , πm ) ∈ D m (Y ) and any σ ∈ D(X) such that σ = 1 on {f ◦ F 6=

0} ∩ spt(T ).
We compute the push-forward of a current [u] as in Proposition 2.6 (stan-
dard example).
22 U. Lang

Lemma 3.7. Suppose u ∈ L1loc (Rm ), m ≥ 1, F ∈ Liploc (Rm , Rm ), and

F |spt(u) is proper. Then F# [u] = [v], where v ∈ L1loc (Rm ) satisfies
X
v(y) = u(x) sign(det(DF (x)))
x∈F −1 {y}

for L m -almost every y ∈ Rm .

Proof. Let (f, π) ∈ D m (Rm ). Then
Z Z
F# [u](f, π) = u(f ◦ F ) det(D(π ◦ F )) dx = h(x) |det(DF (x))| dx
Rm Rm

for h(x) := u(x)f (F (x)) det(Dπ(F (x))) sign(det(DF (x))). By the change of
variables formula (cf. [14, Theorem 3.2.3(2)] or [13, 3.3.3], the case n = m),
Z X Z
F# [u](f, π) = h(x) dy = v(y)f (y) det(Dπ(y)) dy
Rm Rm
x∈F −1 {y}

= [v](f, π).

This proves the lemma.

4 Mass
We now define the mass of a metric current. Our approach is inspired by both
the classical definition (recalled in (5.2)) and [3, Proposition 2.7]. Currents
with locally finite mass will be of particular interest.
Definition 4.1 (mass). For T ∈ Dm (X), m ≥ 0, and every open set V ⊂ X,
we define the mass MV (T ) of T in V as the least number M ∈ [0, ∞] such
that X
T (fλ , π λ ) ≤ M
λ∈Λ

P set, (fλ , π ) = (fλ , π1 , . . . , πm ) ∈ D(X) × [Lip1 (X)] ,

whenever Λ is a finite λ λ λ m

spt(fλ ) ⊂ V , and λ∈Λ |fλ | ≤ 1. The number M(T ) := MX (T ) is the total

mass of T . We denote by
Mm,loc (X)
the vector space of all T ∈ Dm (X) such that MV (T ) < ∞ for every open set
V b X, and we put

Mm (X) := {T ∈ Dm (X) : M(T ) < ∞}.

Furthermore, we define

kT k(A) := inf{MV (T ) : V ⊂ X is open, A ⊂ V }

for T ∈ Dm (X) and every set A ⊂ X.

Local currents in metric spaces 23

Note that for T ∈ D0 (X),

MV (T ) = sup{T (f ) : f ∈ D(X), spt(f ) ⊂ V, |f | ≤ 1},

and by the continuity of T it follows that

MV (T ) = sup{T (f ) : f ∈ D(X), |f | ≤ χV }. (4.1)

If T ∈ Dm (X) and A ⊂ X is open, then clearly

kT k(A) = MA (T ). (4.2)

The mass is lower semicontinuous with respect to weak convergence: If Tn →

T weakly in Dm (X), then

MV (T ) ≤ lim inf MV (Tn ) (4.3)

n→∞

for every open set V ⊂ X. For T, T 0 ∈ Dm (X) and α ∈ R, we have

kαT k = |α|kT k, kT + T 0 k ≤ kT k + kT 0 k, (4.4)

and kT k = 0 if and only if T = 0. In particular, M is a norm on Mm (X).

Proposition 4.2. For m ≥ 0, (Mm (X), M) is a Banach space.

Proof. Let (Tk )k∈N be a Cauchy sequence in (Mm (X), M). For every  > 0
there exists an index k such that

|Tk (f, π) − Tl (f, π)| = |(Tk − Tl )(f, π)| ≤ M(Tk − Tl ) ≤ 

whenever k, l ≥ k and (f, π) ∈ D(X) × [Lip1 (X)]m with |f | ≤ 1. In par-

ticular, (Tk (f, π))k∈N is a Cauchy sequence for every such (f, π). It fol-
lows that there is an (m + 1)-linear function T : D m (X) → R such that
limk→∞ Tk (f, π) = T (f, π) for all (f, π) ∈ D m (X) and T satisfies the locality
condition (Definition 2.1(3)). Moreover, for  > 0 and k as above, we have

|Tk (f, π) − T (f, π)| ≤ 

whenever k ≥ k and (f, π) ∈ D(X) × [Lip1 (X)]m with |f | ≤ 1. It suffices

to verify the continuity of T on the set of all such (f, π). Suppose that
(f, π), (f 1 , π 1 ), (f 2 , π 2 ), . . . belong to this set and (f j , π j ) → (f, π) in D m (X).
Given  > 0, there is an index j such that |Tk (f j , π j ) − Tk (f, π)| ≤  for all
j ≥ j ; then

|T (f j , π j ) − T (f, π)|
≤ |T (f j , π j ) − Tk (f j , π j )| +  + |Tk (f, π) − T (f, π)| ≤ 3.
24 U. Lang

Hence T is a current.
P Whenever Λ is a finite set, (fλ , π λ ) ∈ D(X)×[Lip1 (X)]m
for λ ∈ Λ, and λ∈Λ |fλ | ≤ 1, we have
X
(Tk − Tl )(fλ , π λ ) ≤ M(Tk − Tl ) ≤ 
λ∈Λ

for k, l ≥ k , hence λ∈Λ (Tk − T )(fλ , π λ ) ≤  for k ≥ k . We conclude that

P
M(Tk − T ) ≤  for k ≥ k . Thus T ∈ Mm (X) and Tk → T in (Mm (X), M).

Let U ⊂ Rm be an open set. For T ∈ Dm (U ), it follows from Theorem 2.5

(chain rule) that
MV (T ) = sup{T (f, id) : f ∈ D(X), spt(f ) ⊂ V, |f | ≤ 1}
for every open set V ⊂ U . Given a current [u] as in Proposition 2.6 (standard
example), where u ∈ L1loc (U ), we have
Z  Z
MV ([u]) = sup uf dx : f ∈ D(V ), |f | ≤ 1 = |u| dx (4.5)
V V

for every open set V ⊂ U . In particular, [u] ∈ Mm,loc (U ).

Theorem 4.3 (mass). Let T ∈ Dm (X), m ≥ 0.
(1) The function kT k : 2X → [0, ∞] (cf. Definition 4.1) is a Borel regular
outer measure.
(2) We have spt(kT k) = spt(T ) and kT k(X \ spt(T )) = 0.
(3) For every open set V ⊂ X,
kT k(V ) = sup{kT k(K) : K ⊂ X is compact, K ⊂ V }.

(4) If T ∈ Mm,loc (X), then kT k is a Radon measure, and

m
Y Z


|T (f, π)| ≤ Lip πi |spt(f ) |f | dkT k
i=1 X

for all (f, π) = (f, π1 , . . . , πm ) ∈ D m (X).

In case m = 0, the inequality in (4) reads
Z
|T (f )| ≤ |f | dkT k.
X

In case m ≥ 1, it holds as well with {f 6= 0} ∩ spt(T ) in place of spt(f ), by

Lemma 3.2(3) and the extendability of Lipschitz functions. Properties (1),
(2), and (3) hold in general for every (m+1)-linear functional T : D m (X) → R;
for m = 0 the same argument occurs in the proof of the Riesz representation
theorem (see e.g. [29, Theorem 4.1]). The proof of (4) corresponds essentially
to that of [3, (2.8)].
Local currents in metric spaces 25

Proof. Clearly kT k(∅) = 0. We claim that

∞
X
kT k(V ) ≤ kT k(Vk )
k=1

whenever V, V1 , V2 , . . . ⊂ X are open and V ⊂ ∞ λ

S
k=1 Vk . Let Λ and (fλ , π )
be
SN given as in the definition of MV (T ). S There is an index N such that
k=1 Vk contains the compact PN
set K := λ∈Λ spt(fλ ). Then there exist
ρ1 , . . . , ρN ∈ D(X) such that k=1 ρk = 1 on K and 0 ≤ ρk ≤ 1, spt(ρk ) ⊂ Vk
for k = 1, . . . , N . Then
X N
XX N X
X N
X
T (fλ , π λ ) = T (ρk fλ , π λ ) = T (ρk fλ , π λ ) ≤ kT k(Vk )
λ∈Λ λ∈Λ k=1 k=1 λ∈Λ k=1
P
since spt(ρk fλ ) ⊂ Vk and λ∈Λ |ρk fλ | ≤ 1 for k = 1, . . . , N , proving the
claim. It follows that kT k is an outer measure, and kT k(A ∪ B) = kT k(A) +
kT k(B) whenever inf{d(x, y) : x ∈ A, y ∈ B} > 0. By Carathéodory’s
criterion, every Borel set is kT k-measurable. If A is an arbitrary set, then
A is contained in a Gδ set B with kT k(B) = kT k(A), by the definition of
kT k(A). Thus kT k is Borel regular. This proves (1).
Assertion (2) follows from Lemma 3.2.
For (3), given an open set V , let α < kT k(V ). Then there S exist a finite
Λ and (fλ , π λ ) ∈ D(X) × P
set P [Lip1 (X)]m such that K := λ∈Λ spt(fλ ) ⊂
λ
V, λ∈Λ |fλ | ≤ 1, and s := λ∈Λ T (fλ , π ) ≥ α. For every open set U
containing K we have kT k(U ) ≥ s ≥ α, hence kT k(K) ≥ α.
It remains to prove (4). If T ∈ Mm,loc (X), then kT k is finite on compact
sets, thus T is a Radon measure. For the integral estimate, we first consider
the case m = 0, so that T ∈ M0,loc (X). Assuming without loss of generality
that f ≥ 0, we put fs := min{f, s} and observe that, by (4.1),
|T (ft ) − T (fs )| = |T (ft − fs )| ≤ kT k({f > s})(t − s)
whenever 0 ≤ s < t. Hence s 7→ T (fs ) is a Lipschitz function with
|(d/ds)T (fs )| ≤R kT k({f > s}) for almost every s ≥ 0. Since T (f ) =
∞
T (f ) − T (f0 ) = 0 (d/ds)T (fs ) ds, we conclude that
Z ∞ Z ∞ Z
d
|T (f )| ≤ T (fs ) ds ≤ kT k({f > s}) ds = f dkT k
0 ds 0 X

(for the last step, see e.g. [26, Theorem 8.16]). Now let T ∈ Mm,loc (X),
m ≥ 1. Let first (f, π) ∈ D(X) × [Lip1 (X)]m , and consider Tπ := T b (1, π) ∈
D0 (X). Then kTπ k ≤ kT k, thus Tπ ∈ M0,loc (X), and
Z Z
|T (f, π)| = |Tπ (f )| ≤ |f | dkTπ k ≤ |f | dkT k.
X X
Finally, given (f, π) ∈ D m (X), m

exists π̃ ∈ [Lip(X)] such that π̃ = π on
there
spt(f ) and Lip(π̃i ) = Lip πi |spt(f ) for i = 1, . . . , m. Then T (f, π) = T (f, π̃)
by the strict locality property (2.5), and the result follows.
26 U. Lang

Recall from Sect. 1.2 that we denote by Bc∞ (X) the algebra of all bounded
Borel functions f : X → R such that spt(f ) is compact. From Theorem 4.3(4)
it follows that every T ∈ Mm,loc (X), m ≥ 0, naturally extends to a function

T : Bc∞ (X) × [Liploc (X)]m → R. (4.6)

To see this, note that D(X) is dense in L1 (kT k) (since Cc (X) is, see e.g. [26,
3.14], and since every element of Cc (X) is a uniform limit of a sequence of
Lipschitz functions, cf. Sect. 1.1). Hence, whenever f ∈ Bc∞ (X) ⊂ L1 (kT k)
and U b X is a neighborhood of spt(f ), there is a sequence (gk )k∈N in D(X)
such that gk → f in L1 (kT k) and spt(gk ) ⊂ U for all k. By Theorem 4.3(4),
for every π ∈ [Liploc (X)]m , (T (gk , π))k∈N is a Cauchy sequence whose limit is
independent of the choice of (gk )k∈N . Then T (f, π) is defined to be this limit.

Theorem 4.4 (extended functional). Let T ∈ Mm,loc (X), m ≥ 0. The

extension T : Bc∞ (X) × [Liploc (X)]m → R possesses the following properties:

(1) (multilinearity) T is (m + 1)-linear on Bc∞ (X) × [Liploc (X)]m .

, π j ) → T (f, π) whenever f, f 1 , f 2 , . . . ∈ Bc∞ (X),

(2) (continuity) T (f j S
supj kf k∞ < ∞, j spt(f j ) ⊂ K for some compact set K ⊂ X, f j → f
j

pointwise on X, and π j → π in [Liploc (X)]m .

(3) (locality) In case m ≥ 1, T (f, π) = 0 whenever some πi is constant on

the support of f ∈ Bc∞ (X).

(4) For all (f, π) ∈ Bc∞ (X) × [Liploc (X)]m ,

m
Y Z


|T (f, π)| ≤ Lip πi |spt(f ) |f | dkT k.
i=1 X

Obviously (4) subsumes (3). Moreover, whenever a functional T satis-

fying (4) is linear in the first argument and continuous with respect to the
convergence in [Liploc (X)]m , then T fulfils (2) (cf. (4.9) below). As a conse-
quence of (4) and Theorem 4.3(2), we also have T (f − f˜, π) = 0 and hence

T (f, π) = T (f˜, π) (4.7)

whenever f, f˜ ∈ Bc∞ (X) agree on spt(T ).

Proof. Clearly the extended functional T is (m + 1)-linear, and

m
Y Z
|T (f, π)| ≤ Lip(πi |U ) |f | dkT k (4.8)
i=1 X

whenever (f, π) ∈ Bc∞ (X) × [Liploc (X)]m and U b X is a neighborhood of

spt(f ).
Local currents in metric spaces 27

Given (f, π), (f 1 , π 1 ), (f 2 , π 2 ), . . . and K as in (2), choose a neighborhood

U b X of K. Since π j → π in [Liploc (X)]m , there is a constant l such that
supj Lip(πij |U ) ≤ l and Lip(πi |U ) ≤ l for i = 1, . . . , m. We have

|T (f j , π j ) − T (f, π)| ≤ |T (f j − f, π j )| + |T (f, π j ) − T (f, π)|, (4.9)

and |T (f j − f, π j )| ≤ lm X |f j − f | dkT k → 0, by the bounded convergence

R

theorem. For the second term Ron the right side, given  > 0, choose g ∈ D(X)
such that spt(g) ⊂ U and lm X |f − g| dkT k < /3. For j sufficiently large,
|T (g, π j ) − T (g, π)| < /3 by the continuity of T , hence

|T (f, π j ) − T (f, π)|

≤ |T (f − g, π j )| + |T (f − g, π)| + |T (g, π j ) − T (g, π)| < .

This shows (2).

By the locality of T , clearly T (f, π) = 0 if some πi is constant on a
neighborhood of the support of f ∈ Bc∞ (X). Hence, for the proof of (3),
there is no loss of generality in assuming that πi = 0 on spt(f ). Let βj be
the function defined after (2.5). Then βj ◦ πi = 0 on some neighborhood of
spt(f ), and letting j tend to infinity we obtain T (f, π) = 0 by means of (2).
Finally, (4) follows from (4.8) by (3) and the extendability of Lipschitz
functions.

The extension of T allows to define T b u more generally for locally

bounded Borel functions u, in particular for characteristic functions of Borel
sets. This complements Definition 2.3.

Definition 4.5. For T ∈ Mm,loc (X) and (u, v) ∈ Bloc∞ (X) × [Lip (X)]k ,
loc
where m ≥ k ≥ 0, we define T b (u, v) ∈ Mm−k,loc (X) by the same equation
as in Definition 2.3. For a Borel set B ⊂ X,

T b B := T b χB .

Clearly T b (u, v) is a current, and it follows from Theorem 4.4(4) that

k
Y Z
MV (T b (u, v)) ≤ Lip(vi |V ) |u| dkT k (4.10)
i=1 V
R
for all open sets V ⊂ X (meaning MV (T b u) ≤ V |u| dkT k in case k = 0).
Since the right side is finite if the closure of V is compact, we have T b (u, v) ∈
Mm−k,loc (X).
The next lemma gives information on push-forwards of currents with lo-
cally finite mass.

Lemma 4.6. Suppose T ∈ Mm,loc (X), m ≥ 0, Y is another locally com-

pact metric space, F ∈ Liploc (X, Y ), and F |spt(T ) is proper. Then F# T ∈
Mm,loc (Y ), and the following properties hold:
28 U. Lang

(1) For all (f, π) ∈ Bc∞ (Y ) × [Liploc (Y )]m and any σ ∈ Bc∞ (X) such that
σ = 1 on {f ◦ F 6= 0} ∩ spt(T ),

F# T (f, π) = T (σ(f ◦ F ), π ◦ F ).

(2) For every Borel set B ⊂ Y ,

m
M((F# T )b B) ≤ Lip F |F −1 (B)∩spt(T ) kT k(F −1 (B)).

Proof. Let V b Y be an open set, and choose σ ∈ D(X) such that σ = 1 on

F −1 (V ) ∩ spt(T ). It follows from (3.12) and Theorem 4.3(4) that

m
MV (F# T ) ≤ Lip F |spt(σ) kT k(F −1 (V )).

Hence F# T ∈ Mm,loc (Y ).
To prove (1), fix π ∈ [Liploc (Y )]m and ρ ∈ D(Y ), ρ ≥ 0. Choose τ ∈ D(X)
such that τ = 1 on {ρ ◦ F 6= 0} ∩ spt(T ), and denote by Φ the set of all
f ∈ Bc∞ (Y ) such that |f | ≤ ρ and F# T (f, π) = T (τ (f ◦ F ), π ◦ F ). It
follows from Theorem 4.4(2) that Φ is a Baire class. By (3.12), Φ contains all
f ∈ D(Y ) with |f | ≤ ρ and therefore consists of all f ∈ Bc∞ (Y ) with |f | ≤ ρ,
cf. Sect. 1.2. In view of (4.7), this gives (1).
For (2), suppose (f, π) ∈ D(Y ) × [Lip1 (Y )]m , and let σ be the character-
istic function of F −1 (B) ∩ {f ◦ F 6= 0} ∩ spt(T ). By (1) and Theorem 4.4(4),

((F# T )b B)(f, π) = F# T (χB f, π) = T (σ(f ◦ F ), π ◦ F )

Z

m
≤ Lip F |spt(σ) |f ◦ F | dkT k.
F −1 (B)

This yields the result.

For a current T ∈ Mm,loc (X) and a Borel set B ⊂ X, we always have

M(T b B) ≤ kT k(B). Equality holds, for instance, if spt(T ) is separable; this
is shown by the following lemma.

Lemma 4.7 (characterizing kT k). Suppose T ∈ Mm,loc (X), m ≥ 0, and

B ⊂ X is either a Borel set that is σ-finite with respect to kT k or an open
set. Then kT k(B) is the least number such that
X
T (fλ , π λ ) ≤ kT k(B)
λ∈Λ

whenever Λ is a finite set, (fλ , π λ ) ∈ Bc∞ (X) × [Lip1 (X)]m , and

P
λ∈Λ |fλ | ≤
χB . Moreover
kT kb B = kT b Bk,
in particular kT k(B) = M(T b B).
Local currents in metric spaces 29

In case m = 0, this says that

kT k(B) = sup{T (f ) : f ∈ Bc∞ (X), |f | ≤ χB }. (4.11)

In particular,
kT k({x}) = |T (χ{x} )| (4.12)
for every x ∈ X.
The following simple example illustrates the σ-finiteness assumption in the
lemma. Let R be any uncountable discrete space, and equip X = R × (−1, 1)
with the metric d defined by
(
0 0 |s − s0 | if r = r0 ,
d((r, s), (r , s )) :=
1 if r 6= r0 .

Note that X is locally compact. Let T ∈ M1,loc (X) be the current satisfying
Z
dπ
T (f, π) = f (r, s) (r, s) dH 1 (r, s)
X ds

for all (f, π) ∈ D 1 (X); then kT k = H 1 (compare (4.5)). Since R is uncount-

able, the closed set B := R × {0} is not σ-finite with respect to kT k, hence
kT kb B is not σ-finite. On the other hand, (kT kb B)(K) = 0 for all compact
sets K ⊂ X, and T b B = 0.

Proof of Lemma 4.7. We show the first part. By Theorem 4.4(4), the inequal-
ity always holds. To see that kT k(B) is the least number with this property,
let  > 0, and choose an open set V such that B ⊂ V and kT k(V \ B) ≤ .
Note that this is possible by the assumption on B. Let α < kT k(V ). Then
there exist Λ and (fλ , π λ ) as in the definition of MV (T ) such that
X X X
α≤ T (fλ , π λ ) = T (χB fλ , π λ ) + T (χV \B fλ , π λ )
λ∈Λ λ∈Λ λ∈Λ
≤ kT k(B) + .

This gives the result.

For the second part, it suffices to prove that kT k(B ∩ A) = kT b Bk(A) for
every Borel set A ⊂ X. To verify this equality, apply the result of the first
part to either side.

In (4.6) we extended the elements of Mm,loc (X) to Bc∞ (X)×[Liploc (X)]m .

We now extend currents of finite total mass in another direction. This will
also establish the connection to the metric currents of [3].
Let T ∈ Mm (X). We consider the restriction of T to D(X) × [Lip(X)]m ,
which determines T uniquely (compare Lemma 2.2). Since kT k is finite,
L1 (kT k) contains the algebra B ∞ (X) of bounded Borel functions on X. As
30 U. Lang

in Theorem 4.4, it follows from Theorem 4.3(4) that the restriction of T

extends to a function
T : B ∞ (X) × [Lip(X)]m → R (4.13)
with the following properties:
(1) (multilinearity) T is (m + 1)-linear.
(2) (continuity) T (f j , π j ) → T (f, π) whenever (f, π), (f 1 , π 1 ), (f 2 , π 2 ), . . . ∈
B ∞ (X) × [Lipl (X)]m for some l ≥ 0, supj kf j k∞ < ∞, and (f j , π j ) →
(f, π) pointwise on X.
(3) (locality) In case m ≥ 1, T (f, π) = 0 whenever some πi is constant on
spt(f ).
(4) For all (f, π) ∈ B ∞ (X) × [Lip(X)]m ,
m
Y Z


|T (f, π)| ≤ Lip πi |spt(f ) |f | dkT k.
i=1 X

Furthermore, by (4) and Lemma 4.7, for every Borel set B ⊂ X, kT k(B) is
the least number such that
X
T (fλ , π λ ) ≤ kT k(B) (4.14)
λ∈Λ

whenever Λ is a finite set, (fλ , π λ ) ∈ B ∞ (X) × [Lip1 (X)]m , and λ∈Λ |fλ | ≤
P
χB .
Suppose now, for the remaining part of this section, that X is an arbitrary
metric space. Combining the extension just described with the discussion
of (3.2), we obtain a corresponding normed space (Mm (X), M) of currents
with finite mass and locally compact support in X. By definition, an element
T of Mm (X) is a functional on B ∞ (XT ) × [Lip(XT )]m for some closed and
locally compact set XT ⊂ X. However, T may now equally well be viewed as
a functional on B ∞ (X) × [Lip(X)]m (compare (3.1) and Proposition 3.3). A
current
T ∈ Mm (X)
is then a function as in (4.13) satisfying (1)–(4) (now for arbitrary X), where
kT k is a finite Borel regular outer measure that is concentrated on its locally
compact support spt(kT k) and characterized by (4.14). Since kT k is finite,
spt(kT k) is separable (cf. [14, Theorem 2.2.16]) and hence also σ-compact.
In contrast to Proposition 4.2, the space (Mm (X), M) of all such T , where
M(T ) = kT k(X), is no longer complete in general.
We now denote by (MAK m (X), M) the Banach space of all m-currents in
the sense of Ambrosio and Kirchheim, viewed as functionals on B ∞ (X) ×
[Lip(X)]m , cf. [3, Theorem 3.5]. A current
T ∈ MAK
m (X)
Local currents in metric spaces 31

is a function as in (4.13) satisfying (1)–(3), moreover there exist a σ-compact

set Σ ⊂ X and a finite Borel measure µ on X such that µ(X \ Σ) = 0 and (4)
holds with µ in place of kT k. There is a minimal Borel measure µT with this
property (denoted kT k in [3]), and M(T ) := µT (X). Regarding the existence
of Σ, see [3, Lemma 2.9] and the remark thereafter; note that this lemma
requires completeness of the underlying metric space.
We now verify that (Mm (X), M) is a dense subspace of (MAK m (X), M),
so that
(MAKm (X), M) is the completion of (Mm (X), M). (4.15)
Clearly Mm (X) ⊂ MAK m (X), and for T ∈ Mm (X), (4.14) readily implies
that kT k(B) = µT (B) for every Borel set B ⊂ X, thus the two definitions of
M(T ) agree. Given T ∈ MAK S exist compact sets K1 ⊂ K2 ⊂ . . . in
m (X), there
X such that µT (X \ Σ) = 0 for Σ := ∞ k=1 Kk . The restrictions T b Kk form
a sequence in Mm (X) converging in (MAK m (X), M) to T .
In the sequel, X will again denote a locally compact metric space.

5 Normal currents
We turn to the chain complex of normal currents. In the first part of this
section we prove the compactness theorem for locally normal metric currents.
Then we compare metric currents in an open set U ⊂ Rn with classical cur-
rents, in particular we establish an isomorphism for locally normal currents.

Definition 5.1 (normal current). For T ∈ Dm (X) and every open set V ⊂
X, define
NV (T ) := MV (T ) + MV (∂T )
if m ≥ 1 and NV (T ) := MV (T ) if m = 0, and let N(T ) := NX (T ). The
vector space
Nm,loc (X)
of m-dimensional locally normal currents in X consists of all T ∈ Dm (X)
such that NV (T ) < ∞ for all open sets V b X. An m-dimensional normal
current in X is an element of

Nm (X) := {T ∈ Dm (X) : N(T ) < ∞}.

Note that (Nm (X), N) is a Banach space, cf. Proposition 4.2. If T ∈

Nm,loc (X) and (u, v) ∈ Liploc (X) × [Liploc (X)]k , where m > k ≥ 0, then

∂(T b (u, v)) = (−1)k (∂T )b (u, v) − T b (1, u, v)

by Lemma 3.5. Applying (4.10) twice, we get

k
Y Z 
MV (∂(T b (u, v))) ≤ Lip(vi |V ) |u| dk∂T k + Lip(u|V )kT k(V ) (5.1)
i=1 V
32 U. Lang

for every open set V ⊂ X. Together with (4.10), this shows that T b (u, v) ∈
Nm−k,loc (X). By Lemma 4.6 and (3.9), push-forwards of locally normal cur-
rents are locally normal.
The following result corresponds to [3, Proposition 5.1].

Lemma 5.2 (uniform continuity of normal currents). Let T ∈ Nm,loc (X),

m ≥ 1.

(1) For every (f, g) ∈ D m (X) with g2 , . . . , gm ∈ Lip1 (X),

Z Z
|T (f, g)| ≤ Lip(f ) |g1 | dkT k + |f g1 | dk∂T k.
spt(f ) X

(2) For all (f, g), (f˜, g̃) ∈ D(X) × [Lip1 (X)]m ,
Z
˜
T (f, g) − T (f , g̃) ≤ |f − f˜| dkT k
X
m 
X Z Z 
+ Lip(f ) |gi − g̃i | dkT k + |f ||gi − g̃i | dk∂T k .
i=1 spt(f ) X

Proof. For the proof of (1) we assume m = 1. Let (f, g) ∈ D 1 (X), and choose
σ ∈ D(X) with σ|spt(f ) = 1. By the product rule,

|T (f, g)| ≤ |T (σg, f )| + |T (σ, f g)| = |T (σg, f )| + |∂T (f g)|,

hence Z Z
|T (f, g)| ≤ Lip(f ) |g| dkT k + |f g| dk∂T k.
spt(f ) X

For the proof of (2) we observe that

T (f, g) − T (f˜, g̃)

m
X
= T (f − f˜, g̃) + T (f, g̃1 , . . . , g̃i−1 , gi − g̃i , gi+1 , . . . , gm ),
i=1

then we apply the alternating property and (1) to each summand.

Lemma 5.2 yields the following proposition, which will be used repeatedly
in the sequel.

Proposition 5.3 (convergence criterion). Suppose X is compact, F ⊂

Lip1 (X) is dense in Lip1 (X) with respect to the metric induced by k · k∞ ,
and (Tn )n∈N is a sequence in Nm (X), m ≥ 0, with M := supn N(Tn ) < ∞
and the property that the limit limn→∞ Tn (f, g) exists for all (f, g) ∈ F ×F m .
Then (Tn )n∈N converges weakly to some T ∈ Nm (X).
Local currents in metric spaces 33

Proof. We assume m ≥ 1; the case m = 0 is similar but easier. Let

(f, g), (f˜, g̃) ∈ D(X) × [Lip1 (X)]m , and define
m
X
R(f, g, f˜, g̃) := kf − f˜k∞ + max{kf k∞ , Lip(f )} kgi − g̃i k∞ .
i=1

For every n ∈ N, Lemma 5.2(2) gives

|Tn (f, g) − Tn (f˜, g̃)| ≤ R(f, g, f˜, g̃)N(Tn ) ≤ R(f, g, f˜, g̃)M.

Fix (f, g) ∈ Lip1 (X) × [Lip1 (X)]m for the moment. Given  > 0, there is
(f˜, g̃) ∈ F × F m such that R(f, g, f˜, g̃)M ≤ , and there is an index n0 such
that |Tn (f˜, g̃)−Tn0 (f˜, g̃)| ≤  for all n, n0 ≥ n0 . Then |Tn (f, g)−Tn0 (f, g)| ≤ 3
for all n, n0 ≥ n0 , so (Tn (f, g))n∈N is a Cauchy sequence. Define T : Lip1 (X)×
[Lip1 (X)]m → R such that

T (f, g) = lim Tn (f, g)

n→∞

for all (f, g) ∈ Lip1 (X) × [Lip1 (X)]m . It follows that |T (f, g) − T (f˜, g̃)| ≤
R(f, g, f˜, g̃)M for all (f, g), (f˜, g̃) ∈ Lip1 (X) × [Lip1 (X)]m . Now it is clear
that T extends uniquely to a current T ∈ Dm (X), and Tn → T weakly. By
the lower semicontinuity of mass, T ∈ Nm (X).

We now arrive at a fundamental result for locally normal currents in a

locally compact metric space X.

Theorem 5.4 (Nm,loc compactness). Suppose (Tn )n∈N is a sequence in

Nm,loc (X), m ≥ 0, such that each spt(Tn ) is separable and supn NV (Tn ) < ∞
for every open set V b X. Then there is a subsequence (Tn(i) )i∈N that con-
verges weakly to some T ∈ Nm,loc (X).

Compare [3, Theorem 5.2].

Proof. Assume first that X is compact, so that M := supn N(Tn ) < ∞.

Choose a countable set F ⊂ Lip1 (X) as in Proposition 5.3. For every (f, g) ∈
F × F m,
|Tn (f, g)| ≤ kf k∞ M(Tn ) ≤ kf k∞ M
for all n. A diagonal process then yields a subsequence (Tn(i) )i∈N such that the
limit limi→∞ Tn(i) (f, g) exists for every (f, g) ∈ F × F m . By Proposition 5.3,
this subsequence converges weakly to S some T ∈ Nm (X).
In the general case, the closure of n∈N spt(Tn ) is separable, thus there is
no loss of generality in assuming that X itself is separable. Then there exists
a countable set Σ ⊂ D(X) such that for every compact set K ⊂ X there is a
σ ∈ Σ with σ|K = 1. Note that

N(Tn b σ) ≤ (kσk∞ + Lip(σ))NV (Tn )

34 U. Lang

whenever σ ∈ D(X) and V ⊂ X is an open set containing spt(σ). Hence, for

each σ ∈ Σ, we may apply the first part to the restrictions Tn b σ, temporarily
viewed as currents in the compact set spt(σ). In combination with a diagonal
process, this allows to extract a subsequence (Tn(i) )i∈N such that for every
σ ∈ Σ, the Tn(i) b σ converge weakly to some Tσ ∈ Nm (X). Finally, for every
(f, g) ∈ D m (X), choose σ ∈ Σ with σ|spt(f ) = 1 and put T (f, g) := Tσ (f, g).
This defines a locally normal current T in X, and Tn(i) → T weakly.

In the context of classical currents, there is a similar compactness theorem

for currents with locally finite mass, cf. [29, Lemma 26.14]. Such a result is
not available for metric currents. For instance, let (ηj )j∈N be a sequence of
mollifiers on Rm , as in Sect. 1.5. The corresponding currents [ηj ] ∈ Dm (Rm )
(cf. Proposition 2.6) satisfy M([ηj ]) = 1 and spt([ηj ]) ⊂ U(0, 1/j). However,
no subsequence converges weakly to a current in Dm (Rm ), for there is no
metric m-current for m ≥ 1 whose support is a single point, cf. Lemma 3.2(3).
Viewed as classical currents, the [ηj ] converge weakly to the classical m-
current T̄ satisfying T̄ (f dx1 ∧ . . . ∧ dxm ) = f (0) for all f ∈ Cc∞ (Rm ).
We now examine the relation between metric and classical currents more
closely. To this end we first recall a number of definitions and results from [14].
We adopt the notation from there, except that we put a bar on the symbols
D, M, N, F to distinguish those spaces of forms, currents, and seminorms
from their metric analogues.
For an open subset U of Rn , D̄ m (U ) denotes the vector space of compactly
supported C ∞ m-forms on U , endowed with the usual locally convex C ∞
topology, and D̄m (U ) denotes the dual space, consisting of all m-dimensional
currents in U , cf. [29, p. 131f] and [14, 4.1.7]. The comass kφk of an m-
covector φ ∈ Λm Rn is the supremum of hξ, φi over all simple m-vectors ξ ∈
Λm Rn with Euclidean norm |ξ| ≤ 1, where h·, ·i denotes the dual pairing.
The mass kξk of an m-vector ξ ∈ Λm Rn is the supremum of hξ, φi over all
φ ∈ Λm Rn with comass kφk ≤ 1, cf. [14, 1.8.1]. The comass norm of an
m-form φ ∈ D̄ m (U ) is given by
M̄(φ) = supx∈U kφ(x)k.
Let T ∈ D̄m (U ). For every open set V ⊂ U , put
M̄V (T ) := sup T (φ) : φ ∈ D̄ m (U ), spt(φ) ⊂ V, M̄(φ) ≤ 1 .

(5.2)
The number M̄(T ) := M̄U (T ) ∈ [0, ∞] is the mass of T . As in Theorem 4.3,
one obtains a Borel regular outer measure kT k such that

kT k(A) = inf M̄V (T ) : V ⊂ U is open, A ⊂ V
for every set A ⊂ U . If kT k is locally finite, then it is a Radon measure, and
there exists a kT k-measurable m-vector field ξ on U so that kξ(x)k = 1 for
kT k-almost every x ∈ U and
Z Z
T (φ) = hξ(x), φ(x)i dkT k(x) ≤ kφ(x)k dkT k(x)
U U
Local currents in metric spaces 35

for all φ ∈ D̄ m (U ), cf. [14, 4.1.5 and p. 349] and [29, 26.4–26.8]. The seminorm
N̄ on D̄m (U ) and the space N̄m,loc (U ) of locally normal currents are defined
in analogy to Definition 5.1. For a compact set K ⊂ U , N̄m,K (U ) denotes
the set of all T ∈ D̄m (U ) with spt(T ) ⊂ K and N̄(T ) < ∞, cf. [14, p. 358].
The flat seminorm of a form φ ∈ D̄ m (U ) relative to a compact set K ⊂ U
is given by

F̄K (φ) = sup supx∈K kφ(x)k, supx∈K kdφ(x)k ,

the respective flat seminorm of a current T ∈ D̄m (U ) by

F̄K (T ) = sup T (φ) : φ ∈ D̄ m (U ), F̄K (φ) ≤ 1 .



Note that F̄K (∂T ) ≤ F̄K (T ) for m ≥ 1. If F̄K (T ) < ∞, then spt(T ) ⊂ K. If
T ∈ D̄m (U ) and spt(T ) ⊂ K, then

F̄K (T ) ≤ M̄(T − ∂S) + M̄(S)

for all S ∈ D̄m+1 (U ) with spt(S) ⊂ K, and equality holds for at least one such
S. The F̄K -closure of N̄m,K (U ) in D̄m (U ) is denoted by F̄m,K (U ). The space
F̄m (U ) of flat chains with compact support in U is the union of all F̄m,K (U ).
The space F̄m,loc (U ) of locally flat chains consists of all T ∈ D̄m (U ) such that
T b σ ∈ F̄m (U ) for every σ ∈ Cc∞ (U), cf. [14, 4.1.12].
Finally, we note that the set T ∈ F̄m,K (U ) : M̄(T ) < ∞ equals the
M̄-closure of N̄m,K (U ) in D̄m (U ), cf. [14, 4.1.17], and

F̄m,K (U ) = R + ∂S :R ∈ F̄m,K (U ), M̄(R) < ∞,
S ∈ F̄m+1,K (U ), M̄(S) < ∞ , (5.3)

cf. [14, p. 382].

Theorem 5.5 (comparison map). Let U ⊂ Rn be an open set, n ≥ 1. For

every m ≥ 0, there exists an injective linear map Cm : Dm (U ) → D̄m (U ) such
that
Cm (T )(f dg1 ∧ . . . ∧ dgm ) = T (f, g1 , . . . , gm )

for all (f, g1 , . . . , gm ) ∈ Cc∞ (U ) × [C ∞ (U )]m . The following properties hold:

(1) For m ≥ 1, ∂ ◦ Cm = Cm−1 ◦ ∂.

(2) For all T ∈ Dm (U ), kT k ≤ kCm (T )k ≤ n

m kT k.

(3) The restriction of Cm to Nm,loc (U ) is an isomorphism onto N̄m,loc (U ).

(4) The image of Cm contains the space F̄m,loc (U ) of m-dimensional locally

flat chains in U .
36 U. Lang

For currents with finite mass and compact support in Rn this result was
proved by Ambrosio and Kirchheim in [3, Theorem 11.1]. They conjectured
that the image under Cm of {T ∈ Mm (Rn ) : spt(T ) is compact} coincides
with the space of m-dimensional flat chains with finite mass and compact
support in Rn . In view of (4) and the many analogous properties of Dm (U )
and F̄m,loc (U ), one may similarly ask whether Cm (Dm (U )) = F̄m,loc (U ).

Proof. In case m > n, Dm (U ) = {0} by (2.10), and also D̄m (U ) = {0}. Thus
Cm is the trivial map in this case.
In case m = 0, given T ∈ D0 (U ), C0 (T ) is the functional satisfying
C0 (T )(f ) = T (f ) for all f ∈ D̄ 0 (U ) = Cc∞ (U ). The continuity property
of T implies that C0 (T ) is sequentially continuous on D̄ 0 (U ). This yields
C0 (T ) ∈ D̄0 (U ) (cf. [27, Theorem 6.6]).
P let T ∈ Dm (U ), 1 ≤ m ≤ n. To define∞Cm (T ), write φ ∈ D̄ (U ) as
Now m

φ = λ∈Λ(n,m) φλ dxλ(1) ∧ . . . ∧ dxλ(m) , φλ ∈ Cc (U ), and put

X
Cm (T )(φ) := T (φλ , πλ(1) , . . . , πλ(m) ),
λ∈Λ(n,m)

where πi : U → R is the ith coordinate projection, πi (x) = xi . As above,

due to the continuity of T in the first argument, this defines an element of
D̄m (U ). For (f, g1 , . . . , gm ) ∈ Cc∞ (U )×[C ∞ (U )]m , the coefficients of the form
φ = f dg1 ∧ . . . ∧ dgm are given by φλ = f det[Dλ(k) gi ]m i,k=1 . Thus

X
Cm (T )(φ) = T (φλ , πλ(1) , . . . , πλ(m) ) = T (f, g1 , . . . , gm )
λ∈Λ(n,m)

by Theorem 2.5 (chain rule).

As for the injectivity of Cm in the case 0 ≤ m ≤ n, it suffices to
note that for every nonzero T ∈ Dm (U ) one finds, by approximation, an
(f, g1 , . . . , gm ) ∈ Cc∞ (U ) × [C ∞ (U )]m with

Cm (T )(f dg1 ∧ . . . ∧ dgm ) = T (f, g1 , . . . , gm ) 6= 0.

For (1), let T ∈ Dm (U ) and (f, g1 , . . . , gm−1 ) ∈ Cc∞ (U ) × [C ∞ (U )]m−1 ,

and choose σ ∈ Cc∞ (U ) with σ|spt(f ) = 1. Then

∂(Cm (T ))(f dg1 ∧ . . . ∧ dgm−1 ) = Cm (T )(σ df ∧ dg1 ∧ . . . ∧ dgm−1 )

= T (σ, f, g1 , . . . , gm−1 )
= ∂T (f, g1 , . . . , gm−1 )
= Cm−1 (∂T )(f dg1 ∧ . . . ∧ dgm−1 ).

To prove (2), let T ∈ Dm (U ), and let V ⊂ U be an open set. If Λ

is a finite set, (fλ , g λ ) ∈ Cc∞ (U ) × [C ∞ (U ) ∩ Lip1 (U )]m , spt(fλ ) ⊂ V , and
Local currents in metric spaces 37

≤ 1, then the form φ := λ∈Λ fλ dg1λ ∧ . . . ∧ dgm λ ∈ D̄ m (U ) has

P P
λ∈Λ |fλ |
comass norm M̄(φ) ≤ 1, hence
X
T (fλ , g λ ) = Cm (T )(φ) ≤ M̄V (Cm (T )).
λ∈Λ

P implies that MV (T ) ≤ M̄V (Cm (T

This )). Conversely, for every form φ =
λ∈Λ(n,m) φλ dxλ(1) ∧ . . . ∧ dxλ(m) ∈ D̄ (U ) with spt(φ) ⊂ V and M̄(φ) ≤ 1,
m

we have |φλ | ≤ 1 for all λ ∈ Λ(n, m), so that

 
X n
Cm (T )(φ) = T (φλ , πλ(1) , . . . , πλ(m) ) ≤ MV (T ).
m
λ∈Λ(n,m)

n


Hence M̄V (Cm (T )) ≤ m MV (T ). This yields (2).
We prove (3). From (1) and (2) it follows that Cm (T ) ∈ N̄m,loc (U ) if and
only if T ∈ Nm,loc (U ). Hence, since Cm is injective, it suffices to construct
a map C̄m : N̄m,loc (U ) → Dm (U ) such that Cm (C̄m (T̄ )) = T̄ for all T̄ ∈
N̄m,loc (U ). Let T̄ ∈ N̄m,loc (U ). We first observe that whenever (f, g), (f˜, g̃) ∈
Cc∞ (U ) × [C ∞ (U ) ∩ Lip1 (U )]m , then
Z
˜
T̄ (f dg1 ∧ . . . ∧ dgm ) − T̄ (f dg̃1 ∧ . . . ∧ dg̃m ) ≤ |f − f˜| dkT̄ k
U
m 
X Z Z 
+ Lip(f ) |gi − g̃i | dkT̄ k + |f ||gi − g̃i | dk∂ T̄ k ; (5.4)
i=1 spt(f ) U

this is just the “classical” analogue of Lemma 5.2(2). To define C̄m (T̄ ), let
(f, g) ∈ D(U ) × [Lip1 (U )]m , and choose a sequence ((f k , g k ))k∈N in Cc∞ (U ) ×
[C ∞ (U )∩Lip1 (U )]m such that f k → f in D(U ) and gik → gi locally uniformly
for i = 1, . . . , m. It follows from (5.4) that (T̄ (f k dg1k ∧ . . . ∧ dgm
k ))
k∈N is a
Cauchy sequence whose limit is independent of the choice of the sequence
((f k , g k ))k∈N . Put

C̄m (T̄ )(f, g) := lim T̄ (f k dg1k ∧ . . . ∧ dgm

k
).
k→∞

Let (f˜, g̃) ∈ D(U ) × [Lip1 (U )]m be another such tuple. By choosing the
approximating sequences appropriately, we see that (5.4) holds in the limit,
i.e., C̄m (T̄ )(f, g) − C̄m (T̄ )(f˜, g̃) is less than or equal to the expression on
the right side of (5.4). Now it is clear that C̄m (T̄ ) extends to a current
C̄m (T̄ ) ∈ Dm (U ) satisfying

C̄m (T̄ )(f, g) = T̄ (f dg1 ∧ . . . ∧ dgm )

for all (f, g) ∈ Cc∞ (U )×[C ∞ (U )]m . As the left side of this last equality equals
Cm (C̄m (T̄ ))(f dg1 ∧ . . . ∧ dgm ), we have Cm (C̄m (T̄ )) = T̄ .
It remains to prove (4). First we observe that for every compact set
K ⊂ U , the restriction of C̄m to N̄m,K (U ) naturally extends to a map
38 U. Lang


from the set T̄ ∈ F̄m,K (U ) : M̄(T̄ ) < ∞ , which equals the M̄-closure of
N̄m,K (U ) in D̄m (U ), into Mm (U ). Since the restriction of C̄m is 1-Lipschitz
with respect to M̄ and M, this follows by the completeness of (Mm (U ), M)
(cf. Proposition 4.2). For every T̄ ∈ F̄m,K (U ) with M̄(T̄ ) < ∞, we still
have Cm (C̄m (T̄ )) = T̄ . Next, let T̄ ∈ F̄m,K (U ). Choose R̄ ∈ F̄m,K (U ) with
M̄(R̄) < ∞ and S̄ ∈ F̄m+1,K (U ) with M̄(S̄) < ∞ such that T̄ = R̄ + ∂ S̄,
cf. (5.3), and put C̄m (T̄ ) := C̄m (R̄) + ∂(C̄m+1 (S̄)) ∈ Dm (U ). Then


Cm (C̄m (T̄ )) = Cm (C̄m (R̄)) + ∂ Cm+1 (C̄m+1 (S̄)) = T̄ .

Since Cm is injective, this identity also shows that C̄m (T̄ ) is well-defined.
Finally, suppose that T̄ ∈ F̄m,loc (U ). Given (f, g1 , . . . , gm ) ∈ D m (U ), choose
σ ∈ Cc∞ (U ) such that σ = 1 on some neighborhood of spt(f ) and put
C̄m (T̄ )(f, g1 , . . . , gm ) := C̄m (T̄ b σ)(f, g1 , . . . , gm ). It is readily checked that
this yields a well-defined C̄m (T̄ ) ∈ Dm (U ) with Cm (C̄m (T̄ )) = T̄ .

For an example of a current T ∈ Nm,loc (Rn ) with kT k = 6 kCm (T )k, let

4
T ∈ N2,loc (R ) be defined by
Z


T (f, g, h) = f D1 gD2 h − D2 gD1 h + D3 gD4 h − D4 gD3 h dx
R4

for (f, g, h) ∈ D 2 (R4 ). The corresponding classical current T̄ = C2 (T ) ∈

N̄2,loc (R4 ) is given by
Z
T̄ (φ) = he1 ∧ e2 + e3 ∧ e4 , φi dx
R4

for φ ∈ D̄ 2 (R4 ). Note that these currents have boundary zero. To compute
the mass, use the following inequality: If v, w are vectors in R4 with Euclidean
norm |v|, |w| ≤ 1, then

|v1 w2 − v2 w1 + v3 w4 − v4 w3 | = |h(−v2 , v1 , −v4 , v3 ), wi| ≤ 1,

with equality if and only if w = ±(−v2 , v1 , −v4 , v3 ). It follows that kT k = L 4 .

The same inequality also shows that the form dx1 ∧dx2 +dx3 ∧dx4 has comass
norm 1 (a special case of Wirtinger’s inequality, cf. [14, p. 40]), which leads
to kT̄ k = 2L 4 .

6 Slicing
We now develop the slicing theory for locally normal m-currents in X with
respect to a locally Lipschitz map π : X → Rk , where 1 ≤ k ≤ m. The slices
are currents of dimension m − k in the level sets of π. General references
are [14, 4.2.1 and 4.3.1–5], [29, 28.6–28.10] for classical currents and [3, pp. 31–
36] for metric currents of finite mass. We first treat the case k = 1, which
suffices for most geometric applications. The general case will be relevant in
Sect. 7 and in Theorems 8.4 and 8.5.
Local currents in metric spaces 39

Definition 6.1 (codimension one slices of normal currents). Suppose T ∈

Nm,loc (X), m ≥ 1, π ∈ Liploc (X), and s ∈ R. The left-hand and right-hand
slices of T at s with respect to π are the currents in Dm−1 (X) defined by

hT, π, s−i := ∂(T b {π < s}) − (∂T )b {π < s}

= (∂T )b {π ≥ s} − ∂(T b {π ≥ s}),
hT, π, s+i := ∂(T b {π ≤ s}) − (∂T )b {π ≤ s}
= (∂T )b {π > s} − ∂(T b {π > s}),

respectively.

Note that since T ∈ Mm,loc (X) and ∂T ∈ Mm−1,loc (X), the restrictions
T b B and (∂T )b B are defined for every Borel set B ⊂ X. Moreover, by
Theorem 4.4(1), T b B + T b (X \ B) = T . If (kT k + k∂T k)(π −1 {s}) = 0, then
T b π −1 {s} = 0 and (∂T )b π −1 {s} = 0, thus hT, π, s−i = hT, π, s+i. It follows
that if spt(T ) is separable, then

hT, π, s−i = hT, π, s+i (6.1)

for all but countably many s ∈ R. We now focus on right-hand slices.

From the two representations of hT, π, s+i we see that

spt(hT, π, s+i) ⊂ π −1 {s} ∩ spt(T ) (6.2)

for every s ∈ R. If m ≥ 2, then

h∂T, π, s+i = −∂hT, π, s+i (6.3)

for every s ∈ R.
There is a useful characterization of hT, π, s+i as a weak limit. For this
we approximate the characteristic function χ{π>s} by a family (us,δ )δ>0 in
Liploc (X) such that

0 ≤ χ{π>s} − us,δ ≤ χ{s<π<s+δ}

on X. (A natural choice is us,δ := γs,δ ◦ π for the piecewise affine (1/δ)-

Lipschitz function γs,δ : R → R with γs,δ |(−∞,s] = 0 and γs,δ |[s+δ,∞) = 1.)
By (3.6), T b (1, us,δ ) = (∂T )b us,δ − ∂(T b us,δ ), hence

hT, π, s+i − T b (1, us,δ ) = (∂T )b (χ{π>s} − us,δ ) − ∂(T b (χ{π>s} − us,δ )).

It follows that
hT, π, s+i = lim T b (1, us,δ ) (6.4)
δ→0+

for the weak limit.

Theorem 6.2 (codimension one slices of normal currents). Suppose T ∈

Nm,loc (X), m ≥ 1, and π ∈ Liploc (X).
40 U. Lang

(1) For every s ∈ R and every open set V ⊂ X,

1
khT, π, s+ik(V ) ≤ lim inf kT b (1, π)k(V ∩ {s < π < s + δ}).
δ→0+ δ

(2) For all (f, g) ∈ D m−1 (X),

Z
hT, π, s+i(f, g) ds = (T b (1, π))(f, g).
R

(3) For every kT b (1, π)k-measurable set B ⊂ X with kT b (1, π)k(B) < ∞,
Z
khT, π, s+ik(B) ds = kT b (1, π)k(B).
R

(4) If spt(T ) is separable, then hT, π, s+i ∈ Nm−1,loc (X) for almost every
s ∈ R. If π|spt(T ) is proper, or if T ∈ Nm (X) and π ∈ Lip(X), then
hT, π, s+i ∈ Nm−1 (X) for almost every s ∈ R.
Proof. For every δ > 0, fix a non-decreasing function γδ ∈ C 1,1 (R) such that
γδ |(−∞,0] = 0 and γδ |[δ,∞) = 1. Let s ∈ R. Define γs,δ (t) := γδ (t − s) for t ∈ R,
and put us,δ := γs,δ ◦ π. By Theorem 2.5 (chain rule),
0 0
T b (1, us,δ ) = T b (γs,δ ◦ π, π) = (T b (1, π))b (γs,δ ◦ π).

Hence, for every open set V ⊂ X, (4.10) gives

Z
0
kT b (1, us,δ )k(V ) ≤ γs,δ ◦ π dkT b (1, π)k
V
≤ Lip(γs,δ )kT b (1, π)k(V ∩ {s < π < s + δ}),
0 ◦ π is zero outside {s < π < s + δ}. Choosing the function γ
since γs,δ δ
appropriately, with Lip(γδ ) ≤ 1/δ + δ say, we obtain (1) by (6.4) and the
lower semicontinuity of mass.
For the proof of (2) we assume (f, g) ∈ D(X) × [Lip1 (X)]m−1 . Choose
a ∈ R such that spt(f ) ⊂ {π > a}. If δ > 0 is fixed, then the function
s 7→ T (f, us,δ , g) is continuous, and an approximation argument using simple
functions shows that
Z ∞  Z ∞ 
T (f, us,δ , g) ds = T f, us,δ ds, g .
a a

Now we let δ tend to 0. We know that then T (f, us,δ , g) → hT, π, s+i(f, g)
for every s ∈ R. Moreover, since |us,δ | ≤ 1, Lemma R 5.2(1) yields the
uniform bound |T (f, us,δ , g)| ≤ Lip(f )kT k(spt(f )) + X |f | dk∂T k. Hence,
s 7→ hT, π, s+i(f, g) is a bounded Borel function with compact support, and
Z ∞ Z ∞ Z
lim T (f, us,δ , g) ds = hT, π, s+i(f, g) ds = hT, π, s+i(f, g) ds.
δ→0+ a a R
Local currents in metric spaces 41

R∞ R t−a
On the other hand, the functions t 7→ a γs,δ (t) ds = −∞ γδ (r) dr are 1-
Lipschitz for all δ > 0 and converge uniformly to t 7→ max{t − a, 0}, as δ → 0.
It follows that
 Z ∞ 
lim T f, us,δ ds, g = T (f, π − a, g) = T (f, π, g) = (T b (1, π))(f, g).
δ→0+ a

This proves (2).

Suppose V ⊂ X is an open set such that kT b (1, π)k(V ) < ∞. Then
the function s 7→ kT b (1, π)k(V ∩ {π < s}) is non-decreasing, hence almost
everywhere differentiable. From (1) it follows that
d
khT, π, s+ik(V ) ≤ kT b (1, π)k(V ∩ {π < s}) (6.5)
ds
for almost every s ∈ R. Hence
Z ∗
khT, π, s+ik(V ) ds ≤ kT b (1, π)k(V ),
R
R∗
where R denotes the upper integral. The reverse inequality
Z
khT, π, s+ik(V ) ds ≥ kT b (1, π)k(V )
∗R

for the lower integral is a direct consequence of (2). This shows (3) for every
open set V with kT b (1, π)k(V ) < ∞. Since every compact set K ⊂ X is a
difference of two such open sets, the same identity holds for all compact sets,
and we obtain (3) by approximation.
Finally, (4) follows easily from (3), or just (6.5), together with the corre-
sponding result for ∂hT, π, s+i = −h∂T, π, s+i in case m ≥ 2.

Now we pass to slices of codimension k, for 1 ≤ k ≤ m. Our approach is

similar to that of [14, 4.3.1], where slices are defined for arbitrary locally flat
chains (cf. the last remark on p. 437 in [14]). Definition 6.3 generalizes (6.4)
and applies to all metric currents, however we shall prove the existence of the
weak limits in question only for locally normal currents.
For s ∈ R and δ > 0, let now γs,δ : R → R be the piecewise affine (1/δ)-
Lipschitz function with γs,δ |(−∞,s] = 0 and γs,δ |[s+δ,∞) = 1. Then, for y =
(y1 , . . . , yk ) ∈ Rk and δ > 0, define γy,δ : Rk → Rk such that
γy,δ (z) = (γy1 ,δ (z1 ), . . . , γyk ,δ (zk ))
for all z = (z1 , . . . , zk ) ∈ Rk .
Definition 6.3 (slices). Suppose 1 ≤ k ≤ m, π ∈ Liploc (X, Rk ), and T ∈
Dm (X). We define the slice of T at y ∈ Rk with respect to π as the weak
limit
hT, π, yi := lim T b (1, γy,δ ◦ π)
δ→0+

whenever it exists and defines an element of Dm−k (X).

42 U. Lang

In view of (6.4), in case T ∈ Nm,loc (X) and k = 1 we have

hT, π, yi = hT, π, y+i (6.6)

for every y ∈ R. (We could equally well arrange that hT, π, yi = (hT, π, y−i +
hT, π, y+i)/2, as in [14, 4.3.4], by choosing a more symmetric definition of
hT, π, yi.) Before proving the existence of slices of locally normal currents in
the case k > 1 we discuss a few properties that can be derived directly from
the definition.
Let y = (y1 , . . . , yk ) ∈ Rk , and suppose that hT, π, yi ∈ Dm−k (X) exists.
For δ > 0, consider the cube

C(y, δ) := [y1 , y1 + δ] × . . . × [yk , yk + δ] ⊂ Rk .

A simple Lipschitz partition of unity argument on Rk \ C(y, δ) shows that

spt(T b (1, γy,δ ◦ π)) ⊂ π −1 (C(y, δ)) ∩ spt(T ), (6.7)

so that
spt(hT, π, yi) ⊂ π −1 {y} ∩ spt(T ). (6.8)
If m > k, then (∂T )b (1, γy,δ ◦ π) = (−1)k ∂(T b (1, γy,δ ◦ π)) by (3.7), hence

h∂T, π, yi = (−1)k ∂hT, π, yi. (6.9)

Whenever 0 ≤ l ≤ m − k and (u, v) ∈ Liploc (X) × [Liploc (X)]l , the alternating

property gives (T b (u, v))b (1, γy,δ ◦ π) = (−1)kl (T b (1, γy,δ ◦ π))b (u, v), so

hT b (u, v), π, yi = (−1)kl hT, π, yib (u, v), (6.10)

in particular hT b u, π, yi = hT, π, yib u.

Theorem 6.4 (slicing). Suppose 1 ≤ k ≤ m, π ∈ Liploc (X, Rk ), T ∈

Nm,loc (X), and spt(T ) is separable.

(1) For L k -almost every y ∈ Rk , the slice hT, π, yi ∈ Dm−k (X) exists and
is locally normal.

(2) For all (f, g) ∈ Bc∞ (X) × [Liploc (X)]m−k ,

Z
hT, π, yi(f, g) dy = (T b (1, π))(f, g).
Rk

(3) For every kT b (1, π)k-measurable set B ⊂ X,

Z
khT, π, yik(B) dy = kT b (1, π)k(B).
Rk
Local currents in metric spaces 43

Proof. We proceed by induction on k.

Let k = 1. We know from (6.6) and Theorem 6.2 that (1) holds, (2)
holds for all (f, g) ∈ D m−1 (X), and (3) holds for all kT b (1, π)k-measurable
sets with finite measure. Since spt(T ) is separable by assumption, spt(T )
is σ-compact, hence kT b (1, π)k is σ-finite, and (3) follows. To prove (2) in
the general case, fix g ∈ [Lip1 (X)]m−1 and σ ∈ D(X), 0 ≤ σ ≤ 1, and
let K := spt(σ). Denote by Φ the set of all f ∈ Bc∞ (X) with |f | ≤ σ such
that (2) holds for (f, g). Suppose that f1 , f2 , . . . ∈ Φ, f ∈ Bc∞ (X), and fj → f
pointwise on X. For almostR every y ∈ R, |hT, π, yi(fj , g)| ≤ khT, π, yik(K)
for all j. Moreover, by (3), R khT, π, yik(K) dy = kT b (1, π)k(K) < ∞. Now
the bounded convergence theorem and Theorem 4.4(2) imply that f ∈ Φ, so
that Φ is a Baire class. Since Φ contains all f ∈ D(X) with |f | ≤ σ, Φ is the
class of all f ∈ Bc∞ (X) with |f | ≤ σ, cf. Sect. 1.2. This yields (2).
Now let k ≥ 2, and suppose (1), (2), and (3) hold with k 0 := k − 1 in place
of k. We write π 0 for (π1 , . . . , πk0 ), so that π = (π 0 , πk ). There is a Borel set
0 0 0
D0 ⊂ Rk with L k (Rk \ D0 ) = 0 such that hT, π 0 , y 0 i exists as an element of
Nm−k0 ,loc (X) for every y 0 = (y1 , . . . , yk0 ) ∈ D0 . By (6.6),

Ty := hT, π 0 , y 0 i, πk , yk ∈ Dm−k (X)

then exists for every y = (y 0 , yk ) ∈ D0 × R. Moreover, for fixed (f, g) ∈

D m−k (X), the function y 7→ Ty (f, g) is Borel measurable on D0 × R, since

Ty (f, g) = lim lim T (f, γy0 ,δ0 ◦ π 0 , γyk ,δ ◦ πk , g)

δ→0+ δ 0 →0+

and y 7→ T (f, γy0 ,δ0 ◦ π 0 , γyk ,δ ◦ πk , g) is continuous on Rk for fixed δ 0 , δ > 0.

Since each kTy k is concentrated on the separable set spt(T ), it also follows
that for every open set V ⊂ X the function y 7→ kTy k(V ) ∈ [0, ∞] is Borel
measurable on D0 × R. Hence, by Fubini’s theorem, the result in the case
k = 1, (6.10), and the induction assumption,
Z Z Z
kTy k(V ) dy = hT, π 0 , y 0 i, πk , yk (V ) dyk dy 0
Rk Rk0 R
Z
= khT, π 0 , y 0 ib (1, πk )k(V ) dy 0
Rk0
Z
= khT b (1, πk ), π 0 , y 0 ik(V ) dy 0
Rk0
= k(T b (1, πk ))b (1, π 0 )k(V )
= kT b (1, π)k(V ).

Reasoning as in the proof of Theorem 6.2 and using again the σ-compactness
of spt(T ) we obtain
Z
kTy k(B) dy = kT b (1, π)k(B) (6.11)
Rk
44 U. Lang

for every kT b (1, π)k-measurable set B ⊂ X. Now it follows that Ty ∈

Mm−k,loc (X) for almost every y ∈ Rk , and for fixed (f, g) ∈ D m−k (X),
the function y 7→ Ty (f, g) is in L1 (Rk ). By Fubini’s theorem, the result in
the case k = 1, and the induction assumption,
Z Z Z
Ty (f, g) dy = hT, π 0 , y 0 i, πk , yk (f, g) dyk dy 0
Rk R k0 R
Z
= hT, π 0 , y 0 i(f, πk , g) dy 0
Rk 0
= T (f, π 0 , πk , g)
= (T b (1, π))(f, g).

As in the case k = 1, a Baire class argument then shows that

Z
Ty (f, g) dy = (T b (1, π))(f, g) (6.12)
Rk

for all (f, g) ∈ Bc∞ (X)×[Liploc (X)]m−k . Given φ ∈ Bloc ∞ (Rk ), we may replace

f by (φ ◦ π)f ; since φ ◦ π = φ(y) on spt(Ty ) ⊂ π −1 {y}, this yields the identity

Z
φ(y)Ty (f, g) dy = (T b (φ ◦ π, π))(f, g). (6.13)
Rk

Now we show that for almost every y ∈ Rk , the slice hT, π, yi exists as
an element of Nm−k,loc (X) and coincides with Ty . By (6.12) and (6.11), this
will complete the proof of the theorem.
We first assume that X is compact, so that T is normal and π is Lipschitz.
Then we fix a countable set F ⊂ Lip1 (X) that is dense in Lip1 (X) with
respect to k · k∞ . Put Tπ := T b (1, π) ∈ Nm−k (X) for the moment, and let
y ∈ Rk and δ > 0. Using (6.7), (4.7), and Theorem 4.4 we get

T b (1, γy,δ ◦ π) = T b (χC(y,δ) ◦ π, γy,δ ◦ π)

1 1
= k T b (χC(y,δ) ◦ π, π) = k Tπ b π −1 (C(y, δ)). (6.14)
δ δ
Similarly, if m > k, applying (3.7) twice we obtain

∂ T b (1, γy,δ ◦ π) = (−1)k (∂T )b (1, γy,δ ◦ π)

1 1
= (−1)k k (∂T )b (1, π) b π −1 (C(y, δ)) = k (∂Tπ )b π −1 (C(y, δ)).


δ δ
For every Borel set C ⊂ Rk , put

µ(C) := (kTπ k + k∂Tπ k)(π −1 (C))

if m > k, and µ(C) := kTπ k(π −1 (C)) if m = k. This defines a finite Borel
measure µ on Rk . Let Mµ : Rk → [0, ∞] be the maximal function of µ. Then
1
N T b (1, γy,δ ◦ π) = k µ(C(y, δ)) ≤ αk k k/2 Mµ (y).


δ
Local currents in metric spaces 45

For all (f, g) ∈ D m−k (X), (6.14) and (6.13) give

1

T b (1, γy,δ ◦ π) (f, g) = k T b (χC(y,δ) ◦ π, π) (f, g)
δ Z
1
= k Tz (f, g) dz. (6.15)
δ C(y,δ)

Almost every y ∈ Rk satisfies Mµ (y) < ∞ and is a Lebesgue point of z 7→

Tz (f, g) for all (f, g) ∈ F × F m−k . Applying Proposition 5.3 (convergence
criterion) and (1.4) we conclude that for every such y, the slice hT, π, yi exists
as an element of Nm−k (X) and coincides with Ty .
In the general case, when X is locally compact and spt(T ) is separable,
there is no loss of generality in assuming X itself to be separable. Then we
choose a countable set Σ ⊂ D(X) as in the proof of Theorem 5.4. The above
argument together with (6.10) then shows that for almost every y ∈ Rk , the
slice hT b σ, π, yi exists as an element of Nm−k (X) and coincides with Ty b σ
for every σ ∈ Σ. The general result follows.

The following theorem will be used in the proof of Theorem 8.5 (rectifiable
slices).

Theorem 6.5 (iterated slices). Suppose k, k 0 ≥ 1, m ≥ k + k 0 , π ∈

0
Liploc (X, Rk ), π 0 ∈ Liploc (X, Rk ), T ∈ Nm,loc (X), and spt(T ) is separable.
Then
hT, (π, π 0 ), (y, y 0 )i = hT, π, yi, π 0 , y 0
0 0
for L k+k -almost every (y, y 0 ) ∈ Rk+k .

In case k 0 = 1 the result is clear from the preceding proof. Our argument
for the general case follows [14, Theorem 4.3.5].

Proof. We assume that X is compact, so that T ∈ Nm (X) and (π, π 0 ) ∈

0
Lip(X, Rk+k ). The theorem easily follows from the result in this special
case. We fix a countable set F ⊂ Lip1 (X) that is dense in Lip1 (X) with
respect to k · k∞ .
From Theorem 6.4 and Fubini’s theorem we conclude that there is a
0 0 0
Borel set D1 ⊂ Rk+k with L k+k (Rk+k \ D1 ) = 0 such that whenever
(y, y 0 ) ∈ D1 , both hT, π, yi and hT, (π, π 0 ), (y, y 0 )i exist and are normal cur-
0 0
rents, L k (({y} × Rk ) \ D1 ) = 0, and
Z
N hT, (π, π 0 ), (y, z 0 )i dz 0 < ∞.


(6.16)
Rk0
0
For δ 0 > 0 and (y, y 0 ) ∈ D1 we define a functional Aδ0 (y, y 0 ) on D m−k−k (X)
by Z
0 1
Aδ0 (y, y )(f, g) := 0 k hT, (π, π 0 ), (y, z 0 )i(f, g) dz 0 .
(δ ) C(y0 ,δ0 )
46 U. Lang

0
Now (6.16) with M in place of N shows that hT, (π, π 0 ), (y, ·)i(f, g) ∈ L1 (Rk )
and, in conjunction with the bounded convergence theorem, that Aδ0 (y, y 0 )
satisfies the continuity condition for metric currents. Hence Aδ0 (y, y 0 ) is a
current; since
Z
1
N(Aδ0 (y, y 0 )) ≤ 0 k N hT, (π, π 0 ), (y, z 0 )i dz 0 ,


(6.17)
(δ ) C(y0 ,δ0 )

it is normal by (6.16). Now let δ, δ 0 > 0 and (y, y 0 ) ∈ D1 . For all (f, g) ∈
0
D m−k−k (X), we obtain

T (f, γy,δ ◦ π, γy0 ,δ0 ◦ π 0 , g)

Z
1
= k 0 k hT, (π, π 0 ), (z, z 0 )i(f, g) d(z, z 0 )
δ (δ ) C(y,δ)×C(y0 ,δ0 )
Z
1
= k Aδ0 (z, y 0 )(f, g) dz (6.18)
δ C(y,δ)

by the same argument as for (6.15), and by Fubini’s theorem. Note that the
left side of this identity is continuous in (y, y 0 ), moreover the limit for δ → 0
and for fixed δ 0 > 0 is equal to hT, π, yi(f, γy0 ,δ0 ◦ π 0 , g). Let D2 be the Borel
set of all (y, y 0 ) ∈ D1 where the limit of (6.18) equals Aδ0 (y, y 0 )(f, g) for all
0 0
rational numbers δ 0 > 0 and (f, g) ∈ F × F m−k−k . Then L k+k (D1 \ D2 ) =
0, and for every (y, y 0 ) ∈ D2 , the equality

hT, π, yi(f, γy0 ,δ0 ◦ π 0 , g) = Aδ0 (y, y 0 )(f, g) (6.19)

0
holds for all rational numbers δ 0 > 0 and (f, g) ∈ F × F m−k−k , hence for
0
all δ 0 > 0 and (f, g) ∈ D m−k−k (X) by continuity and multilinearity. Now we
pass to the limit for δ 0 → 0. Let D3 be the Borel set of all (y, y 0 ) ∈ D2 where
the limit of the right side of (6.19) coincides with hT, (π, π 0 ), (y, y 0 )i(f, g) for
0
all (f, g) ∈ F × F m−k−k and where
Z
1
N hT, (π, π 0 ), (y, z 0 )i dz 0 < ∞.


supδ0 >0 0 k (6.20)
(δ ) C(y0 ,δ0 )
0
Then L k+k (D2 \ D3 ) = 0, and
0 0 0


lim
0
hT, π, yib (1, γy 0 ,δ 0 ◦ π ) (f, g) = hT, (π, π ), (y, y )i(f, g)
δ →0

0
for all (y, y 0 ) ∈ D3 and (f, g) ∈ F × F m−k−k . Combining (6.19), (6.17), and
0


(6.20), we have supδ0 >0 N hT, π, yib (1, γy0 ,δ0 ◦ π ) < ∞, so it follows from
Proposition 5.3 (convergence criterion) that

hT, π, yi, π 0 , y 0 = hT, (π, π 0 ), (y, y 0 )i

for all (y, y 0 ) ∈ D3 .

Local currents in metric spaces 47

We conclude this section with a result that will be used in the proof of
Theorem 8.9 (closure theorem).
Proposition 6.6 (slicing convergent sequences). Suppose m ≥ 1, (Tn )n∈N
is a sequence in Nm,loc (X) that converges weakly to some T ∈ Nm,loc (X),
each spt(Tn ) is separable, and supn NV (Tn ) < ∞ for every open set V b X.
Let π ∈ Liploc (X). Then for L 1 -almost every s ∈ R there is a subsequence
(Tn(i) )i∈N such that
hTn(i) , π, si → hT, π, si
weakly and supi NV (hTn(i) , π, si) < ∞ for every open set V b X.
Compare [36, p. 208] and [3, Proposition 8.3].
Proof. We assume without loss of generality that X is compact.
We first show the existence of a subsequence (Tn(i) )i∈N such that for all but
countably many s ∈ R, hTn(i) , π, si → hT, π, si weakly for i → ∞. Consider
the measures µn on R defined by µn (B) := (kTn k + k∂Tn k)(π −1 (B)) for
all Borel sets B ⊂ R. Note that supn µn (R) < ∞. Choose a subsequence
(µn(i) )i∈N that converges weakly∗ to some finite Borel measure µ on R (see
e.g. [13, Sect. 1.9]). Now let s ∈ R be such that µ({s}) = 0. Suppose
(f, g) ∈ D(X) × [Lip1 (X)]m , and |f | ≤ 1. Given  > 0, choose δ > 0 such
that kT k({s < π < s+δ}) < /3 and µ([s, s+δ]) < /3, and let us,δ ∈ Lip(X)
be given as in (6.4). Then
(T b χ{π>s} )(f, g) − (T b us,δ )(f, g) = |T ((χ{π>s} − us,δ )f, g)|
≤ kT k({s < π < s + δ}) < /3.
Since lim supi→∞ µn(i) ([s, s + δ]) ≤ µ([s, s + δ]), it follows similarly that
(Tn(i) b χ{π>s} )(f, g) − (Tn(i) b us,δ )(f, g) ≤ µn(i) ([s, s + δ]) < /3
for all sufficiently large i. Moreover, since Tn(i) → T weakly,
(Tn(i) b us,δ )(f, g) − (T b us,δ )(f, g) < /3
for i sufficiently large. Combining these estimates we conclude that
Tn(i) b χ{π>s} → T b χ{π>s}
weakly for i → ∞, and a completely analogous argument shows that
(∂Tn(i) )b χ{π>s} → (∂T )b χ{π>s}
weakly. By (6.6), hTn(i) , π, si → hT, π, si weakly for i → ∞.
Applying Theorem 6.2(3) and (6.3) we obtain
Z Z
lim inf N(hTn , π, si) ds ≤ lim inf N(hTn , π, si) ds
R n→∞ n→∞ R
≤ Lip(π) supn N(Tn ) < ∞.
Hence, for L 1 -almost every s ∈ R, there is a subsequence (Tn(i) )i∈N such that
supi N(hTn(i) , π, si) < ∞. Together with the first part of the proof this yields
the result.
48 U. Lang

7 Projections to Euclidean spaces

The main purpose of this section is to establish Theorem 7.6, the central piece
of the proof of the rectifiablity criterion used later on, Theorem 8.4. We start
by recalling a few basic facts about functions of bounded variation (cf. [13,
Ch. 5], [29, §6]).
Let U ⊂ Rm be an open set, m ≥ 1, and let u ∈ L1loc (U ). For an open set
A ⊂ U,
Z 
1 m
VA (u) := sup u div(ϕ) dx : ϕ ∈ Cc (A, R ), |ϕ| ≤ 1
A

defines the variation of u in A. The number VA (u) is unchanged if Cc1 (A, Rm )

is replaced by Cc∞ (A, Rm ) or Lipc (A, Rm ) = [D(A)]m . Then u is a function
of locally bounded variation in U if VA (u) < ∞ for every open set A b U ,
and BVloc (U ) denotes the vector space of all such functions. Similarly, u is a
function of bounded variation in U if u ∈ L1 (U ) and VU (u) < ∞, and BV(U )
denotes the vector space of all such functions. Note that for u ∈ Liploc (U ),
the integration by parts formula
Z Z
u div(ϕ) dx = − h∇u, ϕi dx
A A

holds for all ϕ ∈ Cc1 (A, Rm ), which implies that VA (u) = A |∇u| dx. In par-
R

ticular, Liploc (U ) ⊂ BVloc (U ). The representation theorem for u ∈ BVloc (U )

says that there exist a Radon measure on U , denoted by |Du|, and a |Du|-
measurable vector field τ on U such that |τ (x)| = 1 for |Du|-almost every
x ∈ U and Z Z
u div(ϕ) dx = − hϕ, τ i d|Du| (7.1)
U U

for ϕ ∈ Cc1 (U, Rm ) or even ϕ ∈ Lipc (U, Rm ) = [D(U )]m . The Radon measure
|Du| is characterized by |Du|(A) = VA (u) for all open sets A ⊂ U .
Functions of bounded variation will be used through the following known
fact (cf. the proof of [3, Lemma 7.3] and the references there), which expresses
a Lipschitz property in terms of the maximal function of the variation mea-
sure.

Lemma 7.1. Suppose that u ∈ BV(Rm ), m ≥ 1. Whenever x, x0 are two

Lebesgue points of u, then

|u(x) − u(x0 )| ≤ cm M|Du| (x) + M|Du| (x0 ) |x − x0 |

for some constant cm depending only on m.

Thus, by (1.3), u is 2cm s-Lipschitz outside a set of Lebesgue measure at

most 3m s−1 |Du|(Rm ), for any s > 0.
Local currents in metric spaces 49

Proof. SupposeR first that u ∈ C 1 (Rm ) in addition, and let r > 0. Since
1
u(z) − u(0) = 0 h∇u(tz), zi dt for z ∈ Rm , it follows that
1
|u(z) − u(0)|
Z Z Z
dz ≤ |∇u(tz)| dt dz
B(0,r) |z| B(0,r) 0
Z 1Z
= |∇u(tz)| dz dt
0 B(0,r)
Z 1 Z
1
= |∇u(z)| dz dt
0 tm B(0,tr)
≤ αm rm M|Du| (0).

For a general u ∈ BV(Rm ), a smoothing argument then shows that

|u(z) − u(x)|
Z
dz ≤ αm rm M|Du| (x)
B(x,r) |z − x|

for every Lebesgue point x of u (cf. [2, Remark 3.8]).

Now suppose that x, x0 ∈ Rm are two Lebesgue points of u, and r :=
|x − x0 | > 0. Put c := 2αm rm /L m (B(x, r) ∩ B(x0 , r)),

A := z ∈ B(x, r) : |u(x) − u(z)| > cM|Du| (x)|x − z| ,

and define A0 similarly with x0 in place of x. The above estimate implies that
L m (A), L m (A0 ) < αm rm /c and thus

L m (A ∪ A0 ) < 2αm rm /c = L m (B(x, r) ∩ B(x0 , r)).

In particular, B(x, r) ∩ B(x0 , r) \ (A ∪ A0 ) is non-empty; for any z in this set

we get

|u(x) − u(x0 )| ≤ |u(x) − u(z)| + |u(x0 ) − u(z)|

≤ cM|Du| (x) |x − z| + cM|Du| (x0 ) |x0 − z|.

Since |x − z|, |x0 − z| ≤ r = |x − x0 |, the result follows.

The next result relates normal currents of the type described in Propo-
sition 2.6 (standard example) to functions of bounded variation, cf. [29, Re-
mark 26.28] and [3, Theorem 3.7].

Theorem 7.2 (normal m-currents in Rm ). Let U ⊂ Rm be an open set,

m ≥ 1.

(1) If u ∈ L1loc (U ), then MA (∂[u]) = VA (u) for every open set A ⊂ U .

(2) If T ∈ Nm,loc (U ), then T = [u] for some u ∈ BVloc (U ), and k∂T k =

|Du|.
50 U. Lang

In particular, if T ∈ Nm (U ), then T = [u] for some u ∈ BV(U ).

Proof. Let u ∈ L1loc (U ). Denote by π the identity map on U , πk (x) = xk .

Let ϕ ∈ Cc∞ (U, Rm ), and choose σ ∈ D(U ) such that σ|spt(ϕ) = 1. By
Theorem 2.5 (chain rule),
Z
u div(ϕ) dx = [u](div(ϕ), π)
U
m
X
= (−1)k−1 [u](σ, ϕk , π1 , . . . , πk−1 , πk+1 , . . . , πm )
k=1
m
X
= (−1)k−1 ∂[u](ϕk , π1 , . . . , πk−1 , πk+1 , . . . , πm ).
k=1

Assuming that A ⊂ U is an open set with MA (∂[u]) < ∞, we conclude that

VA (u) ≤ mMA (∂[u]) < ∞. In particular u ∈ BV(A), and the representation
formula (7.1) holds on A. Let τ be a corresponding vector field on A. To show
that in fact VA (u) ≤ MA (∂[u]), let v ∈ Rm be a unit vector, and complete
it to an orthonormal basis (v, e1 , . . . , em−1 ) of Rm . Define gi ∈ Lip1 (U ) by
gi (x) := hx, ei i. For f, σ ∈ D(U ) with spt(f ) ⊂ A and σ|spt(f ) = 1, we have
Z Z
hf v, −τ i d|Du| = uDv f dx = [u](σ, f, g1 , . . . , gm−1 )
A A Z
= ∂[u](f, g1 , . . . , gm−1 ) ≤ |f | dk∂[u]k.
A

Now an approximation argument shows that |Du|(A) ≤ k∂[u]k(A), thus

VA (u) ≤ MA (∂[u]).
To prove the reverse inequality, suppose that g1 , . . . , gm−1 ∈ C 2 (U ). Con-
sider the matrix D(g1 , . . . , gm−1 ), and let Mk be the (m − 1) × (m − 1) minor
obtained by deleting the kth column. Let ϕ be the C 1 vector field on U with
components ϕk = (−1)k−1 det(Mk ). Note that div(ϕ) = 0 since

0 = d(dg1 ∧ . . . ∧ dgm−1 ) = div(ϕ) dx1 ∧ . . . ∧ dxm .

Now let f ∈ Cc1 (U ). Then

det(D(f, g1 , . . . , gm−1 )) = h∇f, ϕi = div(f ϕ) − f div(ϕ) = div(f ϕ)

on U . Hence, for any σ ∈ D(U ) with σ|spt(f ) = 1,

Z
∂[u](f, g1 , . . . , gm−1 ) = [u](σ, f, g1 , . . . , gm−1 ) = u div(f ϕ) dx.
U

If, in addition, g1 , . . . , gm−1 ∈ Lip1 (U ), then |ϕ| ≤ 1 by the Cauchy–Binet

formula (see e.g. [2, Proposition 2.69]). It follows easily that MA (∂[u]) ≤
VA (u) for every open set A ⊂ U . This completes the proof of (1).
Local currents in metric spaces 51

Let T ∈ Nm,loc (U ). Specializing Theorem 6.4(2) to the case where X = U ,

k = m, and π : U → Rk is the inclusion map, we get
Z Z
T (f, π) = hT, π, xi(f ) dx = f (x)u(x) dx
U U

for every f ∈ Bc∞ (U ), where u(x) := hT, π, xi(σ) for any σ ∈ D(U ) with
σ(x) = 1 (note that spt(hT, π, xi) ⊂ {x}). By (2.11) and a smoothing argu-
ment it follows that T = [u]. Together with (1), this proves (2).

Next we discuss two auxiliary results for currents T ∈ Mm,loc (X). We

look at π# (T b f ) for f ∈ Bc∞ (X) and π ∈ Lip(X, Rm ). Note that since
spt(T b f ) is compact, the push-forward is defined, moreover π# (T b f ) is an
element of Mm (Rm ) with compact support. In order to gain information
on the local structure of T , we would like to know that π# (T b f ) is of the
standard type [u] for some u ∈ L1 (Rm ). Except for m = 1 and m = 2, it is
an open problem whether every S ∈ Mm (Rm ) is of this form, cf. [3, p. 15 and
p. 21]. However, by Theorem 7.2, this is the case if S is normal, in particular
π# (T b f ) is of standard type if T is locally normal and f ∈ D(X). Thus, for
the purpose of this paper, the following result suffices.
Lemma 7.3. Let T ∈ Mm,loc (X), m ≥ 1. For π ∈ Lip(X, Rm ), each of the
following two statements implies the other:
(1) Whenever f ∈ D(X), then π# (T b f ) = [uf ] for some uf ∈ L1 (Rm ).

(2) Whenever B b X is a Borel set, then π# (T b B) = [uB ] for some uB ∈

L1 (Rm ).
Proof. Suppose first that (fk )k∈N is a sequence in Bc∞ (X) that converges in
L1 (kT k) to some f ∈ Bc∞ (X). Suppose further that π ∈ Lip1 (X, Rm ) and
π# (T b fk ) = [uk ] for some uk ∈ L1 (Rm ). Then [uk − ul ] = [uk ] − [ul ] =
π# (T b fk ) − π# (T b fl ) = π# (T b (fk − fl )), and
Z


|uk − ul | dx = M([uk − ul ]) = M π# (T b (fk − fl ))
Rm Z
≤ M(T b (fk − fl )) ≤ |fk − fl | dkT k
X

by (4.5), Lemma 4.6(2), and (4.10). Since fk → f in L1 (kT k), by the com-
pleteness of L1 (Rm ) there is a function u ∈ L1 (Rm ) such that uk → u
in L1 (Rm ). By passing to the M-limit on either side of the identity
π# (T b fk ) = [uk ], for k → ∞, we conclude that π# (T b f ) = [u]. Now, to
see that (1) implies (2), apply this procedure to the characteristic function
f of a Borel set B b X and an approximating sequence (fk )k∈N in D(X).
To show that (2) implies (1), choose (fk )k∈N such that each fk is a finite
linear combination of characteristic functions of compact sets, and such that
fk → f ∈ D(X) in L1 (kT k).
52 U. Lang

Proposition 7.4 (absolute continuity). Let T ∈ Mm,loc (X), m ≥ 1.

(1) If condition (1) or (2) of Lemma 7.3 holds for some π ∈ Lip(X, Rm ),
then
kT b (1, π)k(π −1 (N ) ∩ B) = 0
whenever N ⊂ Rm is a Borel set with L m (N ) = 0 and B ⊂ X is a Borel
set that is σ-finite with respect to kT k.

(2) If condition (1) or (2) of Lemma 7.3 holds for all π ∈ Lip(X, Rm ), then
kT k(A) = 0 for every set A ⊂ X with H m (A) = 0.

Together with Theorem 7.2, this shows that for every T ∈ Nm,loc (X), kT k
is absolutely continuous with respect to H m (cf. [3, Theorem 3.9]).

Proof. We assume that condition (1) of Lemma 7.3 holds for some π ∈
Lip(X, Rm ). Suppose N ⊂ Rm is a bounded Borel set with L m (N ) = 0.
Let f ∈ D(X), and choose σ ∈ D(X) with σ|spt(f ) = 1. By assumption,
π# (T b f ) = [uf ] for some uf ∈ L1 (Rm ). We abbreviate Tπ := T b (1, π); note
that Tπ ∈ M0,loc (X). Using Lemma 4.6(1) we obtain

(Tπ b π −1 (N ))(f ) = T ((χN ◦ π)f, π) = (T b f )(σ(χN ◦ π), π)

Z
= π# (T b f )(χN , id) = [uf ](χN , id) = uf χN dx = 0.
Rm

As this holds for all f ∈ D(X), we have Tπ b π −1 (N ) = 0. Let B ⊂ X be a

Borel set that is σ-finite with respect to kT k. By Lemma 4.7, kTπ k(π −1 (N ) ∩
B) = 0, and it follows that the same identity holds if N is unbounded.
To prove (2), let A ⊂ X be a Borel set with H m (A) = 0. Then A
is separable and hence σ-finite with respect to kT k. Let π ∈ Lip(X, Rm ).
Since L m (π(A)) = 0, there is a Borel set N ⊂ Rm with L m (N ) = 0 that
contains π(A). Then kTπ k(A) = kTπ k(π −1 (N ) ∩ A) = 0 by (1). Hence
T (f, π) = Tπ (f ) = 0 for all f ∈ Bc∞ (X) with {f 6= 0} ⊂ A. As this holds for
all π ∈ Lip(X, Rm ), we conclude from Lemma 4.7 that kT k(A) = 0.

We conclude this section with two results for locally normal currents that
relate the previous discussion to the slicing theory.

Theorem 7.5 (0-dimensional slices). Suppose T ∈ Nm,loc (X), m ≥ 1, and

π ∈ Liploc (X, Rm ). Whenever f ∈ Bc∞ (X), then π# (T b f ) = [uf ] for some
unique uf ∈ L1 (Rm ), moreover uf ∈ BV(Rm ) if f ∈ D(X). If spt(T ) is
separable, and if f ∈ Bc∞ (X), then

hT, π, yi(f ) = uf (y)

for L m -almost every y ∈ Rm .

Local currents in metric spaces 53

Proof. If f ∈ D(X), then π# (T b f ) ∈ Nm (Rm ), hence π# (T b f ) = [uf ] for

some uf ∈ BV(Rm ) by Theorem 7.2. The argument of the proof of Lemma 7.3
then shows that π# (T b f ) = [uf ] for some uf ∈ L1 (Rm ) in the general case,
when f ∈ Bc∞ (X). Clearly (the equivalence class) uf ∈ L1 (Rm ) is uniquely
determined.
Now suppose spt(T ) is separable, and f ∈ Bc∞ (X). Choose σ ∈ D(X)
with σ|spt(f ) = 1. For all φ ∈ D(Rm ), Theorem 6.4(2) (the case k = m) gives
Z
φ(y)hT, π, yi(f ) dy = (T b (1, π))((φ ◦ π)f )
Rm
= (T b f )(σ(φ ◦ π), π)
= π# (T b f )(φ, id)
= [uf ](φ, id)
Z
= φ(y)uf (y) dy.
Rm

We conclude that hT, π, yi(f ) = uf (y) for L m -almost every y ∈ Rm .

Theorem 7.6 (partial rectifiability). Suppose T ∈ Nm,loc (X), m ≥ 1, spt(T )

is separable, and π ∈ Liploc (X, Rm ). Put

A := x ∈ X : hT, π, π(x)i ∈ M0,loc (X) and khT, π, π(x)ik({x}) > 0 .

Then there exists a countable

S family

of pairwise disjoint compact sets Bi ⊂ A
such that kT b (1, π)k A \ i Bi = 0 and π|Bi is bi-Lipschitz for every i.
Compare [3, Theorem 7.4].

Proof. We assume without loss of generality that X is compact, so that T ∈

Nm (X) and π ∈ Lip(X, Rm ). By Theorem 7.5, for every f ∈ D(X) there is a
function uf ∈ BV(Rm ) such that π# (T b f ) = [uf ]. Hence, using Theorem 7.2,
Lemma 4.7 and Lemma 4.6(2), we obtain

|Duf |(B) = k∂[uf ]k(B) = k∂(π# (T b f ))k(B) = kπ# (∂(T b f ))k(B)

≤ Lip(π)m−1 k∂(T b f )k(π −1 (B))

for every Borel set B ⊂ Rm . Suppose that |f | ≤ 1 and Lip(f ) ≤ 1. Since

∂(T b f ) = (∂T )b f − T b (1, f ), cf. (3.6), it follows that

|Duf |(B) ≤ Lip(π)m−1 (kT k + k∂T k)(π −1 (B)) =: µ(B)

for all Borel sets B ⊂ Rm . Note that the finite Borel measure µ so defined
is independent of f . Now we choose a countable subset F of {f ∈ D(X) :
0 ≤ f ≤ 1, Lip(f ) ≤ 1} with the following property: Whenever x ∈ X and
0 < r ≤ 1, there is an f ∈ F such that
3
f (x) ≥ r, f ≤ r, spt(f ) ⊂ U(x, r). (7.2)
4
54 U. Lang

By Theorem 6.4, Theorem 7.5, and Lemma 7.1 there exists a Borel set N ⊂
Rm with L m (N ) = 0 such that whenever y, y 0 ∈ Rm \ N and f ∈ F , the slice
Ty := hT, π, yi exists as an element of M0 (X), Ty (f ) = uf (y), Mµ (y) < ∞,
and

|Ty (f ) − Ty0 (f )| = |uf (y) − uf (y 0 )| ≤ cm Mµ (y) + Mµ (y 0 ) |y − y 0 |.

By Proposition 7.4(1),

kT b (1, π)k(π −1 (N )) = 0.

Let , δ ∈ (0, 1] ∩ Q, and recall that kTy k is concentrated on π −1 {y}. Denote

by E,δ the Borel set of all x ∈ X such that y := π(x) 6∈ N and
1 
Mµ (y) ≤ , kTy k({x}) ≥ , kTy k(U(x, 2δ) \ {x}) ≤ .
2 4
We have A \ π −1 (N ) = ,δ E,δ . Let x, x0 ∈ E,δ with 0 < r := d(x, x0 ) ≤ δ
S
and put y := π(x), y 0 = π(x0 ). Then choose f ∈ F as in (7.2). Since
U(x, r) ⊂ U(x0 , 2r) \ {x0 }, it follows that
Z

|Ty (f )| ≤
0 f dkTy0 k ≤ r · kTy0 k(U(x0 , 2δ) \ {x0 }) ≤ r.
X 4
Similarly, using (4.12), we obtain
3  
|Ty (f )| ≥ |Ty (χ{x} f )| − |Ty (χX\{x} f )| ≥ r ·  − r · = r.
4 4 2
We conclude that
 
d(x, x0 ) = r ≤ |Ty (f )| − |Ty0 (f )| ≤ |Ty (f ) − Ty0 (f )|
4 4
cm cm
≤ cm Mµ (y) + Mµ (y 0 ) |y − y 0 | ≤ |y − y 0 | = |π(x) − π(x0 )|.


 
This shows that whenever we restrict π to a subset of E,δ with diameter
at most δ, the resulting map is a bi-Lipschitz map into Rm \ N . The result
follows.

8 Integer rectifiable currents

We now turn to integer multiplicity rectifiable currents.
We recall that a subset E of a metric space X is countably m-rectifiable
a countable family of Lipschitz maps Fi : Ai → X, Ai ⊂ Rm , such
if there is S
that E ⊂ i Fi (Ai ). The set E ⊂ X is countably H m -rectifiable if there is a
countably m-rectifiable set E 0 ⊂ X with H m (E \ E 0 ) = 0 (cf. [14, 3.2.14]).
Now we let again X denote a locally compact metric space. Since X is
locally complete, it is not difficult to see that every countably m-rectifiable set
E ⊂ X is contained in a countably m-rectifiable and σ-compact set Ē ⊂ X.
Local currents in metric spaces 55

Definition 8.1 (integer rectifiable current). Let T ∈ Dm (X), m ≥ 0. We

call T a locally integer rectifiable current if

(1) T ∈ Mm,loc (X),

(2) whenever B b X is a Borel set and π ∈ Lip(X, Rm ), then π# (T b B) = [u]

for some u = uB,π ∈ L1 (Rm , Z), and

(3) kT k is concentrated on some countably H m -rectifiable Borel set E ⊂ X.

The set of all m-dimensional locally integer rectifiable currents in X is de-

noted by
Im,loc (X).
An m-dimensional integer rectifiable current in X is an element of

Im (X) := Im,loc (X) ∩ Mm (X).

Note that if a current T ∈ Mm,loc (X) satisfies condition (2), then kT k

is absolutely continuous with respect to H m by Proposition 7.4. Moreover,
condition (3) implies that the support of T is separable. Clearly Im,loc (X)
forms an abelian group, as does Im (X). Push-forwards of locally integer
rectifiable currents under locally Lipschitz maps are again locally integer rec-
tifiable. If T ∈ Im,loc (X), then T b B ∈ Im,loc (X) for every Borel set B ⊂ X.
In case m = 0, condition (2) for T ∈ M0,loc (X) means that T (χB ) ∈ Z
for every Borel set B b X (cf. the remark following Proposition 2.6). If
T ∈ I0,loc (X), then kT k is concentrated on some countable set E ⊂ X
consisting of atoms of kT k, i.e., kT k({x}) > 0 for x ∈ E. By (4.12) and
condition (2), kT k({x}) = |T (χ{x} )| ∈ Z for x ∈ E, so E is discrete, and
X
T (f ) = ax f (x) (8.1)
x∈E

for all f ∈ Bc∞ (X), where ax = T (χ{x} ) ∈ Z \ {0}.

Lemma 8.2 (characterizing I0,loc ). Suppose S ∈ M0,loc (X), spt(S) is sepa-

rable, and S(χK ) ∈ Z for every compact set K ⊂ X. Then S ∈ I0,loc (X).

Proof. Let Σ be the set of all x ∈ X such that kSk({x}) ≥ 1. Since S

has locally finite mass, Σ is discrete. Now let x ∈ X \ Σ. There is an open
neighborhood V of x such that kSk(V ) < 1. Then |S(χK )| ≤ kSk(V ) < 1 and
thus S(χK ) = 0 for every compact set K ⊂ V . By approximation, this implies
that S(f ) = 0 for every f ∈ D(X) with spt(f ) ⊂ V , hence x 6∈ spt(S). This
shows that spt(S) ⊂ Σ. Since spt(S) is separable by assumption, it follows
that S ∈ I0,loc (X).

The next theorem corresponds to [3, Theorem 4.5] in the case of finite
mass. A similar characterization of classical rectifiable currents is given in [14,
Theorem 4.1.28].
56 U. Lang

Theorem 8.3 (parametric representation). Let T ∈ Mm,loc (X), m ≥ 1.

Then T ∈ Im,loc (X) if and only if there exists a countable family of currents
T i , each of the form T i = Fi# [ui ] for some function ui ∈ L1 (Rm , Z) and some
bi-Lipschitz map Fi : Ki → X defined on a compact set Ki ⊂ Rm containing
spt(ui ), such that
X X
kT k(A) = kT i k(A), T (f, π) = T i (f, π)
i i

for all Borel sets A ⊂ X and for all (f, π) ∈ D m (X).

The proof of the ‘only if’ part uses the following fact: If E ⊂ X is an H m -
measurable and countably H m -rectifiable set, then there exists a countable
family of bi-Lipschitz maps Fi : Ki → E, with Ki ⊂ Rm compact, such that
the Fi (Ki ) are pairwise disjoint and

H m E \ i Fi (Ki ) = 0.
S

(8.2)

For X = Rn , this result is contained in [14, Lemma 3.2.18]. For a general

complete metric space X, it is stated in [3, Lemma 4.1]; the argument relies
on the metric differentiability theorem of [23] and generalizes to every locally
complete metric space, in particular to our locally compact metric space X.
We omit the details since the ‘only if’ part of the theorem will not be used
in the sequel. In fact, for locally integer rectifiable currents whose bound-
ary has locally finite mass, the existence of such parametric representations
also follows from Theorem 7.6, cf. the proof of Theorem 8.4 together with
Theorem 8.5(1) (the case k = m).

Proof of Theorem 8.3. Suppose T ∈ Im,loc (X), m ≥ 1. Then kT k is con-

centrated on some countably H m -rectifiable Borel set E ⊂ X. Choose a
countable family of bi-Lipschitz maps Fi : Ki → E, with Ki ⊂ Rm S compact,
such that the sets B i := Fi (Ki ) are pairwise disjoint and H m E \ i B i = 0,

cf. (8.2). Since kT k is absolutely continuous with respect to H m by Propo-

sition 7.4,
kT k X \ i B i = kT k E \ i B i = 0.
S
S

Let π i : X → Rm be a Lipschitz extension of Fi−1 : B i → Rm , and put T i :=

T b B i . Since T ∈ Im,loc (X), we have π#i T i = [u ] for some u ∈ L1 (Rm , Z)
i i
i
with spt(ui ) ⊂ Ki . It follows that T = Fi# (π# i T i ) = F [u ]. For every
i# i
Borel set A ⊂ X,
S
X X
kT k(A) = kT k A ∩ i B i = kT k(A ∩ B i ) = kT i k(A).
i i

S∈D
Let (f, π) m (X), and denote by χ and χ the characteristic functions of
i
i i
B and i B , respectively. Using Theorem 4.4(2) we get
X X
T (f, π) = T (χf, π) = T (χi f, π) = T i (f, π).
i i
Local currents in metric spaces 57

Conversely, suppose that T ∈ Mm,loc (X) admits a representation as in the

theorem. Since kT i k(X S \ Fi (Ki )) = 0, kT k is concentrated on the countably
m-rectifiable Borel set i Fi (Ki ). From Lemma 3.7 it follows that [u Pi ] ∈
Im (Rm ), hence T i = Fi# [ui ] ∈ Im (X). By forming partial sums of i T i
we find a sequence (S k )k∈N in Im (X) such that for every Borel set B b X,
M(T b B −S k b B) → 0 for k → ∞. For π ∈ Lip(X, Rm ), we have π# (S k b B) =
[uk ], uk ∈ L1 (Rm , Z). A similar argument as in the proof of Lemma 7.3 then
shows that π# (T b B) = [u] for some u ∈ L1 (Rm , Z). Thus T ∈ Im,loc (X).

We now supplement the slicing theory for locally normal currents T with
the following two theorems, providing criteria for the rectifiability of T in
terms of the rectifiability of slices, and vice-versa. We refer to [38], [20] for
similar statements in the context of classical currents, and to [3, Theorem 8.1]
for a corresponding result for normal metric currents.

Theorem 8.4 (rectifiability criterion). Suppose T ∈ Nm,loc (X), m ≥ 1,

spt(T ) is separable, and P is a countable subset of Lip1 (X) such that for
every π0 ∈ Lip1 (X) and every compact set K ⊂ spt(T ) there is a sequence
in P converging uniformly on K to π0 . If for each π = (π1 , . . . , πm ) ∈ P m ,
hT, π, yi ∈ I0,loc (X) for L m -almost every y ∈ Rm , then T ∈ Im,loc (X).

Proof. Fix π ∈ P m for the moment. Put


Aπ := x ∈ X : hT, π, π(x)i ∈ M0,loc (X) and khT, π, π(x)ik({x}) > 0 ,

and let Eπ ⊂ Aπ be the respective σ-compact set provided by Theorem 7.6, so

that kT b (1, π)k(Aπ \ Eπ ) = 0. By assumption, for L m -almost every y ∈ Rm ,
khT, π, yik is concentrated on some countable set contained in Aπ . Hence, by
Theorem 6.4(3),
Z
kT b (1, π)k(X \ Aπ ) = khT, π, yik(X \ Aπ ) dy = 0.
Rm

Thus kT b (1, π)k(X \ Eπ ) = 0.

Now let E := π∈P m Eπ . Suppose (f, π) ∈ Bc∞ (X) × P m , and {f 6=
S
0} ⊂ spt(T ) \ E. Since {f 6= 0} ⊂ X \ Eπ and kT b (1, π)k(X \ Eπ ) = 0,
we have T (f, π) = (T b (1, π))(f ) = 0. By the choice of the family P, it
follows that T (f, π) = 0 for all (f, π) ∈ Bc∞ (X) × [Lip1 (X)]m with {f 6= 0} ⊂
spt(T ) \ E. Since kT k is σ-finite, Lemma 4.7 yields kT k(X \ E) = 0. So kT k
is concentrated on the countably m-rectifiable Borel set E.
By the respective property of the sets Eπ , we find aScountable family of
k k


pairwise disjoint compact sets B ⊂ E with kT k E \ k B = 0 and the
property that for every k, there is a π k ∈ P m such that π k |B k is bi-Lipschitz.
By Theorem 7.5, π# k (T b B k ) = [u ] for some u ∈ L1 (Rm ), and
k k

uk (y) = hT, π k , yi(χB k )

58 U. Lang

for L m -almost every y. By assumption, hT, π k , yi(χB k ) ∈ Z for L m -almost

every y, hence uk ∈ L1 (Rm , Z). Now it follows that T admits a parametric
representation as in Theorem 8.3. Hence T ∈ Im,loc (X).

Theorem 8.5 (rectifiable slices). Suppose T ∈ Nm,loc (X), 1 ≤ k ≤ m, and

spt(T ) is separable.

(1) If T satisfies condition (2) of Definition 8.1, in particular if T ∈

Im,loc (X), and if π ∈ Liploc (X, Rk ), then hT, π, yi ∈ Im−k,loc (X) for
L k -almost every y ∈ Rk .

(2) Conversely, if for each π ∈ Lip(X, Rk ), hT, π, yi ∈ Im−k,loc (X) for L k -

almost every y ∈ Rk , then T ∈ Im,loc (X).

Proof. For the proof of (1), we first consider the case k = m, so that
π ∈ Liploc (X, Rm ). Choose a countable family U of open sets U b X such
that for every x ∈ spt(T ) and  > 0 there is a U ∈ U with x ∈ U ⊂ U(x, ),
and such that the union of finitely many elements of U is in U . By as-
sumption, for every U ∈ U there exists a function uU ∈ L1 (Rm , Z) such that
π# (T b U ) = [uU ]. Thus, for L m -almost every y, hT, π, yi ∈ M0,loc (X), and,
by Theorem 7.5,
hT, π, yi(χU ) = uU (y) ∈ Z
for all U ∈ U . It follows that for every such y, hT, π, yi(χK ) ∈ Z for all
compact sets K ⊂ X. Lemma 8.2 then shows that hT, π, yi ∈ I0,loc (X).
Now suppose 1 ≤ k < m, and let π ∈ Liploc (X, Rk ). Choose a countable
set P ⊂ Lip1 (X) as in Theorem 8.4. Fix ρ ∈ P m−k for the moment. We
know from Theorem 6.5 (iterated slices) and the result in the case k = m
that
hT, π, yi, ρ, z = hT, (π, ρ), (y, z)i ∈ I0,loc (X)
for L m -almost every (y, z) ∈ Rk × Rm−k . Hence, there is a set Nρ ⊂ Rk with
L k (Nρ ) = 0 such that for every y ∈ Rk \ Nρ , hT, π, yi ∈ Nm−k,loc (X) and
hT, π, yi, ρ, z ∈ I0,loc (X) for all z ∈ Rm−k \ Ny,ρ
0 , where L m−k (N 0 ) = 0.
y,ρ
Suppose y ∈ Rk \ ρ∈P m−k Nρ , and put Ty := hT, π, yi. For each ρ ∈ P m−k ,
S

hTy , ρ, zi ∈ I0,loc (X) for all z ∈ Rm−k \ Ny,ρ

0 . Hence T ∈ I
y m−k,loc (X) by
Theorem 8.4. Since P is countable, this proves (1).
We show (2). In case k = m, the result holds by Theorem 8.4.
Now suppose 1 ≤ k < m. Let π ∈ Lip(X, Rk ) and ρ ∈ Lip(X, Rm−k ).
Let again U be given as above. It follows from Theorem 6.4 that the set
M of all (y, z) ∈ Rk × Rm−k such that hT, (π, ρ), (y, z)i ∈ M0,loc (X) and
hT, (π, ρ), (y, z)i(χU ) ∈ Z for all U ∈ U is L m -measurable. By assumption
(and Theorem 6.4), hT, π, yi ∈ Im−k,loc (X) ∩ Nm−k,loc (X) for L k -almost ev-
ery y ∈ Rk . Hence, for L k -almost every y ∈ Rk , it follows from Theorem 6.5
and the first part of the proof of (1) that for L m−k -almost every z ∈ Rm−k ,

hT, (π, ρ), (y, z)i = hT, π, yi, ρ, z ∈ I0,loc (X),

Local currents in metric spaces 59

thus L m−k ({z ∈ Rm−k : (y, z) 6∈ M }) = 0. We conclude that L m (Rm \M ) =

0, and, by virtue of Lemma 8.2, that hT, (π, ρ), (y, z)i ∈ I0,loc (X) for all
(y, z) ∈ M . As π and ρ were arbitrary, we have T ∈ Im,loc (X) by the result
in the case k = m.

Finally, we consider the chain complex of integral currents.

Definition 8.6 (integral current). The abelian group of m-dimensional lo-

cally integral currents in X is defined by

Im,loc (X) := {T ∈ Im,loc (X) : ∂T ∈ Im−1,loc (X)}

if m ≥ 1, and I0,loc (X) := I0,loc (X). An m-dimensional integral current in

X is an element of

Im (X) := Im,loc (X) ∩ Nm (X).

The following important result yields a simpler characterization of

Im,loc (X), cf. [29, Theorem 30.3] and [3, Theorem 8.6].

Theorem 8.7 (boundary rectifiability). For all m ≥ 0, Im,loc (X) ∩

Nm,loc (X) = Im,loc (X).

In other words, if T ∈ Im,loc (X), m ≥ 1, and if ∂T has locally finite mass,

then ∂T ∈ Im−1,loc (X).

Proof. In case m = 0 there is nothing to prove.

Now consider the case m = 1. Let T ∈ I1,loc (X) ∩ N1,loc (X), and let
U b X be an open set. If U = X, then ∂T (χU ) = ∂T (1) = T (1, 1) = 0. If
U 6= X, put π(x) := inf{d(x, z) : z ∈ X \ U } for x ∈ X. Then for L 1 -almost
every s > 0,

∂T (χ{π>s} ) = ((∂T )b {π > s})(χU )

= ∂(T b {π > s})(χU ) + hT, π, si(χU )
= hT, π, si(χU ) ∈ Z

by Theorem 8.5(1). By continuity, letting s tend to 0 we conclude that

∂T (χU ) is an integer. As this holds for every open set U b X, it follows that
∂T (χK ) is an integer for every compact set K ⊂ X, and Lemma 8.2 yields
∂T ∈ I0,loc (X).
Now let m ≥ 1 and suppose the result holds in dimension m. Let
T ∈ Im+1,loc (X) ∩ Nm+1,loc (X) and π ∈ Lip(X). For almost every y ∈ R,
hT, π, yi ∈ Im,loc (X) ∩ Nm,loc (X) by Theorem 8.5(1), hence

h∂T, π, yi = −∂hT, π, yi ∈ Im−1,loc (X)

by (6.3) and the induction hypothesis. As this holds for every π ∈ Lip(X),
Theorem 8.5(2) implies that ∂T ∈ Im,loc (X).
60 U. Lang

As a corollary, we obtain another characterization of locally integral cur-

rents (cf. [3, Theorem 8.8(ii)]).
Theorem 8.8. Let T ∈ Nm,loc (X), m ≥ 0. Then T ∈ Im,loc (X) if and only
if spt(T ) is separable and T satisfies condition (2) of Definition 8.1.
Proof. One implication is clear from Definition 8.1. The other follows from
Lemma 8.2 if m = 0 and from Theorem 8.5 (e.g. the case k = m) and
Theorem 8.7 if m ≥ 1.
A similar induction argument as in the proof of Theorem 8.7 also shows
the next result, cf. [29, Theorem 32.2], [36, Sect. 3], and [3, Theorem 8.5].
Theorem 8.9 (closure theorem). If (Tn )n∈N is a sequence in Im,loc (X) con-
verging weakly to T ∈ Dm (X), m ≥ 0, and if supn NV (Tn ) < ∞ for every
open set V b X, then T ∈ Im,loc (X).
P
Proof. Suppose first that m = 0. Each Tn is a locally finite sum i an,i [xn,i ]
with coefficients an,i ∈ Z \ {0}, where [xn,i ](f ) = f (xn,i ) for all f ∈ D(X),
cf. (8.1). For every open set V b X, spt(Tn ) ∩ V contains at most MV (Tn )
points. Since Tn → T , it follows that spt(T ) ∩ V consists P of at most
supn MV (Tn ) points. Hence, T is a locally finite sum j aj [xj ] with co-
efficients aj ∈ R \ {0}. To see that aj ∈ Z, choose f ∈ D(X) so that
0 ≤f ≤ 1, f (xj ) = 1, f (x) = 0 for all x ∈ spt(T ) \ {xj }, and such that the
set n : {0P < f < 1} ∩ spt(Tn ) 6= ∅ is finite. Then, for sufficiently large n,
Tn (f ) = i an,i f (xn,i ) ∈ Z, thus aj = T (f ) = limn→∞ Tn (f ) ∈ Z.
Now let m ≥ 1 and suppose the result holds in dimension m − 1. Let
π ∈ Lip(X). For almost every y ∈ R, we have hTn , π, yi ∈ Im−1,loc (X)
for all n by Theorem 8.5(1) and Theorem 8.7. Note that T ∈ Nm,loc (X)
by the lower semicontinuity of mass. By Proposition 6.6 (slicing convergent
sequences), for almost every y ∈ R, hT, π, yi is the weak limit of some locally
N-bounded subsequence of (hTn , π, yi)n∈N . Hence hT, π, yi ∈ Im−1,loc (X) by
the induction hypothesis. As this holds for every π ∈ Lip(X), Theorem 8.5(2)
and Theorem 8.7 imply that T ∈ Im,loc (X).
The compactness theorem for locally integral currents is now an immedi-
ate consequence. See [29, Theorem 27.3] for the classical result.
Theorem 8.10 (Im,loc compactness). Suppose (Tn )n∈N is a sequence in
Im,loc (X) such that supn NV (Tn ) < ∞ for every open set V b X. Then there
is a subsequence (Tn(i) )i∈N that converges weakly to some T ∈ Im,loc (X).
Note that X stillSdenotes an arbitrary locally compact metric space. How-
ever, the closure of n∈N spt(Tn ), which also contains the support of the limit
T , is separable.
Proof. Combine Theorem 5.4 (Nm,loc compactness) and Theorem 8.9.
In [24] applications of these results to the asymptotic geometry of spaces
of nonpositive curvature will be discussed.
Local currents in metric spaces 61

References
[1] Adams, T., Flat chains in Banach spaces. J. Geometric Anal. 18 (2008),
no. 1, 1–28.

[2] Ambrosio, L., Fusco, N., Pallara, D., Functions of bounded variation
and free discontinuity problems. Oxford Mathematical Monographs. The
Clarendon Press, Oxford University Press, New York, 2000.

[3] Ambrosio, L., Kirchheim, B., Currents in metric spaces. Acta Math. 185
(2000), no. 1, 1–80.

[4] Cheeger, J., Differentiability of Lipschitz functions on metric measure

spaces. Geom. Funct. Anal. 9 (1999), no. 3, 428–517.

[5] Cheeger, J., Kleiner, B., On the differentiability of Lipschitz maps from
metric measure spaces to Banach spaces. Inspired by S. S. Chern, 129–
152, Nankai Tracts Math., 11. World Sci. Publ., Hackensack, NJ, 2006.

[6] Courant, R., Hilbert, D., Methoden der mathematischen Physik, Band
I. Zweite verbesserte Auflage. Berlin, Springer, 1931.

[7] De Giorgi, E., Problema di Plateau generale e funzionali geodetici. Atti

Sem. Mat. Fis. Univ. Modena 43 (1995), 285–292.

[8] De Lellis, C., Some fine properties of currents and applications to distri-
butional Jacobians. Proc. Roy. Soc. Edinburgh Sect. A 132 (2002), no.
4, 815–842.

[9] De Lellis, C., Some remarks on the distributional Jacobian. Nonlinear

Anal. 53 (2003), no. 7-8, 1101–1114.

[10] De Pauw, Th., Hardt, R., Size minimization and approximating prob-
lems. Calc. Var. Partial Differential Equations 17 (2003), no. 4, 405–442.

[11] De Pauw, Th., Hardt, R., Application of scans and fractional power
integrands. Variational problems in Riemannian geometry, 19–31, Progr.
Nonlinear Differential Equations Appl., 59. Birkhäuser, Basel, 2004.

[12] de Rham, G., Variétés différentiables. Formes, courants, formes har-

moniques. Actualités scientifiques et industrielles, 1222. Hermann, Paris,
1955.

[13] Evans, L. C., Gariepy, R. F., Measure theory and fine properties of
functions. Studies in Advanced Mathematics. CRC Press, Boca Raton,
FL, 1992.

[14] Federer, H., Geometric measure theory. Die Grundlehren der mathema-
tischen Wissenschaften, 153. Springer, New York, 1969.
62 U. Lang

[15] Federer, H., Fleming, W. H., Normal and integral currents. Ann. of
Math. (2) 72 (1960), 458–520.

[16] Hardt, R., Rivière, T., Connecting topological Hopf singularities. Ann.
Sc. Norm. Super. Pisa Cl. Sci. (5) 2 (2003), no. 2, 287–344.

[17] Heinonen, J., Lectures on analysis on metric spaces. Universitext.

Springer, New York, 2001.

[18] Heinonen, J., Lectures on Lipschitz analysis. Report. University of

Jyväskylä Department of Mathematics and Statistics, 100. University
of Jyväskylä, Jyväskylä, 2005.

[19] Heinonen, J., Nonsmooth calculus. Bull. Amer. Math. Soc. (N.S.) 44
(2007), no. 2, 163–232.

[20] Jerrard, R. L., A new proof of the rectifiable slices theorem. Ann. Sc.
Norm. Super. Pisa Cl. Sci. (5) 1 (2002), no. 4, 905–924.

[21] Jerrard, R. L., Soner, H. M., Functions of bounded higher variation.

Indiana Univ. Math. J. 51 (2002), no. 3, 645–677.

[22] Keith, S., A differentiable structure for metric measure spaces. Adv.
Math. 183 (2004), no. 2, 271–315.

[23] Kirchheim, B., Rectifiable metric spaces: local structure and regularity
of the Hausdorff measure. Proc. Amer. Math. Soc. 121 (1994), no. 1,
113–123.

[24] Kleiner, B., Lang, U., Quasi-minimizing varieties in spaces of nonpositive

curvature. In preparation.

[25] Mucci, D., Fractures and vector valued maps. Calc. Var. Partial Differ-
ential Equations 22 (2005), no. 4, 391–420.

[26] Rudin, W., Real and complex analysis. Third edition. McGraw-Hill, New
York, 1987.

[27] Rudin, W., Functional analysis. Second edition. International Series in

Pure and Applied Mathematics. McGraw-Hill, New York, 1991.

[28] Sandier, E., Serfaty, S., A product-estimate for Ginzburg-Landau and

corollaries. J. Funct. Anal. 211 (2004), no. 1, 219–244.

[29] Simon, L., Lectures on geometric measure theory. Proceedings of the

Centre for Mathematical Analysis, Australian National University, 3.
Australian National University, Centre for Mathematical Analysis, Can-
berra, 1983.
Local currents in metric spaces 63

[30] Weaver, N., Lipschitz algebras. World Scientific Publishing, River Edge,
NJ, 1999.

[31] Weaver, N., Lipschitz algebras and derivations. II. Exterior differentia-
tion. J. Funct. Anal. 178 (2000), no. 1, 64–112. Erratum, J. Funct. Anal.
186 (2001), no. 2, 546.

[32] Wenger, S., Isoperimetric inequalities of Euclidean type in metric spaces.

Geom. Funct. Anal. 15 (2005), no. 2, 534–554.

[33] Wenger, S., Filling invariants at infinity and the Euclidean rank of
Hadamard spaces. Int. Math. Res. Not. 2006, Art. ID 83090, 33 pp.

[34] Wenger, S., Flat convergence for integral currents in metric spaces. Calc.
Var. Partial Differential Equations 28 (2007), no. 2, 139–160.

[35] Wenger, S., Gromov hyperbolic spaces and the sharp isoperimetric con-
stant. Invent. Math. 171 (2008), no. 1, 227–255.

[36] White, B., A new proof of the compactness theorem for integral currents.
Comment. Math. Helv. 64 (1989), no. 2, 207–220.

[37] White, B., The deformation theorem for flat chains. Acta Math. 183
(1999), no. 2, 255–271.

[38] White, B., Rectifiability of flat chains. Ann. of Math. (2) 150 (1999), no.
1, 165–184.

Urs Lang
[email protected]
Department of Mathematics
ETH Zurich
8092 Zurich
Switzerland