Beruflich Dokumente
Kultur Dokumente
com
In the last twenty years, sub-Riemannian geometry has emerged as an independent Volume I
research domain, with extremely rich motivations and ramifications in several parts of
pure and applied mathematics, such as geometric analysis, geometric measure theory,
stochastic calculus and evolution equations together with applications in mechanics,
Davide Barilari
The aim of the lectures collected here is to present sub-Riemannian structures for the Ugo Boscain
use of both researchers and graduate students.
Mario Sigalotti
Editors
ISBN 978-3-03719-162-0
www.ems-ph.org
www.Ebook777.com
Barilari et al. Vol. I | Rotis Sans | Pantone 287, Pantone 116 | 170 x 240 mm | RB: 16 ?? mm
Free ebooks ==> www.Ebook777.com
www.Ebook777.com
EMS Series of Lectures in Mathematics
Edited by Ari Laptev (Imperial College, London, UK)
Davide Barilari
Ugo Boscain
Mario Sigalotti
Editors
Free ebooks ==> www.Ebook777.com
Editors:
2010 Mathematics Subject Classification: Primary: 53C17; Secondary: 35H10, 60H30, 49J15
Key words: sub-Riemannian geometry, hypoelliptic operators, non-holonomic constraints, optimal control,
rough paths
ISBN 978-3-03719-162-0
The Swiss National Library lists this publication in The Swiss Book, the Swiss national bibliography, and the
detailed bibliographic data are available on the Internet at http://www.helveticat.ch.
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting,
reproduction on microfilms or in other ways, and storage in data banks. For any kind of use permission
of the copyright owner must be obtained.
Contact address:
European Mathematical Society Publishing House
Seminar for Applied Mathematics
ETH-Zentrum SEW A27
CH-8092 Zürich
Switzerland
www.Ebook777.com
Preface
This book, divided into two volumes, collects different cycles of lectures given at
the IHP Trimester “Geometry, Analysis and Dynamics on Sub-Riemannian Mani-
folds”, held at Institut Henri Poincaré in Paris, and the CIRM Summer School “Sub-
Riemannian Manifolds: From Geodesics to Hypoelliptic Diffusion”, held at Centre
Internationale de Rencontres Mathématiques, in Luminy, during fall 2014.
Sub-Riemannian geometry is a generalization of Riemannian geometry, whose
birth dates back to Carathéodory’s 1909 seminal paper on the foundations of Carnot
thermodynamics, followed by E. Cartan’s 1928 address at the International Congress
of Mathematicians in Bologna.
Sub-Riemannian geometry is characterized by non-holonomic constraints: dis-
tances are computed by minimizing the length of curves whose velocities belong
to a given subspace of the tangent space. From the theoretical point of view, sub-
Riemannian geometry is the geometry underlying the theory of hypoelliptic opera-
tors and degenerate diffusions on manifolds.
In the last twenty years, sub-Riemannian geometry has emerged as an indepen-
dent research domain, with extremely rich motivations and ramifications in several
parts of pure and applied mathematics. Let us mention geometric analysis, geomet-
ric measure theory, stochastic calculus and evolution equations together with appli-
cations in mechanics and optimal control (motion planning, robotics, nonholonomic
mechanics, quantum control) and another to image processing, biology and vision.
Even if, nowadays, sub-Riemannian geometry is recognized as a transverse sub-
ject, researchers working in different communities are still using quite different lan-
guage. The aim of these lectures is to collect reference material on sub-Riemannian
structures for the use of both researchers and graduate students. Starting from basic
definitions and extending up to the frontiers of research, this material reflects the
point of view of authors with different backgrounds. The exchanges among the par-
ticipants of the IHP Trimester and of the CIRM school are reflected here by several
connections and interplays between the different chapters. This will hopefully re-
duce the existing gap in language between the different communities and favour the
future development of the field.
The notes of Francesco Serra Cassano give an extensive presentation of geomet-
ric measure theory in Carnot groups. The first part of the notes discusses differen-
tial calculus for maps between Carnot groups in relation with the underlying metric
structure. The text then focuses on differential calculus within Carnot groups and
uses it to investigate intrinsic regular and Lipschitz surfaces in Carnot groups and
their relation with rectifiability. The final section deals with sets of finite perimeter
and with the related notions of reduced and minimal boundary. An application to
minimal graphs in Heisenberg groups is developed.
vi Preface
Davide Barilari
Ugo Boscain
Mario Sigalotti
Contents
1 Some topics of geometric measure theory in Carnot groups . . . . . . . . . 1
Francesco Serra Cassano
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 An introduction to Carnot groups . . . . . . . . . . . . . . . . . . . 3
3 Differential calculus on Carnot groups . . . . . . . . . . . . . . . . 20
4 Differential calculus within Carnot groups . . . . . . . . . . . . . . 34
5 Sets of finite perimeter and minimal surfaces in Carnot groups . . . 83
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
2 Hypoelliptic operators and some aspects of analysis and geometry of sub-
Riemannian spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Nicola Garofalo
1 Sub-Riemannian geometry and hypoelliptic operators . . . . . . . . 123
2 Carnot groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3 Fundamental solutions and the Yamabe equation . . . . . . . . . . . 147
4 Carnot–Carathéodory distance . . . . . . . . . . . . . . . . . . . . 160
5 Sobolev and BV spaces . . . . . . . . . . . . . . . . . . . . . . . . 175
6 Fractional integration in spaces of homogeneous type . . . . . . . . 189
7 Fundamental solutions of hypoelliptic operators . . . . . . . . . . . 202
8 The geometric Sobolev embedding and the isoperimetric inequality . 212
9 The Li–Yau inequality for complete manifolds with Ricci ≥ 0 . . . . 216
10 Heat semigroup approach to the Li–Yau inequality . . . . . . . . . 224
11 A heat equation approach to the volume doubling property . . . . . 233
12 A sub-Riemannian curvature-dimension inequality . . . . . . . . . 239
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian
foliations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Fabrice Baudoin
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
2 Riemannian foliations and their Laplacians . . . . . . . . . . . . . 261
3 Horizontal Laplacians and heat kernels on model spaces . . . . . . 268
4 Transverse Weitzenböck formulas . . . . . . . . . . . . . . . . . . 282
5 The horizontal heat semigroup . . . . . . . . . . . . . . . . . . . . 292
6 The horizontal Bonnet–Myers theorem . . . . . . . . . . . . . . . . 302
7 Riemannian foliations and hypocoercivity . . . . . . . . . . . . . . 308
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
Free ebooks ==> www.Ebook777.com
Chapter 1
Some topics of geometric measure theory in
Carnot groups
To my parents
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 An introduction to Carnot groups . . . . . . . . . . . . . . . . . . . . . . . . . 3
3 Differential calculus on Carnot groups . . . . . . . . . . . . . . . . . . . . . . 20
4 Differential calculus within Carnot groups . . . . . . . . . . . . . . . . . . . . 34
5 Sets of finite perimeter and minimal surfaces in Carnot groups . . . . . . . . . . 83
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
1 Introduction
These notes aim at illustrating some results achieved in geometric measure theory
in Carnot groups. They are an extended version of part of the course Geomet-
ric Measure Theory given during the “Geometry, Analysis and Dynamics on sub-
Riemannian Manifolds” trimester, held in Paris in September 2014 at the Institut
Henri Poincaré.
First of all, I would like to thank the organizers of the trimester Andrei Agrachev,
Davide Barilari, Ugo Boscain, Yacine Chitour, Frederic Jean, Ludovic Rifford, and
Mario Sigalotti, for their kind invitation, as well to IHP for its backing.
It is also a great pleasure for me to acknowledge the help and support of several
friends of mine who have made this work possible: first of all, most of the results
1francesco.serracassano@unitn.it
Dipartimento di Matematica, Università di Trento, Via Sommarive 14, 38123, Trento, Italy.
F.S.C. is supported by MIUR, Italy, GNAMPA of the INdAM, University of Trento, Italy and by MAnET Marie
Curie Initial Training Network Grant 607643–FP7-PEOPLE-2013-ITN. Part of the work was done while F.S.C.
was visiting at the Institut Henri Poincaré, Paris. He wishes to thank the IHP for its hospitality.
www.Ebook777.com
2 Francesco Serra Cassano
presented here have been achieved jointly with Bruno Franchi and Raul Serapioni.
Our long joint work has been always an invaluable source of scientific as well as
human enrichment. Without their contribution and friendship, I would never have
been able to “attack” this hard subject. I have to thank them also for allowing me to
widely quote both our joint papers and their latest ones.
I have appreciated the support of other friends and collaborators with whom I
have shared fruitful discussions during the last 15 years and whose work is men-
tioned here: Andrei Agrachev, Giovanni Alberti, Luigi Ambrosio, Zoltan Balogh,
Stefano Bianchini, Francesco Bigolin, Ugo Boscain, Luca Capogna, Laura Car-
avenna, Jih-Hsin Cheng, Giovanna Citti, Bernd Kirchheim, Enrico Le Donne, Nicola
Garofalo, Piotr Hajłasz, Pekka Koskela, Bruce Kleiner, Gian Paolo Leonardi,
Valentino Magnani, Andrea Malchiodi, Pertti Mattila, Francescopaolo Montefal-
cone, Michele Miranda jr., Roberto Monti, Pierre Pansu, Fabio Paronetto, Scott
Pauls, Andrea Pinamonti, Severine Rigot, Manuel Ritoré, Jeremy Tyson, Davide
Vittone, Paul Yang. . . and many others whom I am likely to have left out, for which
I apologize.
These notes are not intended to provide a thorough—not even a partial—survey
of geometric measure theory in Carnot groups, since they deal with only some topics
of this field. Anyone interested in an exhaustive overview of this subject, as well as
in a full bibliography, sharp statements, and detailed proofs, may refer to the PhD
theses of Magnani [153], Monti [181], and Montefalcone [175], as well as to the
recent ones of Bigolin [38], Vittone [221], and Pinamonti [196]. For more specific
facets we can merely recommend to the reader the monographs [73, 128, 129, 125,
124, 216, 220, 180, 50, 43, 2], and the papers [10, 11, 13, 48, 68, 99, 105, 106, 118,
130, 140, 191, 190, 192, 193, 223] and the references therein.
Since these notes especially focus on some issues of geometric measure the-
ory in Carnot groups, they of course leave out many other important topics, for
which reference will be made to special in-depth studies. By way of example, co-
area formulas (see, for instance, [153, 155]), fractal sets (see, for instance, [217,
23, 27, 30, 209, 81]), variational formulas for minimal surfaces (see, for instance,
[69, 178, 133, 134, 56, 213, 187, 206, 54, 114, 115, 179]), isoperimetric sets (see,
for instance, [148, 71, 206, 182, 185, 205, 116, 184, 87]), Bernstein’s problem for
minimal surfaces (see, for instance, [120, 56, 33, 187, 72, 134, 117]), currents and
Rumin’s complex (see, for instance, [11, 106, 207, 208, 21, 110]), curvatures (see,
for instance, [50, 2, 136, 199, 34, 3]), mass transportation and optimal transport (see,
for instance, [14, 201, 84, 79]), applications to theoretical computer science, ge-
ometry of Banach spaces, mathematical models in neuroscience (see, for instance,
[52, 64]).
Finally we warmly thank Simonetta Rivelli for invaluable support in English.
1 Some topics of geometric measure theory in Carnot groups 3
where [V1 ,Vi ] is the subspace of g generated by the commutators [X,Y ] with X ∈ V1
and Y ∈ Vi . Let mi = dim Vi , for i = 1, . . . , k and hi = m1 + · · · + mi with h0 = 0
and, clearly, hk = n. Choose a basis e1 , . . . , en of g adapted to the stratification, that
is, such that
frame X1 (x), . . . , X m1 (x). This way, a horizontal section φ is identified with a func-
tion φ = (φ1 , . . . , φ m1 ) : Rn → Rm1 . When dealing with two such sections φ and ψ
whose argument is not explicitly written, we drop the index x in the scalar product
writing hψ, φi for hψ(x), φ(x)i x . The same convention is adopted for the norm.
Two important families of diffeomorphisms of G are the so-called intrinsic trans-
lations and the intrinsic dilations of G. For any x ∈ G, the (left) translation τx : G →
G is defined as
z 7→ τx z := x · z.
For any λ > 0, the dilation δ λ : G → G, is defined as
δ λ (x 1 , . . . , x n ) = (λ α1 x 1 , . . . , λ α n x n ), (2.2)
where α i ∈ N is called the homogeneity of the variable x i in G (see [86, Chap. 1] )
and is defined as
α j = i whenever hi−1 + 1 < j ≤ hi , (2.3)
hence 1 = α1 = · · · = α m1 < α m1 +1 = 2 ≤ · · · ≤ α n = k.
Example 2.1 (Euclidean spaces). The Euclidean space Rn can be thought of as a
(trivial) abelian step-1 Carnot group with respect to its structure of a vector space.
Namely, the group law can be defined by the standard sum p + q, if p, q ∈ Rn , and
the dilations defined as λ p, if p ∈ Rn and λ > 0. The standard base of the Lie
algebra E of Rn is given by the left-invariant vector fields X j = ∂ j , j = 1, . . . , n. All
commutator relations in E are trivial.
Example 2.2 (Heisenberg group). The simplest example of a Carnot group is pro-
vided by the Heisenberg group Hn = R2n+1 . The reader can find an exhaustive
introduction to this structure as well as applications in [216]. A point p ∈ Hn is
denoted by p = (p1 , . . . , p2n , p2n+1 ) ∈ R2n+1 . For p, q ∈ Hn , we define the group
operation given by
2n
1X
p · q = * p1 + q1 , . . . , p2n + q2n , p2n+1 + q2n+1 + (pi qi+n − pi+n qi ) +
, 2 i=1 -
and the family of (nonisotropic) dilations
δ λ (p) := (λ p1 , . . . , λ p2n , λ 2 p2n+1 ) ∀p ∈ Hn , λ > 0.
The standard base of the Lie algebra of Hn , denoted h n , is given by the left-invariant
vector fields
pn+ j
X j = ∂j − ∂2n+1 , j = 1, . . . , n,
2
pj
Yj = ∂n+ j + ∂2n+1 , j = 1, . . . , n,
2
T = ∂2n+1 ,
1 Some topics of geometric measure theory in Carnot groups 5
[X j ,Yj ] = T, j = 1, . . . , n.
Thus the vector fields X1 , . . . , X n ,Y1 , . . . ,Yn satisfy Hörmander’s rank condition, and
Hn is a step-2 Carnot group, the stratification of the Lie algebra of left-invariant
vector fields being given by h n = V1 ⊕ V2 ,
(ii) ∀x, y ∈ G,
δ λ x · δ λ y = δ λ (x · y),
that is, that δ λ : G → G is an automorphism of the group and that the inverse x −1 of
an element x = (x 1 , . . . , x n ) ∈ (Rn , ·) has the form
x −1 = (−x 1 , . . . , −x n )
(see [86, Proposition 2.1]). Moreover, the unit element of the group is e = (0, . . . , 0).
Proposition 2.4. The vector fields X j have polynomial coefficients and if h`−1 < j ≤
h` , 1 ≤ ` ≤ k, then
Xn
X j (x) = ∂ j + qi, j (x)∂i , (2.8)
i >h l
where qi, j (x) = ∂Q ∂y j (x, y)| y=0 , so that if h`−1 < j ≤ h` , then qi, j (x) = qi, j (x 1 , . . . ,
i
G = G1 ⊕ G2 ⊕ · · · ⊕ G k , (2.9)
where Qi : Rn × Rn → Rm i , i = 2, . . . , k is defined as
Qi (p, q) := Qh i−1 +1 , . . . , Qh i ,
m1
X
γ̇(s) = c j (s)X j (γ(s)), for a.e. s ∈ [0,T].
j=1
d c (p, q) := inf {T > 0 : there exists a subunit curve γ with γ(0) = p, γ(T ) = q} .
By Chow’s theorem (see [43, Chap. 17]), the set of subunit curves joining p and q
is not empty, furthermore d c is a distance on G that induces the Euclidean topology
(see [43, Theorem 19.1.3] and Proposition 2.15 below).
We stress explicitly that, in general, Carnot–Carathéodory distances are not Eu-
clidean at any scale, and hence not Riemannian. A beautiful proof can be found in
[211] (for a more general statement see also [159]).
Definition 2.11. Given a metric space (X, d) and a curve γ : [a, b] → X, its total
variation is, by definition,
k−1
X
Var(γ) = sup d(γ(t i+1 ), γ(t i )). (2.17)
a ≤t 1 < ···<t k ≤b i=1
The supremum is taken over all finite partitions of [a, b]. If Var(γ) < +∞ the curve
γ is called rectifiable.
Lipschitz curves are rectifiable and the total variation has an integral representa-
tion with respect to the metric derivative (see [17, Theorem 4.1.6] and [181, Theorem
1.3.5]).
Notice that the following proposition holds (see, for instance, [128, Proposition
11.4] or [181, Proposition 1.3.3]).
It is well known that any rectifiable curve in a metric space admits an arc-length
parametrization. Note also that the arc-length parametrization makes the curve 1-
Lipschitz (see [17, Remark 4,11] ) and hence subunit.
A continuous rectifiable curve γ : [0, 1] → Rn is said to be a geodesic, or a
segment, if Var(γ) = d(γ(0), γ(1)). By an arc-length reparametrization, a geodesic
γ can always be reparametrized on the interval [0, Var(γ)] in such a way that d(γ(t),
γ(s)) = |t − s| for all s,t ∈ [0, Var(γ)] (see [17, Theorem 4.2.1]). The existence of
geodesics in a Carnot group endowed with the CC distance is also well known (see,
for instance, [181, Theorem 1.4.4] and [17, Theorem 4.2.3]).
Theorem 2.13. Let G = (Rn , ·) be a Carnot group, endowed with the CC distance
d c . Then for all x, y ∈ G there exists a geodesic connecting them.
Several distances equivalent to d c have been used in the literature. Later on, we
shall use
d ∞ (x, y) = d ∞ (y −1 · x, 0),
which can also be computed explicitly, where, if p = [p1 , . . . , pk ] ∈ Rm1 ×· · ·×Rm k =
Rn , then
1/ j
kpk∞ := d ∞ (p, 0) = max{ε j ||p j ||Rm j , j = 1, . . . , k}. (2.18)
Here ε 1 = 1, and ε 2 , . . . , ε k ∈ (0, 1] are suitable positive constants depending on the
group structure (see [105, Theorem 5.1]).
Remark 2.14. Notice that for the Heisenberg group G = Hn , one can choose ε 1 =
ε 2 = 1, that is, if p = [p1 , p2 ] = [(p1 , . . . , p2n ), p2n+1 ] ∈ R2n+1 , then kpk∞ =
max{||(p1 , . . . , p2n )||R2n , |p2n+1 | 1/2 }.
As above, we shall denote by U∞ (p,r) and B∞ (p,r) respectively the open and
closed balls associated with d ∞ .
Both the CC metric d c and the metric d ∞ are invariant (see Definition 2.6). In
particular, they are equivalent by Proposition 2.7.
Proposition 2.15. Let G = (Rn , ·) be a step-k Carnot group. Let d c and d E denote,
respectively, the CC metric on G ≡ Rn and the Euclidean one on Rn .
(ii) ([43, Corollary 5.15.1]) For each compact A ⊂ Rn (with respect to the Eu-
clidean topology) there exists a positive constant C A such that
C −1
A d E (x, y) ≤ d c (x, y) ≤ C A d E (x, y)
1/k
∀x, y ∈ A.
Remark 2.16. The right inequality in Proposition 2.15(ii) gives the first evidence
that the metric d c is not a Riemannian metric, being not locally equivalent to the
Euclidean distance d E . Moreover, taking this estimate into account, an (incorrect)
feeling could arise that CC (or √
SR) geometry
is of fractal type. Recall that, for
instance, the metric space Rn , k·kRn is of fractal type. Namely, we mean that
each C1 curve γ : [a, b] → Rn is unrectifiable, that is, according to Definition 2.11,
Var(γ) = +∞ (see also the beautiful book on fractals [73]). Indeed, as suggested in
Gromov’s paper [124], the right way to look at CC geometry is “from within”.
Remark 2.17. Observe that the inequalities in Proposition 2.15(ii) are local, that is,
the constant C A does actually depend in general on the set A. Indeed, for instance,
let us consider
the first Heisenberg group H1 ≡ R3 and the subunit curve γ : [0, 1] →
a
H1 , γ(s) := s, a, − s . Then, it is immediate to see that, if x = γ(0) and y = γ(1),
2
s
a2
d c (x, y) ≤ 1 and d E (x, y) = 1 + .
4
This implies that the left inequality in Proposition 2.15(ii) must be local.
Remark 2.18. By Proposition 2.7, it follows that, given an invariant distance d on
G, then statements (i), (ii) and (iii) of Proposition 2.15 stay true when replacing d c
with d.
The integer
n
X k
X
Q= αj = i dim Vi (2.19)
j=1 i=1
Ln (Bd (p,r)) = Ln (Ud (p,r)) = r Q Ln (Bd (p, 1)) = r Q Ln (Bd (0, 1)). (2.20)
Proof. Let us simply define the balls Bd ≡ B and Ud ≡ U. Notice that, by Proposi-
tion 2.15, Ln (Bd (p,r)) < ∞ for each p ∈ G, r > 0. Let us consider the diffeomor-
phisms τx : Rn → Rn and δ λ : Rn → Rn . Because of the group structure, it is easy
to see that
det J (τx ) = 1, det J (δ λ ) = λ Q ,
Free ebooks ==> www.Ebook777.com
2.3.1 Hausdorff measures in metric spaces. Let us recall below three different
notions of Hausdorff measures in a metric space. The notions of Hausdorff and
spherical Hausdorff measures are classical and they are obtained by Carathéodory’s
construction as in [83, Section 2.10.2]; the one of centered Hausdorff measure is
more recent (see [209]).
Let us introduce some notation and notions. Throughout this paper (X, d) will be
a separable metric space, and
B(x,r) := {y ∈ X : d(x, y) ≤ r }
is the closed ball with center a and radius r > 0. The diameter of a set E ⊂ X is
denoted as
diam(E) := sup {d(x, y) : x, y ∈ E} .
If µ is an outer measure in X and A ⊂ X, then the restriction of µ to A is denoted as
µ A(E) = µ( A ∩ E) if E ⊂ X.
We will assume the following condition on the diameter of closed balls: there
exist constants 0 < %0 ≤ 2 and δ0 > 0 such that, for all r ∈ (0, δ0 ) and x ∈ X,
diam(B(x,r)) = %0 r. (2.21)
β m := α m %−m
0 . (2.22)
Remark 2.20. In the case where m is an integer, the constant α m turns out to be equal
to the m-dimensional Lebesgue measure of a unit closed ball of Rm . The reason for
this normalization is to get equality (2.47) in the Euclidean case.
www.Ebook777.com
12 Francesco Serra Cassano
(X [ )
m
Hd,δ ( A) = inf β m diam(Ei ) m : A ⊂ Ei , diam(Ei ) ≤ δ .
i i
(X [
m
Sd,δ ( A) = inf β m diam(B(x i ,r i )) m : A ⊂ B(x i ,r i ),
i i
)
diam(B(x i ,r i )) ≤ δ .
for F , ∅,
(X [
m
Cd,δ (F) = inf β m diam(B(x i ,r i )) m : F ⊂ B(x i ,r i ),
i i
)
x i ∈ F, diam(B(x i ,r i )) ≤ δ .
Also the measures Cdm are metric measures in any metric space, but this fact, proved
in [81], is not as immediate as for H m and S m .
It holds that (see [81])
Hdm ≤ Sdm ≤ Cdm ≤ 2m Hdm . (2.23)
In particular, the three measures Hdm , Sdm , and Cdm are equivalent.
Remark 2.22. Notice that if d 1 and d 2 are two equivalent metrics on X, then Hd1
and Hd2 are equivalent, as well as Sd1 and Sd2 , Cd1 and Cd2 .
To state the definition in other words, Hdim( A) is the unique number (it may be ∞
in some metric space) for which
s < Hdim( A) ⇒ H s ( A) = ∞, (2.26)
s > Hdim( A) ⇒ H s ( A) = 0. (2.27)
At the borderline case s = Hdim( A) we cannot have any general nontrivial infor-
mation about the value H s ( A): all three cases H s ( A) = 0, 0 < H s ( A) < ∞,
and H s ( A) = ∞ are admissible. But if for some given A we can find s such that
0 < H s ( A) < ∞, then s = Hdim( A).
Let us now propose a simple criterion in order to calculate the Hausdorff dimen-
sion in a metric space.
Definition 2.25 (Ahlfors regularity). Let (X, d) be a metric space and let µ be a
Radon measure on X. We say that the metric measure space (X, d, µ) is Ahlfors
regular of dimension Q if
a1 r Q ≤ µ(B(x,r)) ≤ a2 r Q ∀x ∈ X,r > 0 (2.28)
for suitable positive constants ai (i = 1, 2).
14 Francesco Serra Cassano
Hdim(X ) = Q.
For the proof of this result we will need a basic Vitali covering result in metric
space, the proof of which can be found in [168].
Theorem 2.27. Let (X, d) be a separable metric space and let F be an arbitrary
family of balls (closed or open) in X such that
∪ B ∈F B ⊆ ∪ B ∈F 0 5B,
where 5B denotes the ball with same center as B and radius five times the radius
of B.
Proof of Theorem 2.26. It suffices to prove that, for a given x 0 ∈ X,
From (2.29), it will follow that Hdim(B(x 0 ,r)) = Q for each r > 0. Therefore,
since X = ∪∞ h=1 B(x 0 , h), by (2.25), we can get the desired conclusion.
Fix r > 0 and let 0 < δ < r. Consider a covering of closed balls F = {Bi : i ∈
N} of B(x 0 ,r) with Bi = B(yi ,r i ), diam(Bi ) = %0 r i ≤ δ for each i ∈ N. Without
loss of generality, we can assume that B(x 0 ,r) ∩ Bi , ∅ for each i ∈ N. From this
assumption, we can infer that
Applying Theorem 2.27 to the family of balls F , we get that there exists a countable,
disjoint subfamily of F , {Bi h :, h = 1, . . . , N }, where N can be a finite integer or
infinity, such that
B(x 0 ,r) ⊂ ∪∞i=1 Bi ⊂ ∪h=1 5Bi h .
N
(2.31)
From the definition of the spherical Hausdorff measure, (2.31), (2.28), and (2.30),
it follows that
N
X N
X
SdQ (B(x 0 ,r)) ≤ βQ (5%0 ) Q r iQh ≤ C1 µ(Bi h )
i=1 i=1 (2.32)
≤ C1 µ(∪i=1
N
Bi h ) ≤ C1 µ(B(x 0 , 2r)) ≤ C1 a2 2 r
Q Q
< ∞,
1 Some topics of geometric measure theory in Carnot groups 15
Q
0 < a1 r Q ≤ µ(B(x 0 ,r)) ≤ C2 Sd,δ (B(x 0 ,r)) ∀δ > 0.
Taking the limit as δ → 0 in the previous inequality we get
0 < a1 r Q ≤ µ(B(x 0 ,r)) ≤ C2 SdQ (B(x 0 ,r)) ∀r > 0. (2.34)
Remark 2.28. Notice that, from (2.32) and (2.34), it follows that if (X, d, µ) is
an Ahlfors metric measure space of dimension Q, so is the metric measure space
(X, d, SdQ ).
Finally let us estimate the metric dimension of a Lipschitz curve in a metric
space, which is the simplest example of a 1-dimensional regular submanifold in a
metric space. More precisely, the following theorem holds.
Theorem 2.29. Suppose that γ : [a, b] → (X, d) is a Lipschitz curve and let Γ =
γ([a, b]). Then
H1d (Γ) ≤ Sd1 (Γ) ≤ Cd1 (Γ) ≤ Var(γ) < +∞,
and equality holds if γ is injective.
Proof. We have only to show that
Cd1 (Γ) ≤ Var(γ). (2.35)
Indeed, from [17, Theorem 4.4.2], it follows that
H1d (Γ) = Var(γ) if γ is injective.
Thus, from (2.23) and (2.35), we get the desired conclusion.
Let us prove (2.35). By a reparametrization of γ (see [17, Theorem 4.2.1]), we
can assume that a = 0, b = Var(γ) with metric derivative of γ, | γ̇| = 1 a.e. on [a, b].
In particular, observe that γ is 1-Lipschitz and
Var(γ, [t, s]) = s − t ∀ a ≤ t ≤ s ≤ b. (2.36)
16 Francesco Serra Cassano
Given δ > 0, choose N ∈ N such that Var(γ)/N < δ and let h := Var(γ)/N and
Ji := [i h, (i+1) h], i = 0, . . . , N −1. Let Bi := B(pi , h/%0 ) with pi = γ((2i+1) h/2),
i = 0, . . . , N − 1, where %0 is the constant in (2.21). Then, by (2.21),
diam(Bi ) = h, i =, 0, . . . , N − 1 (2.37)
and
N −1
Γ ⊆ ∪i=1 Bi . (2.38)
Indeed, notice that, by (2.36), for each i = 0, . . . , N − 1,
2i + 1 2i + 1
!!
h h
d γ(t), γ h ≤ t − h ≤ ≤ ∀ t ∈ Ji .
2 2 2 % 0
2.3.3 Applications to Carnot groups. Now we are going to apply Hausdorff mea-
sures and metric dimension issues to the case when X = G is a Carnot group
equipped with an invariant distance d.
Let G be a Carnot group equipped with an invariant distance d. Let us first
observe that (2.21) is satisfied with %0 = 2 by Proposition 2.8. Moreover, because
of the invariance of distance d, all three Hausdorff measures are left-invariant with
respect to the family of left translations τx (x ∈ G), and m-homogeneous with respect
to the family of 1-parameter dilations δ λ (λ > 0), that is, if µ = Hdm , or µ = Sdm , or
µ = Cdm , then, for each A ⊂ G,
µ(τx ( A)) = µ( A), µ(δ λ ( A)) = λ m µ( A) ∀x ∈ G, λ > 0. (2.40)
We denote by H m the m-dimensional Hausdorff measure obtained from the Eu-
clidean distance in Rn ' G, by Hcm the m-dimensional Hausdorff measure obtained
from the distance d c in G, and by H∞ m the m-dimensional Hausdorff measure ob-
Theorem 2.30. Let G = (Rn , ·) be a Carnot group endowed with an invariant metric
d. Then
Hdim(G) = Q,
where Q denotes the homogeneous dimension of G; see (2.19).
Remark 2.31. Observe that n ≤ Q and the equality holds iff G is a step-1 Carnot
group, that is, G ≡ Rn is isomorphic to the Euclidean space Rn . In the other case,
namely when G is a Carnot group of step k ≥ 2, the topological dimension n is
strictly less than the metric dimension Q. For instance, in the case of the Heisenberg
group Hn , Hdim(Hn ) = Q = dim V1 + 2 dim V2 = 2n + 2. This gap is more
evidence that CC geometry differs from the Euclidean one and a new feeling for the
CC geometry requirements. In particular, we should change the Euclidean feeling
that a regular submanifold of a Carnot group still keeps both the topological and
metric dimensions equal. We will discuss this issue later in Section 4.
Proof of Theorem 2.30. From (2.20) it follows that the metric measure space (G, d,
Ln ) is Ahlfors regular of dimension Q. Thus, from Theorem 2.26 we get the desired
conclusion.
(i) ([202, Proposition 2.1]) SdQ (B) = βQ diam(B) Q for each ball B ⊂ G.
Proof. (iii) For the sake of simplicity, define SdQ = S Q and CdQ = S Q . From (i) and
the definition of CdQ , we get at once
for each ball B ⊂ G. Let µ be a normalized Haar measure on G such that µ(B) =
βQ diam(B) Q for each closed ball B. In order to accomplish the proof it will suffice
to prove that
C Q (U) ≤ µ(U), (2.42)
for each open ball U. Indeed, since C Q is a left-invariant Radon measure, diam(U) =
diam(U), from Proposition 2.8, if B = U with U an open ball; then
C Q (B) = C Q (U),
and, by (2.42), we get the desired inequality.
Let us prove (2.42). Given an open ball U, E ⊂ U, and δ > 0, let
( )
F := B(x,r) : x ∈ E, B(x,r) ⊂ U, diam(B(x,r)) < δ .
i=1 Bi ) = 0.
C Q (E \ ∪∞ (2.43)
and, by (2.43),
CδQ (E \ ∪∞ Q
i=1 Bi ) ≤ C0 (E \ ∪i=1 Bi ) ≤ C (E \ ∪i=1 Bi ) = 0.
∞ Q ∞
(2.45)
Taking first the limit as δ → 0 in (2.46) and then the supremum on all sets E ⊆ U,
(2.42) follows.
Corollary 2.33. Under the same assumptions as Theorem 2.32, it holds that SdQ =CdQ .
1 Some topics of geometric measure theory in Carnot groups 19
Proof. Since there exists a positive constant β such that SdQ = β CdQ , and SdQ (B) =
CdQ (B) on all balls, we can infer that β = 1.
Remark 2.34. Notice that, by choosing βQ = 2−Q Ln (Bd (0, 1)), we can get SdQ =
CdQ = Ln . We also point out that the agreement between Sdm (E) and Cdm (E) may
not occur for a dimension 1 < m < Q and for a general invariant metric d, even if
E ⊂ G is very regular (see Remark 4.31).
The constant Cd in (2.41) is related to the so-called isodiametric problem. Let us
recall that the classical isodiametric problem in the Euclidean space Rn says that a
ball maximizes the volume among all sets with prescribed diameter,
Ln ( A) ≤ β n diam( A) n ,
where β n is the constant defined in (2.22) (that is, the volume of a unit ball of Rn )
with %0 = 2 (see, for instance, [8, Proposition 2.52]). This gives nontrivial infor-
mation about the geometry of the Euclidean space and about the relation between
the Euclidean metric and the Lebesgue measure. In particular, a well-known con-
sequence of the isodiametric inequality is the following agreement between the n-
dimensional Lebesgue measure and the n-dimensional Hausdorff measure defined
with respect to the Euclidean distance:
Ln = H n . (2.47)
SdQ = HQ
d
.
In particular, it was proved that, given a nonabelian Carnot group G, one can always
find some invariant distance on G, namely the d ∞ distance as well as some CC dis-
tances, for which the sharp isodiametric inequality (2.48) does not hold (see [202,
20 Francesco Serra Cassano
Theorems 3.4, 3.5 and 3.6]). Moreover, an interesting relationship with some topics
of geometric measure theory in Carnot groups was established. Indeed, as a conse-
quence of those results, the pure Q-unrectifiability of Carnot groups can be obtained
(see [201, Theorem 4.1] and Definition 4.98 below), which we will discuss later.
Note that it is not difficult to see that one can always find sets achieving the
supremum in (2.41), called isodiametric sets (see [201, Theorem 3.1]). In [149], for
the case of the Heisenberg group Hn equipped with its CC distance d c , the regularity
of the isodiametric sets has been studied, as well as their characterization under
some symmetry assumption. The characterization of a general isodiametric set in a
Heisenberg group is still open.
Finally let us point out the paper [2] about the study of volume measure in sub-
Riemannian manifolds by means of Hausdorff measures and [45] concerning the
spherical Hausdorff measure and the Popp measure in sub-Riemannian manifolds.
differentiability in Carnot groups are basically due to Pansu ([190]), or are inspired
by his ideas.
The notion of P-differentiability for functions acting between Carnot groups was
introduced by Pansu in [190].
Definition 3.1. Let G1 , G2 be Carnot groups, with homogeneous norms k · k1 , k · k2
and dilations δ1λ , δ2λ . We say that L : G1 → G2 is H-linear, or is a homogeneous
homomorphism, if L is a group homomorphism such that
L(δ1λ g) = δ2λ L(g) ∀g ∈ G1 and λ > 0.
Let us recall the following characterization of an H-linear map between Carnot
groups (see also [153, Proposition 3.1.3 and Theorem 3.1.12]). We thank Nicolussi
Golo and Le Donne for pointing out to us a more complete statement of this charac-
terization than a previous version, as well thanking Le Donne for its proof and other
suggestions on the topic.
Theorem 3.2. Let (Gi , k · ki ) (i = 1, 2) be two Carnot groups. Denote by gi (i = 1, 2)
their Lie algebras and by expi : gi → Gi (i = 1, 2) their exponential maps. Let
g1 = V1 ⊕ · · · ⊕ Vk , g2 = W 1 ⊕ · · · ⊕ W l ,
G1 = V1 ⊕ V2 ⊕ · · · ⊕ Vk , G2 = W1 ⊕ W2 ⊕ · · · ⊕ Wl
be the stratifications defined according to, respectively, (2.1) and (2.9). Let L : G1 →
G2 be a homomorphism. Then the following are equivalent:
(i) L is H-linear;
(ii) L : (G1 , k · k1 ) → (G2 , k · k2 ) is C∞ and Lipschitz;
(iii) L is C∞ and it satisfies the contact property, that is, dL(V1 ) ⊂ W1 , where
dL : T (G1 ) → T (G2 ) denotes the differential on tangent bundles;
(iv) L is C∞ and L(V j ) ⊆ W j for each j.
Let us recall some well-known facts about Lie groups and homomorphisms.
Lemma 3.3. Under the same assumptions as Theorem 3.2, we have the following
results.
(i) Let ϕ : G1 → G2 be a continuous homomorphism. Then ϕ is C∞ .
(ii) Let ϕ : G1 → G2 be a C∞ homomorphism. Then the differential map dϕ :
g1 → g2 is a Lie algebra homomorphism, that is, it is a linear map such
that dϕ([X,Y ]) = [dϕ(X ), dϕ(Y )] for each X, Y ∈ g1 . Moreover, the map
ϕ̃ := exp−1
2 ◦ϕ◦exp1 : g1 ≡ Te (G1 ) → g2 ≡ Te (G2 ) agrees with the differential
of ϕ at e, dϕ(e) : Te (G1 ) → Te (G2 ). In particular, ϕ̃ : g1 → g2 is also a Lie
algebra homomorphism.
22 Francesco Serra Cassano
Proof of Lemma 3.3. (i) The fact that ϕ is C∞ follows by [224, Theorem 3.39].
(ii) Since ϕ is C∞ , by a classical result about homomorphisms between Lie
groups (see, for instance, [43, Theorem 2.1.50(ii)]), it follows that dϕ : g1 → g2
is a Lie algebra homomorphism. On the other hand, because Gi ((i = 1, 2) are con-
nected and simply connected Lie groups, we can apply [43, Theorem 2.1.61] and,
for all X ∈ g1 , Y = ϕ̃(X ) ∈ g2 is the only left-invariant vector field on G2 defined by
Proof of Theorem 3.2. (i) ⇐⇒ (ii): It follows by Lemma 3.3 and [153, Proposition
3.1.8].
(ii)=⇒(iii): Let v ∈ V1 and let us consider the C1 curve γ : R → G2 , γ(t) :=
L exp1 (tv) . Then, by Lemma 3.3(i) and (3.1),
dL(e)(Vj ) ⊂ W j
and, by linearity,
Given two Carnot groups (Gi , k·ki ), i = 1, 2, where k·ki is a homogeneous norm
on Gi , denote by L H (G1 , G2 ) the set of H-linear functions L : G1 → G2 endowed
with the norm
f (x) − f (x 0 ) − L(x −1
0 · x)
lim =0
x→x 0 d(x, x 0 )
(see also [140]).
Remark 3.11. The above definition is equivalent to the following one: there exists a
homomorphism L from G to (R, +) such that
f (τx0 (δ λ v)) − f (x 0 )
lim = L(v)
λ→0+ λ
uniformly with respect to v belonging to compact sets in G. In particular, L is unique
and we shall write L = d G f (x 0 ). Notice that this definition of a differential de-
pends only on G and not on the particular choice of the canonical generating vector
fields. Indeed, any two Carnot–Carathéodory distances induced by different choices
of (equivalent) scalar products in HG are equivalent as distances.
Definition 3.12. If Ω is an open set in G, we shall denote by C1G (Ω, HG) the set of
all sections φ of HG with canonical coordinates φ j ∈ C1G (Ω) for j = 1, . . . , m1 .
Remark 3.13. We recall that C1 (Ω) ⊂ C1G (Ω) and that the inclusion may be strict;
for an example see [102, Remark 5.9]. We point out that a function f ∈ C1G (Ω) may
be very irregular from a Euclidean point of view (see [139]).
1 Some topics of geometric measure theory in Carnot groups 25
For a proof see [186, Remark 3.3]. The following proposition can be proved via
an approximation argument as in [102, Proposition 5.8].
Free ebooks ==> www.Ebook777.com
3.2 Area formula between Carnot groups. The content of this section relies
on [153, Chapter 4] (see also [10, 193, 223]).
An important tool in Euclidean geometric measure theory is the so-called area
formula for Lipschitz maps, which turns out to be true also for Lipschitz maps be-
tween Carnot groups.
Definition 3.18 (H-Jacobian). Given two Carnot groups (Gi , d i ), i = 1, 2, endowed
with an invariant distance d i , let L : G1 → G2 be an H-linear map. Let Q denote
the homogeneous dimension of G1 . The Jacobian of L is defined by
HQ
d2
(L(B1 ))
JQ (L) := ,
HQ
d1
(B1 )
where N ( f , A, y) denotes the multiplicity function of f , that is, the number of ele-
ments of set f −1 (y).
The area formula turns out to be a useful tool in order to study low-dimensional
surfaces in Carnot groups (see, for instance, Section 4.5) as well as for proving some
unrectifiability results for Carnot groups (see Section 4.8).
3.3 Whitney theorem for maps between Carnot groups. It is well known
that a useful tool for the approximation of a function in a Euclidean setting turns out
to be Whitney’s extension theorem. In the setting of Carnot groups we mean the
www.Ebook777.com
1 Some topics of geometric measure theory in Carnot groups 27
Assume
% K (δ) → 0 as δ → 0 for every compact set K ⊂ F.
Then there exist f˜ : G → R, f˜ ∈ C1 (G) such that
G
f˜|F = f , ∇G f˜|F = k.
To our knowledge, the only case with a target space G2 (nonabelian) Carnot group
that has been studied, is when G1 = R and G2 = Hn E where E denotes the Engel
group (see Example 4.7). More precisely, Speight [215] was able to give a positive
answer for a curve γ : F ⊂ R → Hn horizontal by proving a C1 extension theorem
like Theorem 3.20 (see [215, Proposition 2.7]) and by proving a Lusin approxima-
tion result for absolutely continuous horizontal curves by means of C1 horizontal
curves (see [215, Theorem 1.2]). Meanwhile he disproved a Lusin approximation
result in the setting of an Engel group by means of a very interesting example (see
[215, Theorem 3.2]). Finally we have been informed that Hajłasz is also working on
similar arguments.
In the case of a Carnot group G, the vector fields X1 , . . . , X m will agree with a
system of generating vector fields X1 , . . . , X m1 of the group.
We shall denote by X ∗j the operator formally adjoint to X j in L 2 (Rn ), that is, the
operator which for all φ, ψ ∈ C0∞ (Rn ) satisfies
Z Z
φ(x)X j ψ(x) dx = ψ(x)X ∗j φ(x) dx.
Rn Rn
Definition 3.21. Let Ω be an open subset of Rn . One can define the Sobolev space
1, p
W X (Ω), 1 ≤ p ≤ ∞ associated with the family X as the space of all functions with
finite norm kukW 1, p = kuk p + k Xuk p , where |Xu| 2 = |X j u| 2 and the derivatives
P
X
X j u are understood in the sense of distributions. The L p -norms should be meant
with respect to Lebesgue measure.
Throughout this paper, if E ⊂ Rn , both |E| and Ln (E) denote its Lebesgue
measure. Analogously, if µ is a measure in a set X, we write µ(E) or |E| µ for the
µ-measure of the set E ⊂ X.
3.4.3 Poincaré inequality. The starting point is the following Poincaré inequality
in Carnot groups ([219, 135]).
Thanks to the Poincaré inequality (3.5), by [95] and [112] it follows that we can
represent a continuously differentiable function f in terms of a fractional integral of
∇G f . We have
for x ∈ U, where τU is the ball concentric with U of radius τr (U) and Q denotes
the homogeneous dimension of G.
Theorem 3.27. Suppose 0 < α < Q, and denote by Iα the Riesz potential in G of
order α, i.e., for x ∈ G, put
Z
|g(y)|
Iα g(x) := dy.
G d(x, y) Q−α
kIα f k L q ≤ C p k f k L p .
ii) if p = 1, then
! Q/(Q−1)
k∇G f k L 1
|{x ∈ G : | f (x) − f U |(x) > λ}| ≤ C p , (3.8)
λ
for λ > 0.
Inequality (3.7) is “almost” the Sobolev–Poincaré inequality in the group for p >
1 (the nongeometric case). We say “almost” because of the presence of the dilation
factor τ > 1. But we shall get rid of τ pretty soon. On the other hand, (3.8) is not
a Sobolev–Poincaré-type inequality in G because of the presence of the weak-type
norm on the left. Since p = 1 is an endpoint for the weak-type estimates of Theorem
3.27(ii), we cannot rely on the Marcinckiewicz interpolation theorem to pass from
Free ebooks ==> www.Ebook777.com
www.Ebook777.com
32 Francesco Serra Cassano
and µ be doubling. Suppose that f and g are measurable functions on Ω and for each
ball U with τU ⊂ Ω there exists a constant gU such that
||g − gU || L q (U ) ≤ A|| f || L p (τU )
dµ dν
for all φ ∈ C10 (Ω, HG). Finally the notion of the gradient ∇G can be extended from
regular functions to functions f ∈ BVG defining ∇G f as the vector-valued measure
∇G f := −σ f ||∇G f || = −(σ f )1 ||∇G f ||, . . . , −(σ f )m1 ||∇G f || ,
where (σ f ) j are the components of σ f with respect to the moving base X j .
1 Some topics of geometric measure theory in Carnot groups 33
It is well known that the usefulness of these definitions for the calculus of vari-
ations relies mainly on the validity of the two following theorems. In the context of
sub-Riemannian geometries they are proved respectively in [118] and [99].
p
Theorem 3.34 (Compactness). BVG,loc (G) is compactly embedded in L loc (G) for
Q
1 ≤ p < Q−1 where Q, defined in (2.19), is the homogeneous dimension of G.
Remark 3.36. Fine differentiability properties for BVG functions have been studied
in [13, 177, 15].
Definition 3.37. A measurable set E ⊂ G is of locally finite G-perimeter in Ω (or is
a G-Caccioppoli set) if the characteristic function 1 E ∈ BVG,loc (Ω). In this case we
call the perimeter of E the measure
Remark 3.38. This remark is analogous to Remark 3.14. The symbol |∂E|G is some-
how incorrect, indeed the value of the G-perimeter depends on the choice of the
generating vector fields X1 , . . . , X m1 precisely through the bound |φ| ≤ 1 in (3.11).
The values of the perimeters induced by two different families of generating vector
fields coincide only if the two families are mutually orthonormal; nevertheless the
perimeters induced by different families are equivalent as measures and, as a conse-
quence, the notion of being a G-Caccioppoli set is an intrinsic one depending only
on the group G.
Proposition 3.39. If E is a G-Caccioppoli set with C 1 boundary, then the G-peri-
meter has the following representation:
v
m1
Z u
tX
|∂E|G (Ω) = hX j , NE i2Rn d H n−1 ;
∂E∩Ω j=1
Remark 3.40. The G-perimeter is invariant under group translations, that is,
|∂E|G ( A) = |∂(τp E)|G (τp A), ∀p ∈ G, and for any Borel set A ⊂ G;
indeed divG is invariant under group translations and the Jacobian determinant of
τp : G → G equals 1. Moreover, the G-perimeter is homogeneous of degree Q − 1
with respect to the dilations of the group, that is,
also, this fact is elementary and can be proved by changing variables in formula
(3.11).
Fundamental estimates in geometric measure theory are the so-called relative
and global isoperimetric inequalities for Caccioppoli sets. They were obtained by
Garofalo and Nhieu (see also [90, 94] and the monograph [50]).
Theorem 3.41 (Isoperimetric inequalities, [118, Theorem 1.18]). There exists a pos-
itive constant CI such that for any G- Caccioppoli set E ⊂ G, for any x ∈ G, r > 0,
min{|E ∩ Uc (x,r)|, |Uc (x,r) \ E|} (Q−1)/Q ≤ CI |∂E|G (Uc (x,r)) (3.15)
and
min{|E|, |G \ E|} (Q−1)/Q ≤ CI |∂E|G (G), (3.16)
where Uc denotes the (open) ball with respect to the CC distance d c .
Finally we point out that an extension of the notion of BV functions to a general
metric space was introduced in [173].
G = M · H,
G = H · M,
Free ebooks ==> www.Ebook777.com
Example 4.4 ([108, Proposition 4.1]). Let G be the Heisenberg group Hn (see Exam-
ple 2.2). All homogeneous subgroups of Hn are either horizontal, that is, contained
in the horizontal fiber HHen , or vertical, that is, containing the subgroup T. A hori-
zontal subgroup has linear dimension and metric dimension k, with 1 ≤ k ≤ n, and
it is algebraically isomorphic and isometric to Rk . A vertical subgroup can have any
dimension d, with 1 ≤ d ≤ 2n + 1, its metric dimension is d + 1, and it is a normal
subgroup. All couples V, W of complementary subgroups of Hn are of the type
www.Ebook777.com
1 Some topics of geometric measure theory in Carnot groups 37
considered as Lie groups with Haar measure Ln−1 , satisfying (2.20) with Q ≡ Q − 1
and closed with respect to the family of dilations of G. Therefore (M, d, Ln−1 ) is
a measure metric Ahlfors-regular space of dimension Q − 1: as a by-product, its
metric dimension is Q − 1. In conclusion we get the splitting G = M · H, with M an
(n − 1,Q − 1)-subgroup and H a (1, 1)-subgroup.
Example 4.7. The Engel group is E = (R4 , ·, δ λ ), where the group law is defined as
x 1 + y1
x y
*. 1 +/ *. 1 +/ *.. x 2 + y2 +/
x y
2 2
.. // · .. // = .. x 3 + y3 + (x 1 y2 − x 2 y1 )/2
//
. x 3 / . y3 / .. x + y + [(x y − x y ) + (x y − x y )]/2///
4 4 1 3 3 1 2 3 3 2
, x 4 - , y4 - , +(x 1 − y1 + x 2 − y2 )(x 1 y2 − x 2 y1 )/12 -
and the family of dilation is
δ λ (x 1 , x 2 , x 3 , x 4 ) = (λ x 1 , λ x 2 , λ 2 x 3 , λ 3 x 4 ).
A basis of left-invariant vector fields is X1 , X2 , X3 , X4 defined as
X1 (x) := ∂x1 − (x 2 /2) ∂x3 + −x 3 /2 − (x 1 x 2 + x 22 )/12 ∂x4 ,
X2 (x) := ∂x2 + (x 1 /2) ∂x3 + −x 3 /2 + (x 21 + x 1 x 2 )/12 ∂x4 ,
X3 (x) := ∂x3 − ((x 1 + x 2 )/2) ∂x4 ,
X4 (x) := ∂x4 .
The commutation relations are [X1 , X2 ] = X3 , [X1 , X3 ] = [X2 , X3 ] = X4 and all the
other commutators are zero.
Inside the Engel group there are two families of complementary subgroups. The
first one is formed by 1-dimensional horizontal subgroups and 3-dimensional sub-
groups containing all the vertical directions. This family gives a semidirect splitting
of E. The second family is formed by 2-dimensional subgroups that are not nor-
mal subgroups. All the computations can be easily done directly, preferably using a
symbolic computation program.
The homogeneous subgroups
Mγ, β := {(γt, βt, 0, 0) : t ∈ R}, Nγ,δ := {(γt, δt, x 3 , x 4 ) : t, x 3 , x 4 ∈ R}
are complementary subgroups in E, provided that γδ − βγ , 0. Moreover, Nγ,δ is a
normal subgroup, hence E is the semidirect product of Mγ, β and Nγ,δ .
The second family, for γ + β , 0, is given by
K := {(x 1 , −x 1 , x 3 , 0) : x 1 , x 3 ∈ R}, Hγ, β := {(γt, βt, 0, x 4 ) : t, x 4 ∈ R}.
One can compute directly that K and Hγ, β are complementary subgroups in E and
that neither K nor Hγ, β are normal subgroups. Hence E = K · Hγ, β , but the product
is not a semidirect product.
38 Francesco Serra Cassano
Proof. The right-hand side is the triangle inequality. For the left-hand side observe
that M∩∂B(0, 1) and H∩∂B(0, 1) are disjoint compact sets, hence they have positive
distance. The general statement follows by dilation.
The following estimates will be crucial later in the paper (see also [160, Lemma
3.9 and Remark 3.10]).
Lemma 4.9 ([98, Lemma 2.13]). Let G be a step-κ Carnot group. Then there exists
a positive constant C = C(G) such that
1/κ (k−1)/κ 1/κ (k−1)/κ
kp−1 q−1 pqk∞ ≤ C kpk∞ kqk∞ + kqk∞ kpk∞ ∀p, q ∈ G, (4.2)
and, consequently,
1/κ (k−1)/κ 1/κ (k−1)/κ
kq−1 pqk∞ ≤ kpk∞ + C kpk∞ kqk∞ + kqk∞ kpk∞ ∀p, q ∈ G. (4.3)
G, then M will always be the first “factor”, and H the second one and gM ∈ M and
gH ∈ H are the unique elements such that
g = gM gH . (4.5)
We stress that this notation is ambiguous because each component gM and gH de-
pends on both the complementary subgroups M and H and also on the order in which
they are taken. The projection maps PM : G → M and PH : G → H are defined as
(iii) PH g = B1 g 1 , B2 (g 2 − Q2 ( A1 g 1 , B1 g 1 )), . . . ,
Bκ (gκ − Qκ ( A1 g 1 , . . . , Bκ−1 gκ−1 )) ;
V = {x = (x 1 , 0, 0) : x 1 ∈ R} , W = {x = (0, x 2 , x 3 ) : x 2 , x 3 ∈ R} .
Let q = (1, 0, 0) and pε = (1, ε, ε/2), then PW q = (0, 0, 0) and PW pε = (0, ε, ε).
Hence
√
kq−1 pk∞ = k(0, ε, 0)k∞ ≈ ε, k(PW q) −1 PW pε k∞ = k(0, ε, ε)k∞ ≈ ε
PM (g −1 ḡ) = PM (h−1 m−1 m̄ h̄) = PM (m−1 m̄ m̄−1 mh−1 m−1 m̄ h̄) = m−1 m̄.
Even if the projections are not Lipschitz we have the following control on the
measure of projected sets.
Lemma 4.14 ([98, Lemma 2.20]). Let M and H be complementary subgroups of G.
Denote by d t ≤ d m respectively, the topological and the metric dimensions of M.
Then there is c = c(M, H) > 0 such that Ld t (PM (B∞ (p,r))) = c r d m for all balls
B∞ (p,r) ⊂ G.
Free ebooks ==> www.Ebook777.com
In our notation such an S is a T-graph (or simply t-graph), where T is the center of
Hn . The left cosets of T are parametrized over the set A = HHen . We recall that the
center T has no complementary subgroup in Hn (see Example 4.4), and in general
there isn’t a couple of complementary subgroups M, H of Hn and a ψ : M → H such
that S = graph (ψ), even locally.
If a set S ⊂ G is an intrinsic graph then it stays an intrinsic graph after left
translations or group dilations.
www.Ebook777.com
42 Francesco Serra Cassano
= q · graph (φ).
Remark 4.18. From (4.8) and the continuity of the projection maps PM and PH it
follows that the continuity of a function is preserved by translations. Precisely, given
q = qM qH and f : M → H, then the translated function f q is continuous in m ∈ M if
and only if the function f is continuous in the corresponding point (q−1 m)M .
Remark 4.19. The algebraic expression of φq in Proposition 4.17 can be made more
explicit when G is a semidirect product of M, H. Precisely,
(i) if M is normal in G then φq (m) = qH φ (q−1 m)M , for m ∈ Eq = qE (qH ) −1 ;
G-regular surfaces have a unique tangent plane at each point. This follows from
a Taylor formula for functions in C1G that is basically proved in [190].
g
If S = {x : f (x) = 0} ⊂ G is a G-regular hypersurface, the tangent group TG S(x)
to S at x is
m1
X
g
( )
TG S(x) := v = (v1 , . . . , vn ) ∈ G : X j f (x)v j = 0 . (4.10)
j=1
g
By (2.3), TG S(x) is a proper subgroup of G. We can define the tangent plane to S at
x as
g
TG S(x) := x · TG S(x). (4.11)
We stress that this is a good definition. Indeed, the tangent plane does not depend on
the particular function f defining the surface S because of point (iii) of the implicit
function theorem, Theorem 4.24 below, that yields
g
TG S(x) = {v ∈ G : hνE (x), π x vi x = 0},
44 Francesco Serra Cassano
where νE is the generalized inward unit normal defined in (3.13), and π x (v) =
j=1 v j X j (x). Notice that the map v 7→ π x (v), for x ∈ G fixed,
Pm
m1
X
π x (v) = v j X j (x), (4.12)
j=1
m m
SG+ (0, v 0 ) := x ∈ G :
( X ) ( X )
x i vi > 0 and SG− (0, v 0 ) := x ∈ G : x i vi < 0 .
i=1 i=1
( X m )
Π(0, v 0 ) := x : x i vi = 0 .
i=1
If v =
Pm
i=1 vi X i (y) ∈ HGy , SG± (y, v) and Π(y, v) are the translated sets
where v and v 0 have the same components vi with respect to the left-invariant basis
X i . Hence
( m
X )
SG± (y, v) = x ∈ G : (x i − yi )vi > 0(< 0) . (4.13)
i=1
Note also that the class of G-regular hypersurfaces is different from the class of
Euclidean C 1 embedded surfaces in Rn . On the one hand, G-regular surfaces can
have “ridges” because continuity of the derivatives of the defining functions f is
required only in the horizontal directions; on the other hand, a Euclidean C 1 sur-
face can have so-called characteristic points, i.e., points p ∈ S where the Euclidean
tangent plane Tp S contains the horizontal fiber HG p .
The points of Char(S) are, in many aspects, irregular points of S. Note indeed
that the tangent group does not exist in these points. It is also well known that these
points are “few” on smooth hypersurfaces. Precise estimates of the HQ−1 c measure of
the characteristic sets of C 1 surfaces in general Carnot groups Hn have been obtained
by Magnani [157], extending previous results of Balogh [22] in the Heisenberg group
and Franchi, Serapioni, & Serra Cassano [105] in step-2 Carnot groups. Note that
the study of the size of the characteristic set has a long history. We refer to the
contributions of Derridj [80], Franchi & Wheeden [111], and Danielli, Garofalo, &
Nhieu [68]. Magnani’s result reads as follows.
Theorem 4.23. If S is a Euclidean C1 -smooth hypersurface in a Carnot group G
with homogeneous dimension Q and equipped with an invariant distance d, then
HQ−1
d
(Char(S)) = 0. (4.15)
We can state now our implicit function theorem, holding that a G-regular hy-
persurface S = { f (y) = 0} boundary of the set E = { f (y) < 0} can be locally
parametrized through a function Φ : Rn−1 → Rn so that the G-perimeter of E can be
written explicitly in terms of ∇G f and Φ. In view of the notion of the intrinsic graph
introduced in Definition 4.15, we can also say that S is locally an intrinsic graph in
the direction of a 1-dimensional horizontal subgroup (see Corollary 4.25).
Theorem 4.24 (Implicit function theorem, [103, Theorem 2.1]). Let Ω be an open
set in G = (Rn , ·), 0 ∈ Ω, and let f ∈ C1G (Ω) be such that f (0) = 0 and X1 f (0) > 0.
Define
E = {x ∈ Ω : f (x) < 0}, S = {x ∈ Ω : f (x) = 0},
and, for δ > 0, h > 0,
Then there exist δ, h > 0 such that the map (t, ξ) → γ(t, ξ) is a diffeomorphism of a
neighborhood of Jh × Iδ onto an open subset of Rn , and, if we denote by U ⊂⊂ Ω
the image of Int(Jh × Iδ ) through this map, we have
(i) E has finite G-perimeter in U ;
(ii) ∂E ∩ U = S ∩ U ;
∇G f (x)
(iii) νE (x) = − ∀x ∈ S ∩ U ;
|∇G f (x)| x
46 Francesco Serra Cassano
where νE is the generalized inner unit normal defined by (3.13), that can be identified
with a section of HG with |ν(x)| x = 1 for |∂E|G -a.e. x ∈ U . In particular, νE can
be identified with a continuous function and |ν| ≡ 1. Moreover, there exists a unique
function
φ = φ(ξ) : Iδ → Jh
such that the following parametrization holds: if ξ ∈ Iδ , put Φ(ξ) = γ(φ(ξ), ξ), then
(iv) S ∩ Ū = {x ∈ Ū : x = Φ(ξ), ξ ∈ Iδ };
(v) φ is continuous;
By Dini’s Theorem 4.24 and Example 4.6, it follows that if H := {(x 1 , 0, . . . , 0)} ≡
R and
M = {(0, x 2 , . . . , x n ) : x i ∈ R, i = 2, . . . , n} ≡ Rn−1 ,
then
(i) M and H are, respectively, (n − 1,Q − 1)- and (1, 1)-subgroups of G, and they
are complementary, that is, G = M · H;
Remark 4.26. Notice that, taking the Euclidean case into account, the approach of
defining a regular hypersurface S ⊂ G = (Rn , ·) by means of a parametrization
map cannot work in a nonabelian Carnot group. Namely, suppose that there is a map
φ ∈ C1H (U , G), whose P-differential d P φ(p) : Rn−1 → G is injective for each p ∈ U ,
such that the image S = φ(U ) with U ⊂ Rn−1 . If this is the case, let us consider
the injective H-linear map L̃ := d P φ(p) : Rn−1 → g read as a homomorphism
of Lie algebras. In view of Theorem 3.2, since it is a contact map, it follows that
L̃(Rn−1 ) ⊆ V1 , where V1 is the first horizontal layer of the stratification of g (see
(2.1)). Thus V1 must contain a (trivial) (n − 1)-dimensional algebra, which implies
that g must also be trivial.
1 Some topics of geometric measure theory in Carnot groups 47
Our next theorem is a mild regularity result. Roughly speaking, it states that G-
regular hypersurfaces do not have cusps or spikes if they are studied with respect
to the intrinsic Carnot–Carathéodory distance, while they can be very irregular as
Euclidean submanifolds. To make precise the former statement we recall the notion
of essential boundary (or of measure-theoretic boundary) ∂∗ F of a set F ⊂ G:
( n
L (F ∩ U (x,r)) Ln (F c ∩ U (x,r))
)
∂∗ F := x ∈ G : lim sup min , > 0 . (4.16)
r →0+ Ln (U (x,r)) Ln (U (x,r))
Notice that the definition above makes sense in any metric measure space and
that the essential boundary does not change if, in the definition (4.16), the distance
d is substituted by an equivalent distance d 0.
Theorem 4.27. Let Ω ⊂ G be a fixed open set, and E be such that ∂E ∩ Ω = S ∩ Ω,
where S is a G-regular hypersurface. Then
∂E ∩ Ω = ∂∗ E ∩ Ω. (4.17)
measure µ and a set A ⊂ H1 and two constants 0 < k 1 < k2 such that µ is absolutely
continuous w.r.t. Sc2 ,
Θ2 (µ, x) = k1 < k 2 = Θ∗2
F (µ, x) ∀x ∈ A
and
µ( A) > t Sc2 ( A) ∀t ∈ (k1 , k 2 ). (4.20)
In particular, from what was said before, it follows that the general statement, given
A ⊂ X and k > 0,
Θm (µ, x) = k ∀x ∈ A ⇒ µ A = k Sm A, (4.21)
may fail.
1 Some topics of geometric measure theory in Carnot groups 49
Therefore Magnani stressed the need for an alternative proof of the assertions
in [102, 105, 103, 106], that the perimeter measure |∂E|G agrees, up to a constant,
Q−1
with the (Q − 1)-dimensional spherical Hausdorff measure S∞ restricted to a G-
regular boundary ∂E, where X = G is a Carnot group of Hausdorff dimension Q,
and µ = |∂E|G is the perimeter measure of a set E. Meanwhile he proved
Theorem 4.30 ([162, Theorem 5.2]). Let G be a Carnot group endowed with the
invariant distance d ∞ in (2.18) and let Q be its homogeneous dimension. Let E ⊂ G
be such that its topological boundary ∂E is a G-regular hypersurface; then there is
a constant k = k (G) > 0 for which
Remark 4.31. Theorem 4.30 actually holds for more general invariant distances than
d ∞ : it also holds for each invariant distance d for which the unit closed ball Bd (e, 1)
is convex. Observe also that for a general invariant distance d, ΘQ−1 (|∂E|G , ·) and
Θ∗FQ−1 (|∂E|G , ·) may differ. For instance, in a very recent note [163], Magnani
proved that, inside the Heisenberg group G = H1 ≡ R3 equipped with its sub-
Riemannian metric d = d c , given E ⊂ H1 such that its topological boundary ∂E is a
G-regular hypersurface, then there are two positive constants k i (i = 1, 2) for which
We thank Magnani for sharing with us his unpublished note as well as for some
useful suggestions on the argument.
From Theorems 4.29 and 4.30 we get the following
Corollary 4.32. Let G be a Carnot group endowed with the invariant distance d ∞ .
Let E ⊂ G with topological boundary ∂E a G-regular hypersurface; then there is a
constant k = k (G) > 0 such that
Remark 4.35. The reverse assertion is false: there exist G-regular hypersurfaces in
G ≡ Rn that have Euclidean Hausdorff dimension greater than n − 1: indeed in [139]
it was shown that there exist G-regular hypersurfaces in the Heisenberg group H1
(Q = 4, n = 3) with Euclidean Hausdorff dimension 2.5.
By the previous Magnani example, it turns out that Federer densities play a priv-
ileged role when dealing with the spherical measures S m . On the other hand, these
densities are frequently harder to compute than the centered ones. Therefore, in
[109] we studied whether the aforementioned area formula stays true with “centered”
densities Θ∗ m and measures Sdm replaced by equivalent ones. Centered Hausdorff
measures Cdm (see Definition 2.21(iii)) actually play this role. Indeed, we have
Theorem 4.36 ([109, Theorem 3.1]). Let µ be a Borel regular measure in X such
that there exists a countable open covering of X whose elements have µ finite mea-
sure; let A ⊂ X be a Borel set. If Cdm ( A) < ∞ and µ A is absolutely continuous
with respect to Cdm A, then
Remark 4.37. Since Cdm and Sdm are equivalent, then Cdm ( A) < ∞ if and only if
Sdm ( A) < ∞ and µ A is absolutely continuous with respect to Cdm if and only if
µ A is absolutely continuous with respect to Sdm .
Remark 4.38. According to Theorem 4.36, in the differentiation results for measure
stated in [103, Theorem 3.4] and [106, Theorem 4.3] the centered Hausdorff measure
has to be considered in place of the spherical one.
Corollary 4.39. Under the same assumptions as Theorem 4.36, if, for A ⊆ X, there
is a constant k > 0 such that
Θ∗m (µ, x) = k ∀x ∈ A,
then
µ A = k Cm A.
Free ebooks ==> www.Ebook777.com
HQ−1
d
S = ζ1 SdQ−1 S = ζ2 CdQ−1 S,
Theorem 4.42 ([109]). Let G be a Carnot group and d be an invariant distance. Let
Ω ⊂ G be a fixed open set and let E ⊂ G such that ∂E ∩ Ω = S ∩ Ω, where S is a
G-regular hypersurface.
Then
n−1 U (0, 1) ∩ T g S(x)
L d
ΘQ−1 (|∂E|G , x) =
G
∀x ∈ S, (4.25)
αQ−1
g
where TG S(x) denotes the tangent group to S at x according to (4.10), and Ud (0, 1)
denotes the unit (open) ball defined in (2.16). In particular,
www.Ebook777.com
52 Francesco Serra Cassano
4.4.1 Regular surfaces in Heisenberg groups [106]. Let us start with some com-
ments about possible notions of regular submanifolds of a group. It is barely worth
saying that Euclidean regular submanifolds of Hn ≡ R2n+1 are not a satisfactory
choice for many reasons. Indeed, Euclidean regular submanifolds need not be group
regular; this is clear for low-dimensional submanifolds: 1-dimensional group-regular
objects are horizontal curves that are a small subclass of C1 lines, but, also a low-
codimensional Euclidean submanifold need not be group regular due to the presence
of the so-called characteristic points where no intrinsic notion of tangent space to
the surface exists (see [22, 157]). On the other hand, as seen in the previous sec-
tion, in Carnot groups 1-codimensional G-regular hypersurfaces, that we will call
H-regular in the setting of Heisenberg groups, can be highly irregular as Euclidean
objects but enjoy intrinsic regularity properties, so that it is natural to think of them
as 1-codimensional regular submanifolds of the group.
What do we mean by “intrinsic regularity” properties? We have already stated
what intrinsic should mean here. We believe that the most natural requirements to be
made on a subset S ⊂ Hn to be considered as an intrinsic regular submanifold are
(i) S has, at each point, a tangent “plane” and a normal “plane” (or better, a
“transversal plane”);
(ii) tangent “planes” depend continuously on the point;
the notion of “plane” has to depend only on the group structure and on the differ-
ential structure as given by the horizontal bundle. Since subgroups are the natural
counterpart of Euclidean subspaces, accordingly it seems natural to ask that
(iii) both the tangent “plane” and the transversal “plane” are subgroups (or bet-
ter, cosets of subgroups) of Hn ; Hn is the direct product of them (see [106,
Section 3.2]);
Free ebooks ==> www.Ebook777.com
(iv) the tangent “plane” to S in a point is the limit of group dilations of S centered
in that point (see [106, Definition 3.4]).
Notice that the requirement that the limit of a blow-up procedure is a subgroup comes
out naturally even in much more general settings than Hn (see [169]). Moreover, the
explicit requirement of existence of both a tangent space and a transversal space is
not pointless, because there are subgroups in Hn , as the T axis for example, that is,
T = {p = (p1 , . . . , p2n+1 ) ∈ R2n+1 : p1 = · · · = p2n = 0}, without a complementary
subgroup, i.e., a subgroup M ⊂ Hn such that M ∩ T = 0 and Hn = M · T. Finally,
the distinction between normal and transversal planes is natural, because a normal
subgroup does not necessarily exist at each point, even if a (possibly not normal
but) transversal subgroup exists. Condition (iv) entails that the tangent plane has
the natural geometric meaning of “surface seen at infinite scale”, the scale however
being meant with respect to intrinsic dilations. Notice that if S is both a Euclidean
smooth manifold and a group-regular manifold the intrinsic tangent plane is usually
different from the Euclidean one. On the other hand, there are sets, ‘bad” from the
Euclidean point of view, that behave as regular sets with respect to group dilations.
It is natural to check whether requirements (i)–(iv) are met by the classes of
regular submanifolds of Hn considered in the literature.
• C1 horizontal curves: they are Euclidean C1 curves; their (Euclidean) tangent
space in a point is a 1-dimensional affine subspace contained in the horizontal
fiber through the point, hence it is also a coset of a 1-dimensional subgroup
of Hn . The normal space is the complementary subspace of the tangent space,
and it is again a subgroup. Clearly both of them depend continuously on the
point. It can also be shown (see [106, Theorem 3.5]) that the Euclidean tangent
lines are also limits of group dilation of the curve, so that they are also tangent
in the group sense.
• Legendre submanifolds: they are n-dimensional, hence maximal-dimensional,
integral manifolds of the horizontal distribution (see [19]). The tangent spaces
are n-dimensional affine subspaces of the horizontal fiber that are also cosets
of subgroups of Hn . The complementary affine subspaces are the normal sub-
groups. As before the tangent spaces are limits of intrinsic dilations of the
surface (see [106, Theorem 3.5]).
• H-regular hypersurfaces: we recall that, locally, they are given as level sets
of C1H (Hn ) functions from Hn to R, with P-differential of maximal rank (see
Definition 4.20). We showed before that H-regular hypersurfaces have a natu-
ral normal space (i.e., the span of the horizontal normal vector) at each point,
hence they are cosets of a 1-dimensional subgroup contained in the horizontal
fiber; that the natural tangent space is a subgroup obtained as the limit of in-
trinsic dilations of the surface; and finally, notwithstanding that these surfaces
www.Ebook777.com
54 Francesco Serra Cassano
In conclusion, all the surfaces in these examples are intrinsically regular subman-
ifolds in the sense that they satisfy requirements (i)–(iv). Notice that C1 horizon-
tal curves have topological dimension 1, Legendre submanifolds have topological
dimension n, and H-regular hypersurfaces have topological dimension 2n (the sys-
tematic specification “topological” is not pointless, because, as noticed before, other
dimensions play a role in the geometry of Carnot groups). Our aim now is to fill the
picture, finding other classes of intrinsically regular submanifolds of arbitrary topo-
logical dimension. Notice that, from the analytical point of view, horizontal curves
and Legendre surfaces are given locally as images in Hn respectively of intervals
I ⊂ R or of open sets in Rn through P-differentiable maps with injective differen-
tials. On the other hand, 1-codimensional H-regular surfaces are given locally as
level sets of P-differentiable functions with surjective differentials. The first idea
that comes to mind, and the one we take here, is to generalize both these approaches.
Notice that, even if in the Euclidean setting they are fully equivalent, this is no longer
true in Heisenberg groups. Thus we define
S ∩ U = φ(V ).
and the cone at a point p is given as TanH ( A, p) := τp TanH (τ−p ( A), 0).
Theorem 4.47 ([106, Theorem 3.5]). Let S be a k-dimensional H-regular surface,
1 ≤ k ≤ n.
(i) S is a Euclidean k-dimensional submanifold of R2n+1 of class C1 .
(ii) Let Tan S denote the Euclidean tangent bundle of S. Then
k
(iii) S∞ S is comparable with Hk S.
Remark 4.48. It is easy to check that low-dimensional H-regular surfaces are graphs
because they are Euclidean C1 submanifolds and because low-dimensional intrinsic
graphs in Hn turn out to be Euclidean graphs.
Definition 4.49. Let S = {x : φ(x) = 0} be a k-codimensional H-regular surface
with φ ∈ C1H (U , Rk ) and 1 ≤ k ≤ n and let p ∈ S. The tangent group to S at p,
g
denoted TH (S, p), is the subgroup defined as
g
TH (S, p) := ker d P φ(p).
S ∩ U = {Φ(ξ) := ξ · φ(ξ) : ξ ∈ V } .
(iii) ([106, Theorem 4.1]) By the previous claim (i) and with the notation therein,
there exists a positive continuous function ζ ∈ C0 (U ), depending only on f
which can be explicitly represented (see [106, formula (48)]), such that
Z
C∞2n+2−k
(S ∩ U ) = ζ ◦ Φ d H2n+1−k . (4.28)
V
ous Magnani example and according to Remark 4.38, the centered Hausdorff mea-
2n+2−k has to be used in place of the spherical one.
sure C∞
that an H-regular surface stays H-regular after a (left) translation (see the definition
given in [107]). We propose here a geometric definition. We say that f : M → H
is intrinsic Lipschitz continuous if, at each p ∈ graph ( f ), there is an (intrinsic)
closed cone with vertex p, axis H, and fixed opening, intersecting graph ( f ) only in
p. The equivalence of this definition and others, more algebraic, is the content of
Proposition 4.56. Notice also that M and H are metric spaces, being subsets of G,
hence it makes sense to speak also of metric Lipschitz continuous functions from M
to H. As usual, f : M → H is said to be (metric) Lipschitz if there is L > 0 such
that, for all m, m 0 ∈ M,
The notions of intrinsic Lipschitz continuity and of Lipschitz continuity are differ-
ent ones (see Example 4.58) and we will try to convince the reader that intrinsic
Lipschitz continuity seems more useful in the context of functions acting between
subgroups of a Carnot group.
CM,H (q, β) = q · CM,H (e, β), where CM,H (e, β) = {p : kpM k∞ ≤ βkpH k∞ } .
Remark 4.52. If 0 < β1 < β2 , CM,H (q, β1 ) ⊂ CM,H (q, β2 ), and CM,H (e, 0) = H.
Moreover, for all t > 0,
The Lipschitz constant of f in E is the infimum of the L > 0 such that (4.30) holds.
An intrinsic Lipschitz (continuous) function, with Lipschitz constant not exceeding
L > 0, is called an L-Lipschitz function.
We will call a set S ⊂ G an intrinsic Lipschitz graph if there exists an intrinsic
Lipschitz function f : E ⊂ M → H such that S = graph ( f ) for suitable comple-
mentary subgroups M and H. We will call the topological dimension of M and H,
respectively, the dimension and codimension of graph ( f ).
Because of Proposition 4.17 and Definition 4.51, left translations of intrinsic
Lipschitz H-graphs, or of intrinsic L-Lipschitz functions, stay intrinsic Lipschitz
H-graphs, or intrinsic L-Lipschitz functions. We state these facts in the following
theorem.
Theorem 4.55. Let G be a Carnot group; then for all q ∈ G, f : E ⊂ M → H is
intrinsic L-Lipschitz if and only if f q : Eq ⊂ M → H is intrinsic L-Lipschitz,where
f q is the function defined in (4.8).
The geometric definition of intrinsic Lipschitz graphs has equivalent algebraic
forms (see also [18, 107, 98]).
Proposition 4.56. Let M, H be complementary subgroups in G, f : E ⊂ M → H and
L > 0. Then(i) to (iv) are equivalent.
(i) f is intrinsic L-Lipschitz in E .
Proof. The equivalence between (i) and (ii) follows from the definition (4.51), ob-
serving that if q̄ ∈ graph ( f ), then CM,H ( q̄, 1/L) ∩ graph ( f ) = { q̄} is equivalent to
CM,H (e, 1/L) ∩ graph ( f q̄ −1 ) = {e}.
The equivalence of (ii) and (iii) follows once more from the definition of a cone and
from the left invariance of the definition. Indeed, we recall that
−1
f q̄ −1 (m) = ( f ( m̄)m)H
f m̄( f ( m̄)m)M ;
then
Changing variables, (iv) follows from (iii). Let m2 := ( f ( m̄)m)M , that is, m =
( f ( m̄) −1 m2 )M ; then
−1
· f m̄( f ( m̄)m)M = f ( m̄)( f ( m̄) −1 m2 )M −1
f ( m̄)m H H · f m̄m2
= f ( m̄) f ( m̄) −1 m2 f ( m̄) −1 m2 −1
−1
H H · f m̄m2
= f ( m̄) −1 m2 H · f m̄m2 .
1 −1
k q̄−1 q k q̄ qk∞ − k q̄−1 q M k∞ ≤ L k q̄−1 q M k∞ .
H k∞ ≤
c0
Conversely,
Hence in this case intrinsic Lipschitz functions are the same as the usual metric
Lipschitz functions from (M, d ∞ ) to (H, d ∞ ).
Example 4.58 ([18, Example 3.21]). Here we show that condition (4.29) is not in-
variant under left translations of the graph. It follows that neither intrinsic Lips-
chitz continuity implies (metric) Lipschitz continuity nor the opposite. Let G =
H1 = W · V where V := {v = (v1 , 0, 0)} and W = {w = (0, w2 , w3 )}. Notice that
kwk∞ = max{|w2 |, |w3 | 1/2 }, if w ∈ W and kvk∞ = |v1 | if v ∈ V.
(i) φ : W → V, defined as φ(w) := (1 + |w3 | 1/2 , 0, 0), satisfies (4.29) with L = 1,
hence φ is Lipschitz. On the other hand, φ is not intrinsic Lipschitz. Indeed,
let p := (1, 0, 0) ∈ graph (()φ); from Proposition 4.17 we have φ p −1 (w) =
(|w2 + w3 | 1/2 , 0, 0). For φ p −1 , Proposition 4.56(iii) does not hold. This shows
also that condition (4.29) is not invariant under graph translations.
(ii) ψ : W → V, defined as ψ(w) := (1 + |w3 − w2 | 1/2 , 0, 0), is intrinsic Lipschitz;
indeed, with p = (1, 0, 0) and φ(w) := (|w3 | 1/2 , 0, 0) we have ψ(w) = φ p (w),
so that ψ is intrinsic Lipschitz because φ is intrinsic Lipschitz. On the other
hand, ψ is not Lipschitz, in the sense of (4.29), as can be easily observed.
Analogously,
(iii) the constant function φ : V → W defined as φ(v) := (0, 1, 0) for all v ∈ V, is
Lipschitz but it is not intrinsic Lipschitz;
(iv) ψ : V → W defined as ψ(v) := (0, 1, −v1 ) for all v ∈ V, is intrinsic Lipschitz
continuous but it is not Lipschitz.
It is natural to ask whether intrinsic Lipschitz functions are metric Lipschitz func-
tions provided that appropriate choices of the metrics in the domain or in the target
spaces are made. The answer is almost always negative. Nevertheless, something
relevant can be stated (see [98, Remark 3.6]).
Free ebooks ==> www.Ebook777.com
However, we will also refer to the function % f as a quasi distance for a general
function f : E ⊂ M → H, even if it is not intrinsic Lipschitz. Observe that in this
case % f still satisfies (QD1 ) and (QD2 ).
The quasi distance % f , introduced in [16] in the setting of Heisenberg groups,
has been called graph distance since in this case it is equivalent to the metric d ∞
restricted to graph ( f ). More precisely, let us define the parametric graph function
Φ f induced through f ,
where f : E ⊂ M → H.
1
% f (m, m̄) ≤ d ∞ Φ f (m), Φ f ( m̄) ≤ C % f (m, m̄) ∀m, m̄ ∈ E . (4.33)
C
Proof. The fact that the map Φ f : (E , % f ) → (G, d ∞ ) is a metric Lipschitz function,
that is, the right inequality, has been proved in [98]. The left one immediately follows
by Proposition 4.8.
www.Ebook777.com
62 Francesco Serra Cassano
Since
k Φ f ( m̄) −1 Φ f (m) k∞ ≤ 2 % f (m, m̄) ∀m, m̄ ∈ E ,
M
we get (4.34) by choosing L 2 = 2L 1 .
(ii) ⇒ (i): We will first need a useful estimation of the quantity % f which holds
for a general f : E ⊂ M → H (even if not intrinsic Lipschitz). More precisely, let
us prove that there exist two positive constants Ci = Ci (G) (i = 1, 2) such that
1/κ (κ−1)/κ 1/κ (κ−1)/κ
% f (m, m̄) ≤ C1 kpk∞ + C2 kpk∞ kqk∞ + kqk∞ kpk∞ (4.36)
if p = Φ f ( m̄) −1 Φ f (m) and q = Φ f ( m̄) −1 Φ f (m) for all m, m̄ ∈ E , where
M H
κ denotes the group step. Notice that, given g ∈ G, if g = gM gH , then g −1 =
(g −1 )M (g −1 )H , and g̃ := (g −1 )M (g −1 )H gH . Then we get
By (4.1), (4.3), and (4.37), it follows that there exist two positive constants C = C(G)
and c0 = c0 (G) such that
1 1 −1 −1
k(g −1 )M k∞ ≤ k g̃k∞ = kg g gH k∞
c0 c0 H M
1 C 1/κ (κ−1)/κ 1/κ (κ−1)/κ
≤ kgM k∞ + kgM k∞ kgH k∞ + kgH k∞ kgM k∞ .
c0 c0
(4.38)
0 +1
By choosing g = Φ f ( m̄) −1 Φ f (m) in (4.38), (4.36) follows with C1 = c2c 0
and
C2 = 2c0 . Notice now that, by (4.36), applying Young’s inequality in a standard way,
C
for all m, m̄ ∈ E , ε > 0. By (4.34) and (4.39), by choosing ε = ε(L 2 ) small enough,
there exists a positive constant C4 = C4 (G, L 2 ) such that
% f (m, m̄) ≤ C4 k Φ f ( m̄) −1 Φ f (m) k∞ ∀m, m̄ ∈ E . (4.40)
M
Remark 4.61. From Propositions 4.8, 4.59, and Theorem 4.60 it follows that f :
E ⊂ M → H is intrinsic L-Lipschitz in E if and only if (4.33) holds.
We stress that in general it is impossible to find a unique quasi distance that
works for all intrinsic Lipschitz functions. Notice that this is possible exactly when
H is a normal subgroup.When M is a normal subgroup then not only Φ f but f it-
self is a metric Lipschitz function from (M, % f ) → (H, d ∞ ). Indeed, the following
proposition holds.
Proof. (i) This equivalence has already been proved in [98, Proposition 3.7]. On the
other hand, it is an immediate consequence of Remark 4.61, noticing that, since H is
normal,
Φ f ( m̄) −1 Φ f (m) = m̄−1 m.
M
(ii) Since M is a normal subgroup,
Φ f ( m̄) −1 Φ f (m) = f ( m̄) −1 m̄−1 m f ( m̄) ∀m, m̄ ∈ E , (4.43)
M
and
Φ f ( m̄) −1 Φ f (m) = f ( m̄) −1 f (m) ∀m, m̄ ∈ E . (4.44)
H
Therefore, by Theorem 4.60 the equivalence follows.
Free ebooks ==> www.Ebook777.com
Remark 4.63. If H is not a normal subgroup Proposition 4.62(i) can be false: even if
f is very regular, the “natural” parametrization of graph ( f ) given by Φ f may not be
metric Lipschitz (see [98]).
Finally we notice that it is an open problem to understand if and when metric
Lipschitz parametrizations of graph ( f ), different from the “natural” parametrization
Φ f , exist. This problem was addressed in [67] where the authors proved that, if
S is a noncharacteristic C1 hypersurface in H1 ≡ R3 , then a metric bi-Lipschitz
parametrization of S exists from the Euclidean plane R2 , endowed with the so-called
parabolic metric, which is the restriction of distance d ∞ to a vertical plane W :=
{0, y,t) : y,t ∈ R2 } ≡ R2 . On the other hand, it was proved in [42] that there are
no bi-Lipschitz parametrizations for a given H-regular surface in H1 , defined on an
open set of R2 , endowed with the parabolic metric.
We conclude this section stressing that intrinsic Lipschitz functions, even if non-
metric Lipschitz, nevertheless are metric Hölder continuous.
Proposition 4.64 ([98, Proposition 3.8]). Let M, H be complementary subgroups in
a step-κ group G. Let L > 0 and f : E ⊂ M → H be an intrinsic L-Lipschitz
function. Then
(i) f is bounded on bounded subsets of E : precisely, for all R > 0 there is C1 =
C1 (M, H, f , R) > 0 such that
www.Ebook777.com
1 Some topics of geometric measure theory in Carnot groups 65
this stronger statement (stronger in the sense that it holds also for lower-dimensional
domains E) is false in general as the following easy example in [98] shows: in the
Heisenberg group H1 := (R3 , ·) let V and W be the complementary homogeneous
subgroups defined as V := {v = (v1 , 0, 0) : v1 ∈ R} and W := {w = (0, w2 , w3 ) :
w2 , w3 ∈ R}. Let f : W → V be the intrinsic Lipschitz function (constant) defined
as w = (0, w2 , w3 ) 7→ f (w) := (1, 0, 0). Let E := {(0, w2 , 0) : w2 ∈ R}. Since E is a
horizontal curve, its metric dimension equals 1. But graph ( f |E ) = {(1, w2 , −w2 /2) :
w2 ∈ R} is not a horizontal curve anymore, hence its metric dimension is larger than
1, indeed it equals 2.
Notice that the proofs of upper and lower bounds on the Hausdorff measure of
a Lipschitz graph are trivially true in Euclidean spaces. Indeed, if f : Rk → Rn−k
is Lipschitz then the map Φ f : Rk → Rn defined as Φ f (x) := (x, f (x)) is a Lips-
chitz parametrization of the Euclidean graph of f ; this gives the upper bound on the
dimension of the graph. On the other hand, the projection Rn ≡ Rk × Rn−k → Rk
is Lipschitz continuous, with Lipschitz constant 1, yielding the lower bound. Such
a proof cannot work here. On the one hand, the projections PM or PH are not Lip-
schitz continuous (see Example 4.12 and [24]), on the other hand, the “natural”
parametrization Φ : M → G, Φ f (m) := m · f (m) is almost never a Lipschitz con-
tinuous map between the two metric spaces M and G (see [98, Remark 3.6]). Notice
that the following result contains a stronger statement: it states that a Lipschitz graph
parametrized on a homogeneous subgroup of dimension d m is (locally) Alhfors d m -
regular.
Let us point out that, in the particular case of a Heisenberg group Hn , an explicit
characterization of intrinsic linear functionals can be provided, in terms of an H-
linear functional. Indeed, we have
kd f m̄ (m) −1 · f p̄ −1 (m)k∞
(i) lim km k∞ →0 = 0.
kmk∞
The intrinsic linear map d f m̄ is called the intrinsic differential of f at m̄.
Writing explicitly the expression for f p̄ −1 (see (4.8)) in (i), we can equivalently state
that f is intrinsic differentiable at m̄ ∈ A if
(ii) ⇒ (i): It is sufficient to prove that there exist two positive constants C0 and r 0
such that
Let us first prove that there exist two positive constants C1 and r 0 such that
f ( m̄) −1 f (m)
−1
= d f m̄ f ( m̄) −1 m̄−1 m f ( m̄) d f m̄ f ( m̄) −1 m̄−1 m f ( m̄) · f ( m̄) −1 · f (m) .
Thus, by Proposition 4.69(ii) and (4.48), (4.50) follows. Now applying (4.39),
(4.43), (4.44), and (4.50), there exists a suitable positive constant C0 such that (4.49)
holds.
Remark 4.77. Let us recall Pansu’s differentiability (see Definition 3.1): a function
f , acting between two nilpotent groups G1 and G2 , is P-differentiable at ḡ ∈ G1 if
there is a homogeneous homomorphism L : G1 → G2 such that
(ii) d f m̄ depends continuously on m̄, that is, for each compact K ⊂ A, there is
η = η K : [0, +∞) → [0, +∞), with limt→0+ η(t) = 0 such that, for each m̄1 ,
m̄2 ∈ A,
sup kd f m̄1 (m) −1 · d f m̄2 (m)k∞ ≤ η k m̄1−1 · m̄2 k∞ ;
m ∈K
(iii) for each compact K ⊂ A, there is ε = ε A, K : [0, +∞) → [0, +∞), with
limt→0+ ε(t) = 0 and such that for all m ∈ K p̄ −1 , and for all m̄ ∈ K,
where p̄ = m̄ · f ( m̄).
Proof. Let p̄ = m̄ · f ( m̄) ∈ graph ( f ). Since A is relatively open there exists r > 0
such that K := M ∩ B( m̄,r) ⊂ A. Observe now that
and, by Propositions 4.68(i) and 4.69(ii), there exists a positive constant C1 > 0 such
that
kd f m̄ (m)k∞ ≤ C1 kmk∞ ∀m ∈ M ∩ B(e,r). (4.52)
Since
f p̄ −1 (m) = d f m̄ (m) · d f m̄ (m) −1 · f p̄ −1 (m),
then by the triangle inequality, (4.52), (4.51), Definition 4.81(iii), and Proposition
4.56(iii), we get the desired conclusion.
www.Ebook777.com
72 Francesco Serra Cassano
they are not kernels of homogeneous homomorphisms, while they are tangent spaces
to low-dimensional submanifolds; on the other hand, injective homogeneous homo-
morphisms Rd → Hn do not exist, if d ≥ n + 1 (see Remark 4.45).
From Proposition 4.82 and Theorem 4.83 we have
Corollary 4.86. Each H-regular surface S ⊂ Hn is locally a graph of an intrinsic
Lipschitz function.
An interesting result as far as the calculus within Carnot groups is concerned is a
Rademacher-type theorem for 1-codimensional Lipschitz graphs within the Heisen-
berg group. More precisely, we have
Theorem 4.87 ([108, Theorem 4.29]). Let Hn = W · V with V a (1, 1)-dimensional
subgroup. If U ⊂ W is relatively open, and f : U → V is intrinsic Lipschitz, then f
is intrinsic differentiable in U , a.e. with respect to the Haar measure of W.
Remark 4.88. It is an open problem whether a Rademacher-type theorem still holds
for higher-codimensional Lipschitz graphs in Heisenberg groups. For instance, as-
sume that H2 ≡ R5 = W · V with W := exp(span{Y1 ,Y2 ,T }) = {(0, 0, w3 , w4 , w5 )}
and V := exp(span{X1 , X2 }) = {(v1 , v2 , 0, 0, 0)} complementary subgroups, accord-
ing to the notation of Example 2.2. Note that the subgroup W is isomorphic to the
first Heisenberg group H1 ≡ R3 and V to R2 . Given an intrinsic Lipschitz map
f = ( f 1 , f 2 , 0, 0, 0) : W → V it is unknown whether it is intrinsic differentiable a.e.
with respect to the Haar measure of W.
Remark 4.89. Franchi, Marchi, and Serapioni recently extended Theorem 4.87 to
1-codimensional intrinsic Lipschitz graphs within a more general class of Carnot
groups called type ?, which contains step-2 Carnot groups (see Definition 5.23 and
[96]).
For the sake of simplicity, we will restrict to the first Heisenberg group H1 in
order to introduce the two remaining characterizations. Moreover, we will consider
the so-called X1 -graphs in H1 , that is, we will consider H1 ≡ R3 = W · V with W :=
exp(span{Y1 ,T }) = {(0, y,t)} and V := exp(span{X1 }) = {(x, 0, 0)} complementary
subgroups, according to the notation of Example 2.2. Note that subgroup W is a
normal subgroup, isomorphic to the (abelian) group R2 and V to R. Moreover, each
function φ : A ⊂ W → V can be meant as a function φ(0, y,t) = ( f (y,t), 0, 0) with
f : A ⊂ R2 → R, that is, we can identify φ with f and a point w = (0, y,t) ∈ W
with a point A = (y,t) ∈ R2 .
www.Ebook777.com
74 Francesco Serra Cassano
α L ∈ R (see Proposition 4.70 and [16, Proposition 2.15]); thus the notion
of intrinsic differentiability for a function φ : A ⊂ W → V (see Definition
4.71) can be equivalently stated in terms of the quasi distance %φ (see (4.31)
and Proposition 4.76) as follows: φ is intrinsic differentiable at a point w̄ =
(0, ȳ, t¯) ∈ A (or we will also say that f is intrinsic differentiable at Ā = ( ȳ, t¯))
if and only if there exists a unique α = α( Ā) ∈ R such that
f ( A) − f ( Ā) − α (y − ȳ)
lim =0 (4.53)
A→ Ā %̃ f ( A, Ā)
where A = (y,t),
1 (4.54)
= max{|y − ȳ|, σ f ( A, Ā)} + max{|y − ȳ|, σ f ( Ā, A)}
2
and
σ f ( A, Ā) := |t − t¯ − f ( Ā)(y − ȳ)| 1/2 ; (4.55)
we will call α( Ā) the ∇ f -gradient of f at Ā and will denote it by ∇ f f ( Ā).
d f (γ(s)) − f (γ(0))
∃ f (γ(s))|s=0 = lim =α
ds s→0 s
and we define α = ∂ f f ( Ā) (see [39, Definition 2.11]). It can be proved that,
if f is intrinsic differentiable at a given Ā ∈ A and given an integral curve
γ : (−δ, δ) → A of ∇ f with γ(0) = Ā for which (δ, δ) 3 s → f (γ(s)) is of
class C1 , then there exists ∂ f f ( Ā) = ∇ f f ( Ā) (see [16, Proposition 3.7]).
∇f f = g in A, (4.56)
(B.1) ĝ = g L2 -a.e. in A;
(B.2) for every integral curve γ ∈ C1 ((−δ, δ); A) of the vector field ∇ f it
holds that (−δ, δ) 3 s 7→ f (γ(s)) is absolutely continuous and
d
f (γ(s)) = ĝ(γ(s)) a.e. s ∈ (−δ, δ);
ds
Theorem 4.90 ([16, Theorem 5.1], [41, Theorem 1.2]). Let H1 = W · V, with W =
{(0, y,t)} ≡ R2 and V = {(x, 0, 0)} ≡ R and let φ = ( f , 0, 0) : A ⊂ W → V, where
A is a relatively open set in W. Then the following are equivalent:
(i) S = graph (φ) = {w · φ(w) : w ∈ A} is an H-regular hypersurface;
Remark 4.91. Notice that a distributional continuous solution f of (4.56) with con-
tinuous source g can be very irregular. For instance, it may occur that f < BVloc (A)
(see [139]).
76 Francesco Serra Cassano
Theorem 4.92 ([62], [39, Theorem 1.1]). Let H1 = W · V, with W = {(0, y,t)} ≡ R2
and V = {(x, 0, 0)} ≡ R and let φ = ( f , 0, 0) : A ⊂ W → V, where A is a relatively
open set in W. Then the following are equivalent:
(i) φ is (locally) intrinsic Lipschitz;
(ii) there exists a measurable (locally) bounded function g : A → R such that f
is a broad solution of the balance law (4.56);
(iii) there exists a measurable (locally) bounded function g : A → R, such that f
is a distributional continuous solution of the balance law (4.56).
Example 4.93. In H1 , with the notation of Theorem 4.90, fix 1/2 < α < 1 and
consider φ : W → V, φ(0, y,t) = (|t| α , 0, 0). Then, since f (y,t) = |t| α is a
distributional continuous solution of (4.56) with g(y,t) = 2α |t| 2α−2 t, then S =
graph (φ) is an H-regular hypersurface. Moreover, one can easily check that S is
not a Euclidean graph in any neighborhood of the origin (see [106, Example 3.8]).
Arguing in the same way, if α = 1/2, by Theorem 4.92, S turns out to be an intrinsic
(locally) Lipschitz graph.
Remark 4.94. The approximation, by means of regular functions, and a Poincaré
inequality for 1-codimensional intrinsic Lipschitz graphs in Heisenberg groups has
been obtained, respectively, in [62] and [63].
As a by-product of Theorems 4.90 and 4.92 we can give a characterization of H-
regular graphs in terms of a ∂ f -derivative and a ∇ f -gradient, similar to the Euclidean
one for C1 graphs.
Theorem 4.95. Let H1 = W · V, with W = {(0, y,t)} ≡ R2 and V = {(x, 0, 0)} ≡ R
and let φ = ( f , 0, 0) : A ⊂ W → V, where A is a relatively open set in W. Then the
following are equivalent:
(i) S = graph (φ) = {w · φ(w) : w ∈ A} is an H-regular hypersurface;
(ii) for f ∈ C0 (A), there exists ∂ f f ( A) for each A ∈ A and ∂ f f : A → R is
continuous;
(iii) f is intrinsic differentiable in A and ∇ f f : A → R is continuous.
Moreover, in cases (ii) and (iii), ∂ f f ( A) = ∇ f f ( A) for each A ∈ A.
Proof. (ii) ⇒ (i): Let γ : (a, b) → A be an integral curve of the vector field ∇ f .
Then the function (a, b) 3 s → f (γ(s)) is of class C1 . Indeed, since there exists
∂ f f ( A) for each A ∈ A and A 3 A 7→ ∂ f f ( A) is continuous, it follows that
d
∃ f (γ(s)) = ∂ f f (γ(s)) ∀s ∈ (a, b)
ds
Free ebooks ==> www.Ebook777.com
Again applying Theorem 4.90(ii), we can pass to the limit as ε → 0+ in the previous
identity and we get that
Z s
f (γ(s)) − f (γ(0)) = ∇ f f (γ(r)) dr ∀s ∈ (−δ, δ). (4.58)
0
∃ ∂ f f ( A) = ∇ f f ( A) ∀A ∈ A (4.59)
Observe that, since φ is intrinsic differentiable at (0, ȳ, t¯), there exist two positive
constants C1 and r 0 such that
www.Ebook777.com
78 Francesco Serra Cassano
Arguing as in the proof of Proposition 4.76 (see (4.49)), by (4.61) there exists
C0 > 1 such that
kφ( m̄) −1 m̄−1 mφ( m̄)k∞ ≤ % f (m, m̄) ≤ C0 kφ( m̄) −1 m̄−1 mφ( m̄)k∞ (4.62)
for each m ∈ B∞ ( m̄,r 0 ) ∩ W. We can also read (4.61) and (4.62) in terms of f ,
respectively, as follows:
γ : [−δ1 , δ1 ] → BW ( Ā,r 0 ) ⊂ A;
moreover, by (4.63),
Z s 1/2
σ f (γ(s), Ā) = f (γ(z)) − f ( Ā) dz
0
! 1/2
≤ C01/2 |s| 1/2 max %̃ f (γ(s), Ā) (4.66)
s ∈[−δ 1,δ 1 ]
1
≤ 2C0 |s| + max %̃ f (γ(s), Ā) ∀s ∈ [−δ1 , δ1 ].
2 s ∈[−δ 1,δ 1 ]
Remark 4.97. It is also well known that, in the Euclidean setting (that is, when X =
Rn ), the countable k-rectifiability can be also expressed in terms of k-dimensional
surfaces, by the Whitney extension theorem. Namely, it holds that E is countably
Hk -rectifiable if and only if there is a sequence of k-dimensional C1 surfaces (Si )i ⊂
Rn such that Hk (E \ ∪∞ i= Si ) = 0.
Definition 4.98. A metric space (X, d) is said to be purely k-unrectifiable if for each
Lipschitz function f : A ⊂ (Rk , k·kRk ) → (X, d),
Hdk ( f ( A)) = 0.
Theorem 4.99 ([10, Theorem 7.2]). The first Heisenberg group (H1 , d) is purely
k-unrectifiable for k = 2, 3, 4, for each invariant distance d.
Theorem 4.100 ([156, Theorem 1.1]). Let G be a Carnot group and let V1 denote
the first horizontal layer in its stratification (see (2.1)). Then (G, d) is purely k-
unrectifiable iff V1 does not contain a (trivial) Lie subalgebra of dimension k.
Therefore, taking the previous unrectifiability results into account, a new, suitable
notion of rectifiability in Carnot groups is needed, better fitting the new geometry.
setting of Heisenberg groups [102] and then step-2 Carnot groups for studying the
structure of sets of finite G-perimeter [105].
We recall that De Giorgi’s celebrated structure theorem in Euclidean spaces ([74,
75]) states that if E ⊂ Rn is a set of locally finite perimeter, then the associated
perimeter measure |∂E| is concentrated on a portion of the topological boundary ∂E,
the so-called reduced boundary ∂ ∗ E ⊂ ∂E. In addition, ∂ ∗ E is H n−1 -rectifiable, i.e.,
∂ ∗ E, up to a set of (n − 1)-Hausdorff measure zero, is a countable union of compact
subsets of C1 submanifolds and the perimeter measure is the (n − 1)-Hausdorff mea-
sure of the reduced boundary. Roughly speaking, this says that the perimeter measure
is supported on a portion of the topological boundary ∂E, that can be expressed—
after removing a negligible set of “bad points”—as the countable union of compact
subsets of “good hypersurfaces”.
If in the spirit of De Giorgi’s theorem we want to describe the structure of sets of
finite intrinsic perimeter in a Carnot group G, we need a natural notion of rectifiable
subsets, and in this perspective, the correct definition of “good hypersurfaces”, i.e.,
of intrinsic C 1 -regular submanifolds of G given in Definition 4.20. Keeping in mind
this notion, the following definition is the natural counterpart of the corresponding
Euclidean definition (see [103] and [96]).
Definition 4.101. Let G be a Carnot group and let Q denote its homogeneous di-
mension.
(i) E ⊂ G is said to be (Q−1)-dimensional G-rectifiable if there exists a sequence
of G-regular hypersurfaces (Si )i such that
∞
[
Q−1
S∞ E\ Si = 0. (4.67)
i=1
Remark 4.102. Franchi, Marchi, and Serapioni recently proved that, for Carnot
groups of type ? (see Definition 5.23), (Q − 1)-dimensional G-rectifiability and
(Q − 1)-dimensional G L -rectifiability are equivalent (see [96, Proposition 4.4.4]).
Before we enter the study of the rectifiability of the reduced boundary (whatever
this means, as we shall see below), let us point out the relationships between our
definition in Carnot groups and the standard Euclidean notion. The following result
proved in [105] yields that “negligible” subsets of codimension 1 in a Carnot group
Free ebooks ==> www.Ebook777.com
with respect to the Euclidean distance are “negligible” subsets of codimension 1 with
respect to Carnot–Carathéodory distance.
Proposition 4.103 ([105, Proposition 4.4]). Let G = (Rn , ·) be a Carnot group.
For any α ≥ 0 and R > 0 there is a constant c(α, R) > 0 such that for any set
E ⊂ G ∩ U∞ (0, R),
α+Q−n
H∞ (E) ≤ c(α, R) Hα (E), α ≥ 0. (4.69)
In particular, for all E ⊂ G,
α+Q−n
Hα (E) = 0 =⇒ H∞ (E) = 0, α ≥ 0. (4.70)
Proposition 4.103 combined with Theorem 4.23 yields
Theorem 4.104. Let G = (Rn , ·) be a Carnot group. Then, if S is an (n − 1)-dimen-
sional Euclidean rectifiable set of Rn then S is also (Q−1)-dimensional G-rectifiable.
On the other hand, there are (Q − 1)-dimensional G-rectifiable sets in Carnot
groups G = (Rn , ·) that are not (n − 1)-dimensional Euclidean rectifiable. Indeed, in
[28] a set N ⊂ R3 is constructed, such that for an appropriate ε > 0,
3
H∞ (N ) = 0 and H2+ε (N ) > 0.
Hence N is (trivially) 3 = (Q − 1)-dimensional H1 -rectifiable, but it is not 2-dimen-
sional Euclidean rectifiable because its Euclidean Hausdorff dimension is strictly
larger than 2. As we mentioned above, a sharper result in this direction is contained
in [139]: there exist G-regular hypersurfaces in the Heisenberg group H1 (Q = 4,
n = 3) with Euclidean Hausdorff dimension 2.5. We recall here that relationships
between Euclidean and intrinsic Hausdorff measures in Heisenberg groups have been
deeply investigated in [28], where sharp results were also obtained. This study was
extended more recently to a general Carnot group in [30].
Finally let us stress that each H-regular t-graph in Hn ≡ R2n+1 , that is, each
graph with respect to the subgroup T = {(0, . . . , 0, p2n+1 )}, keeps the same Euclidean
metric dimension. Indeed, we have
Theorem 4.105 ([213, Corollary 1.2]). Let S ⊂ Hn be an H-regular hypersurface
that is not (Euclidean) countably H2n -rectifiable; then S is not a t-graph. In partic-
ular, each H regular t-graph has Euclidean metric dimension 2n.
www.Ebook777.com
82 Francesco Serra Cassano
i=1 f i ( Ai )) = 0;
(E \ ∪∞
k
S∞
i=1 Si ) = 0,
(E \ ∪∞
k+1
S∞
where Wi and Vi are, respectively, (k, k + 1)- and (2n + 1 − k, 2n + 1 − k)-
subgroups, which are complementary.
Remark 4.107. We recall that, by a combination of Proposition 4.103 with Theorem
4.23, for n + 1 ≤ k ≤ 2n, a k-dimensional Euclidean rectifiable set E ⊂ R2n+1 ≡ Hn
is always a (k, H)-rectifiable set, while the converse is false. On the other hand, for
1 ≤ k ≤ n, a (k, H)-rectifiable set is k-dimensional Euclidean rectifiable while the
opposite is false: for instance, the vertical T axis in H1 ≡ R3 gives the simplest
example of a Euclidean 1-dimensional rectifiable set that is not 1-dimensional H-
rectifiable.
Remark 4.108. Definition 4.106(i) is nontrivial only if 1 ≤ k ≤ n, because, for
n+1 ≤ k and f : A ⊂ Rk → Hn Lipschitz, then S∞ k ( f ( A)) = 0 always (see Theorem
km
Theorem 4.110. Let E ⊂ Hn be a set which has locally finite S∞ measure.
(i) ([170, Theorem 3.14]) Let 1 ≤ k ≤ n. Then E is (k, H L )-rectifiable if and
k a.e. The latter means
only if it has an approximate tangent subgroup T p , S∞
k
that, for S∞ a.e. p ∈ E, there exists a homogeneous subgroup T p , of topolog-
ical and metric dimension k, such that
( )
S∞k E ∩ B(p,r) ∩ q : d (p−1 · q, T ) > s d (p, q)
∞ p ∞
lim = 0,
r →0 rk
for each s > 0.
(ii) ([170, Theorem 3.15]) Let n + 1 ≤ k ≤ 2n. Then E is (k, H)-rectifiable if
k a.e. The latter
and only if it has an approximate tangent subgroup T p , S∞
k
means that, for S∞ a.e. p ∈ E, there exists a homogeneous subgroup T p , of
topological dimension k and metric dimension k + 1, such that
( )
k+1 E ∩ B(p,r) ∩ q : d (p−1 · q, T ) > s d (p, q)
S∞ ∞ p ∞
lim = 0,
r →0 r k+1
for each s > 0, and
k+1 (E ∩ B(p,r))
S∞
lim inf > 0. (4.71)
r →0 r k+1
Remark 4.111. We point out that some other notions of rectifiability in Carnot groups
were introduced by Pauls [193] and, recently, by Ghezzi & Jean [121]. However, to
our knowledge, there is no comparison among these notions of rectifiability.
www.Ebook777.com
84 Francesco Serra Cassano
introduced it in the pioneering papers [74, 75], strongly inspired by some previous
ideas of Caccioppoli (see [7] for an interesting account of Euclidean sets of finite
perimeter). Later, De Giorgi [78] successfully used the sets of finite perimeter in
order to establish existence and regularity of minimal surfaces for 1-codimensional
surfaces in Rn (see [122, 8, 152]). We will apply this approach in Section 5.3 in
order to get existence of the so-called nonparametric minimal surfaces in the setting
of the simplest Carnot group, namely the Heisenberg group Hn .
Caccioppoli’s primitive idea, then refined by De Giorgi through sets of finite
perimeter, considered oriented hypersurfaces, which (at least locally) are boundaries
of sets, and exploited techniques of measure theory. Thus one of the main problems
studied by De Giorgi was the structure of the perimeter’s measure support, which is
a typical issue of geometric measure theory.
Definition 5.1. Let (X, d) be a separable metric space and let µ be a Borel outer
measure on X. The set
De Giorgi’s strategy was to single out a smaller set within the topological bound-
ary, where the perimeter measure is concentrated, which is also regular from the
viewpoint of measure theory, that is, is countably rectifiable (see Definition 4.96).
De Giorgi introduced to this goal the notion of reduced boundary, which can easily
be extended to a Carnot group.
Definition 5.5 (De Giorgi’s reduced boundary). Let d be an invariant metric on a
Carnot group G (see Definition 2.6). Let E be a G-Caccioppoli set and let νE denote
the generalized inward horizontal normal defined in (3.13); we say that a point x ∈ G
belongs to the G-reduced boundary of E, written x ∈ ∂G∗ E, if
(i) |∂E|G (Bd (x,r)) > 0 for any r > 0;
(ii) there exists limr →0 B (x,r ) νE d|∂E|G ;
R
d
where SdQ−1 denotes the (Q − 1)-dimensional spherical Hausdorff measure (see Def-
inition 2.21(ii)) and ci = ci (G) (i=1,2) are positive geometric constants.
From Theorems 5.7, 4.29, and 4.36, the following consequence immediately fol-
lows.
Corollary 5.8. Under the same assumptions as Theorem 5.7,
(i) w(x) = Θ∗FQ−1 (|∂E|G , x) for SdQ−1 -a.e. x ∈ ∂G∗ E, where Θ∗FQ−1 (|∂E|G , ·)
denotes the (Q − 1)-Federer density introduced in Definition 4.28(ii);
(ii) |∂E|G = Θ∗ Q−1 (|∂E|G , ·) CdQ−1 ∂G∗ E, where Θ∗ Q−1 (|∂E|G , ·) denotes the
(Q − 1)-spherical upper density introduced in Definition 4.28(i) and CdQ−1
denotes the (Q − 1)-dimensional centered Hausdorff measure (see Definition
2.21(iii)).
Remark 5.9. Let us recall that Θ∗FQ−1 (|∂E|G , x) and Θ∗ Q−1 (|∂E|G , x) may differ for
x ∈ ∂G∗ E, if d is a general invariant distance on G (see Remark 4.31).
Remark 5.10. The intrinsic rectifiability of the reduced boundary ∂G∗ E is still an open
problem if G is a Carnot group of step greater than 2. We will deal with this issue in
the next section.
It is well known that, if E ⊂ G := (Rn , ·) is of locally finite Euclidean perime-
ter (that is, it is a set of locally finite G-perimeter for the trivial Carnot group G =
(Rn , +)), then it is also of locally finite G-perimeter. For such a set one can rep-
resent its G-perimeter measure both with respect to Euclidean (n − 1)-dimensional
Hausdorff measure H n−1 and (sub-Riemannian) (Q − 1)-spherical Hausdorff mea-
sure SdQ−1 . This representation is well known when ∂E is regular (see Proposition
3.39) and E is a set of locally finite perimeter in G = Hn (see [213, Proposition
2.10]). We now extend the result to any Carnot group.
We denote by ∂ ∗ E and NE (P), respectively, the Euclidean reduced boundary of
E and the generalized Euclidean inward normal to E at P ∈ ∂ ∗ E, that is, according
to Definition 5.5, NE = νE when considering the trivial Carnot group G := (Rn , +).
www.Ebook777.com
1 Some topics of geometric measure theory in Carnot groups 87
then
SdQ−1 (Char(E)) = 0. (5.3)
Remark 5.12. Proposition 5.11 extends Proposition 3.39 and Theorem 4.23.
Proof. It is well known that
X j χ E = hX j , NE i|∂E| = hX j , NE iH n−1 ∂∗ E
holds in the sense of distributions for any j = 1, . . . , m1 , |∂E| being the Euclidean
perimeter of E. The first equality in (5.2) immediately follows. Moreover, from
Theorem 5.7,
|NEG | H n−1 ∂ ∗ E = |∂E|G = w SdQ−1 ∂G∗ E (5.4)
and thus the second equality in (5.2) will follow if we show that
Notice that, by (5.4), since 0 < c1 ≤ w(x) ≤ c2 for each x ∈ ∂G∗ E, we have
and Z
|NEG | d H n−1 = 0 if F1 = ∂ ∗ E \ ∂G∗ E. (5.7)
F1
From (5.6), (5.5) follows provided we show that
or
NEG = 0, H n−1 -a.e. on ∂ ∗ E \ ∂G∗ E. (5.10)
88 Francesco Serra Cassano
If (5.9) holds, because of (4.69), since d ∞ and d are equivalent, (5.8) follows at once.
Thus, assume that (5.10) holds.
Since ∂ ∗ E is locally (n − 1)-countably rectifiable in the Euclidean sense (see, for
instance, [8, Theorem 3.59]), there exists a family (S j ) j ∈N of (Euclidean) C1 surfaces
Q−1 ∗
j=0 S j ) = 0 (whence also Sd
in G such that H n−1 (∂ ∗ E \∪∞ (∂ E \∪∞j=0 S j ) = 0, again
because of (4.69) ) and
NS j being the Euclidean unit normal to S j . Theorem 4.23 ensures that for any j,
Taking into account the fact that (5.11) holds also SdQ−1 -a.e. on ∂ ∗ E ∩ S j (recall
(4.69)), we deduce that SdQ−1 {P ∈ ∂ ∗ E ∩ S j : NEG (P) = 0} = 0 for any j, i.e.,
(5.3).
(5.7) now holds with F1 =
By (5.3) and using again (5.4) as before, we get that
(∂ ∗ E \ char(E)) \ ∂G∗ E. This now implies that H n−1 (∂ ∗ E \ char(E)) \ ∂G∗ E = 0,
which in combination with (4.69) yields
Corollary 5.13. Under the same assumptions as Proposition 5.11, the G-reduced
boundary ∂G∗ E is (Q − 1)-dimensional G-rectifiable (see Definition 4.101(i)), that is,
Q−1 ∗
there exists a sequence of G-regular hypersurfaces (S ∗j ) j of G such that S∞ (∂G E \
∪ j=1 S j ) = 0.
∞ ∗
Proof. Since d and d ∞ are equivalent, it is sufficient to prove that there exists a
sequence of G-regular hypersurfaces (S ∗j ) j of G such that
SdQ−1 (∂G∗ E \ ∪∞
j=1 S j ) = 0.
∗
(5.13)
Q = m1 + 2(n − m1 ).
of G,
g
TG (v) := {x : hπ x0 x, vi x0 = 0}.
Theorem 5.14 (Blow-up theorem, [105]). Let G be a step-2 Carnot group. If E is a
G-Caccioppoli set, x 0 ∈ ∂G∗ E and νE (x 0 ) ∈ HG x0 is the inward normal as defined
in (3.13) then
lim 1 E r, x = 1SG+ (ν E (x0 )) in L 1loc (G) (5.16)
r →0 0
As we have already pointed out, Theorem 5.14 fails to hold in general Carnot
groups of step k > 2. In fact, the core of the following example consists in showing
that, in Carnot groups of step greater than 2, cones can exist (i.e., dilation-invariant
sets) that are not flat (they are not of the form SG± (v) for some horizontal vector v)
but nevertheless with vertex belonging to the reduced boundary.
The following counterexample was inspired by Martin Reimann, and then Rob-
erto Monti found a preliminary form of the counterexample itself.
Free ebooks ==> www.Ebook777.com
Example 5.15. Let us recall the definition of an Engel algebra and group. Let
E = (R4 , ·) be the Carnot group whose Lie algebra is g = V1 ⊕ V2 ⊕ V3 with
V1 = span {X1 , X2 }, V2 = span {X3 }, and V3 = span {X4 }, the only nonzero commu-
tation relations being
[X1 , X2 ] = −X3 , [X1 , X3 ] = −X4 .
Notice that the Engel algebra introduced here is different from the one introduced
in Example 4.7. However, they are isomorphic, as well as their associated Carnot
groups. This choice is motivated only for simplicity of calculation.
In exponential coordinates the group law takes the form
4
X 4
X
x·y=H x i Xi , yi X i ,
i=1 i=1
x2 x3 x1 x2 x1 x2
X1 = ∂1 + ∂3 + − ∂4 , X2 = ∂2 − ∂3 + 1 ∂4 ,
2 2 12 2 12
x1
X3 = ∂3 − ∂4 , X4 = ∂4 .
2
Let E = {x ∈ R4 : f (x) ≥ 0}, where
1 1
f (x) = x 2 (x 21 + x 22 ) − x 1 x 3 + x 4 .
6 2
Since ∂E = {x ∈ R4 : f (x) = 0} is a smooth Euclidean manifold, then E is a
G-Caccioppoli set (see Proposition 3.39). Moreover,
1
∇E f (x) = 0, (x 21 + x 22 ) ,
2
and, by the implicit function theorem (Theorem 4.24),
∇E f (x)
νE (x) = − = (0, −1)
|∇E f (x)|
for all x ∈ ∂E \ N, where N = {x ∈ E : x 1 = x 2 = 0}. Since |∂E|E (N ) = 0,
then the origin belongs to the reduced boundary of E. On the other hand, since
f (δ λ x) = λ 3 f (x) for λ > 0, it follows that Eλ,0 = δ λ E = E, so that (5.16) fails to
be true since E is not a vertical halfspace.
www.Ebook777.com
92 Francesco Serra Cassano
Even if we do not enter into the details of the proof of Theorem 5.14, we want
to stress the technical point where the assumption on the step of G is used. In the
∂f ∂f
Euclidean setting an elementary statement says that ∂x 2
= · · · = ∂x n
= 0 implies
f = f (x 1 ). In Carnot groups the corresponding statement should be that the vanish-
ing of X2 f to X m1 f yields that f is a function of just one variable. But this is false
as examples in the Heisenberg group H1 show (see Example 5.17 below). What is
possible to prove in step-2 groups is that if Y1 , . . . ,Ym1 are left-invariant smooth or-
thonormal (horizontal) sections, if Y2 f = · · · = Ym1 f = 0 and if Y1 f is positive, then
f is an increasing function of one variable. Example 5.15 shows that in groups of
step 3 or larger, even this last weaker statement is false.
Lemma 5.16 ([105, Lemma 3.6]). Let G be a step-2 group and let Y1 , . . . ,Ym1 be left
invariant smooth orthonormal sections of HG. Assume that g : G → R satisfies
Then the level lines of g are “vertical hyperplanes orthogonal to Y1 ”, that is, sets
that are group translations of
Example 5.17 ([12, Remark 5.4]). The following simple example shows that the
sign condition is essential for the validity of the conclusions of Lemma 5.16, even
in the first Heisenberg group H1 . Let p = (x, y,t) ∈ H1 , and the vector fields X1 :=
∂x − y2 ∂t and Y1 := ∂y + x2 ∂t ; the function
g(x, y,t) := γ t − 12 x y
(with γ : R → R smooth) satisfies X1 g(x, y,t) = −y γ 0 t − 12 x y and Y1 g = 0.
Therefore the sets Et := {g < t} are Y1 -invariant and are not halfspaces. The same
example can be used to show that there is no local version of Lemma 5.16, because
the sets Et locally may satisfy X1 1 E t ≥ 0 or X1 1 E t ≤ 0 (depending on the sign of
γ 0 and y), but are not locally halfspaces.
We can now state our main structure theorem for G-Caccioppoli sets.
Theorem 5.18 (Structure of G-Caccioppoli sets, [105]). Let G be a step-2 Carnot
group, endowed with the invariant distance d ∞ . Let E ⊂ G be a G-Caccioppoli set
and let ∂G∗ E denote its reduced boundary. Then
(i) |∂E|G = |∂E|G ∂G∗ E;
(ii) there exists a constant ϑ0 = ϑ0 (G) > 0 such that
Remark 5.21. Ambrosio, Kleiner, and Le Donne partially extended Theorem 5.14 in
a general Carnot group [12]. More precisely, they proved that, at each point of the
reduced boundary, a vertical halfspace belongs to the family of tangent spaces at that
point, but, a priori, the family could be not a singleton.
Remark 5.22. Marchi fully extended Theorem 5.18 to a special class of Carnot
groups, called Carnot groups of type ? (see [167]).
94 Francesco Serra Cassano
[X j , [X j , X i ]] = 0 ∀ i, j = 1, . . . , m1 .
for any open set Ω0 b Ω and any measurable F ⊂ Rn such that E∆F b Ω0.
Theorem 5.27 ([33, Theorem 5.15]). Let G = (Rn , ·) be a Carnot group. Let E, Ω
be respectively a measurable and open set of Rn , and define by νE : Ω → Rm1 the
horizontal inward normal to E in Ω. Let us assume
(i) E has locally finite G-perimeter in Ω;
1 Some topics of geometric measure theory in Carnot groups 95
νE ≡ ν0 , |∂E|G -a.e. in Ω
for a suitable constant vector ν0 ∈ Rm1 . Then, thanks to Theorem 5.27, it is straight-
forward to check that E is a minimizer for the G-perimeter.
Observe that many interesting questions, such as regularity and rectifiability, are
open even in this quite simple class of sets: see [36, 37] and also Section 5.4.
Example 5.29 (t-subgraphs in Hn , [33, Example 2.6]). Let G = Hn ≡ R2n+1 and
X = (X1 , . . . , X n , X n+1 , . . . , X2n ) = (X1 , . . . , X n ,Y1 , . . . ,Yn ) (see Example 2.2). Let
u ∈ C2 (ω) for a suitable open set U ⊂ R2n and let E be a t-subgraph, that is,
Minimal t-graphs. Let us first deal with the class of t-graphs. Let us recall that
S ⊂ Hn is called a t-graph in Hn (see Remark 4.16) if it is a graph with respect to
the nonhorizontal vector field T, namely if there exists a function u : U → R such
that
Hereafter, by U we will denote a fixed open and bounded subset of the 2n-dimen-
sional plane
When clear from the context we will canonically identify Π with R2n , and accord-
ingly we will write (x, y) instead of (x, y, 0). By U × R we will mean the t-cylinder
For maps u with Sobolev regularity, the area functional for t-graphs At : W 1,1 (U ) →
R is Z
At (u) := |∂Eut |H (U × R) = |∇u + X∗ |d L2n , (5.27)
U
We have defined
∇u + X∗
N (u) := on Unc (u), (5.29)
|∇u + X∗ |
where Unc (u) := U \ Char(u) and Char(u) is the set of characteristic points of u
defined as
Char(u) := {(x, y) ∈ U : ∇u(x, y) + X∗ (x, y) = 0}, (5.30)
i.e., the projection on Π of Char(Sut ) (see (4.14)).
The solutions of (5.28) are called H-minimal. One is not allowed to deduce that
(5.28) is satisfied on U in the sense of distributions even when L2n (Char(u)) = 0;
moreover, the size of Char(u) may be large even for u ∈ W 1,1 (U ) (see [22]). These
problems have been studied in detail in [58] and a suitable minimal surface equation
was obtained.
The study of the relaxed functional At : L 1 (U ) → [0, +∞] of At with respect
to the L 1 -topology and its representation formula on its domain has been studied in
[213]. We therefore introduce
( Z )
At (u) := inf lim inf |∇uk + X∗ | d L2n : uk ∈ W 1,1 (U ),uk → u in L 1 (U ) .
k→∞ U
Definition 5.30. We say that u ∈ L 1 (U ) belongs to the class BVt (U ) of maps with
bounded t-variation if |∂Eut |H (U × R) < +∞.
We say that u ∈ BVt,loc (U ) if Eut has finite H-perimeter in U 0 × R for any open set
U0 b U.
98 Francesco Serra Cassano
The structure of the space BVt (U ) and several different notions of “area” for the
boundary of t-subgraphs Eut of u ∈ L 1 (U ) have been studied in [213, Section 3]. In
particular, among them it has been proved that the perimeter |∂Eut |H (U × R) and the
relaxed functional At of At agree on L 1 (U ).
Theorem 5.31 ([213, Theorem 3.2]). Let U ⊂ R2n be a bounded open set. Then
It turns out that BVt (U ), which is the finiteness domain of these functionals, coin-
cides with the classical space BV (U ) of functions with bounded variation in U .
Theorem 5.32 ([213, Theorem 1.2]). Let U ⊂ R2n be a bounded open set. Then
BVt (U ) = BV (U ).
Definition 5.33. Let U ⊂Π be a bounded open set with Lipschitz regular boundary.
We say that u∈BV (U ) is a t-minimizer of the area functional (briefly, t-minimizer) if
Z Z
|Du + X∗ | ≤ |Dv + X∗ |
U U
Remark 5.35. The last integral in (5.32) equals both the Euclidean and sub-Rieman-
nian areas of that part of the cylinder ∂ U × R between the graphs of u and ϕ, hence
it can be seen as a penalization for not taking the boundary values ϕ on ∂ U . See also
[213, Proposition 3.7 and Remark 3.8].
Theorem 5.34 extends the existence results contained in [193] and [58] because for-
mulation (5.32) allows more general domains U to be considered.
Remark 5.36. A minimizer of the penalized functional (5.32) might not take the pre-
scribed boundary value ϕ: this situation can be illustrated by explicitly constructing
an example where t-minimizers do not exist (see [213, Example 3.6]). In particular,
the existence of solutions for the Dirichlet minimum problem for At is not guaran-
teed even when the boundary ∂ U and the datum ϕ are very regular: in this sense,
Theorem 5.34 does not extend the results in [193] and [58].
An existence result for continuous BV t-minimizers for continuous boundary
data on smooth parabolically convex domains has been obtained in [53].
In [197], existence, uniqueness, and Lipschitz regularity of t-minimizers (assum-
ing the prescribed boundary datum) is proved under the assumption that the boundary
datum ϕ satisfies the so-called bounded slope condition.
Definition 5.37. We say that a function ϕ : ∂Ω → R satisfies the bounded slope
condition with constant Q > 0 (Q-B.S.C. for short, or simply B.S.C. when the con-
stant Q does not play any role) if for every z0 ∈ ∂Ω there exist two affine functions
wz+0 and wz−0 such that
|z − z0 | 2 ≤ C dist(z, Πz 0 ) ∀z ∈ ∂ U ,
This property seems to say that the B.S.C. is quite a restrictive assumption. Anyhow
the following one, due to Miranda [172] (see also [123, Theorem 1.1]), shows that
the class of functions satisfying the B.S.C. on a uniformly convex set is quite large.
Remark 5.40. It can be proved that the previous theorem is sharp, at least for n = 1,
in the sense that minimizers might not be better than Lipschitz regular (see [197,
Examples 6.6 and 6.7]).
In particular, Theorem 5.39 extends Theorem 5.34 as well as some related results
in [193] and [58].
Finally a result on boundedness for local t-minimizer holds.
Theorem 5.41 (Local boundedness of minimal t-graphs, [213, Theorem 1.5]). Let
u ∈ BV (U ) be a local t-minimizer. Then u ∈ L ∞
loc (U ).
As a consequence, we also obtain a local boundedness result for the weak solutions
u ∈ W 1,1 (U ) of the minimal surface equation (5.28) (see [213, Theorem 3.14]).
Theorem 5.41 is sharp at least in the first Heisenberg group H1 . Indeed, a minimal
t-graph can be induced by a function u ∈ L ∞ 0
loc (U ) \ C (U ) (see [213, Section 3.4]). It
is an open problem whether a similar example can also be constructed in Hn , n ≥ 2.
Minimal intrinsic graphs. We are now going to deal with minimal surfaces which
are intrinsic graphs, that is, graphs of functions φ : ω ⊂ W → V acting between a
(2n, 2n+1)-subgroup W and (1, 1)-subgroup V of Hn , which are also complementary
(see Definitions 4.1 and 4.2, Example 4.4). Without loss of generality, we will always
consider X1 -graphs, i.e., intrinsic graphs along the X1 -direction. Let us introduce
some preliminary notation. If n ≥ 2, we identify the maximal subgroup
with R by writing s instead of (s, 0, . . . , 0). Let ω denote a fixed open bounded subset
of W; the intrinsic cylinder ω · R is defined by
ω · R := { A · s ∈ Hn : A ∈ ω, s ∈ R},
Φ( A) := A · φ( A), A ∈ ω. (5.35)
S := Φ(ω) = { A · φ( A) : A ∈ ω} .
The X1 -subgraph and the X1 -epigraph (or X1 -upper-graph) of φ are defined, respec-
tively, as
Eφ := { A · s : A ∈ ω, s < φ( A)} (5.36)
and
E φ := { A · s : A ∈ ω, s > φ( A)} . (5.37)
Let Lip(ω) be the classical space of Lipschitz functions on ω ⊂ W ≡ R2n . The area
functional AW : Lip(ω) → R is
Z q
AW (φ) := |∂Eφ |H (ω · R) = 1 + |∇φ φ| 2 d L2n , (5.38)
ω
where
1 1
W φ φ := Y1 φ + T (φ2 ) = ∂y1 φ + ∂t (φ2 ).
2 2
1 Some topics of geometric measure theory in Carnot groups 103
We agree that, when φ is not regular, the differential operators appearing in (5.39)
will be understood in the sense of distributions. The intrinsic gradient ∇φ was intro-
duced and studied in [16]; see also [61, 40, 41, 62]. Observe that we can write
Z
AW (φ) = L W ( A, φ( A), ∇φ( A)) d L2n ( A),
ω
if n ≥ 2, while
1/2
L W ( A, φ, ξ) := 1 + hY1 ( A) + φ T ( A), ξi2 (5.41)
if n = 1. The vector fields X j ( j = 2, . . . , n),Yj ( j = 1, . . . , n), and T are tangent to
W ≡ R2n and therefore can be viewed as elements of R2n . The scalar products in
(5.40) and (5.41) are the usual ones between vectors in R2n .
When φ ∈ C2 (ω) is a local minimizer of the functional AW , the first variation of
the functional AW gives the minimal surface equation for X1 -graphs,
∇φ φ
∇φ · * p +=0 in ω. (5.42)
φ 2
, 1 + |∇ φ| -
It was pointed out in [72] that AW is not convex for n = 1. Indeed, for any α > 0
the function
2α y t
φ(y,t) :=
2 + α y2
satisfies (5.42) on R2 , while (see [70]) φ is not a local minimizer for AW : v(ω) → R
on a suitable bounded open set ω ⊂ R2 . In particular, AW cannot be convex on v(ω)
because the stationary point φ is not a minimum. The presence of stationary points
that are not minimizers for AW is an interesting open question in the case n ≥ 2.
Nevertheless, the nonconvexity of AW : v(ω) → R can also occur in the higher-
dimensional case (see [213, Proposition 4.1]).
A study of the C2 minimizers of AW was carried out in [70, 33, 69, 72] also in
connection with the Bernstein problem for intrinsic graphs. First and second vari-
1,1
ations for minimizers in WW (ω), a suitable class of intrinsic graphs with Sobolev
regularity introduced in [187] (see [213, Definition 2.8]), have been studied in [187].
The regularity of Lipschitz continuous vanishing viscosity solutions of the minimal
surface equation for intrinsic graphs has been studied in [46, 47, 31]. We have to
mention that, in the first Heisenberg group H1 , there are minimizers of AW whose
regularity is not better than 21 -Hölder: see [187, Theorem 1.5].
104 Francesco Serra Cassano
Definition 5.42. We say that φ ∈ L 1 (ω) belongs to the class BVW (ω) of functions
with intrinsic bounded variation if |∂Eφ |H (ω · R) < +∞.
We say that φ belongs to BVW,loc (ω) if Eφ is a set with finite H-perimeter in ω 0 · R
for any open set ω 0 b ω.
The class BVW (ω) is deeply different from BV (ω): for instance, it is not even a
vector space (see [213, Remark 4.2]). In spite of these differences, BVW (ω) shares
with BV (ω) several properties:
• The functional φ 7→ |∂Eφ |H (ω · R) coincides with the relaxed one AW of AW
on L 1 (ω) (see [213, Theorem 4.7]).
• Each function in BVW (ω) can be approximated by a sequence of C ∞ regular
functions (φ j ) j such that
then any sequence (φ j ) j ⊂ BVW (ω) bounded in the k · k BVW “norm” and such
that
sup |∂Eφ j |H (H+n ) + |∂E φ j |H (H−n ) < +∞
(5.43)
j
We are now going to deal with the problem of an existence result for minimal X1 -
graphs on ω with prescribed “boundary datum”. Let ω0 c ω be a bounded open set
and ϑ ∈ BVW (ω0 ) be such that
When φ ∈ BVW (ω) has a trace in a generalized sense, then it possesses an extension
ϑ ∈ BVW (ω0 ), on a suitable ω0 c ω, satisfying (5.45): if this is the case, then
problem (5.46) can be viewed as that of minimizing area with boundary datum given
by φ.
Theorem 5.43 (Existence of minimal X1 -graphs, [213, Theorem 1.8]). Problem
(5.46) attains a minimum in BVW (ω0 ).
Also a local boundedness result for minimal X1 -graphs holds.
Theorem 5.44 (Local boundedness of minimal X1 -graphs, [213, Theorem 1.8]). Let
loc (ω) be such that Eφ is a local minimizer of the H-perimeter in ω · R. Then
φ ∈ L 2n+1
φ ∈ L∞loc (ω).
This result is not the exact counterpart of Theorem 5.41 for minimal t-graphs.
We do not know whether the additional (2n + 1)-summability is only a technical
problem or if there exist minimal X1 -graphs φ < L ∞ loc (ω). Moreover, in Theorem
5.41 we prove the local boundedness of t-minimizers using the fact that they are also
H-perimeter minimizing sets. In Theorem 5.44 we instead require the subgraph Eφ
to be H-perimeter minimizing: as far as we know, there is no geometric arrange-
ment, similar to the one for t-graphs given by [213, Theorem 3.15], ensuring that the
subgraph of a minimal intrinsic graph is also H-perimeter minimizing.
One of the most important results in this setting claims that the portion of
reduced boundary ∂ ∗ E ∩ Ω of a minimizer E in Ω is (locally) an analytic
hypersurface (see, for instance, [122, Theorem 8.4]). However there may be
minimizers with irregular topological boundary ∂E, which are cones, in di-
mension n ≥ 8 (see, for instance, [122, Theorems 11.7 and 11.8]).
Another relevant result concerns the (global) regularity of a nonparametric
minimal surface, that is, minimal boundaries of sets which are subgraphs.
More precisely, assume E := {(x 1 , . . . , x n−1 , s) ∈ U ×R : s < u(x 1 , . . . , x n−1 )}
is a minimizer in Ω := U × R and u is a function of Euclidean bounded vari-
ation, that is, u ∈ BVG (U ) and G = (Rn−1 , +) in accordance with Example
2.1 and Section 3.5. Then u is an analytic function and therefore the boundary
∂E ∩ Ω is an analytic hypersurface (see, for instance, [122, Theorem 14.13]).
3. Engel group. Here we mean that G = E = (R4 , ·) endowed with the sub-
Riemannian structure introduced in Example 4.7. Let us recall that G is a
step-3 Carnot group and its Lie algebra g is generated by four left-invariant
vector fields X j , j = 1, 2, 3, 4. Such an algebra is stratified as g = V1 ⊕ V2 ⊕ V3
with V1 = span {X1 , X2 }, V2 = span {X3 }, and V3 = span {X4 }, the only
nonzero commutation relations being [X1 , X2 ] = X3 , [X1 , X3 ] = X4 .
In this setting the structure and regularity of a set E of finite G-perimeter with
constant generalized normal νE have been studied in [36]. Namely, we assume
there exists a constant section ν0 = a X1 + b X2 ∈ V1 such that νE (x) = ν0
|∂E|G -a.e. x ∈ R4 .
Let us recall that, by Example 5.28, such a set is a (global) minimizer for a
G-perimeter in Ω = G. Notice also that such a set, in a step-2 Carnot group, is
always equivalent to a vertical hyperplane orthogonal to ν0 (see Lemma 5.16).
It is not the case in the Engel group as already shown in Example 5.15.
The general regularity results proved in [36] claim that each set E ⊂ G of
locally G-finite perimeter with constant normal is equivalent to a (Euclidean)
Lipschitz regular set Ẽ ⊂ G, that is, L4 (E∆ Ẽ) = 0.
In [36] the possibility of expressing a set with constant normal as an intrinsic
upper-graph is also studied, when using the model of the Engel group corre-
sponding to the so-called exponential coordinates of the second kind, and what
regularity to expect for such a graph. The authors proved that there are exam-
ples of sets with constant horizontal normal such that, if one writes the set as
an intrinsic upper-graph in the direction of the normal, the function giving the
graph is not even continuous. However, all constant normal sets are intrinsic
Lipschitz upper-graphs in other horizontal directions. This last feature is pe-
culiar to the Engel group. In a forthcoming paper [37], they will prove that
in general Carnot groups sets with constant horizontal normal might fail to be
intrinsic Lipschitz upper-graphs in every horizontal direction. We thank Le
Donne for sharing his unpublished results with us.
108 Francesco Serra Cassano
Bibliography
[1] A. Agrachev, D. Barilari, U. Boscain, Lecture notes on Riemannian and sub-Riemannian
geometry. Preprint, 2013.
[2] A. Agrachev, D. Barilari, U. Boscain, On the Hausdorff volume in sub-Riemannian geome-
try. Calc. Var. and PDE, 43 (2012), no. 3-4, 355–388. Zbl 1236.53030.
[3] A. Agrachev, D. Barilari, L. Rizzi, The curvature: A variational approach. Preprint, 2013.
[4] L. Ambrosio, Some fine properties of sets of finite perimeter in Ahlfors regular metric measure
spaces. Adv. in Math., 159 (2001), 51–67. Zbl 1002.28004.
[5] L. Ambrosio, Fine properties of sets of finite perimeter in doubling metric measure spaces.
Calculus of variations, nonsmooth analysis and related topics. Set-Valued Anal. 10 (2002),
no. 2-3, 111–128. Zbl 1037.28002.
[6] L. Ambrosio, Personal communication. March 2002.
[7] L. Ambrosio, La teoria dei perimetri di Caccioppoli-De Giorgi e i suoi più recenti sviluppi.
Rend. Lincei Mat. Appl. 21 (2010), 275–286. Zbl 1195.49052.
[8] L. Ambrosio, N. Fusco, D. Pallara, Functions of bounded variation and free discontinuity
problems. Oxford Mathematical Monographs, Oxford University Press (2000).
Zbl 0957.49001.
[9] L. Ambrosio, R. Ghezzi, V. Magnani, BV functions and sets of finite perimeters in sub-
Riemannian manifolds. Ann. Inst. H. Poincaré Anal. Non Linéaire, to appear.
[10] L. Ambrosio, B. Kirchheim, Rectifiable sets in metric and Banach spaces. Math. Ann. 318
(2000), 527–555. Zbl 0966.28002.
[11] L. Ambrosio, B. Kirchheim, Currents in metric spaces. Acta Math. 185 (2000), 1–80.
Zbl 0984.49025.
[12] L. Ambrosio, B. Kleine, E. Le Donne, Rectifiability of sets of finite perimeter in Carnot
groups: Existence of a tangent hyperplane. J. Geom. Anal. 19 (2009), no. 3, 509–540.
Zbl 1187.28008.
[13] L. Ambrosio, V. Magnani, Weak differentiability of BV functions on stratified groups. Math.
Z. 245 (2003), 123–153. Zbl 1048.49030.
[14] L. Ambrosio, S. Rigot, Optimal mass transportation in the Heisenberg group. J. Funct. Anal.
208 (2004), no. 2, 261–301. Zbl 1076.49023.
[15] L. Ambrosio, M. Scienza, Locality of the perimeter in Carnot groups and chain rule. Ann.
Mat. Pura Appl. (4) 189 (2010), no. 4, 661–678. Zbl 1203.28001.
[16] L. Ambrosio, F. Serra Cassano, D. Vittone, Intrinsic regular hypersurfaces in Heisenberg
groups. J. Geom. Anal. 16 (2006), 187–232. Zbl 1085.49045.
[17] L. Ambrosio, P. Tilli, Topics on analysis in metric spaces. Oxford University Press, Oxford,
2004. Zbl 1080.28001.
[18] G. Arena, R. Serapioni, Intrinsic regular submanifolds in Heisenberg groups are differen-
tiable graphs. Calc. Var. 35 (2009), no. 4, 517–536. Zbl 1225.53031.
Bibliography 109
[19] V.I. Arnold, Mathematical methods of classical mechanics. Second ed., Springer, 1989.
Zbl 0692.70003.
[20] D. Bakry, T. Coulhon, M. Ledoux, L. Saloff-Coste, Sobolev inequalities in disguise. Indiana
U. Math. J. 44 (1995), 1033–1074. Zbl 0857.26006.
[21] A. Baldi, B. Franchi, N. Tchou, M.C. Tesi, Compensated compactness for differential forms
in Carnot groups and applications. Adv. Math., 223 (5) (2010), 1555–1607. Zbl 1184.43007.
[22] Z.M. Balogh, Size of characteristic sets and functions with prescribed gradient. J. reine
angew. Math., 564 (2003), 63–83. Zbl 1051.53024.
[23] Z.M. Balogh, R. Berger, R. Monti, J.T. Tyson, Exceptional sets for self-similar frac-
tals in Carnot groups. Math. Proc. Cambridge Philos. Soc. 149 (2010), no. 1, 147–172.
Zbl 1195.22005.
[24] Z.M. Balogh, E. Durand-Cartagena, K. Fässler, P. Mattila, J.T. Tyson, The effect of projec-
tions on dimension in the Heisenberg group. Rev. Mat. Iberoam. 29 (2013), no. 2, 381–432.
Zbl 1279.28007.
[25] Z.M. Balogh, K.S. Fässler, Rectifiability and Lipschitz extensions into the Heisenberg group.
Math. Z. 263 (2009), no. 3, 673–683. Zbl 1177.53032.
[26] Z.M. Balogh, K. Fässler, P. Mattila, J.T. Tyson, Projection and slicing theorems in Heisen-
berg groups. Adv. Math. 231 (2012), no. 2, 569–604. Zbl 1260.28007.
[27] Z.M. Balogh, H. Höfer-Isenegger, J.T. Tyson, Lifts of Lipschitz maps and horizontal frac-
tals in the Heisenberg group. Ergodic Theory Dynam. Systems 26 (2006), no. 3, 621–651.
Zbl 1097.22005.
[28] Z.M. Balogh, M. Rickly, F. Serra Cassano, Comparison of Hausdorff measures with re-
spect to the Euclidean and the Heisenberg metric. Publ. Mat. 47 (2003), no. 1, 237–259.
Zbl 1060.28002.
[29] Z.M. Balogh, K. Rogovin, T. Zürcher, The Stepanov differentiability theorem in metric mea-
sure spaces. J. Geom. Anal. 14 (2004), no. 3, 405–422. Zbl 1069.28001.
[30] Z.M. Balogh, J.T. Tyson, B. Warhurst, Sub-Riemannian vs. Euclidean dimension com-
parison and fractal geometry on Carnot groups. Adv. Math. 220 (2009), no. 2, 560–619.
Zbl 1155.22011.
[31] D. Barbieri, G. Citti, Regularity of minimal intrinsic graphs in 3-dimensional sub-
Riemannian structures of step 2. J. Math. Pures Appl. (9) 96 (2011), no. 3, 279–306.
Zbl 1231.53030.
[32] D. Barilari, U. Boscain, R.W. Neel, Small-time heat kernel asymptotics at the sub-
Riemannian cut locus. J. Differential Geom. 92 (2012), no. 3, 373–416. Zbl 1270.53066.
[33] V. Barone Adesi, F. Serra Cassano, D. Vittone, The Bernstein problem for intrinsic graphs
in Heisenberg groups and calibrations. Calc. Var. PDEs, 30 (2007), 17–49. Zbl 1206.35240.
[34] F. Baudoin, M. Bonnefont, N. Garofalo, A sub-Riemannian curvature-dimension inequality,
volume doubling property and the Poincaré inequality. Math. Ann. (2014) 358 833–860.
doi 10.1007/s00208-013-0961-y.
110 Francesco Serra Cassano
[52] J. Cheeger, B. Kleiner. Differentiating maps into L 1 and the geometry of BV functions. Ann.
of Math. (2) 171 (2006), no. 2, 1347–1385. Zbl 1194.22009.
[53] J.H. Cheng, J.-F. Hwang, Properly embedded and immersed minimal surfaces in the Heisen-
berg group. Bull. Austr. Math. Soc., 70 (2004), 507–520. Zbl 1062.35046.
[54] J.H. Cheng, J.-F. Hwang, Variations of generalized area functionals and p-area minimizers
of bounded variation in the Heisenberg group. Bull. Inst. Math. Acad. Sin. (N.S.) 5 (2010),
no. 4, 369–412. Zbl 1232.35178.
[55] J.H. Cheng, J.-F. Hwang, Uniqueness of p-area minimizers and integrability of a horizontal
normal in the Heisenberg group. Calc. Var. (2014) 50 579–597.
doi 10.1007/s00526-013-0647-z. Zbl 1293.49102.
[56] J.H. Cheng, J.-F. Hwang, A. Malchiodi, P. Yang, Minimal surfaces in pseudohermitian ge-
ometry and the Bernstein problem in the Heisenberg group. Ann. Sc. Norm. Pisa Cl. Sci. (5),
4 (2005), no. 1, 129–177. Zbl 1158.53306.
[57] J.H. Cheng, J.-F. Hwang, A. Malchiodi, P. Yang, A Codazzi-like equation and the singular set
for C 1 smooth surfaces in the Heisenberg group. J. reine angew. Math. 671 (2012), 131–198.
Zbl 1255.53026.
[58] J.H. Cheng, J.-F. Hwang, P. Yang, Existence and uniqueness for p-area minimizers in the
Heisenberg group. Math. Ann. 337 (2007), no. 2, 253–293. Zbl 1109.35009.
[59] J.H. Cheng, J.-F. Hwang, P. Yang, Regularity of C 1 smooth surfaces with prescribed p-mean
curvature in the Heisenberg group. Math. Ann. 344 (2009), no. 1, 1–35. Zbl 1171.35026.
[60] G. Citti, M. Manfredini, Blow-up in non homogeneous Lie groups and rectifiability. Houston
J. Math. 31, 2 (2005), 333–353. Zbl 1078.49030.
[61] G. Citti, M. Manfredini, Implicit function theorem in Carnot-Carathéodory spaces. Com-
mun. Contemp. Math. 8, 5 (2006), 657–680. Zbl 1160.53017.
[62] G. Citti, M. Manfredini, A. Pinamonti, F. Serra Cassano, Smooth approximation for the
intrinsic Lipschitz functions in the Heisenberg group. Calc. Var. Partial Differential Equations
49 (2014), no. 3-4, 1279–1308. Zbl 1291.22011.
[63] G. Citti, M. Manfredini, A. Pinamonti, F. Serra Cassano, Poincaré- type inequality for
intrinsic Lipschitz continuous vector fields in the Heisenberg group. To appear in J. Math.
Pures Appl.
[64] G. Citti, A. Sarti, A cortical based model of perceptual completion in the roto-translation
space. J. Math. Imaging Vision 24, 3 (2006), 307–326. Zbl 1088.92008.
[65] D.L. Cohn, Measure theory. Birkhäuser, 1980. Zbl 0436.28001.
[66] R.R. Coifman, G. Weiss, Analyse harmonique non-commutative sur certains espaces homo-
genes. Lecture Notes in Math., Vol. 242, Springer, New York/Berlin, 1971. Zbl 0224.43006.
[67] D. Cole, S.D. Pauls, C 1 hypersurfaces of the Heisenberg group are N-rectifiable. Houston J.
Math. 32, 3 (2006), 307–326. Zbl 1110.53024.
[68] D. Danielli, N. Garofalo, D.M. Nhieu, Traces inequalities for Carnot-Carathéodory
spaces and applications. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 27 (1998), 195–252.
Zbl 0938.46036.
112 Francesco Serra Cassano
[87] V. Franceschi, R. Monti, Isoperimetric problems in H-type groups and Grushin spaces.
Preprint, 2014.
[88] B. Franchi, Stime subellittiche e metriche Riemanniane singolari II. Seminario di Analisi
Matematica, Università di Bologna (1983), VIII-1–VIII-17.
[89] B. Franchi, Weighted Sobolev-Poincaré inequalities and pointwise estimates for a class of
degenerate elliptic equations. Trans. AMS 327 (1991), 125–158. Zbl 0751.46023.
[90] B. Franchi, S. Gallot, R.L. Wheeden, Sobolev and isoperimetric inequalities for degenerate
metrics. Math. Ann., 300 (1994), 557–571. Zbl 0830.46027.
[91] B. Franchi, C.E. Gutiérrez, R.L. Wheeden, Weighted Sobolev-Poincaré inequalities for
Grushin type operators. Comm. Partial Differential Equations, 19 (1994), no. 3-4, 523–604.
Zbl 0822.46032.
[92] B. Franchi, P. Hajłasz, P. Koskela, Definitions of Sobolev classes on metric spaces. Ann.
Inst. Fourier (Grenoble) 49 (1999), no. 6, 1903–1924. Zbl 0938.46037.
[93] B. Franchi, E. Lanconelli, Hölder regularity theorem for a class of linear nonuniformly
elliptic operators with measurable coefficients. Ann. Sc. Norm. Super. Pisa, Cl. Sci., IV. Ser.
10, 523–541 (1983). Zbl 0552.35032.
[94] B. Franchi, G. Lu, R.L. Wheeden, Representation formulas and weighted Poincaré inequal-
ities for Hörmander vector fields. Ann. Inst. Fourier, 49 (1995), 577–604.
Zbl 0820.46026.
[95] B. Franchi, G. Lu, R.L. Wheeden, A relationship between Poincaré-type inequalities and rep-
resentation formulas in spaces of homogeneous type. Internat. Math. Res. Notices, 1 (1996),
1–14. Zbl 0856.43006.
[96] B. Franchi, M. Marchi, R.P. Serapioni, Differentiability and approximate differentiability
for intrinsic Lipschitz functions in Carnot groups and a Rademacher theorem. Anal. Geom.
Metr. Spaces 2 (2014), 258–281. Zbl 1307.22007.
[97] B. Franchi, C. Pérez, R.L. Wheeden, Self-improving properties of John–Nirenberg and
Poincaré inequalities on spaces of homogeneous type. J. Functional Analysis 153 (1998),
108–146. Zbl 0892.43005.
[98] B. Franchi, R.P. Serapioni, Intrinsic Lipschitz graphs within Carnot groups. Journal of Geo-
metric Analysis 05/2015. doi 10.1007/s12220-015-9615-5.
[99] B. Franchi, R. Serapioni, F. Serra Cassano, Meyers-Serrin type theorems and relaxation
of variational integrals depending on vector fields. Houston Journal of Mathematics, 22, 4
(1996), 859–889. Zbl 0876.49014.
[100] B. Franchi, R. Serapioni, F. Serra Cassano, Approximation and imbedding theorems for
weighted Sobolev spaces associated with Lipschitz continuous vector fields. Boll. Un. Mat.
Ital., (7) 11-B (1997), 83–117. Zbl 0952.49010.
[101] B. Franchi, R. Serapioni, F. Serra Cassano, Sur les ensembles des périmètre fini dans
le groupe de Heisenberg. C.R. Acad. Sci. Paris, Ser. I, Math. 329 (1999), 183–188.
Zbl 1033.49045.
114 Francesco Serra Cassano
[102] B. Franchi, R. Serapioni, F. Serra Cassano, Rectifiability and perimeter in the Heisenberg
group. Math. Ann. 321 (2001), 479–531. Zbl 1057.49032.
[103] B. Franchi, R. Serapioni, F. Serra Cassano, Regular hypersurfaces, intrinsic perimeter and
implicit function theorem in Carnot groups. Comm. Anal. Geom., 11 (2003), no. 5, 909–944.
Zbl 1077.22008.
[104] B. Franchi, R. Serapioni, F. Serra Cassano, Rectifiability and perimeter in step 2 groups.
Proceedings of Equadiff10, Prague Aug. 2001, Math. Bohemica 127 (2002), 219–228.
Zbl 1018.49029.
[105] B. Franchi, R. Serapioni, F. Serra Cassano, On the structure of finite perimeter sets in step
2 Carnot groups. Journal of Geometric Analysis. 13 (2003), no. 3, 421–466. Zbl 1064.49033.
[106] B. Franchi, R. Serapioni, F. Serra Cassano, Regular submanifolds, graphs and area formula
in Heisenberg groups. Advances in Math, 211, 1 (2007), 152–203.
Zbl 1125.28002.
[107] B. Franchi, R. Serapioni, F. Serra Cassano, Intrinsic Lipschitz graphs in Heisenberg groups.
J. Nonlinear Convex Anal. 7 (2006), no. 3, 423–441. Zbl 1151.58005.
[108] B. Franchi, R. Serapioni, F. Serra Cassano. Differentiability of intrinsic Lipschitz functions
within Heisenberg groups. J. Geom. Anal. 21 (2011), no. 4, 1044–1084. Zbl 1234.22002.
[109] B. Franchi, R. Serapioni, F. Serra Cassano. Area formula for centered Hausdorff measure
in metric spaces. Nonlinear Analysis (2015). doi 10.1016/j.na.2015.02.008.
[110] B. Franchi, M.C. Tesi, Wave and Maxwell’s equations in Carnot groups. Commun. Con-
temp. Math. 14 (2012) no. 5 . Zbl 1251.35175.
[111] B. Franchi, R.L. Wheeden, Compensation couples and isoperimetric estimates for vector
fields. Coll. Math. 74 (1997), 1–27. Zbl 0915.46028.
[112] B. Franchi, R.L. Wheeden, Some remarks about Poincaré type inequalities and representa-
tion formulas in metric spaces of homogeneous type. J. of Inequal. Appl. 3 (1999), 65–89.
Zbl 0934.46037.
[113] O.K. Friedrichs, The identity of weak and strong extensions of differential operators. Trans.
Amer. Math. Soc. 55 (1944), 132–151. Zbl 0061.26201.
[114] M. Galli, First and second variation formulae for the sub-Riemannian area in three-
dimensional pseudo-Hermitian manifolds. Calc. Var. Partial Differential Equations 47 (2013),
no. 1-2, 117–157. Zbl 1268.53039.
[115] M. Galli, On the classification of complete area-stationary and stable surfaces in the sub-
riemannian Sol manifold. Pacific J. Math. 271 (2014), no. 1, 143–157. Zbl 1303.53043.
[116] M. Galli, M. Ritoré, Existence of isoperimetric regions in contact sub-Riemannian mani-
folds. J. Math. Anal. Appl. 397 (2013), no. 2, 697–714. Zbl 1296.53059.
[117] M. Galli, M. Ritoré, Area-stationary and stable surfaces of class C1 in the sub-Riemannian
Heisnberg group H1 . Preprint, 2014. arXiv:1410.3619v1.
[118] N. Garofalo, D.M. Nhieu, Isoperimetric and Sobolev inequalities for Carnot-Carathéodory
spaces and the existence of minimal surfaces. Comm. Pure Appl. Math., 49 (1996), 1081–
1144. Zbl 0880.35032.
Bibliography 115
[119] N. Garofalo, D.M. Nhieu, Lipschitz continuity, global smooth approximations and extension
theorems for Sobolev functions in Carnot Carathéodory spaces. J. Anal. Math. 74 (1998),
67–97. Zbl 0906.46026.
[120] N. Garofalo, S. Pauls, The Bernstein problem in the Heisenberg group. Preprint, 2005.
arXiv:math/0209065v2.
[121] R. Ghezzi, F. Jean, A new class of H1 -rectifiable subsets of metric spaces. Comm. Pure Appl.
Anal. 12 (2013), no. 2, 881–898. Zbl 1268.28005.
[122] E. Giusti, Minimal surfaces and functions of bounded variation. Birkhäuser, Basel, 1984.
Zbl 0545.49018.
[123] E. Giusti, Direct methods in the calculus of variations. World Scientific Publishing Co.,
River Edge, NJ, 2003. Zbl 1028.49001.
[124] M. Gromov, Carnot-Carathéodory spaces seen from within. In Subriemannian Geometry,
Progress in Mathematics, 144. Ed. by A. Bellaïche and J. Risler, Birkhäuser, Basel, 1996.
Zbl 0864.53025.
[125] M. Gromov, Metric structures for Riemannian and non-Riemannian spaces. Progress in
Mathematics, 152, Birkhäuser, Boston (1999). Zbl 0953.53002.
[126] P. Hajłasz, Sobolev spaces on an arbitrary metric space. Potential Anal., 5 (1996), no. 4,
403–415. Zbl 0859.46022.
[127] P. Hajłasz, Geometric approach to Sobolev spaces and badly degenerated elliptic equations.
Kenmochi, N. (ed.) et al., Proceedings of the Banach Center minisemester on nonlinear analy-
sis and applications, Warsaw, Poland, November 14–December 15, 1994. Tokyo: Gakkotosho
Co., Ltd. GAKUTO Int. Ser., Math. Sci. Appl. 7, 141–168 (1995). Zbl 0877.46024.
[128] P. Hajłasz, P. Koskela, Sobolev met Poincaré., Memoirs of the American Mathematical
Society, 688 (2000). Zbl 0954.46022.
[129] J. Heinonen, Calculus on Carnot groups. Ber. Univ. Jyväskylä Math. Inst., 68 (1995), 1–31.
Zbl 0863.22009.
[130] J. Heinonen, Nonsmooth calculus. Bull. AMS, 44 (2007), 163–232. Zbl 1124.28003.
[131] S. Helgason, Differential geometry, Lie groups, and symmetric spaces. Academic Press, New
York, 1978. Zbl 0451.53038.
[132] I.N. Herstein, Topics in algebra. Blaisdell Publishing Company, London, 1964.
Zbl 0122.01301.
[133] R.K. Hladky, S.D. Pauls, Constant mean curvature surfaces in sub-Riemannian geometry.
J. Differential Geom. 79 (2008), no. 1, 111–139. Zbl 1156.53038.
[134] A. Hurtado, M. Ritoré, C. Rosales, The classification of complete stable area-sta-
tionary surfaces in the Heisenberg group H1 . Adv. Math. 224 (2010), no. 2, 561–600.
Zbl 1192.53038.
[135] D. Jerison, The Poincaré inequality for vector fields satisfying Hörmander’s condition. Duke
Math. J., 53 (1986), 503–523. Zbl 0614.35066.
116 Francesco Serra Cassano
[136] N. Juillet, Geometric inequalities and generalized Ricci bounds in the Heisenberg group.
Int. Math. Res. Not. (2009), no. 13, 2347–2373. Zbl 1176.53053.
[137] B. Kirchheim, Rectifiable metric spaces: Local structure and regularity of the Hausdorff
measure. Proc. AMS, 121 (1994), 113–123. Zbl 0806.28004.
[138] B. Kirchheim, V. Magnani, A counterexample to metric differentiability. Proc. Ed. Math. Soc.
46 (2003), 221–227. Zbl 1059.26007.
[139] B. Kirchheim, F. Serra Cassano, Rectifiability and parameterization of intrinsic regular
surfaces in the Heisenberg group. Ann. Scuola Norm. Sup. Pisa, Cl. Sc. (5) III (2004), 871–
896. Zbl 1170.28300.
[140] A. Korányi, H.M. Reimann, Foundations for the theory of quasiconformal mappings on the
Heisenberg group. Advances in Mathematics, 111 (1995), 1–87. Zbl 0876.30019.
[141] A. Kozhevnikov, Propriétés métriques des ensembles de niveaux des applications différen-
tiables sur le groupes de Carnot. Metric Geometry [math.MG]. Université Paris Sud - Paris
XI, 2015.
[142] E. Lanconelli, D. Morbidelli, On the Poincaré inequality for vector fields. Ark. Mat. 38
(2000), 327–342. Zbl 1131.46304.
[143] E. Le Donne, Lecture notes on sub-Riemannian geometry. Preprint, 2010.
[144] E. Le Donne, A metric characterization of Carnot groups., Proc. Am. Math. Soc., 143 (2015),
no. 2, 845–849. Zbl 1307.53027.
[145] E. Le Donne, V. Magnani Measure of submanifolds in the Engel group. Rev. Mat. Iberoamer-
icana 26 (2010), no. 1, 333–346. Zbl 1198.58007.
[146] E. Le Donne, S. Rigot Besicovitch covering property for homogeneous distances in Heisen-
berg groups. Preprint, 2014.
[147] G.P. Leonardi, V. Magnani Intersections of intrinsic submanifolds in the Heisenberg group.
J. Math. Anal. Appl. 378 (2011), no. 1, 98–108. Zbl 1220.53074.
[148] G.P. Leonardi, S. Rigot, Isoperimetric sets on Carnot groups. Houston J. Math. 29 (2003),
no. 3, 609–637. Zbl 1039.49037.
[149] G.P. Leonardi, S. Rigot, D. Vittone, Isodiametric sets in the Heisenberg group. Rev. Mat.
Iberoam. 28 (2012), no. 4, 999–1024. Zbl 06109703.
[150] R. Long, F. Nie, Weighted Sobolev inequality and eigenvalue estimates of Schrödinger oper-
ators. Lect. Notes in Math. 1494 (1990), 131–141. Zbl 0786.46034.
[151] G. Lu, R.L. Wheeden, An optimal representation formula for Carnot-Carathéodory vector
fields. Bull. London Math. Soc. 30 (1998), no. 6, 578–584. Zbl 0931.31003.
[152] F. Maggi, Sets of finite perimeter and geometric varational problems. Cambridge Studies in
Advanced Mathematics 135, Cambridge University Press, New York, 2012. Zbl 1255.49074.
[153] V. Magnani, Elements of geometric measure theory on sub-Riemannian groups. PhD Thesis,
Scuola Normale Superiore, Edizioni della Normale, Pisa, 2002. Zbl 1064.28007.
Bibliography 117
[154] V. Magnani, A blow-up theorem for regular hypersurfaces in nilpotent groups. Manuscripta
Math. 110 (2003), no. 1, 55–76. Zbl 1010.22010.
[155] V. Magnani, The coarea formula for real-valued Lipschitz maps on stratified groups. Math.
Nachr., 278 (2005), no. 14, 1689–1705. Zbl 1079.49030.
[156] V. Magnani, Unrectifiability and rigidity in stratified groups. Arch. Math., 83 (2004), no. 6,
568–576. Zbl 1062.22019.
[157] V. Magnani, Characteristic points, rectifiability and perimeter measure on stratified groups.
J. Eur. Math. Soc. 8 (2006), no. 4, 585–609. Zbl 1107.22004.
[158] V. Magnani, Contact equations, Lipschitz extensions and isoperimetric inequalities. Calc.
Var. 39 (2010), 233–271. Zbl 1195.49053.
[159] V. Magnani, An area formula in metric spaces. Colloq. Math. 124 (2011), 275–283.
Zbl 1231.28007.
[160] V. Magnani, Towards differential calculus in stratified groups. J. Aust. Math. Soc. 95 (2013),
no. 1, 76–128. Zbl 1279.22011.
[161] V. Magnani, On a measure theoretic area formula. To appear in Proc. Roy. Soc. Edinburgh
Sect. A. Available at http://cvgmt.sns.it/paper/2331/. arXiv:1401.2536.
[162] V. Magnani, A new differentiation, shape of the unit ball and perimeter measure. Preprint,
2014.
[163] V. Magnani, Some remarks on densities in the Heisenberg group. Preprint, 2015.
[164] V. Magnani, J.T. Tyson, D. Vittone, On transversal submanifolds and their measure. J. Anal.
Math., 125 (2015), 319–351. Zbl 06454333.
[165] V. Magnani, D. Vittone, An intrinsic measure for submanifolds in stratified groups. J. reine
angew. Math. 619 (2008), 203–232. Zbl 1146.53022.
[166] P. Maheux, L. Saloff–Coste, Analyse sur les boules d’un opérateur sous–elliptique. Math.
Ann. 303 (1995), 713–740. Zbl 0836.35106.
[167] M. Marchi, Regularity of sets with constant intrinsic normal in a class of Carnot groups.
Ann. Inst. Fourier (Grenoble) 64 2 (2014), 429–455. Zbl 1310.28003.
[168] P. Mattila, Geometry of sets and measures in Euclidean spaces. Cambridge University Press,
Cambridge (1995). Zbl 0819.28004.
[169] P. Mattila, Measures with unique tangent measures in metric groups. Math. Scand. 97
(2005), 298–308. Zbl 1093.28002.
[170] P. Mattila, R. Serapioni, F. Serra Cassano. Characterizations of intrinsic rectifiability in
Heisenberg groups. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (5) Vol. IX (2010), 687–723.
Zbl 1229.28004.
[171] M. Miranda, Superfici cartesiane generalizzate ed insiemi di perimetro localmente finito sui
prodotti cartesiani. Ann. Scuola Norm. Sup. Pisa III, 18 (1964), 515–542. Zbl 0152.24402.
[172] M. Miranda, Un teorema di esistenza e unicitá per il problema dell’area minima in n vari-
abili. Ann. Sc. Norm. Sup. Pisa III, 19 (1965), 233–249. Zbl 0137.08201.
118 Francesco Serra Cassano
[173] M. Miranda Jr, Functions of bounded variation on “good” metric spaces. J. Math. Pures
Appl. (9), 82 (2003), no. 8, 975–1004. Zbl 1109.46030.
[174] J. Mitchell, On Carnot-Carathèodory metrics. J. Differ. Geom. 21 (1985), 35–45.
Zbl 0554.53023.
[175] F. Montefalcone, Some remarks in differential and integral geometry of Carnot groups. PhD
Thesis, University of Bologna, 2004.
[176] F. Montefalcone, Sets of finite perimeter associated with vector fields and polyhedral ap-
proximation. Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei. Mat. Appl. 14
(2003), no. 4, 279–295. Zbl 1072.49031.
[177] F. Montefalcone, Some relations among volume, intrinsic perimeter and one-dimensional
restrictions of BV functions in Carnot groups. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 4
(2005), no. 1, 79–128. Zbl 1150.49022.
[178] F. Montefalcone, Hypersurfaces and variational formulas in sub-Riemannian Carnot
groups. J. Math. Pures Appl. (9) 87 (2007), no. 5, 453–494. Zbl 1123.49035.
[179] F. Montefalcone, Stable H-minimal hypersurfaces. J. Geom. Anal. 25 (2015), no. 2, 820–
870. Zbl 1316.49050.
[180] R. Montgomery, A tour of subriemannian geometries, their geodesics and applications.
Mathematical Surveys and Monographs 91, American Mathematical Society, Providence,
RI, 2002. Zbl 1044.53022.
[181] R. Monti, Distances, boundaries and surface measures in Carnot-Carathéodory spaces.
PhD Thesis, University of Trento, 2001.
[182] R. Monti, Heisenberg isoperimetric problems. The axial case. Adv. Calc. Var. 1 (2008),
93–121. Zbl 1149.49036.
[183] R. Monti, Lipschitz approximation of H-perimeter minimizing boundaries. Calc. Var. Partial
Diff. Eq. 50 (2014), 171–198. Zbl 1290.49088.
[184] R. Monti, Isoperimetric problem and minimal surfaces in the Heisenberg group. Lecture
notes ERC course, Pisa, 2013. Zbl 06438519.
[185] R. Monti, M. Rickly, Convex isoperimetric sets in the Heisenberg group. Ann. Sc. Norm.
Super. Pisa Cl. Sci. (5) 8 (2009), no. 2, 391–415. Zbl 1170.49037.
[186] R. Monti, F. Serra Cassano, Surface measures in Carnot-Carathéodory spaces. Calc. Var.
Partial Diff. Eq. 13 (2001), 339–376. Zbl 1032.49045.
[187] R. Monti, F. Serra Cassano, D. Vittone, A negative answer to the Bernstein problem for
intrinsic graphs in the Heisenberg group. Boll. Unione Mat. Ital. (9) 1 (2008), no. 3, 709–727.
Zbl 1204.53048.
[188] R. Monti, D. Vittone, Sets with finite H-perimeter and controlled normal. Math. Z. 270
(2012), no. 1-2, 351–367. Zbl 1257.49047.
[189] R. Monti, D. Vittone, Height estimate and slicing formulas in the Heisenberg group. Anal.
PDE, 8 (2015), no. 6, 1421–1454. Zbl 06488453.
Bibliography 119
[191] P. Pansu, Une inégalité isopérimétrique sur le groupe de Heisenberg. C.R. Acad. Sci. Paris
I, 295 (1982), 127–130. Zbl 0502.53039.
[192] P. Pansu, Geometrie du group d’Heisenberg. These pour le titre de Docteur 3ème cycle,
Universite Paris VII (1982).
[193] S.D. Pauls, A notion of rectifiability modeled on Carnot groups. Indiana Univ. Math. J. 53
(2004), no. 1, 49–81. Zbl 1076.49025.
[194] S.D. Pauls, Minimal surfaces in the Heisenberg group. Geom. Dedicata 104 (2004), 201–
231. Zbl 1054.49029.
[195] S.D. Pauls, H-minimal graphs of low regularity in H1 . Comment. Math. Helv. 81 (2006),
337–381. Zbl 1153.53305.
[196] A. Pinamonti, Intrinsic Lipschitz graphs in Heisenberg groups and non linear sub-elliptic
PDEs. PhD thesis University of Trento, 2011, available http://eprints-phd.biblio.unitn.it/666/.
[197] A. Pinamonti, F. Serra Cassano, G. Treu, D. Vittone, BV minimizers of the area functional
in the Heisenberg group under the bounded slope condition. Ann. Sc. Norm. Super. Pisa Cl.
Sci. (5) XIV (2015), no. 3, 907–935. doi 10.2422/2036-2145.201305_004. Zbl 1330.49044
[199] L. Rifford, Ricci curvatures in Carnot groups. Math. Control and Rel. Fields 3 (2013), no.
4, 467–487. Zbl 1275.53034.
[200] S. Rigot, Counter example to the Besicovitch covering property for some Carnot groups
equipped with their Carnot-Carathéodory metric. Math. Z. 248 (2004), no. 4, 827–848.
Zbl 1082.53030.
[201] S. Rigot, Mass transportation in groups of type H. Commun. Contemp. Math. 7 (2005), no.
4, 509–537. Zbl 1089.49041.
[202] S. Rigot, Isodiametric inequality in Carnot groups. Ann. Acad. Sci. Fenn. Math. 36 (2011),
no. 1, 245–260. Zbl 1222.28008.
[203] S. Rigot, S. Wenger, Lipschitz non-extension theorems into jet space Carnot groups. Int.
Math. Res. Not. IMRN 2010, no. 18 (2010), 3633–3648. Zbl 1203.53028.
[206] M. Ritoré, C. Rosales, Area-stationary surfaces in the Heisenberg group H1 . Adv. Math.
219 (2008), no. 2, 633–671. Zbl 1158.53022.
120 Francesco Serra Cassano
[207] M. Rumin, Differential geometry on C-C spaces and application to the Novikov-Shubin num-
bers of nilpotent Lie groups. C. R. Acad. Sci. Paris Ser. I Math., 329 (1999), 985–990.
Zbl 0982.53029.
[208] M. Rumin, Around heat decay on forms and relations of nilpotent Lie groups. In Séminaire
de Théorie Spectrale et Géométrie, 19, Univ. Grenoble I (2001), 123–164. Zbl 1035.58021.
[209] X. Saint Raymond, C. Tricot, Packing regularity of sets in n-space. Math. Proc. Cambridge
Philos. Soc. 103 (1988), 133–145. Zbl 0639.28005.
[210] E. Sawyer, R.L. Wheeden, Weighted inequalities for fractional integrals on Euclidean and
homogeneous spaces. Amer. J. Math., 114 (1992), 813–874. Zbl 0783.42011.
[212] S. Semmes, Derivatives and difference quotients for Lipschitz or Sobolev functions on various
spaces. In Lecture Notes on Analysis in Metric Spaces, ed. by L. Ambrosio and F. Serra
Cassano, Edizioni Scuola Normale Superiore, Pisa (2000), 71–103. Zbl 1071.46042.
[213] F. Serra Cassano, D. Vittone, Graphs of bounded variation, existence and local bounded-
ness of non-parametric minimal surfaces in Heisenberg groups. Adv. Calc. Var. 7 (2014),
409–492. doi 10.1515/acv-2013-0105. Zbl 1304.49084.
[215] G. Speight, Lusin approximation and horizontal curves in Carnot groups. Preprint, 2014.
arXiv:1412.2531.
[216] E.M. Stein, Harmonic analysis. Princeton University Press, 1993. Zbl 0821.42001.
[217] R.S. Strichartz, Self-similarity on nilpotent Lie groups. In Geometric analysis (Philadelphia,
PA, 1991), vol. 140 of Contemp. Math. Amer. Math. Soc., Providence, RI, 1992, 123–157.
Zbl 0797.43004.
[218] A.D. Ukhlov, S.K. Vodop’yanov, Approximately differentiable transformations and change
of variables on nilpotent groups. Siberian Math. J. 37 (1996), no. 1, 62–78. Zbl 0870.43005.
[219] N.Th. Varopoulos, Analysis on Lie groups. J. Funct. Anal., 76 (1988), 346–410.
Zbl 0634.22008.
[220] N.Th. Varopoulos, L. Saloff-Coste, T. Coulhon, Analysis and geometry on groups. Cam-
bridge University Press, Cambridge, 1992. Zbl 0813.22003.
[221] D. Vittone, Submanifolds in Carnot groups. Tesi di Perfezionamento, Scuola Normale Su-
periore, Edizioni della Normale, Pisa, 2008. Zbl 1142.53046.
[222] D. Vittone, Lipschitz surfaces, perimeter and trace theorems for BV functions in Carnot-
Carathéodory spaces. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 11 (2012), no. 4, 939–998.
Zbl 1270.53068.
Bibliography 121
[223] S.K. Vodop’yanov, P-differentiability on Carnot groups in different topologies and related
topics. Proc. on Analysis and Geometry, 603–670, Sobolev Institute Press, Novosibirsk,
2000. Zbl 0992.58005.
[224] F.W. Warner, Foundations of differentiable manifolds and Lie groups. Springer, New York,
1983. Zbl 0516.58001.
Chapter 2
Hypoelliptic operators and some aspects
of analysis and geometry of
sub-Riemannian spaces
Nicola Garofalo1
Contents
whereas the second half focuses primarily on global aspects having to do with curva-
ture, and on the pervasive role that heat equation techniques play in connection with
it (the interested reader should be aware that this second part is introductory to the
lectures notes of Fabrice Baudoin, in this same volume).
Having said this, it is appropriate to start with a preliminary motivational dis-
cussion. Sub-Riemannian geometry is an extension of Riemannian geometry which
was created to model media with non-holonomic constraints: motion at any point is
allowed only along a limited set of directions which are prescribed by the physical
problem at hand. When the set of directions coincides with the whole tangent space
we obtain Riemannian geometry. Examples of systems with non-holonomic con-
straints include the Foucault pendulum; in robotics, when the controllable degrees
of freedom are less than the total degrees of freedom of a robot; a falling cat; and a
rolling sphere.
For instance, when studying the physics of semiflexible polymers one encounters
the so-called Mumford operator on M × (0, ∞), where we have denoted M = S1 × R,
then at every point of M × (0, ∞) the determinant of the matrix formed with the
coefficients of {X0 , X1 , X2 , X3 } is given by
1 0 0 0
0cos ϑ sin ϑ −1//
*. +
det .. = −1
− sin ϑ cos ϑ
.0 0 //
,0 − cos ϑ − sin ϑ 0 -
and thus at every point of R4 we have
operators of Hörmander type. Before stating the relevant result about this class let
us recall that, given an open set Ω ⊂ R N , a partial differential operator P with
coefficients in C ∞ (Ω) is said to be hypoelliptic in Ω if, given f ,u ∈ D 0 (Ω), such that
P u = f , one has u ∈ C ∞ (ω) for any open set ω ⊂ Ω in which f ∈ C ∞ (ω).
Examples of hypoelliptic operators are the Laplace operator ∆ in R N (this is the
content of the celebrated Caccioppoli–Cimmino–Weyl lemma,2 the heat equation
∆ − ∂t in R N +1 , and every strictly elliptic, or parabolic, second-order partial differ-
ential operator with C ∞ coefficients. According to the next fundamental result, the
class of second-order hypoelliptic operators is in fact much wider. We will need the
following
Definition 1.1. Let Ω ⊂ R N be an open set and consider a system of vector fields
{X0 , X1 , . . . , X m } in C ∞ (Ω). We say that they satisfy the finite rank condition in Ω if
at every point of Ω one has
Condition (1.3) means that at every point of Ω the vector fields and a sufficiently
high number of their commutators
X j1 , [X j1 , X j2 ], [X j1 , [X j2 , X j3 ]], . . . , [X j1 , [X j2 , [X j3 , . . . , X j k ]]], . . . , ji = 0, 1, . . . , m,
generate the whole of R N , i.e., the tangent space. In other words, at every point of
Ω among such differential operators, there exist N which are linearly independent.
One has the following celebrated result; see [Ho67].
Theorem 1.2 (Hörmander (1967)). Suppose that {X0 , X1 , . . . , X m } satisfy the finite
rank condition (1.3) in Ω, and let c ∈ C ∞ (Ω). Then, the second-order differential
operator
Xm
L= X 2j + X0 + c (1.4)
j=1
is hypoelliptic in Ω.
1.2 Mumford and the roto-translation group. Since the Mumford operator
can be written as M = X12 + X0 , by virtue of the property (1.2) above and of Theorem
1.2, M is hypoelliptic on the whole of M × R. Furthermore, we can endow R4 with
2This fundamental lemma states that, given an open set Ω ⊂ R N and a distributional solution U of ∆U = 0
in Ω, there exists a u ∈ C ∞ (Ω) such that U = u as distributions. It is usually referred to as Weyl’s lemma;
see [W40]. However, as pointed out by Miranda in his monograph [Mi70, p.122], “This nomenclature is inap-
propriate because the same proposition can be found in a more general form in Caccioppoli’s memoir [C37].
And even the extension to the case of arbitrary M given by G. Cimmino in [Ci38/1], [Ci38/2] precedes Weyl’s
memoir by some years. To be accurate, Caccioppoli and Cimmino considered the case N = 2, Weyl the case
N = 3.”
126 Nicola Garofalo
a group law ◦ with respect to which the operator M is left invariant. The group law
is given by
m RT = Lie(X0 , X1 ),
the Lie algebra generated by the vector fields X0 , X1 , then the corresponding Lie
group, known as the roto-translation group, is given by RT = (R4 , ◦); see [M94].
Notice that m RT is not nilpotent. As a consequence, RT is not nilpotent either.
As we will see, such a lack of nilpotency is closely connected to a lack of dilation
invariant structure.
If we let g = (ϑ, x, y,t) and we define the left translation by L g (g 0 ) = g ◦ g 0, then
it is easy to check that for every g, g 0 ∈ R4 we have
M ( f ◦ L g ) = (M f ) ◦ L g ,
hypoellipticity more than thirty years before Hörmander’s work [Ho67]. It is worth
mentioning that the motivating example in the opening of [Ho67] is precisely equa-
tion (1.7), and some generalizations of it. We also mention that a scale-invariant
Harnack inequality for the operator K was first proved in the work [GL90].
As we saw in Section 1.2 the roto-translation Lie group RT associated with the
Mumford operator is not nilpotent. There is however a nilpotent group which, at
the infinitesimal level, plays for the roto-translation group RT the same role played
by the tangent space of a smooth manifold. Such a nilpotent group is generated by
vector fields whose coefficients are the Taylor polynomials of order 2 of those in the
expression of the Mumford operator M (note that we need commutators of order 3
to generate the Lie algebra m RT ). They are given by
x2
X 1 = ∂x , X0 = ∂y + x∂z − ∂t , (1.8)
2
and the approximating operator is given by
L = X12 + X0 . (1.9)
For a partial differential equation related to (1.9) one should see the seminal work of
Kac [Kac49]. The operator in (1.9) is modeled on the Kolmogorov operator K above,
but with an important difference: it contains superlinear powers of the variable x, and
the highest degree of such powers is an even number. This makes the analysis of the
operator L more difficult than that of K. A scale-invariant Harnack inequality for the
operator (1.9) will appear in [GMP15].
Notice that one has
at every point (x, y, z,t) ∈ R4 , the vector fields {X0 , . . . , X3 } generate the Lie algebra
k2 = Lie{X0 , X1 }.
L ( f ◦ δ λ ) = λ 2 (L f ) ◦ δ λ ,
128 Nicola Garofalo
Q = 1 + 4 + 3 + 2 = 10.
1.4 The exponential map and the group law. Since the vector fields X0 , X1
in (1.8) have real analytic coefficients, according to a general result by Bonfiglioli
and Lanconelli (see [BL12, Theorem 1.1]), there exists a group law g ◦ g 0 in R4 such
that, if we define the left-translation operator L g (g 0 ) = g ◦ g 0, then the vector fields
X0 , X1 are left invariant with respect to L g . This group law is found in (1.15) below.
To actually compute it we proceed as follows. With X1 , X0 as in (1.8) and
X2 , X3 as in (1.10), we set Y1 = X1 ,Y2 = X0 ,Y3 = X2 ,Y4 = X3 . Given a point
u = (u1 , . . . ,u4 ) ∈ R4 , we have
This gives
γ1 (s) = x + su1 ,
3 u1 u3 2
γ2 (s) = y + u22 sx 2 + s3 u12 + s2 u1 x + u3 sx + + su4 ,
2 s
2
γ3 (s) = z + su2 x + s2 u1 u2 + su3 ,
γ4 (s)
= t − su2 ,
thus obtaining
x + u1
u2 u1 u3
y+ x2 + 1 2
3 u1 + u1 x + u3 x + + u4 //
*. +
Exp(u · Y )(g) = γ(1) = .. 2 2
// . (1.12)
. z + u3 + u2 x + 12 u1 u2
, t − u2 -
2 Hypoelliptic operators, etc. 129
We thus find
u1
*. u12 u2 u1 u3
Exp(u · Y )(0) = .. 6 + u2 1 u2+ u4 // .
+/
. u3 + 2 /
, −u 2 -
If we set v = Exp(u · Y )(0), then the inverse mapping is given by
v1
*. −v4 +/
Log(v) = .. v3 + v12v4
// . (1.13)
. /
v1 v3 v 12 v 4
,v2 − 2 − 12 -
g ◦ g 0 = exp(Log(g 0 ) · X )(g).
u0 x0
*. 10 +/ *. −t 0
u
+/
u 0 = .. 20 // = .. 0 0 // .
.u / . 3
z 0 + x2t /
0 0 x0 z0 (x 0 ) 2 t 0
,u4 - , y − 2 − 12 -
Applying (1.12) with u replaced by this u 0, we finally have after some elementary
computations,
x + u10
*. u 20 u 10 u 30 +
y+ x 2 + 13 (u10 ) 2 + u10 x + u30 x + + u40 //
g ◦ g = ...
0 2 2 (1.14)
z + u30 + u20 x + 12 u10 u20
//
. /
, t − u20 -
x + x0
0 2+
y + y 0 + xz 0 − t 2x //
*.
= .. .
. z + z 0 − t 0 x //
, t + t0 -
Equation (1.14) is the sought-for group law. If we write it in the more traditional
form
t 0 x2
g◦g 0 = (x, y, z,t)◦(x 0, y 0, z 0,t 0 ) = (x+ x 0, y+ y 0 + xz 0 − , z+z 0 −t 0 x,t +t 0 ), (1.15)
2
130 Nicola Garofalo
the identity element with respect to ◦ is e = (0, 0, 0, 0), and the inverse of g with
respect to ◦ is given by
x2
g −1 = − x, −y + xz + t, −z − xt, −t . (1.16)
2
If we define the left-translation operator as L g (g 0 ) = g ◦ g 0, then its differential is
given by
1 0 0 0
2+
0 1 x − x2 //
*.
dL g = .. , and so det dL g ≡ 1 ∀g ∈ R4 . (1.17)
.0 0 1 −x //
,0 0 0 1 -
Furthermore, by (1.17) one sees that Lebesgue measure is invariant with respect
to L g .
Proposition 1.3. If we denote K2 = (R4 , ◦, δ λ ), with δ λ given by (1.11), then K2 is a
four-dimensional nilpotent Lie group of step r = 3. Furthermore, Lebesgue measure
is invariant with respect to the left translations on K2 . If k2 denotes the Lie algebra
of K2 , then we have
k2 = Lie{X0 , X1 },
and the operator L = X12 + X0 is left invariant on K2 and δ λ -homogeneous of
degree 2.
Once we have the explicit expression of the group law (1.15), we can a posteriori
verify the left invariance of X0 , X1 directly. Considering in fact
t 0 x2
F (g 0 ) = F (x 0, y 0, z 0,t 0 ) = f ◦ L g (g 0 ) = f x + x 0, y + y 0 + xz 0 − , z + z 0 −t 0 x,t +t 0 ,
2
we have
x2
Fx 0 = f x , Fy 0 = f y , Fz 0 = x f y + f z , Ft 0 = − fy − x fz + ft .
2
This gives
X1 ( f ◦ L g ) = (X1 f ) ◦ L g , (1.18)
(x 0 ) 2
X0 ( f ◦ L g ) = Fy 0 + x 0 Fz 0 − Ft 0 (1.19)
2
(x 0 ) 2 x2
= f y + x 0 (x f y + f z ) − f t + fy + x fz
2 2
(x + x 0 ) 2
= f y + (x + x 0 ) f z − f t = (X0 f ) ◦ L g .
2
Formulas (1.18) and (1.19) show that the vector fields X0 , X1 are left invariant with
respect to (1.15), and therefore so is L in (1.9) above.
2 Hypoelliptic operators, etc. 131
2 Carnot groups
The above discussion of the Mumford operator (1.1) and of its “approximating”
higher-order Kolmogorov operator (1.9) brings us to introduce a class of Lie groups
which, besides supporting notable families of differential operators which are left-
translation invariant, also share with Euclidean R N a so-called homogeneous struc-
ture associated with the existence of dilations. As we saw in the discussion in Section
1.4 above, such groups naturally arise as approximating “tangent spaces” to the sub-
Riemannian spaces associated with a Hörmander-type operator. This guiding idea
was in fact at the basis of a visionary program which Stein presented in his address
at the 1970 International Congress of Mathematicians in Nice [S71]. The program
culminated in the celebrated works of Rothschild and Stein [RS76], and of Nagel,
Stein, and Wainger [NSW85]. The reader who wants to become acquainted with the
theory of Carnot groups should consult the classical books [FS82], [CGr90], and the
more recent exhaustive contribution [BLU07] which also contains notable applica-
tions to the potential theory associated with such groups.
We recall that a Carnot group of step r is a connected, simply connected Lie
group G whose Lie algebra g admits a stratification
g = V1 ⊕ · · · ⊕ Vr ,
From the assumptions (2.1) on the Lie algebra one immediately sees that any basis
of the first layer V1 bracket generates the whole Lie algebra g. Because of the special
role played by V1 , this vector subspace of the Lie algebra is usually called the hor-
izontal layer of the stratification. In the case in which r = 1 we are in the Abelian
situation in which g = V1 , and thus G is isomorphic to Rm , where m = dim V1 . We
are thus back in Rm , there is no non-commutative geometry involved, and everything
is classical. The case in which r = 2 is the first genuinely non-Abelian case. We will
analyze some important instances of such Carnot groups of step 2.
A Carnot group of step r is also known in the literature as a stratified nilpotent
Lie group of step r; see [Fo75]. As a rule, we will use letters g, g 0, g 00, g0 for points in
G, whereas we will reserve the letters ξ, ξ 0, ξ 00, ξ0 , η, for elements of the Lie algebra
g. We will always denote by e ∈ G the group identity. The stratification (2.1) of
the Lie algebra g allows a natural one-parameter family of non-isotropic dilations to
be introduced on g. To do this we assign to each element of the layer Vj the formal
degree j. Accordingly, if ξ = ξ1 + · · · + ξr ∈ g, with ξ j ∈ Vj , one defines dilations
on g by the rule
∆ λ ξ = λξ1 + · · · + λ r ξr . (2.2)
132 Nicola Garofalo
Let us note right away that | · |g is homogeneous of degree 1 with respect to the group
dilations (2.2) above, i.e.,
|∆ λ ξ |g = λ|ξ |g . (2.4)
Using the exponential map exp : g → G, the anisotropic dilations (2.2) can be
transferred to the group G as follows:
One can also use the exponential map and (2.3) to define a so-called gauge on G
by letting for g = exp ξ,
|g|G = |ξ |g . (2.6)
We observe that
|δ λ (g)|G = |δ λ (exp ξ)|G = exp ◦∆ λ ◦ exp−1 (exp ξ) = | exp(∆ λ ξ)|G (2.7)
G
= (by (2.6)) |∆ λ ξ |g = (by (2.4)) λ|ξ |g = λ|g|G ,
The motivation for this name comes from the important equation (2.11) below.
In the non-Abelian case r > 1, one clearly has Q > N.
A fundamental tool in differential geometry is the exponential map, which sends
the tangent space at a point to the manifold itself. When, in addition, the manifold
2 Hypoelliptic operators, etc. 133
is a real Lie group, then the exponential map possesses much deeper properties.
For a Carnot group G a remarkable fact is that the exponential map exp : g → G
defines an analytic diffeomorphism of the Lie algebra g onto G; see, e.g., the book
by Varadarajan [V87, Sec. 2.10 onwards]. Perhaps, the most fundamental property
of the exponential map is the Baker–Campbell–Hausdorff formula (see, e.g., [V87,
Sec. 2.15])
1 1
exp(ξ) exp(η) = exp ξ + η + [ξ, η] + [ξ, [ξ, η]] − [η, [ξ, η]] + · · · , (2.9)
2 12
where the dots indicate commutators of order 4 and higher. Furthermore, since by
(2.1) above all commutators of order r and higher are trivial, in every Carnot group
the Baker–Campbell–Hausdorff series on the right-hand side of (2.9) is finite.
It should be clear to the reader that, using (2.9), one can recover the group law
g ◦ g 0 in G from the knowledge of the algebraic commutation relations between the
elements of its Lie algebra. We will respectively denote by
L g (g 0 ) = g ◦ g 0, Rg (g 0 ) = g 0 ◦ g, (2.10)
X j (g) = (L g )∗ (e j ), j = 1, . . . , m,
i.e., the vector fields X1 , . . . , X m are invariant with respect to the left translations on
G. Furthermore, they are homogeneous of degree 1 with respect to the non-isotropic
dilations (2.5), i.e.,
X j (δ λ F) = λδ λ (X j F), j = 1, . . . , m. (2.16)
Definition 2.2. The sub-Laplacian associated with the basis {e1 , . . . , em } is defined
by the formula
Xm
∆H u = X 2j u. (2.17)
j=1
∆ H (δ λ u) = λ 2 δ λ (∆ H u).
Let π j : g → Vj denote the projection onto the jth layer of g. Since the ex-
ponential map exp : g → G is a global analytic diffeomorphism, we can define
analytic maps ξ j : G → Vj , j = 1, . . . ,r by letting ξ j = π j ◦ exp−1 . The notation
{e j,1 , . . . , e j, m j }, j = 1, . . . ,r will indicate a fixed orthonormal basis of the jth layer
Vj . For g ∈ G, the projections of the exponential coordinates of g onto the layer Vj ,
j = 1, . . . ,r are defined as
div X i = 0, i = 1, . . . , m, (2.22)
where div X i indicates the divergence of X i with respect to the exponential coordi-
nates.
2.2 Carnot groups of step r = 2. Since Carnot groups of step r = 2 often play
a special role in analysis and geometry, it will be convenient to have a simplified
notation for objects in the horizontal layer V1 , and in the first vertical layer V2 . For
simplicity, we set m = m1 , k = m2 , and let
Notice that for a Carnot group of step r = 2 the homogeneous dimension of the
group, defined in (2.8) above, is given by the number
Q = m + 2k.
We indicate with
so that
k
X
[ei , e j ] = bisj ε s , i, j = 1, . . . , m. (2.26)
s=1
Since [ei , e j ] = −[e j , ei ], it should be obvious that bisj = −bsji . As in (2.23)
we use special notation for the left-invariant vector fields associated with the basis
{e1 , . . . , em } and {ε 1 , . . . , ε k }, and let
Proposition 2.6. Let G be a Carnot group of step 2. Then, in the exponential coor-
dinates (z,t) one has
k k m
1X ∂ 1 XX s ∂
X i = ∂z i + hJ (ε ` )z, ei i∂t ` = − bi j z j , (2.28)
2 `=1 ∂z i 2 s=1 `=1 ∂t s
∂
Ts = , (2.29)
∂t s
k k
1 X X
∆ H = ∆z + hJ (ε ` )z, J (ε `0 )zi∂t ` ∂t `0 + ∂t ` Θ` , (2.30)
4 `,`0 =1 `=1
m m
b`i j z j ∂z i .
X X
Θ` = hJ (ε ` )z, ei i∂z i = − (2.31)
i=1 i, j=1
hn = V1 ⊕ V2 ,
where V1 = R2n
x, y × {0}t and {0}R2n × Rt . If we adopt the canonical basis
and thus the Heisenberg algebra hn is a stratified Lie algebra which is nilpotent of
step r = 2. Consequently, the Heisenberg group provides an example of a Carnot
group of step r = 2. For a beautiful introduction to the group Hn we refer the
reader to the book [CDPT07]. One should also see the monumental works [S93] and
[BLU07].
138 Nicola Garofalo
The subspace V1 is called the horizontal layer, whereas V2 is called the vertical
layer of the Heisenberg algebra. It is clear that V2 constitutes the center of hn with
respect to (2.33).
If we now identify g, g 0 ∈ Hn respectively with x 1 e1 + · · · + x n en + y1 en+1 + · · · +
yn e2n + tε 1 and g 0 = x 10 e1 + · · · + x n0 en + y10 en+1 + · · · + yn0 e2n + t 0 ε 1 , then from (2.32)
and the Baker–Campbell–Hausdorff formula (2.9) above, which in the present case
reads
1
exp(ξ) exp(η) = exp ξ + η + [ξ, η] ,
2
we easily obtain that the non-Abelian group multiplication on Hn is given by
[X i , X n+ j ] = T δ i j , i, j = 1, . . . , n. (2.35)
The sub-Laplacian on Hn with respect to the canonical basis {e1 , . . . , e2n } is given
by
n
X
∆H = (X 2j + X n+
2
j ).
j=1
This operator has a non-negative characteristic form, but it fails to be elliptic at every
point g = (x, y,t) ∈ Hn . This circumstance can be verified by computing the matrix
A(x, y,t) = A(x, y) associated with the principal symbol of ∆ H and verifying that
the eigenvalues of the matrix are all non-negative, but that one of them vanishes
identically. In fact, in the real coordinates (x, y,t) we have
n
|x| 2 + |y| 2 2 X
∆ H = ∆ x, y + ∂t + ∂t x i ∂y i − yi ∂x i .
4 i=1
2 Hypoelliptic operators, etc. 139
It is easy to see that, when n = 1 and we are dealing with the three-dimensional
Heisenberg group H1 , the matrix A(x, y) has eigenvalues
|x| 2 + |y| 2
λ 1 = 0, λ 2 = 1, λ3 = 1 + .
4
If we assign to the elements of Vj , j = 1, 2 the formal degree j, then the associated
non-isotropic dilations of Hn are given by
all other brackets being assumed trivial. The relations (2.37) show that
so that the Engel group E is a Carnot group of step r = 3. We observe that the
homogeneous dimension of the Engel group E is
Q = m1 + 2m2 + 3m3 = 7.
140 Nicola Garofalo
We will denote with (x, y), t, and s respectively the variables in V1 , V2 , and V3 ,
so that any ξ ∈ e can be written as ξ = xe1 + ye2 + te3 + se4 . If g = exp(ξ), we
will identify g = (x, y,t, s). The group law in E is given by the Baker–Campbell–
Hausdorff formula (2.9). In exponential coordinates, if g = exp(ξ), g 0 = exp(ξ 0 ),
we have
1 1
g ◦ g 0 = ξ + ξ 0 + [ξ, ξ 0] + [ξ, [ξ, ξ 0]] − [ξ 0, [ξ, ξ 0]] .
2 12
Using such formula, a computation based on (2.37) gives (see also [CGr90, ex.
1.2.5])
g ◦ g 0 = x + x 0, y + y 0,t + t 0 + P3 , s + s 0 + P4 ,
where
1
P3 = x y 0 − yx 0 ,
(2.38)
2
1 0 1 2 0 0
P4 = (xt − t x 0 ) + x y − x x 0 (y + y 0 ) + yx 2 . (2.39)
2 12
Once we have the group law, we define the left-translation operator L g (g 0 ) =
g ◦ g 0 on E. Denoting by (L g )∗ its differential, we can associate with the basis
{e1 , e2 , e3 , e4 } the following left-invariant vector fields on E:
X2 = ∂y + 2 ∂t + 12 ∂s ,
x x
∂ ∂
(2.41)
T = ∂t + x2 ∂s ,
S = ∂ .
∂s
all other commutators being trivial. We note that the action of X1 , X2 ,T on a function
on E which is independent of the variable s reduces to the action of the corresponding
vector fields in H1 . The sub-Laplacian on E with respect to the basis {e1 , e2 } of the
horizontal layer V1 is given by
∆ H = X12 + X22 .
Free ebooks ==> www.Ebook777.com
t s = t s (g) = ht(g), ε s i.
hJ (t)z, zi = 0, z ∈ V1 ,t ∈ V2 .
| J (t)z| ≥ B|z||t|, z ∈ V1 , t ∈ V2 .
Definition 2.8. A Carnot group of step 2 is called of Heisenberg type if for every
t ∈ V2 such that |t| = 1, the mapping J (t) is orthogonal. This is equivalent to saying
that for every z, z 0 ∈ V1 , and every t ∈ V2 , one has
www.Ebook777.com
142 Nicola Garofalo
We will need the following simple, yet crucial, result about groups of Heisenberg
type.
Proposition 2.9. If G is of Heisenberg type, then for every t,t 0 ∈ V2 and any z ∈ V1
one has
hJ (t)z, J (t 0 )zi = |z| 2 ht,t 0i.
In particular, we obtain for every `, ` 0 = 1, . . . , k,
Proof. By polarization,
1( )
hJ (t)z, J (t 0 )zi = | J (t)z + J (t 0 )z| 2 − | J (t)z − J (t 0 )z| 2
4
1( )
= | J (t + t 0 )z| 2 − | J (t − t 0 )z| 2
4
|z| 2 ( )
= |t + t 0 | 2 − |t − t 0 | 2
4
= |z| 2 ht,t 0i,
0 −1 0 0
1 0 0 0 //
*. +
A=.. .
.0 0 0 −2//
,0 0 2 0 -
Since det A , 0, the matrix A is non-singular and therefore G is a Metivier group.
But G is not of H-type since A < O(4).
The following expressions in exponential coordinates of the vector fields X i , and
of the sub-Laplacian ∆ H will be useful; see Proposition 2.6 above.
2 Hypoelliptic operators, etc. 143
Lemma 2.11. Let G be a Carnot group of step 2. Then, in the exponential coordi-
nates (z,t) one has
k
1X
X i = ∂z i + hJ (ε ` )z, ei i∂t ` , (2.46)
2 `=1
k k
1 X X
∆ H = ∆z + hJ (ε ` )z, J (ε `0 )zi∂t ` ∂t `0 + ∂t ` Θ` , (2.47)
4 `,`0 =1 `=1
k
|z| 2 X
∆ H = ∆z + ∆t + ∂t ` Θ` . (2.49)
4 `=1
Remark 2.12. Groups of Metivier type are important since they are the largest known
class of Carnot groups in which every sub-Laplacian is real-analytic hypoelliptic.
Thus, sub-Laplacians on Metivier groups share the same fundamental property of
the classical Laplace operator in R N .
To better elucidate Remark 2.12 we recall that in the framework of Carnot groups
of step 2, Helffer proved in [Hel] that if a sub-Laplacian in G is real-analytic hypoel-
liptic, then G must be a Metivier group. Thanks to a result of Metivier [Me2] (see
also [Me1]), it is known that if G is a Metivier group, then a sub-Laplacian ∆ H
is real-analytic hypoelliptic if and only if it is hypoelliptic. Since by Hörmander’s
Theorem 1.2 above every sub-Laplacian is hypoelliptic, it follows that in Metivier
groups all sub-Laplacians are real-analytic hypoelliptic. However, outside groups of
step 2 there is (to the best of the writer’s knowledge) no known example of a real-
analytic sub-Laplacian. It is plausible (for this conjecture see Rothschild in [R84])
that no sub-Laplacian on a Carnot group which is not a Metivier group should be
real-analytic hypoelliptic, but this question seems to be open.
These considerations lead us to a fundamental question.
Open problem: Is it true that in a Carnot group every sub-Laplacian possesses the
unique continuation property, or even the strong unique continuation property?
It is known that a real-analytic function cannot vanish to infinite order at one
point of a connected open set, without being identically zero. This is known as the
144 Nicola Garofalo
strong unique continuation property (sucp). From what we said above, since in every
Metivier group harmonic functions are real-analytic, they possess the sucp. What
happens with the sucp beyond Metivier groups? We can ask for a weaker property:
can harmonic functions in a Carnot group vanish in an open subset of a connected
open set? These important questions presently constitute terra incognita!
2.6 Groups of step 2: some useful computations. Every Carnot group can
be endowed with a non-isotropic gauge which respects the stratification of its Lie
algebra; see, e.g., [Fo75] and also [FS82]. In the special case of groups of step r = 2
it is particularly convenient to introduce the gauge % : G → [0, ∞),
% ◦ δ(g) = λ %(g),
i.e., % is positively homogeneous of degree 1. We note explicitly that, except for the
normalizing factor 16, (2.50) is precisely the non-isotropic gauge defined by (2.3),
(2.6).
Lemma 2.13. Let G be a Carnot group of step 2, and consider the gauge (2.50).
Then,
1
|∇ H %| 2 = 6 |z| 6 + 16| J (t)z| 2 ;
(2.51)
%
see (2.14) above. In particular, if G is of Heisenberg type, then
|z| 2
|∇ H %| 2 = . (2.52)
%2
% = (r 4 + 16s2 ) 1/4 .
Lemma 2.15. Let G be a group of Heisenberg type. The following formulas hold:
k 2
∆ H (|t| 2 ) = |z| , (2.60)
2
|∇ H (|t| 2 )| 2 = |z| 2 |t| 2 , (2.61)
h∇ H (|z| ), ∇ H (|t| )i = 0.
2 2
(2.62)
Proof. Using the second formula in (2.57) we find
k
X k
X k
X
∆ H (|t| ) = 2
2
t ` ∆ H (t ` ) + 2 |∇ H (t ` )| = 2 2
|∇ H (t ` )| 2 .
`=1 `=1 `=1
Next, equation (2.55) gives
m 2 1 Xm m
X 1X
|∇ H (t ` )| 2 = X j (t ` ) = h[z, e j ], ε ` i2 = hJ (ε ` )z, e j i2
j=1
4 j=1
4 j=1
= 41 | J (ε ` )z| 2 = 41 |z| 2 ,
where in the second to the last equality we have used (2.45). Substitution of this
equation into the previous one gives
k 2
∆ H (|t| 2 ) = |z| .
2
Next, we have
m
X 2 m X
X k 2 m X
X k 2
|∇ H (|t| 2 )| 2 = X j (|t| 2 ) = X j (t 2` ) =4 t ` X j (t ` )
j=1 j=1 `=1 j=1 `=1
m X
X k 2 m
X 2 m
X 2
= t ` h[z, e j ], ε ` i = h[z, e j ],ti = hJ (t)z, e j i
j=1 `=1 j=1 j=1
Q−1
∆H % = |∇ H %| 2 in G \ {e}.
%2
Proof. Consider the function K = %4 = |z| 4 + 16|t| 2 . The chain rule gives ∇ H K =
4%3 ∇ H %, and therefore
Next, we have
where in the last term we have used (2.60) above. By (2.56) and (2.58) we find
Substituting this equation into (2.63), and using (2.52) above, we conclude that
Q − 1 |z| 2
∆H % = ,
%2 %2
where CN = 1
(N −2)σ N −1 , and σ N −1 represents the surface measure of the unit sphere
S N −1 ⊂ R N .
The proof of Theorem 3.1 will be divided into several steps. We start as in [Fo73],
but soon after we take an interesting, different path which has the advantage of un-
covering a beautiful connection with the Yamabe equation from CR geometry. We
emphasize that we could have presented a completely different proof of Theorem
3.1 following more classical arguments based on representation formulas. Although
such an approach also has its obvious appeal, had we followed it we would not have
met the beautiful Yamabe equation.
As in [Fo73] for every ε > 0 we consider the function
1/4
%ε (z,t) = (ε 2 + |z| 2 ) 2 + 16|t| 2 . (3.3)
|z| 2
|∇ H %ε | 2 = ,
%2ε
Q−1 mε 2
∆ H %ε = |∇ H %ε | 2 + 3 .
%ε %ε
Substitution in (3.4) gives the first part of the lemma. Next we compute ∆ H K.
Applying Lemma 2.15 again one finds
∆ H K = ∆ H (ε 2 + |z| 2 ) 2 + 16∆ H (|t| 2 ) = ∆ H (ε 2 + |z| 2 ) 2 + 8k |z| 2 .
Finally, replacing (3.6) and (3.7) in (3.5) we obtain the second part of the lemma.
We now consider a function h ∈ C 2 ([0, ∞)) and form a new function v(g) =
h( %ε (g)). The chain rule gives
∆ H v = h 00 ( %ε )|∇ H %ε | 2 + h 0 ( %ε )∆ H %ε .
Q−1 mε 2
∆ H v = h 00 ( %ε )|∇ H %ε | 2 + h 0 ( %ε ) |∇ H %ε | 2 + h 0 ( %ε ) 3
%ε %ε
2
" #
Q−1 0 mε
= h 00 ( %ε ) + h ( %ε ) |∇ H %ε | 2 + h 0 ( %ε ) 3 .
%ε %ε
Q−1 0
If we now choose h(t) = t 2−Q , then it is clear that h 00 (t) + t h (t) = 0, and we
obtain from the previous equation
mε 2 0
∆H v = h ( %ε ) = −(Q − 2)mε 2 %−(Q+2)
ε = −(Q − 2)mε 2 v (Q+2)/(Q−2) .
%3ε
Theorem 3.4. Let G be a group of Heisenberg type and for any ε > 0 consider the
function vε = %2−Q
ε . Then, v solves the semilinear Yamabe equation in G,
We now return to the
Proof of Theorem 3.1. We begin by observing that for every ε > 0 we have
%ε = εδε −1 ◦ %1 ,
www.Ebook777.com
152 Nicola Garofalo
Taking the claim for granted for the moment we now observe that for every g ∈ G
we have
%ε (g) → %(g), as ε → 0.
Therefore, for every ϕ ∈ C0∞ (G) one has in the duality of L 2 (G),
then we have proved that for every ϕ ∈ C0∞ (G) one has
where δ denotes the Dirac delta. This can be expressed by saying that the distribution
Γ = C(m, k) %2−Q satisfies the equation
∆ H Γ = −δ, in D 0 (G).
To complete the proof of the theorem we thus need to establish the Claim. Now,
since v1 ∈ C ∞ (G), there is no problem with local integrability. It will thus suffice to
prove that
v1(Q+2)/(Q−2) ∈ L 1 (G \ { % < 1}).
But,
v1(Q+2)/(Q−2) = %−(Q+2)
1 ≤ %−(Q+2) in G \ { % < 1},
2 Hypoelliptic operators, etc. 153
where C > 0 is the universal constant defined in Theorem 3.1. Then, for every g ∈ G
the function g 0 → Γ(g 0, g) is a fundamental solution of −∆ H with singularity at g,
i.e.,
∆ H Γ(·, g) = −δ g , in D 0 (G).
v1 = %2−Q
1
then it is easy to recognize that the function u = λv1 solves the normalized Yamabe
equation in G,
∆ H u = −u (Q+2)/(Q−2) . (3.13)
154 Nicola Garofalo
Notice that if we integrate u∆ H u on G, and use the fact that u → 0 at infinity, one
easily obtains by an integration by parts,
Z Z
u∆ H u dg = − |∇ H u| 2 dg.
G G
Q+2 2Q
On the other hand, since Q−2 +1= Q−2 , we see from (3.13) that
Z Z
u∆ H u dg = − u2Q/(Q−2) dg.
G G
2Q
We note explicitly that the exponent 2∗ = Q−2 satisfies the equation
1 1 1
− = .
2 2∗ Q
The number 2∗ constitutes the so-called Sobolev exponent relative to 2. The general
question involved is the following: does there exist a universal constant S2 > 0 such
that for every function ϕ ∈ C0∞ (G) one has
Z 1/2∗ Z 1/2
2∗
|ϕ| dg ≤ S2 |∇ H ϕ| 2 dg ? (3.15)
G G
This question is highly non-trivial, especially since, as we have seen, the hori-
zontal gradient ∇ H ϕ of the function ϕ is degenerate and at every point it misses a
good amount of directions in the tangent space. As a consequence of a basic result
of Folland and Stein [Fo75], the inequality (3.15) does hold in greater generality for
every Carnot group G. However, the question of the existence of the best constant
in (3.15) in much subtler, and it is intimately connected to that of the classification
of all positive solutions of the Yamabe equation (3.13). In this context the partial
differential equation in (3.13) arises in the study of the following problem.
Definition 3.7 (CR Yamabe problem). Given a compact, strictly pseudo-convex CR
manifold M, find a choice of contact form ϑ for which the Webster–Tanaka pseudo-
hermitian scalar curvature Rϑ is constant.
For the relevant notions from CR geometry we refer the reader to the beautiful
monograph [DT06]. In the important papers [JL87, JL88] Jerison and Lee solved
the CR Yamabe problem in the case in which n ≥ 2 and (M, ϑ) is not locally CR
2 Hypoelliptic operators, etc. 155
equivalent to the sphere S2n+1 of Cn . After more than a decade, in [GY01] Gamara
and Yacoub solved the CR Yamabe problem for the case in which (M, ϑ) is locally
CR equivalent to the sphere S2n+1 for all n. Subsequently, Gamara in [G01] solved
the CR Yamabe problem in the remaining case n = 1, thus completing the resolution
of the CR Yamabe conjecture for all dimensions.
Similarly to the classical situation, a crucial step in the analysis of the CR Yamabe
problem was the explicit computation of the extremal functions in the horizontal
Sobolev embedding (3.15) in the special situation when G is the Heisenberg group
Hn . Jerison and Lee made the deep discovery that, up to group translations and
dilations, a suitable multiple of the function
−(Q−2)/4
u(z,t) = (1 + |z| 2 ) 2 + t 2 (3.16)
is the only positive entire solution of (3.13) in Hn . Here, we have denoted by (z,t),
z = (x, y) ∈ R2n , t ∈ R the variable point in Hn . In this connection, it is worth
mentioning the following basic question.
Open problem: Generalize Jerison and Lee’s result for the Heisenberg group in
[JL88] to all groups of Heisenberg type. In other words, compute the best constant
in the horizontal Sobolev embedding (3.15) for all groups of Heisenberg type. Equiv-
alently, prove that in every such group, all positive solutions of equation (3.13) are
given by left translations of a suitable multiple of the function
! (Q−2)/4
m(Q − 2)ε 2
Kε (g) = , g ∈ G.
(ε 2 + |x(g)| 2 ) 2 + 16|y(g)| 2
This problem is considerably harder than its already difficult Heisenberg group
predecessor. Some interesting partial progress is contained in [GV01, BU04,
IMV10].
Similarly to the classical case, a necessary condition for the validity of (3.17) is
obtained by means of the group non-isotropic dilations. Suppose in fact that (3.17)
does hold, and for λ > 0 consider the function ϕ λ = ϕ ◦ δ λ ∈ C0∞ (G). Substituting
156 Nicola Garofalo
this function into (3.17), and noting that ∇ H (ϕ λ ) = λδ ◦ ∇ H ϕ (see (2.16) above),
we find
Z ! 1/q Z ! 1/p
q
|ϕ(δ λ g)| dg ≤ S p,q λ |∇ H ϕ(δ λ g)| dg
p
.
G G
It is then clear that, in order for (3.17) to hold with a universal constant S p,q , one
must have
1 1 1
− = . (3.18)
p q Q
Let us note immediately that the necessary condition (3.18) for (3.17) forces
1
p − q1 > 0, hence one must have the gain in integrability 1 ≤ p < q. Furthermore,
since p1 − Q1 = q1 > 0, we must also have 1 ≤ p < Q. Finally, let us observe that,
given a number 1 ≤ p < Q, equation (3.18) completely determines the exponent q
in terms of p and Q, and in fact it results in
pQ
q= .
Q−p
This number is usually denoted by p∗ and it is called the Sobolev exponent relative
to p.
The fact that (3.18) is also sufficient for (3.17) is the content of the following
embedding due to Folland and Stein; see [Fo75]. In fact, these authors proved this
result only in the range 1 < p < Q.
but it was established later by Varopoulos, and with different techniques. It is also
important to observe that the geometric inequality (3.19) implies in a trivial way the
Folland–Stein inequality (3.17). In fact, if we apply (3.19) to the function |ϕ| α , for
2 Hypoelliptic operators, etc. 157
pQ p(Q−1)
If we now choose α such that (α − 1)p0 = Q−p , which means α = Q−p , we obtain
αQ pQ
Q−1 = Q−p , and thus (3.17) follows from the latter inequality.
1, p
the linear space W H (Ω) becomes a Banach space, which is reflexive when 1 < p <
1,2 1, p
∞. When p = 2, the space W H (Ω) is a Hilbert space. We indicate with W H,loc (Ω)
p 1, p
the space of all functions in L loc (Ω) such that for every ω ⊂⊂ Ω one has W H (ω).
1, p
We say that a function u ∈ W H,loc (Ω) is a critical point of the p-energy in Ω if
for any given ϕ ∈ C0∞ (Ω) we have
d
E p (u + tϕ) = 0.
dt t=0
This is equivalent to saying that
Z
d 1 p/2
|∇ H u| 2 + 2th∇ H u, ∇ H ϕi + t 2 |∇ H ϕ| 2 dg
dt t=0 p Ω
Z
= |∇ H u| p−2 h∇ H u, ∇ H ϕidg = 0.
Ω
158 Nicola Garofalo
1, p
Therefore, a function u ∈ W H,loc (Ω) is a critical point of the p-energy in Ω if it
satisfies the condition
Z
|∇ H u| p−2 h∇ H u, ∇ H ϕidg = 0, for every ϕ ∈ C0∞ (Ω). (3.20)
Ω
Equation (3.20) is the weak version of the quasilinear partial differential equation
∆ H, p u = div H (|∇ H u| p−2 ∇ H u) = 0, (3.21)
known as the horizontal p-Laplacian. Here, for a horizontal vector field ζ = ζ1 X1 +
· · · + ζ m X m , we have indicated its horizontal divergence with
m
X
div H ζ = Xj ζj.
j=1
In conclusion, critical points of the p-energy are weak solutions of the horizontal
p-Laplacian (3.21). In the case when p = Q the quasilinear equation (3.21) plays an
important role in Mostow’s celebrated work [Mo73] on rigidity. When 1 < p < Q it
is, as we have seen, intimately connected to the Folland–Stein horizontal embedding
Theorem 3.8.
for suitable universal constants α, β > 0. By Theorem 3.10 we have for every g , e,
Since by [Fo75, Lemma 1.2] the gauge sphere |g|G = 1 is compact with respect to
the Riemannian topology of G, and from Theorem 1.2 above Γ ∈ C ∞ (G \ {e}), we
conclude that the latter two inequalities must hold.
160 Nicola Garofalo
We stress that, unlike Theorem 3.10, now the relevant differential operator is no
longer hypoelliptic, and it is non-linear.
In connection with the horizontal p-Laplacian and the Folland–Stein embedding
in the case p , 2 we now state another fundamental problem which is fully open.
Open problem: Understand the minimizers in the Folland–Stein embedding Theo-
rem 3.8 when p , 2 and G = Hn , the Heisenberg group.
We have mentioned that in the case in which p = 2 this problem was solved by
Jerison and Lee, who proved that up to rescaling and left translations the unique min-
imizer is given by the function (3.16). What happens when p , 2? We recall that in
the classical Sobolev embedding in Rn for 1 < p < n the minimizers were indepen-
dently found by Talenti [T76] and Aubin [A76]. They are given by all translations
and dilations of the function
Cn, p
u(x) = ,
(1 + |x| p/(p−1) ) (Q−p)/p
where Cn, p > 0 is an explicitly computed constant.
By adapting the concentration of the compactness method of Lions, the exis-
tence of minimizers in the Folland–Stein embedding (3.17) above was established
by Vassilev in [Va06]; see Theorem 3.1 in that paper. However, the value of the best
constant is not provided by this method. Remarkably, to date no guess exists about
the specific form of the minimizer. We note explicitly that, suitably normalized, a
minimizer for the Folland–Stein embedding (3.17) would solve the equation
∆ H, p u = −u ((p−1)Q+p)/(Q−p) ,
which is a generalization of Yamabe’s equation (3.13) above. Here, ∆ H, p is the
horizontal p-Laplacian in (3.21) above.
4 Carnot–Carathéodory distance
In this section we develop some elementary background topological material which
plays a basic role in sub-Riemannian geometry. Although for the sake of simplicity
2 Hypoelliptic operators, etc. 161
N
X ∂
Xj = bkj , j = 1, . . . , m,
k=1
∂x k
0,1
and we assume that the real-valued functions bkj ∈ Cloc (R N ), i.e., they are locally
Lipschitz continuous. Let B = [bkj ] ∈ MN ×m , and consider the matrix A = BB t ∈
MN ×N . It is clear that At = A, and that A ≥ 0. One has in fact, for every ξ ∈ R N ,
hAξ, ξi = hB t ξ, B t ξi = |B t ξ | 2 ≥ 0.
` s (γ) = T.
For later use, let us notice that if γ is subunitary, then for every t ∈ [0,T] at which
γ 0 (t)
exists, one has
|γ 0 (t)| ≤ |X (γ(t))|, (4.2)
2 1/2 .
P
where we have defined |X (x)| = m
j=1 |X j (x)|
Given two points x, y ∈ R N we denote by
S (x, y) , ∅. (4.3)
Definition 4.5. The Carnot–Carathéodory distance associated with the system {X1 ,
. . . , X m } is defined by
Be (x,r) = {y ∈ R N | |x − y| < r }.
i : (R N , d) → (R N , | · |)
is continuous.
2 Hypoelliptic operators, etc. 163
i : (R N , | · |) → (R N , d)
Theorem 4.10 (Chow (1939) [C39] and Rashevsky (1938) [R38]). Let Ω ⊂ R N be a
connected open set and suppose that X i ∈ C ∞ (Ω), i = 1, . . . , m satisfy the finite rank
condition (1.3) in Definition 1.1 above. Then, for every x, y ∈ there exists γ ∈ S(x, y)
such that {γ} ⊂ Ω. In particular, Hypothesis 4.4 holds.
Furthermore, from the results in [NSW85] we can see that Hypothesis 4.9 also
holds.
Proposition 4.11. Suppose that the vector fields X i ∈ C ∞ (R N ) and satisfy the finite
rank condition (1.3). Then, the Carnot–Carathéodory space (R N , d) fulfills Hypoth-
esis 4.9. From this result and from Proposition 4.6 we infer that the metric topology
is equivalent to the underlying (Euclidean) topology of R N .
Proof. Let us recall the following basic estimate obtained in [NSW85]: for every
connected Ω ⊂⊂ R N there exist C, ε > 0 such that
C|x − y| ≤ d Ω (x, y) ≤ C −1 |x − y| ε , x, y ∈ Ω.
d(x, y) ≤ C −1 |x − y| ε , x, y ∈ Ω.
This implies the continuity of the inclusion i : (R N , |·|) → (R N , d), or, equivalently,
the (Euclidean) openness of the metric balls.
δ = sup |x − y| < ∞,
x, y ∈U
define
δ
R = inf .
x ∈U M (x, δ)
If R = 0, then by the (Euclidean) compactness of U we can find a sequence {x j } j ∈N
and a point x 0 ∈ U such that |x j − x 0 | → 0 and M (x j , δ) → ∞. But this is impossible
since M (x j , δ) ≤ M (x 0 , 2δ) < ∞. Therefore, it must be that R > 0 and then (4.5)
gives, for any x ∈ U and 0 < r < R,
δ
!
B(x,r) ⊂ B x, ⊂ B e (x, δ),
M (x, δ)
We claim that
γε ⊂ B e (x,r). (4.6)
166 Nicola Garofalo
To see this suppose that there exists y0 ∈ {γε } such that |y0 − x| > r. By connected-
ness there would exist t ε ∈ [0, ` s (γε )) such that γε (t) ∈ Be (x,r) for 0 ≤ t < t ε and
for which γε (t ε ) ∈ ∂Be (x,r). We thus have
Z tε Z tε
r = |x − γε (t ε )| ≤ 0
|γε (t)|dt ≤ |X (γε (t))|dt
0 0
r
≤ M (x,r)` s (γε ) < M (x,r)(d(x, y) + ε) < M (x,r) = r.
M (x,r)
This contradiction proves (4.6). But then we have
Z ` s (γ ε )
|x − y| ≤ |γε0 (t)|dt ≤ (by (4.6)) M (x,r)` s (γε )
0
r
< M (x,r)(d(x, y) + ε) < M (x,r) = r.
M (x,r)
This shows that y ∈ Be (x,r), thus establishing (4.5).
Concerning Proposition 4.12, if the system {X1 , . . . , X m } is globally Lipschitz,
then the local information concerning the compactness of small balls can be con-
verted to global.
Proposition 4.13. Suppose the vector fields {X1 , . . . , X m } have coefficients in
C 0,1 (R N ). Then, for every x ∈ R N and every r > 0 the closed ball B(x,r) is
compact.
Proof. By assumption there exists M > 0 such that for every x ∈ R N one has
Fix x ∈ R N , r > 0. Our objective is to show that there exists a constant C = C(M) >
0 such that q !
B(x,r) ⊂ Be 0, eCr (1 + |x| 2 ) . (4.8)
From (4.8) we would conclude that B(x,r) is Euclidean compact, hence d-compact.
In order to prove (4.8) let y ∈ B(x,r). By Hypothesis 4.4 there exists γ ∈ S (x, y).
Letting ξ (t) = |γ(t)| 2 , we find
which gives
1 + ξ (t)
!
log ≤ Ct,
1 + ξ (0)
and therefore
ξ (t) ≤ (1 + ξ (t)) ≤ eC t (1 + ξ (0)). (4.9)
Since by assumption d(y, x) < r, for any 0 < ε < r − d(y, x) there exists
γε ∈ S (x, y) such that
Applying (4.9) with γ = γε , letting t → ` s (γε ) in it, and recalling that ξ (0) =
|γε (0)| 2 = |x| 2 , ξ (` s (γε )) = |γε (` s (γε ))| 2 = |y| 2 , we conclude that
We next show that the opposite inequality holds. To this end it is enough to prove
that if γ : [0,T] → R is such that γ(0) = x, γ(T ) = y, and γ 0 (t) = α(γ(t))X (γ(t))
for some |α(t)| ≤ 1, then
T ≥ | tan−1 y − tan−1 x| = tan−1 y − tan−1 x.
Now, any such curve must have the form
Z t
γ(t) = tan( β(t) + C), with β(t) = α(s)ds, | β 0 (t)| ≤ 1.
t0
The conditions γ(0) = x, γ(T ) = y force C = tan−1 x − β(0) and β(T ) = tan−1 y −
tan−1 x + β(0). Since
Z T Z T
| β(T ) − β(0)| = β 0 (s)ds ≤ | β 0 (s)|ds ≤ T,
0 0
we have
` s (γ) = T ≥ | β(T ) − β(0)| = | tan−1 y − tan−1 x|.
By taking the infimum on all γ’s we obtain
d(x, y) ≥ | tan−1 y − tan−1 x|.
π
In conclusion, we have proved (4.10). If now we choose r ≥ 2, we obtain from the
latter identity,
B(0,r) = R,
which shows that the metric balls B(0,r), with r ≥ π2 , are not Euclidean bounded.
We observe explicitly that (R, d) is not a complete metric space. In fact, the sequence
x k = k is Cauchy with respect to d, but of course it admits no limit point in (R, d).
Proposition 4.17. Assume that Hypothesis 4.4 holds. Then, H (x, y) , ∅ for every
x, y ∈ R N .
Proof. Let γ ∈ S (x, y) (by Hypothesis 4.4 at least one such γ does exist), and define
σ : [0, 1] → R N by letting
σ(t) = γ(tT ).
Since γ 0 (t) is subunit whenever it exists, by Proposition 4.2 we infer that for all t ∈
[0, 1], but a finite number of them, there exist a1 (t), . . . , a m (t), with m 2
P
j=1 a j (t) ≤ 1,
such that
Xm
γ (t) =
0
a j (t)X j (γ(t)).
j=1
ã j (t) = T a j (tT ),
then
m
X m
X
σ (t) = T γ (tT ) = T
0 0
a j (tT )X j (γ(tT )) = ã j (t)X j (σ(t)).
j=1 j=1
The proof of the next result can be accomplished by adapting the ideas in [JSC87].
Proposition 4.18. The Carnot–Carathéodory distance d(x, y), and the distances
% p (x, y), 1 ≤ p ≤ ∞, are all equal.
170 Nicola Garofalo
4.4 Length-spaces. In this section we will see that the situation depicted in the
above Example 4.15 should not be surprising. What we mean by this is that the lack
of completeness of the metric space and the failure of compactness of the metric balls
are two equivalent properties in view of a powerful generalization of the classical
theorem of Hopf–Rinow which is due to Cohn-Vossen.
We recall that given a continuous curve γ : [a, b] → S in a metric space (S, d),
its metric length is defined by
p
X
`(γ) = sup d(γ(t i ), γ(t i+1 )),
σ i=1
where the supremum is taken over all the partitions σ = {a = t 1 < · · · < t p+1 = b}
of the interval [a, b]. The curve γ is said to be rectifiable if `(γ) < ∞. We denote by
R (x, y) the collection of all continuous, rectifiable curves joining x to y.
Definition 4.19. A metric space (S, d) is called a length-space if for any x, y ∈ S
one has
d(x, y) = inf `(γ). (4.12)
γ ∈R (x, y)
of the interval [a, b], one trivially has from the triangle inequality,
p
X
d(x, y) ≤ d(γ(t i ), γ(t i+1 )).
i=1
Fix x, y ∈ R N . To prove (4.13) it suffices to show that for every γ ∈ S (x, y) one has
inf `(γ) ≤ inf `(γ) ≤ (by (4.14)) inf ` s (γ) = d(x, y),
γ ∈R (x, y) γ ∈S (x, y) γ ∈S (x, y)
This gives
p
X p
X p
X
d(γ(t i ), γ(t i+1 )) = d(x i , x i+1 ) = (t i+1 − t i ) = t p+1 − t 1 = T = ` s (γ).
i=1 i=1 i=1
Taking the supremum over all partitions σ of the interval [0,T] we obtain (4.14).
If for any x, y ∈ S there exists a segment joining x to y, then (S, d) is said to have
the segment property.
www.Ebook777.com
172 Nicola Garofalo
For a proof of the next result we refer the reader to the classical, beautiful book
by Busemann [Bu70].
Theorem 4.23. Let (S, d) be a metric space which is also a length-space. Then, the
following are true.
(i) If for x ∈ S the closed metric ball B(x,r) is compact, then for any y ∈ B(x,r)
there exists a segment contained in B(x,r) joining x to y.
(ii) If for any x ∈ S and r > 0 the closed ball B(x,r) is compact, then (S, d)
possesses the segment property.
(iii) In particular, if (S, d) is complete, then by Theorem 4.21 the closed balls are
compact, and therefore thanks to (ii) (S, d) has the segment property.
Proof. It follows from Propositions 4.12, 4.20 and from Theorem 4.23(i).
The next result provides us with a basic consequence of the theory developed
so far.
Proposition 4.25. Suppose the vector fields {X1 , . . . , X m } are C 0,1 (R N ), and that
they generate a Carnot–Carathéodory metric d for which Hypothesis (4.9) holds.
Then, (R N , d) is complete and satisfies the segment property.
Proof. According to Proposition 4.13 the closed metric balls B(x,r) are compact
for every x ∈ R N and r > 0. Since by Proposition 4.20 (R N , d) is a length-space,
we can apply Theorem 4.21 to conclude that (R N , d) is complete. By of Theorem
4.23(ii) we conclude that (R N , d) possesses the segment property.
(i) for every g ∈ G and r > 0, the closed metric ball B(g,r) is compact;
We postpone the proof of Proposition 4.26 until later in this section. It should be
obvious from the definition of Carnot–Carathéodory distance that if g, g 0, g 00 ∈ G,
one has
d(L g (g 0 ), L g (g 00 )) = d(g 0, g 00 ). (4.16)
Proof. Using (4.16) we see that proving the lemma is equivalent to showing
To establish (4.17) we invoke Proposition 4.18 and work with horizontal, rather than
subunitary, curves. For any curve σ ∈ H (0, g) consider the new curve defined by
σ λ = δ λ ◦ σ.
Claim: σ λ ∈ H (e, δ λ (g)), and furthermore ` h,1 (σ λ ) = λ` h,1 (σ).
If the claim were true we would infer
and the desired conclusion (4.17) would follow from Proposition 4.18 again.
Let then
Xm
σ (t) =
0
a j (t)X j (σ(t)),
j=1
174 Nicola Garofalo
This establishes the claim, thus completing the proof of the lemma.
Proposition 4.28. Let G be a Carnot group. There exist universal constants C1 ,C2 >
0 such that for g, g 0 ∈ G one has
C1 |g −1 ◦ g 0 |G ≤ d(g, g 0 ) ≤ C2 |g −1 ◦ g 0 |G .
Proof. In view of (4.16) it suffices to prove that for every g ∈ G one has for suitable
constants C1 ,C2 > 0,
C1 |g|G ≤ d(g, e) ≤ C2 |g|G . (4.18)
Thanks to Lemma 4.27, inequality (4.18) is equivalent to
C1 ≤ d δ |g |−1 (g), e ≤ C2 .
G
Therefore, it suffices to show that there exist constants C1 ,C2 > 0 such that for every
g ∈ G, with |g|G = 1, one has
C1 ≤ d(g, e) ≤ C2 . (4.19)
At this point we observe that in light of Proposition 4.11 the metric topology is equiv-
alent to the underlying one generated by the Riemannian distance on G. Therefore,
in particular, the function g → d(g, e) is continuous with respect to the underlying
topology of G. Since by [Fo75, Lemma 1.2] the gauge sphere |g|G = 1 is com-
pact with respect to the Riemannian topology of G, we conclude that (4.19) must
hold.
2 Hypoelliptic operators, etc. 175
%(g, g 0 ) = |g −1 ◦ g 0 |G ,
BG (g,r) = {g 0 ∈ G | |g −1 ◦ g 0 |G ≤ r }
Since by Lemma 4.30 the pseudo-ball BG (g,C1 r) is compact with respect to the
Riemannian topology, therefore, according to Proposition 4.31, with respect to the
d-topology, we conclude that B(g,r) is d-compact. This proves (i).
By Propositions 4.12 and 4.20, (G, d) is a locally compact length-space. Since by
(i) the closed metric balls are compact, we conclude from Cohn-Vossen’s Theorem
4.21 that (G, d) must be complete, thus proving (ii).
Finally, by Theorem 4.23(iii) we infer that (G, d) possesses the segment property.
This establishes (iii).
where
N
X ∂ k
X ∗j = − bj ·
k=1
∂x k
denotes the formal adjoint of X j . We notice that X ∗j is well defined under the mere
0,1
assumption that bkj ∈ Cloc (Ω). Once the definition of the weak derivative X j is
given, we then define the weak derivative of order r, X I , inductively.
`, p
Definition 5.1. Let 1 ≤ p ≤ ∞, ` ∈ N. The weak Sobolev space W X associated
with the system X = {X1 , . . . , X m } is defined as
`, p
W X (Ω) = { f ∈ L p (Ω) | X I f ∈ L p (Ω), |I | ≤ `}.
`, p
We endow W X (Ω) with the norm
X
|| f ||W `, p (Ω) = ||X I f || L p (Ω) .
X
| I | ≤`
We also define
`, p ||·|| `, p
W X,0 (Ω) = C0∞ (Ω) W
X
(Ω)
.
`, p
The strong Sobolev space S X (Ω) associated with the system X = {X1 , . . . , X m } is
defined as
||·|| `, p
`, p `, p
S X (Ω) = C ∞ (Ω) ∩ W X (Ω) .
W (Ω)
X
2 Hypoelliptic operators, etc. 177
`, p `, p
The local spaces W X,loc (Ω), S X,loc (Ω) are defined in the usual way. For instance,
`, p `, p
u ∈ W X,loc (Ω) if for any ω ⊂⊂ Ω one has u ∈ W X (ω).
As in the classical case one can show the following
Proposition 5.2. Let 1 ≤ p ≤ ∞. Endowed with the norm || · ||W `, p (Ω) the spaces
X
`, p `, p
W X (Ω) and S X (Ω) are Banach spaces. Furthermore, they are Hilbert spaces
`, p
when p = 2. When 1 < p < ∞ the space W X (Ω) is a reflexive Banach space.
It should be obvious from Definition 5.1 that
`, p `, p
S X (Ω) ,→ W X (Ω).
was established by Friedrichs in [F44], a fact which is not generally known, and
which was rediscovered (more than fifty years later!) in [GN96] (see Theorem A.2
in that paper) and independently in [FSS97].
A first basic application of Theorem 5.3 is the following chain rule for the space
W X1,1 (Ω) which plays a basic role in the analysis of regularity properties of variational
PDEs arising from vector fields. We will assume in this result that the vector fields
0,1
have coefficients in Cloc (Ω).
1, p
Proposition 5.4. Let f ∈ C 1 (R) with | f 0 | ≤ M. For any u ∈ W X (Ω) with 1 ≤ p <
1, p
∞ one has f ◦ u ∈ W X (Ω) and moreover,
X j ( f ◦ u) = ( f 0 ◦ u)X j u in D 0 (Ω).
Xju
+
a.e. in {x ∈ Ω | u(x) > 0},
Xju =
0
otherwise;
178 Nicola Garofalo
−X j u
a.e. in {x ∈ Ω | u(x) < 0},
X j u− =
0
otherwise;
Xju a.e. in {x ∈ Ω | u(x) > 0},
X j |u| = a.e. in {x ∈ Ω | u(x) < 0},
−X j u
a.e. in {x ∈ Ω | u(x) = 0}.
0
1, p
Proof. Let u ∈ W X (Ω) and let f be as in the statement of the proposition. By
1, p
Theorem 5.3 it will suffice to show that f ◦ u ∈ S X (Ω). Now, by Theorem 5.3 there
1, p
exists a sequence uk ∈ C ∞ (Ω) ∩ W X (Ω) such that as k → ∞ one has
uk → u, X j uk → X j u, in L p (Ω).
Furthermore,
Z
| f 0 ◦ uk X j uk − f 0 ◦ uX j u| p dx
Ω
Z Z
≤ Cp | f ◦ uk X j uk − f ◦ uk X j u| dx + C p
0 0 p
| f 0 ◦ uk X j u − f 0 ◦ uX j u| p dx
Ω Ω
Z Z
≤ Cp | f (uk )| |X j uk − X j u| dx + C p
0 p p
| f 0 (uk ) − f 0 (u)| p |X j u| p dx
Ω Ω
Z Z
≤ Cp M p
|X j uk − X j u| dx + C p
p
| f 0 (uk ) − f 0 (u)| p |X j u| p dx.
Ω Ω
R
We clearly have Ω
|X j uk − X j u| p dx → 0 as k → ∞. Furthermore, we have for a.e.
x ∈ Ω,
| f 0 (uk (x)) − f 0 (u(x))| p |X j u(x)| p → 0,
and moreover,
1, p
This proves at once that f ◦ uk ∈ C 1 (Ω) ∩ W X (Ω) and moreover, f ◦ uk → f ◦ u in
1, p
L p (Ω), and X j ( f ◦uk ) → X j ( f ◦u) in L p (Ω). This establishes that f ◦u ∈ S X (Ω),
and that its weak derivative along X j is given by X j ( f ◦ u) = f 0 (u)X j u. In view of
Theorem 5.3 this proves the first part of the lemma. √
To establish the second part we consider the function f ε (t) = ε 2 + t 2 − ε if
t ≥ 0, f ε (t) = 0 if t ≤ 0. Clearly, f ε ∈ C 1 (R) and | f ε0 | ≤ 1. By the first part we
conclude that f ε ◦ u ∈ W X1,1 (Ω). Moreover, for every ϕ ∈ C0∞ (Ω) we have
Z Z
f ε0 (u)X j uϕ dx = f ε (u)X ∗j ϕ dx.
{x ∈Ω |u(x)>0} Ω
Letting ε → 0 we obtain
Z Z
X j uϕ dx = u+ X ∗j ϕ dx.
{x ∈Ω |u(x)>0} Ω
the space
BV(Ω) = {u ∈ L 1 (Ω) | Var(u, Ω) < ∞}
1,1
becomes a Banach space. If u ∈ S X (Ω), then by definition there exists uk ∈ C 1 (Ω)∩
W X1,1 (Ω) such that uk → u and X j uk → X j u in L 1 (Ω) as k → ∞. If now ζ ∈ F (Ω),
then an integration by parts gives
Z Xm Z m
X
uk X ∗j ζ j dx = X j uk ζ j dx.
Ω j=1 Ω j=1
Remarkably, for the space BV X (Ω) one has an approximation theorem similar to
Theorem 5.3 above.
Theorem 5.7 (See [GN96, Theorem 1.14]). For any function u ∈ BV X (Ω) there
exists a sequence {uk }k ∈N in C ∞ (Ω) such that the following hold:
(a) limk→∞ kuk − uk L 1 (Ω) = 0.
(b) limk→∞ Var X (uk ; Ω) = Var X (u; Ω).
Given a set A ⊂ R N we denote by 1 A its indicator function.
Definition 5.8. Given a measurable set E ⊂ R N its X-perimeter with respect to Ω is
defined as
PX (E; Ω) = Var X (1 E ; Ω).
2 Hypoelliptic operators, etc. 181
5.2 Metric Lipschitz functions and the space W X1,∞ (Ω). In this subsection
we present a remarkable sufficient condition for a function to be a member of the
space W X1,∞ (Ω), namely that the function be Lipschitz continuous with respect to the
Carnot–Carathéodory metric d. We need some preliminary material. In what follows
0,1
we consider a vector field Y in R N with coefficients in Cloc (R N ). Given a bounded
open set Ω ⊂ R N we consider the flow Θ(t, x) = ΘY (t, x) associated with Y . In other
words, given any x ∈ Ω we consider the unique solution of the Cauchy problem
Θ(·, ·) : (−T,T ) × ω → R N .
182 Nicola Garofalo
Then, Z t
t
f (t) ≤ g(t) +
R
K (s) exp{ s
K (τ)dτ}g(s)ds.
a
In particular, if g(t) ≡ c and K (t) ≡ K, then we obtain
f (t) ≤ ce K (t−a) , a ≤ t ≤ b.
2 Hypoelliptic operators, etc. 183
0,1
Proof of Lemma 5.11. Since Y ∈ Cloc (Ω) and ω ⊂⊂ Ω, there exists L = L(ω) > 0
such that
|Y (x) − Y (y)| ≤ L|x − y|, x, y ∈ ω.
Since for every x ∈ ω and |t| ≤ T we have
Z t
Θ(t, x) = x + Y (Θ(s, x))ds,
0
we thus find
Z t
|Θ(t, x) − Θ(t, y)| ≤ |x − y| + |Y (Θ(s, x)) − Y (Θ(s, y))|ds
0Z
t
≤ |x − y| + L |Θ(s, x) − Θ(s, y)|ds .
0
If we let f (t) = |Θ(t, x) − Θ(t, y)|, then the latter inequality reads
Z t
f (t) ≤ |x − y| + L f (s)ds .
0
By Lemma 5.12 we conclude that
which establishes (5.1). In particular, (5.1) shows that for every fixed |t| ≤ T the map
x → Θ(t, x) is Lipschitz continuous in ω. By the theorem of Rademacher–Stepanov,
the function is thus differentiable at a.e. x ∈ ω. Fix a point x of differentiability.
Then, for |h| small we have
Θ(t, x + he j ) − Θ(t, x)
Z tf
1 g
= ej + Y (Θ(s, x + he j )) − Y (Θ(s, x)) ds, (5.6)
h h 0
where {e1 , . . . , e N } denotes the standard basis of R N . For j = 1, . . . , N set
Z tf
1 g
η j (t, x; h) = Y (Θ(s, x + he j )) − Y (Θ(s, x)) ds.
h 0
Θ(t, x+he j )−Θ(t, x)
Since lim h exists, then from (5.6) we conclude that
h→0
def
lim η j (t, x; h) = η j (t, x)
h→0
∂Θ
(t, x) = e j + η j (t, x),
∂x j
with η j (t, x) satisfying (5.7). This establishes (5.2) with (5.3). From these latter
equations and linear algebra, the equations (5.4), (5.5) follow.
Definition 5.13 (Hölder and Lipschitz spaces). Given a system {X1 , . . . , X m } of vec-
0,1
tor fields in Cloc (R N ), suppose that Hypothesis 4.4 is satisfied, so that d(x, y) defines
a Carnot–Carathéodory metric in R N . If Ω ⊂ R N is an open set, and α ∈ (0, 1], we
denote by Γ0,αX (Ω) the linear space of all functions f : Ω → R for which there exists
L ≥ 0 such that
| f (x) − f (y)| ≤ Ld(x, y) α , x, y ∈ Ω. (5.8)
If f ∈ Γ0,α
X (Ω) we introduce the seminorm
| f (x) − f (y)|
[ f ]Γ0, α (Ω) = sup .
X x, y ∈Ω, x,y d(x, y) α
Proof. Let us note that by (5.9) the function f is d-continuous. By Hypothesis 4.9
we thus see that f is also continuous with respect to the Euclidean distance, and
therefore it is measurable. To show that f ∈ L ∞ (Ω) we fix x 0 ∈ Ω. We can assume
without restriction that f (x 0 ) = 0, otherwise we consider g = f − f (x 0 ). Then, for
every x ∈ Ω we have from (5.9),
and so
|| f || L ∞ (Ω) ≤ Ld Ω . (5.11)
We next fix j = 1, . . . , m. If we show that the distribution X j f ,
hX j f , ϕi = h f , X ∗j ϕi,
it will suffice to show that for every ϕ ∈ C0∞ (Be (x, ε)) one has
Z
f X ∗j ϕ dy ≤ C||ϕ|| L 1 (B e (x,ε)) . (5.12)
B e (x,ε)
Inequality (5.12) would prove that X j f ∈ L ∞ (B(x,r)). From the arbitrariness of
x ∈ Ω we see that (5.12) would establish X j f ∈ L ∞ (Ω), and ||X j f || L ∞ (Ω) ≤ C.
In order to prove (5.12) we fix a function ϕ ∈ C0∞ (Be (x, ε)) and let ω ⊂⊂
Be (x, ε) be such that supp ϕ ⊂ ω. For a fixed j = 1, . . . , m we write for simplicity
Y = X j , and consider the flow Θ(t, x) = ΘY (t, x) as in Lemma 5.11. By compactness
we may assume that for |t| sufficiently small we have ω ⊂ Be (x, ε) ∩ Θ(t, Be (x, ε)).
We have
Z Z Z
hY f , ϕi = f Y ϕ dy = −
∗
f Y ϕ dy − f (divY )ϕ dy.
B e (x,ε) B e (x,ε) B e (x,ε)
186 Nicola Garofalo
Now,
Z
f (div Y )ϕ dy ≤ || f || L ∞ (B(x,ε)) || div Y || L ∞ (Ω) ||ϕ|| L 1 (B e (x,ε)) .
B e (x,ε)
Next, we want to show
Z
f Y ϕ dy ≤ L + C|| f || L ∞ (B e (x,ε)) ||ϕ|| L 1 (B e (x,ε)) .
(5.13)
B (x,ε)
e
Assume for a moment that we have proved (5.13). Then, by combining (5.12)
with (5.13) we would conclude that Y f ∈ L ∞ (B(x, ε)) and that
||Y f || L ∞ (B(x,ε)) ≤ L + C|| f || L ∞ (B e (x,ε)) . (5.14)
We assume now, as it is not restrictive, that f (x) = 0 and that furthermore |X f (x)| >
0 and |X f | is approximately continuous at x. Consider the vector field defined by
m
X X j f (x)
Z (y) = X j (y).
j=1
|X j f (x)|
This estimate would establish (5.10), thus completing the proof of the theorem. We
are thus left with obtaining (5.14).
By Lebesgue dominated convergence we find
Z Z ϕ(Θ(t, x)) − ϕ(x)
f (x)Y ϕ(x)dx = f (x)lim dx
B e (x,ε) B e (x,ε) t→0 t
ϕ(Θ(t, x)) − ϕ(x)
Z
= lim f (x) dx
t→0 B e (x,ε) t
Z ϕ(Θ(t, x)) − ϕ(x)
= lim f (x) dx .
t→0 B e (x,ε) t
To establish (5.13) it will thus suffice to show that there exists C > 0 such that
Z ϕ(Θ(t, x)) − ϕ(x)
dx ≤ L + C|| f || L ∞ (B e (x,ε)) ||ϕ|| L 1 (B e (x,ε)) .
lim f (x)
t→0 B e (x,ε) t
(5.15)
For t , 0 we now have
ϕ(Θ(t, x)) − ϕ(x)
Z Z Z
1 1
f (x) dx = f (x)ϕ(Θ(t, x))dx − f (x)ϕ(x)dx.
Ω t t Ω t Ω
2 Hypoelliptic operators, etc. 187
In the first integral on the right-hand side we make the change of variable y = Θ(t, x),
for which
dy = | det J (Θ(t, x))|dx = (by (5.4)) [1 + σ(t, x)]dx,
with σ(t, x) verifying (5.5). Since Θ−1 (t, y) = Θ(−t, Θ(t, x)) = x, we thus obtain
Z Z
f (x)ϕ(Θ(t, x))dx = f (Θ(−t, y))ϕ(y)[1 + σ̃(t, y)]dy,
Ω Ω
def
We now observe that for any given y ∈ ω the curve t → Θ(t, y) = γ(t) is a sub-
unitary curve joining y to Θ(t, y). As a consequence, we have ` s (γ) ≤ t. This
gives
d(y, Θ(t, y)) ≤ ` s (γ) ≤ t.
σ̃(t, y)
Z
f (Θ(−t, y))ϕ(y) dy
Ω t
σ̃(t, y)
Z Z
f (Θ(−t, y)) − f (y)
= ϕ(y) σ̃(t, y)dy + f (y)ϕ(y) dy.
Ω t Ω t
Remark 5.15. A interesting situation is when the set Ω in Theorem 5.14 is a metric
ball B(x,r). If we are in the situation in which the metric ball is bounded in the
underlying Euclidean distance (remember that this is not always the case; see Ex-
ample 4.15), then Theorem 5.14 applies. A case of special relevance is when the
d-Lipschitz function is f (y) = d(y, x). Notice that in such case || f || L ∞ (B(x,r )) ≤ r.
(ii) |X ϕ| ≤ C
t−s a.e. in R N ;
1, p
(iii) ϕ ∈ W X (R N ) for every 1 ≤ p < ∞.
2 Hypoelliptic operators, etc. 189
However, in order to simplify the exposition we will assume that d is a true metric.
Let us note the following simple, yet important, fact.
Proposition 6.2. Hypothesis 6.1 implies the following: for every x ∈ S, r > 0 and
0 < t ≤ 1 one has
ν(B(x,tr)) ≥ C2 t Q ν(B(x,r)), (6.2)
with C2 = C1−1 , and
log C1
Q= . (6.3)
log 2
Proof. Suppose that (6.1) holds. For 0 < t ≤ 1 pick k ∈ N such that 2−k < t ≤
2−k+1 . This gives
Definition 6.3. The number Q in (6.3) is called the homogeneous dimension of the
space (S, d, ν).
ν(B(x, sr)) s Q
≥ C2 , (6.4)
ν(B(x,tr)) t
for every x ∈ S, r > 0, 0 < s < t.
In what follows, given a set Ω ⊂ S we let
d Ω = d(x, y) .
x, y ∈Ω
Proposition 6.4. Let Ω ⊂ S be such that d Ω < ∞. Then, for every x ∈ Ω and
0 < r ≤ d Ω one has
!Q
ν(B(x,r)) r
≥ C2 .
ν(Ω) dΩ
Definition 6.5. Given a number A > 0, a measurable set Ω ⊂ S is called of A-type
if for every x ∈ Ω and 0 < r < d Ω one has
If Ω is of A-type, then from (6.1) one infers for every x ∈ Ω and 0 < r < d Ω ,
The notion of corkscrew was introduced in the fundamental paper of Jerison and
Kenig on NTA (non-tangentially accessible) domains; see [JK82]. This notion guar-
antees the existence of regions of non-tangential accessibility for a given domain and
is tailor made for the application to boundary value problems in partial differential
equations.
We note in particular that if Ar (x 0 ) is a corkscrew for x 0 ∈ ∂Ω, then for every
0 < r < R0 one has
B(x 0 ,r) ⊂ B( Ar (x 0 ), 2r). (6.7)
In fact, if d(y, x 0 ) < r, from the triangle inequality and from the right-hand side of
the former inequality in (6.6), one has
Proposition 6.7. If Ω satisfies the uniform interior corkscrew condition, then there
exists a constant A = A(C1 , K ) > 0 such that Ω is of A-type.
www.Ebook777.com
192 Nicola Garofalo
Proof. By the assumption there exist K, R0 > 0 such that given any x 0 ∈ ∂Ω, condi-
tion (6.6) is fulfilled for any 0 < r < R0 . By (6.8) we have
r
ν(Ω ∩ B(x 0 , 2r)) ≥ ν B Ar (x 0 ),
2K
≥ (by the doubling condition (6.1)) C(C1 , K )ν(B( Ar (x 0 ), 2r))
≥ (by (6.7)) C(C1 , K )ν(B(x 0 ,r))
≥ (again by the doubling condition (6.1)) C(C1 , K )ν(B(x 0 , 2r)).
Changing r into r/2 in the above inequality, we reach the desired conclusion.
From Proposition 6.7 and Proposition 4.8 we obtain the following result which
will prove important in the study of the Sobolev embedding.
0,1
Corollary 6.8. Given a system {X1 , . . . , X m } of vector fields in Cloc (R N ), suppose
that Hypothesis 4.4 is satisfied, so that d(x, y) defines a Carnot–Carathéodory met-
ric in R N . Assume the doubling condition (6.1) above. Then, every metric ball
B(x,r) is of A-type.
One easily verifies that L p (S, dν) ,→ L p,∞ (S, dν) but that the inclusion is strict.
Given 1 < p ≤ ∞ there exists CM, p > 0, depending on p and C1 , such that for any
f ∈ L p (S, dν) one has
Z Z
p
M f (x) dν(x) ≤ CM, p | f (x)| p dν(x).
S S
rλ 1/p
Z
|| f || M p, λ (Ω, dν) = || f ||Ω, p, λ = sup | f | p dν
x ∈Ω,0<r <d Ω ν(Ω ∩ B(x,r)) Ω∩B(x,r )
< ∞.
One easily checks that || · || M p, λ (Ω, dν) defines a norm and that, equipped with it,
M p, λ (Ω, dν) is a Banach space. It is also easy to see that if Ω is d-bounded, then
and we have
d Q 1/p d Q 1/p
1/p
Ω
|| f || L p (Ω, dν) ≤ || f || M p, Q (Ω) ≤ A−1/p C1 Ω
|| f || L p (Ω, dν) .
ν(Ω) ν(Ω)
(6.10)
Combining Proposition 6.12 with Corollary 6.8 we obtain the following result.
0,1
Corollary 6.13. Given a system {X1 , . . . , X m } of vector fields in Cloc (R N ), suppose
that Hypothesis 4.4 is satisfied, so that d(x, y) defines a Carnot–Carathéodory met-
ric in R N . Assume the doubling condition (6.1) above. Then, given a bounded set
U ⊂ R N there exists R0 = R0 (U, X ) > 0 such that for every x ∈ U and 0 < r < R0
one has
M p,Q (B(x,r)) = L p (B(x,r)), 1 ≤ p < ∞.
We are now ready to establish the basic continuity theorem for the maximal oper-
ator between Morrey spaces. We will need the following auxiliary results. The next
lemma is a generalization of a result of Fefferman and Stein.
Lemma 6.14. Let Ω ⊂ S be d-bounded. For any 1 < p ≤ ∞ there exists C =
C(p,C1 ) > 0 such that for any measurable f and ϕ ≥ 0 in Ω one has
Z Z
M f ϕ dν ≤ C
p
| f | p Mϕ dν. (6.11)
Ω Ω
2 Hypoelliptic operators, etc. 195
This shows that | f (x)| ≤ t for µ-a.e. x ∈ S (recall, f = 0 in Ωc ), and hence the same
conclusion holds for ν-a.e. x ∈ S. Therefore M f (x) ≤ t for every x ∈ Ω, which
implies || M f || L ∞ (Ω, dζ) ≤ t. By the arbitrariness of t > || f || L ∞ (Ω, dµ) , we conclude
that
|| M f || L ∞ (Ω, dζ) ≤ || f || L ∞ (Ω, dµ) .
This proves (6.14). We next establish (6.13). To this end, it suffices to show that
there exists a constant C = C(C1 ) > 0 such that for every f ∈ L 1 (Ω, dµ) one has
Z
sup t ζ ({x ∈ Ω| M f (x) > t}) ≤ C | f |dµ. (6.15)
t >0 Ω
By Lemma 6.9 there exists a sequence of points x j ∈ Ω such that the balls B(x j ,
r (x j )) are pairwise disjoint, and for which
[
{x ∈ Ω| M f (x) > t} ⊂ B(x j , κr (x j )),
j ∈N
Next, for any x ∈ B(x j ,r (x j )) the triangle inequality gives B(x j , κr (x j )) ⊂ B(x,
2κr (x j )). We thus have
Z
1
ϕ(y)dν(y)
ν(B(x j , κr (x j ))) B(x j ,κr (x j ))
Z
1
≤ ϕ(y)dν(y)
ν(B(x j , κr (x j ))) B(x,2κr (x j ))
ν(B(x, 2κr (x j )))
Z
1
= ϕ(y)dν(y) ≤ C Mϕ(x),
ν(B(x j , κr (x j ))) ν(B(x, 2κr (x j ))) B(x,2κr (x j ))
This proves (6.15), and thus (6.13). As we have mentioned above, the desired con-
clusion (6.11) now follows by the theorem of Marcinckiewicz.
2 Hypoelliptic operators, etc. 197
Lemma 6.15. Let Ω ⊂ S be d-bounded. Then, for every fixed x ∈ Ω one has
ν(Ω ∩ B(x,r))
M1Ω∩B(x,r ) (y) ≤
ν(B(y, d(x, y) − r))
for every y ∈ Ω such that d(x, y) > r.
Proof. It suffices to observe that if d(x, y) > r, then B(y, %) ∩ B(x,r) , ∅ if and
only if % ≥ d(x, y) − r. Hence,
ν(B(y, %) ∩ B(x,r) ∩ Ω) ν(Ω ∩ B(x,r))
M1Ω∩B(x,r ) (y) = sup ≤ .
% >0 ν(B(y, %)) ν(B(y, d(x, y) − r))
The next theorem provides the main result about the mapping properties of the
maximal operator between Morrey spaces.
Theorem 6.16. Let Ω ⊂ S be a d-bounded set. For every λ > 0 there exists C =
C(C1 , λ) > 0 such that for every y ∈ Ω, r < d Ω , and any t > 0 one has, for
f ∈ M 1, λ (Ω, dν),
ν(Ω ∩ B(y,r))
sup t ν ({x ∈ Ω | M f (x) > t} ∩ B(y,r)) ≤ C || f || M 1, λ (Ω, dν) .
t >0 rλ
(6.16)
Given 1 < p < ∞, λ > 0, there exists C = C(C1 , p, λ) > 0 such that for any
f ∈ M p, λ (Ω, dν) one has
|| M f || M p, λ (Ω, dν) ≤ C|| f || M p, λ (Ω, dν) . (6.17)
Proof. We first analyze the case 1 < p < ∞. Consider f ∈ M p, λ (Ω), and extend it
with zero in Ωc . For x ∈ Ω and 0 < r < d Ω we let R = d Ω , and fix k0 ∈ N such that
2k0 ε ≤ 2R < 2k0 +1 ε.
One has
Z
M f (y) p dν(y)
Ω∩B(x,r )
Z
= M f (y) p 1Ω∩B(x,r ) (y)dν(y) (by Lemma 6.14)
Ω
Z
≤C | f (y)| p M1Ω∩B(x,r ) (y)dν(y)
Ω
Z
=C | f (y)| p dν(y)
Ω∩B(x,r )
k0 Z
X ν(Ω ∩ B(x,r))
+ | f (y)| p dν(y) ,
k=0 Ω∩ B(x,2k +1 r )\B(x,2k r )
ν(B(y, d(y, x) − r))
198 Nicola Garofalo
where in the estimate of the second term in the right-hand side we have used Lemma
6.15. In order to establish (6.17) we multiply both sides of the latter inequality by
rλ
the quantity ν (Ω∩B(x,r )) . The first term in the right-hand side is trivially estimated
p
by || f || M p, λ (Ω) . The second term, instead, is estimated by
k0
X ν(Ω ∩ B(x, 2k+1 r)) (2k+1 r) λ
ν(Ω ∩ B(x, 2k+1 r))
k=0
2 (k+1)λ ν(Ω ∩ B(x, 2 k+1 r))
| f (y)| p
Z
× dν(y).
Ω∩ B(x,2k +1 r )\B(x,2k r ) ν(B(y, 2
k−1 r))
p
≤ C|| f || M p, λ (Ω) ,
where C = C(C1 , λ) > 0. This proves (6.17).
Finally, (6.16) follows from the weak inequality (6.12) in Lemma 6.14. Taking
in fact ϕ = M1Ω∩B(y,r ) in (6.12) we find
Z
t ν({x ∈ Ω ∩ B(y,r) | M f (x) > t}) = t 1Ω∩B(y,r ) (x)dν(x)
{x ∈Ω | M f (x)>t }
Z
≤C | f (x)| M1Ω∩B(y,r ) (x)dν(x)
ZΩ
=C | f (x)|dν(x)
Ω∩B(y,r )
k0 Z
X ν(Ω ∩ B(y,r))
+C | f (x)| dν(x)
k=0 Ω∩
B(y,2k +1 r )\B(y,2k r )
ν(B(x, d(y, x) − r))
ν(Ω ∩ B(y,r))
≤C || f || M 1, λ (Ω) .
rλ
For k = 0, 1, . . . , k0 denote
Rk (x) = Ω ∩ B(x, 2k+1 ε) \ B(x, 2k ε) .
Iα2 f (x)
d(x, y) α
Z
≤ | f (y)| dν(y)
Ω∩(B(x,2R)\B(x,ε)) ν(B(x, d(x, y)))
k0 Z
X d(x, y) ασ p 1/p
≤ | f (y)| p dν(y)
k=0 R k (x) ν(B(x, d(x, y))) σ p
d(x, y) α(1−σ) p
Z 0 1/p0
× dν(y)
R k (x) ν(B(x, d(x, y)))
(1−σ) p 0
k0 (
(2k+1 ε) ασ (2k+1 ε) λ 1/p
X Z
p
≤ | f (y)| dν(y)
k=0
ν(B(x, 2k ε)) σ ν(Ω ∩ B(x, 2k+1 ε)) Ω∩B(x,2k +1 ε)
ν(Ω ∩ B(x, 2k+1 ε)) 1/p (2k+1 ε) α(1−σ)
)
1/p 0
× ν(Ω ∩ B(x, 2 k+1
ε))
(2k+1 ε) λ/p ν(B(x, 2k ε) 1−σ
k0
X λ ν(Ω ∩ B(x, 2k+1 ε))
≤ (2k+1 ε) −( p −α) || f || M p, λ (Ω, dν) (by (6.1))
k=0
ν(B(x, 2k ε)
∞
−( λp −α) −( λp −α) λ λ
X
≤ C1 2 ε || f || M p, λ (Ω, dν) 2−k ( p −α) = C ∗∗ ε −( p −α) || f || M p, λ (Ω, dν) ,
k=0
where C ∗∗ = C ∗∗ (C1 , p, α, λ). Combining the estimates for Iα1 f (x) and Iα2 f (x), we
find for every ε > 0,
λ
Iα f (x) ≤ C ∗ ε α M f (x) + C ∗∗ ε −( p −α) || f || M p, λ (Ω, dν) . (6.21)
g(ε) = Aε α + Bε −β
with
λ
α+ β= ,
p
λ
so that β = p − α > 0. Since g 0 (ε) = 0 if and only if
βB
ε α+β = ,
αA
Free ebooks ==> www.Ebook777.com
i.e.,
! 1/(α+β) ! p/λ
βB βB
ε= = ,
αA αA
we obtain
! α p/λ ! λp ( λp −α)
βB βB
gmin =A +B .
αA αA
Comparing with (6.21) we see that taking A=C ∗ M f (x) and B=C ∗∗ || f || M p, λ (Ω, dν) ,
then the previous calculation gives with a C = C(C1 , p, α, λ) > 0,
α p/λ
Iα f (x) ≤ C|| f || M p, λ (Ω, dν) M f (x) 1−(α p/λ) , x ∈ Ω. (6.22)
rλ
Z ! 1/q
q
|Iα f | dν
ν(Ω ∩ B(x,r)) Ω∩B(x,r )
rλ
Z ! 1/q
α p/λ q(1− αλp )
≤ C|| f || M p, λ (Ω, dν) |M f | dν .
ν(Ω ∩ B(x,r)) Ω∩B(x,r )
If we now invoke (6.17) in Theorem 6.16, we reach the desired conclusion (6.18)
from the latter inequality by simply observing that
αp p
+ = 1.
λ q
To prove (6.20) we use the case p = 1 of (6.22) which gives for t > 0 (note that
we can assume without loss of generality that || f || M 1, λ (Ω, dν) , 0),
www.Ebook777.com
202 Nicola Garofalo
( λ/(λ−α) )!
=t νq
x ∈ Ω ∩ B(y,r) | M f (x) > tC −1
|| f || −α/λ
M 1, λ (Ω, dν)
ν(Ω ∩ B(y,r)) q
=C || f || M 1, λ (Ω, dν) .
rλ
This completes the proof of the theorem.
It is worth stating explicitly the following basic corollary of Theorem 6.18, Pro-
position 6.12, and Corollary 6.13.
Theorem 6.19. Let 0 < α < Q, 1 < p < Q α . Given a d-bounded set of A-type Ω ⊂ S,
there exists a constant C = C(C1 , A, p, α) > 0 such that for any f ∈ L p (Ω, dν) one
has !α
dΩ
||Iα f || L q (Ω, dν) ≤ C || f || L p (Ω, dν) , (6.23)
ν(Ω) 1/Q
provided that
1 1 α
= − . (6.24)
q p Q
If instead p = 1, then there exists C = C(C1 , A, d Ω ) > 0 such that for any y ∈ Ω and
Q
r > 0 one has with q = Q−α ,
! 1/q Q
ν(B(y,r)) dΩ
sup t ν ({x ∈ Ω ∩ B(y,r) | Iα f (x) > t}) 1/q
≤C || f || L 1 (Ω, dν) .
t >0 rQ ν(Ω)
(6.25)
where X ∗j denotes the adjoint of the vector field X j . When the vector fields X j are
those associated with a basis of the horizontal layer of a Carnot group, we have
X ∗j = −X j (see [Fo75]), and the operator in (7.1) coincides with (2.17) in Definition
2.2 above.
Definition 7.1. A distribution u ∈ D 0 (Ω) is called harmonic in Ω if ∆ H u = 0.
Since by Hörmander’s Theorem 1.2 the operator ∆ H is hypoelliptic, if u ∈ D 0 (Ω) is
harmonic there exists U ∈ C ∞ (Ω) such that u = U in D 0 (Ω).
7.1 The theorem of Nagel, Stein, and Wainger. We denote by Y1 , . . . ,Y` the
collection of the X j ’s and of those commutators which are needed to generate R N . A
“degree” is assigned to each Yi , namely the corresponding order of the commutator.
If I = (i 1 , . . . ,i N ), 1 ≤ i j ≤ ` is an N-tuple of integers, following [NSW85] one
defines
XN
d(I) = deg(Yi j ) and a I (x) = det[Yi 1 , . . . ,Yi N ].
j=1
Q = sup {d(I) | |a I (x)| , 0, x ∈ U}, Q(x) = inf {d(I) | |a I (x)| , 0}, (7.2)
3 ≤ N ≤ Q(x) ≤ Q. (7.3)
204 Nicola Garofalo
The numbers Q and Q(x) are respectively called the local homogeneous dimen-
sion of U, and the pointwise homogeneous dimension at x, with respect to the system
{X1 , . . . , X m }.
Remark 7.3. In a Carnot group G the Nagel–Stein–Wainger polynomial is simply a
monomial, and it does not depend on the point. In other words,
Λ(g,r) ≡ Cr Q , g ∈ G, r > 0,
rQ
≤ C ∗ R0Q , x ∈ U, 0 < r ≤ R0 . (7.8)
|B(x,r)|
2 Hypoelliptic operators, etc. 205
Proposition 7.6. The polynomial function Λ(x,r) in Definition 7.2 satisfies the fol-
lowing property. Given a bounded set U ⊂ R N one has
Λ(x,r) Λ(x,r 2 ) − Λ(x,r 1 ) Λ(x,r)
Q(x) ≤ ≤Q
r r2 − r1 r
for any x ∈ U, 0 < r 1 < r 2 < R0 , and some r = r (x) ∈ (r 1 ,r 2 ). Here, R0 is the
characteristic local parameter of U and Q is its local homogeneous dimension (7.2).
Proof. We begin by observing that, from Definition 7.2, for any bounded set U ⊂
R N one has
rΛ0 (x,r)
Q(x) ≤ ≤ Q, for every x ∈ U, 0 < r < R0 , (7.9)
Λ(x,r)
where Q and Q(x) are as in (7.2). We fix x ∈ U and 0 < r 1 < r 2 ≤ R0 , and apply the
mean value theorem to the function Λ(x, ·) to reach the conclusion from (7.9).
Finally, we recall the following definition from [NSW85, p. 123]. For x ∈ R N ,
and r > 0, we set
`
X
Box(r) = {x ∈ R N
| x = exp( u j Yj ) with |u j | < r d j }, (7.10)
j=1
where we have let d j = deg(Yj ). Here, exp denotes the exponential mapping as-
sociated with the vector fields Y1 , . . . ,Y` . For its definition and main properties we
refer the reader to the appendix of [NSW85]. The following result is contained in
[NSW85, Theorem 7].
Theorem 7.7 (Ball-box). Given a bounded set U ⊂ R N there exist η ∈ (0, 1), and
R0 > 0, such that for any x ∈ U, and 0 < r < R0 , one has
Remark 7.8. One can be more precise about the shape of the sets B(x,r). They have
size r in the directions of the X j ’s, whereas they have size r 2 in the directions of the
commutators [X i , X j ], and so on (see [NSW85]).
for some universal constants α, β > 0. Combining this estimate with Proposition
4.28 we obtain
α∗ β∗
≤ Γ(g −1
◦ g 0
) ≤ ,
d(g, g 0 ) Q−2 d(g, g 0 ) Q−2
where α ∗ , β ∗ > 0 are appropriate universal constants, and d(g, g 0 ) indicates the Carn-
ot–Carathéodory distance on G. On the other hand, Lemma 4.27 and a change of
variable imply the existence of a universal constant ω > 0 such that
Using (7.11) in the above estimate, we see that the latter can be recast in the form
d(g, g 0 ) 2 d(g, g 0 ) 2
C ≤ Γ(g −1
◦ g 0
) ≤ C −1
, (7.12)
|B(g, d(g, g 0 ))| |B(g, d(g, g 0 ))|
for a certain universal constant C > 0.
The following fundamental result which was established independently by Nagel,
Stein, and Wainger in [NSW85], and by Sánchez-Calle in [SCa84], represents a
(local) generalization of (7.12) to operators ∆ X of the form (7.1) above. Given such
an operator on a bounded open set Ω ⊂ R N , we indicate with Γ(x, y) its (positive)
fundamental solution in Ω, as constructed in [RS76] and [NSW85]. This means that
for every x ∈ Ω we have Γ(·, x) ∈ D 0 (Ω) ∩ C ∞ (Ω \ {x}), and ∆ X Γ(·, x) = δ x in
D 0 (Ω), where δ x indicates the Dirac delta function concentrated in x. Since ∆ X is
self-adjoint, one can prove that
d(x, y) 2−s
|X j1 X j2 . . . X j s Γ(x, y)| ≤ C −1 , (7.13)
|B(x, d(x, y))|
d(x, y) 2
Γ(x, y) ≥ C .
|B(x, d(x, y))|
In the first inequality in (7.13), one has ji ∈ {1, . . . , m} for i = 1, . . . , s, and X j i is
allowed to act on either x or y.
balls, but display some distinguishing features which make them better suited than
the metric balls to the study of various fine questions which specifically involve the
intrinsic geometry of operators such as (7.1).
Given a bounded open set Ω ⊂ R N with characteristic local parameters C1 , R0 >
0, for every x ∈ Ω and 0 < r ≤ R0 we introduce the quantity
Λ(x,r)
E(x,r) = . (7.14)
r2
We observe that, because of (7.3), for every fixed x ∈ Ω the function r → E(x,r) is
strictly increasing on (0, R0 ). We denote by F (x, ·) the inverse function of E(x, ·), so
that F (x, E(x,r)) = r. We remark explicitly that in a Carnot group one has
One of the main geometric properties of the X-balls is that they are equivalent to the
Carnot–Carathéodory balls. To see this, we observe that in view of (7.6), (7.13), it
is easy to recognize that there exists a > 1, depending on Ω and X, such that for
x ∈ Ω, 0 < r ≤ R0 ,
B(x, a−1 r) ⊂ B X (x,r) ⊂ B(x, ar). (7.16)
We observe that, as a consequence of (7.5), and of (7.13), one has
!
1
Cd(x, y) ≤ F x, ≤ C −1 d(x, y), (7.17)
Γ(x, y)
Another key property of the X-balls is that, since Γ(·, x) ∈ C ∞ (Ω\{x}), by Sard’s
theorem for a.e. r ∈ (0, R0 ] the set ∂B X (x,r) is a C ∞ manifold of codimension 1 in
R N . Therefore, we can use the tools of calculus on these sets. For instance, we have
the following basic representation formula for a smooth function which will prove
quite useful in the sequel. In what follows, if D ⊂ R N denotes a bounded open set,
with characteristic local parameters C1 , R0 , we will let
[
D R0 = {Ω(x,r) | x ∈ D, 0 < r < R0 } .
Theorem 7.11 (Representation formula; see [CGL93]). Let ψ ∈ C ∞ (D R 0 ). For
every x ∈ D and 0 < t ≤ R0 , one has
|X y Γ(x, y)| 2
Z Z f 1g
ψ(x) = ψ(y) dH N −1 (y) + ∆ X ψ(y) Γ(x, y) − dy.
∂Ω(x, t ) |DΓ(x, y)| Ω(x, t ) t
(7.18)
The same approach as that of the proof of Theorem 7.11 leads to the following
result which, among other things, plays a crucial role in establishing an important
geometric embedding and an isoperimetric inequality. For these aspects the reader
should consult Section 8.
Corollary 7.12. Let U ⊂ R N be a bounded open set. Then, there exists a constant
C = C(U, X ) > 0 such that for any function f ∈ C00,1 (U), and any x ∈ U one has
Z
d(x, y)
| f (x)| ≤ C |X f (y)| dy = CI1 (|X f |)(x).
U |B(x, d(x, y))|
Proof. Using the fact that supp f is a compact subset of U, we can proceed as in
the proof of Theorem 7.11 and, after some integration by parts, obtain the following
identity: Z
f (x) = − hX f (y), X Γ(x, y)idy, x ∈ U.
U
From this identity the desired conclusion follows if one invokes the basic estimate
(7.13) in Theorem 7.9 above.
|X y Γ(x, y)| 2
!
1
K R (x, y) = f R ,
Γ(x, y) Γ(x, y) 2
whenever the right-hand side makes sense.
2 Hypoelliptic operators, etc. 209
Theorem 7.14 (See [CDG97]). Let D ⊂ R N be bounded and open and suppose that
∆ X u = 0 (u is harmonic) in D. There exists R0 > 0, depending on D and X, such
that for any x ∈ D, and every 0 < R ≤ R0 for which Ω(x, 2R) ⊂ D, one has
u(x) = JR u(x).
|X y Γ(x, y)| 2
Z
u(x) = u(y) dH N −1 (y). (7.21)
∂Ω(x, t ) |DΓ(x, y)|
We are now going to use (7.21) to complete the proof. The idea is to start from
the definition of JR u(x), and then to use Federer’s co-area formula [Fe69]. One finds
∞ |X y Γ(x, y)| 2
Z Z
JR u(x) = f R (t) u(y) dH N −1 (y) dt.
0 ∂Ω(x, t ) |DΓ(x, y)|
This equality, (7.21), and the fact that R f R (s)ds = 1, imply the conclusion.
R
Theorem 7.15 (See [CDG97]). Fix a bounded open set D ⊂ R N . There exists a
constant R0 > 0, depending only on D and on the system X = {X1 , . . . , X m }, such
that for any u ∈ L 1loc (R N ), x ∈ D, 0 < R ≤ R0 , and s ∈ N, one has for some
C = C(D, X, s) > 0,
Z
C 1
|X j1 X j2 . . . X j s JR u(x)| ≤ |u(y)|dy.
R F (x, R) 2+s Ω(x, R)
210 Nicola Garofalo
Proof. We first consider the case s = 1. From (7.13), and from the support property
(7.20) of K R (x, ·), we can differentiate under the integral sign in (7.19), to obtain
Z
|X JR u(x)| ≤ |u(y)||X x K R (x, y)|dy.
B(x,2R)
By the definition of K R (x, y) and the chain rule it is easy to recognize that the com-
ponents of its subgradient X x K R (x, y) are estimated as follows:
To control the three terms in the right-hand side of the above inequality we use the
size estimates (7.13), along with the observation that, due to the fact that on the
support of K R (x, ·) one has
1 1
< Γ(x, y) ≤ ,
2R R
then Theorem 7.9 and (7.17) give for all x ∈ D, 0 < R ≤ R0 , and y ∈ Ω(x, 2R) \
Ω(x, R),
d(x, y)
C≤ ≤ C −1 . (7.22)
F (x, R)
Using (7.13), (7.22), one obtains that for i = 1, 2, 3,
C
sup |I Ri (x, y)| ≤
y ∈Ω(x,2R)\Ω(x, R) RF (x, R) 3
for any x ∈ D, provided that 0 < R ≤ R0 . This completes the proof in the case
s = 1. The case s ≥ 2 is handled recursively by similar considerations based on
Theorem 7.9, and we omit details. It may be helpful for the interested reader to note
that Theorem 7.9 implies
Theorem 7.16 (See [CDG97]). Let D ⊂ Rn be a bounded open set and suppose that
u is harmonic in D. There exists R0 > 0, depending on D and X, such that for every
x ∈ D and 0 < r ≤ R0 for which B X (x,r) ⊂ D, one has for any s ∈ N
C
|X j1 X j2 . . . X j s u(x)| ≤ max |u|,
r s B X (x,r )
for some constant C = C(D, X, s) > 0. In the above estimate, for every i = 1, . . . , s,
the index ji runs in the set {1, . . . , m}.
Remark 7.17. We emphasize that Theorem 7.16 cannot be established similarly to its
classical ancestor for harmonic functions, where one uses the mean value theorem
coupled with the trivial observation that any derivative of a harmonic function is
harmonic. In the present non-commutative setting, derivatives of harmonic functions
are no longer harmonic!
Proof of Theorem 7.16. We observe explicitly that the assumption states that with
R = E(x,r)/2, then Ω(x, 2R) = B X (x,r) ⊂ D. By Theorem 7.14, and by (7.23), we
find
To complete the proof we need to observe only that Ω(x, R) = B X (x,r), and that,
thanks to Theorem 7.9, (7.17), one has
C |B(x,r)| C −1
≤ ≤ .
rs RF (x, R) 2+s rs
www.Ebook777.com
212 Nicola Garofalo
Then, there exists S1 > 0, depending on C1 in (6.1) and on C in (8.1), such that for
1,1
every f ∈ W X,0 (B(x,r)) one has
Z ! (Q−1)/Q Z !
1 1
|f| Q/(Q−1)
dy ≤ S1 r |X f |dy . (8.2)
|B(x,r)| B(x,r ) |B(x,r)| B(x,r )
We note explicitly that inequality (8.2) in Theorem 8.1 cannot be derived directly
from the hypothesis (8.1) since, as we saw in Theorem 6.18, the fractional integration
operator I1 does not map L 1 (B(x,r), dν) into L q (B(x,r), dν), with q = Q/(Q − 1),
but it maps instead L 1 (B(x,r), dν) into L q,∞ (B(x,r), dν). Our strategy for proving
Theorem 8.1 consists, in fact, in first establishing the following weak version of it.
Theorem 8.2 (Weak geometric Sobolev embedding). Under the assumptions of The-
0,1
orem 8.1, there exists S1 > 0 such that for every f ∈ W X,0 (B(x,r)) one has with
q = Q/(Q − 1),
Z
r
sup t |{y ∈ B(x,r) | | f (y)| > t}| 1/q ≤ S1 |X f |dy. (8.3)
t >0 |B(x,r)| 1/Q B(x,r )
We next show how Theorem 8.2 implies a remarkable, sharp isoperimetric in-
equality.
Theorem 8.3 (Isoperimetric inequality). Under the assumptions in Theorem 8.2, let
U and B(x,r) be as in its statement. Let E ⊂ E ⊂ B(x,r) be a C 1 open set. Then,
with the same constant S1 > 0 as in equation (8.3) above, one has
r
|E| (Q−1)/Q ≤ S1 PX (E; B(x,r)), (8.4)
|B(x,r)| 1/Q
where PX (E; B(x,r)) denotes the relative X-perimeter of E with respect to B(x,r)
(see Definition 5.8 above).
Proof. Let
r (x) = inf |x − y| = diste (x, E).
y ∈E
Set δ0 = 1
2 diste (E, B(x,r)), and for 0 < δ ≤ δ0 consider the function
!+
r (x)
uδ (x) = 1 − .
δ
214 Nicola Garofalo
Applying formula (8.4) in Theorem 8.3 to the set Et we conclude from the last
equation,
Z 1/Q Z ∞
−1 |B(x,r)|
|X f |dy ≥ (S1 ) |Et | (Q−1)/Q dt. (8.5)
B(x,r ) r 0
The proof of this inequality is a simple consequence of the fact that the function
t → |Et | is non-increasing. Using this inequality in (8.5), and the real analysis result
which states that
Z Z ∞
Q
|f| Q/(Q−1)
dy = t 1/(Q−1) |Et |dt,
B(x,r ) Q−1 0
we finally obtain the desired conclusion (8.2) by a density argument.
Remark 8.4. We stress that the above proofs show that the weak geometric Sobolev
embedding is in fact equivalent to the strong one since, trivially, we have that Theo-
rem 8.1 =⇒ Theorem 8.2, and we have proved that
Theorem 8. 2 =⇒ Theorem 8. 3 =⇒ Theorem 8. 1 .
An important consequence of the geometric Sobolev embedding Theorem 8.1
is the following result which constitutes a generalization of the celebrated Sobolev
embedding theorem.3
Theorem 8.5 (Non-geometric Sobolev embedding). Under the assumptions of The-
orem 8.1, given any number 1 < p < Q there exists a constant S p > 0, depending on
1, p
C1 in (6.1), C in (8.1), and p, such that for any function f ∈ W X,0 (B(x,r)) one has
Z ! 1/q Z ! 1/p
1 1
| f | q dy ≤ Sp r |X f | p dy , (8.6)
|B(x,r)| B(x,r ) |B(x,r)| B(x,r )
Remark 8.6. As a consequence of Theorem 7.9 and Corollary 7.12 we see that when
X1 , . . . , X m ∈ C ∞ (R N ) is a system satisfying the finite rank condition (1.3) above,
then the assumption (8.1) is fulfilled. Furthermore, thanks to Theorem 7.4 above, the
doubling condition (7.6) is also satisfied. It follows that Theorems 8.1 and 8.3 are
both valid for such systems. These facts play an important role in the study of the
regularity of (weak) solutions of partial differential equations of subelliptic type.
In closing we mention that, under very general assumptions, a comprehensive
theory of isoperimetric and Sobolev inequalities for BV functions was developed in
the paper [GN96], which also contains an existence result for minimal surfaces in
Carnot–Carathéodory spaces. The interested reader should consult that source for
all the relevant results.
9.1.1 The Li–Yau inequality. A fundamental result of Li and Yau (see [LY86]),
states that in every complete, boundaryless, n-dimensional Riemannian manifold M
with Ricci ≥ 0, for any positive solution u of the heat equation on M one has for
f = log u,
n
|∇ f | 2 − f t ≤ in M × (0, ∞). (9.1)
2t
The proof of this result follows easily from the following deep a priori gradient
estimate.
Let us see how the Li–Yau inequality (9.1) can be derived from Theorem 9.1. But
let us first give the formal statement.
Before discussing the original proof of the Li–Yau inequality (9.3) we briefly
recall the two fundamental tools from Riemannian geometry which constitute its key
ingredients:
T (X,Y ) = ∇ X Y − ∇Y X − [X,Y ].
R(X,Y ) Z = ∇ X ∇Y Z − ∇Y ∇ X Z − ∇[X,Y ] Z.
if v =
Pn
i=1 vi Ei .
9.1.3 The Bochner identity. Henceforth in these notes we will adopt the summa-
tion convention over repeated indices. Given a function f ∈ C ∞ (M) we define its
gradient as the vector field ∇ f ∈ X(M) defined by
g(∇ f ,V ) = d f (V ), ∀V ∈ X(M),
d f (V ) = V f .
Proof. The proof follows easily from the following commutation identity known as
a special case of Ricci’s formulas (see, e.g., [CLN06, Lemma 1.36]),
∆∇i f = ∇i ∆ f + Ri j ∇ j f . (9.5)
∆F 2 = 2F∆F + 2|∇F | 2 .
Ric(v, v) ≥ 0. (9.8)
∆u − ut = 0, in M × (0, ∞).
Since u is positive we can consider the function f = log u. With h(t) = log t, and by
means of the formula
∆u − ut |∇u| 2
∆ f − ft = − 2 = −|∇ f | 2 . (9.9)
u2 u
One has the following crucial result.
Lemma 9.8. For any α ≥ 1, let
F = t(|∇ f | 2 − α f t ).
Then,
2t
∆F − Ft ≥ −2h∇ f , ∇Fi + (|∇ f | 2 − f t ) 2 − (|∇ f | 2 − α f t ). (9.10)
n
Proof. The proof of this lemma is based exclusively on the Bochner identity (9.4).
For its details the reader should see the original paper of Li and Yau, or the upcoming
book [BG16].
We can now establish the main result of Li and Yau.
Proof of Theorem 9.1. Let f = log u as before, and let F have the same meaning as
in Lemma 9.8. Since
|∇u| 2 ut F
2
−α = ,
u u t
proving the theorem is equivalent to showing that
F α4 nα 2
sup ≤C 2 + .
BR t (α − 1)R2 2t
www.Ebook777.com
222 Nicola Garofalo
(ψ 0 ) 2
ψ 0 ≤ 0, ψ 00 ≥ −C1 , ≤ C2 .
ψ
and so
|φ| 2 (ψ 0 ) 2 C2
= |∇%| 2 ≤ 2 .
φ ψ R2 R
In order to estimate ∆F from below we are going to use Lemma 9.8 (which
ultimately relies on the Bochner formula), whereas to estimate the term ∆φ from
below the Laplacian comparison Theorem 9.7 plays a crucial role. We recall that,
under the assumption Ric(M) ≥ 0, the latter gives
n−1
∆% ≤ .
%
Since the support of ψ 0 is concentrated on the interval [1, 2], from the formula
ψ 00 ψ0
∆φ = |∇%| 2
+ ∆%,
R2 R
we conclude that
C1 (n − 1)|ψ 0 |
∆φ ≥ − − .
R2 R2
We thus find
|φ| 2
0 ≥φ∆F + F∆φ − 2F
φ
C1 + 2C2 + (n − 1)|ψ 0 |
≥φ∆F − F.
R2
Using Lemma 9.8 we now obtain
( )
2t
0 ≥ φ −2h∇ f , ∇Fi + (|∇ f | 2 − f t ) 2 − (|∇ f | 2 − α f t ) + Ft
n
C1 + 2C2 + (n − 1)|ψ 0 |
− F
R2
h∇ f , ∇φi
( )
2t
≥φ 2 F + (|∇ f | − f t ) − (|∇ f | − α f t )
2 2 2
φ n
C1 + 2C2 + (n − 1)|ψ 0 |
− F,
R2
since φFt = (φF)t ≥ 0 at (x 0 ,t 0 ). At this point we set for notational simplicity
C1 + 2C2 + (n − 1)max |ψ 0 |
A= .
R
R2
224 Nicola Garofalo
From this point on, after several clever algebraic manipulations one arrives at the
inequality
2 nα 2 C2 t
0 ≥ − [1 + At](φF) + 2
(φF) 2 − (φF)
nα 4(α − 1)R2
nα 2 C2 t
( " #)
2
=(φF) (φF) − 1 + At + .
nα 2 4(α − 1)R2
nα 2 nα 2 C2 t
" #
φF ≤ 1 + At + .
2 4(α − 1)R2
nα 2 nα 2 nα 2 C2
!
(φF)(x,T ) ≤ + A+ T.
2 2 4(α − 1)R2
C 1 +2C 2 +(n−1)max |ψ 0 |
Since φ ≡ 1 on BR , recalling that A = R2
R
, we conclude that
nα 2 α 2 C5 α2
!
F (·,T )
sup ≤ + 1 + . (9.11)
BR T 2T R2 α−1
Γ( f ) = Γ( f , f ) = |∇ f | 2 , (10.3)
1
Γ2 ( f ) = Γ2 ( f , f ) = ∆|∇ f | 2 − h∇ f , ∇(∆ f )i (10.4)
2
1
= ∆Γ( f ) − Γ( f , ∆ f ).
2
By means of the forms Γ and Γ2 we now introduce the following definition.
Definition 10.1. We say that a Riemannian manifold (M, g) satisfies the curvature-
dimension inequality CD( %, n), for % ∈ R and n > 0, if for every f ∈ C ∞ (M) one
has
1
Γ2 ( f ) ≥ (∆ f ) 2 + %Γ( f ), f ∈ C ∞ (M). (10.5)
n
It is remarkable that the curvature-dimension inequality (10.5) perfectly captures
the notion of the Ricci curvature lower bound. This will be evident after we state the
following result.
Theorem 10.2. On a complete n-dimensional Riemannian manifold M the inequality
CD( %, n) is equivalent to Ric(M) ≥ %.
Proof. We begin with proving the simple direction, namely that Ric(M) ≥ % =⇒
CD( %, n). For this direction, in fact, we do not need M to be complete. We apply
Bochner’s formula (9.4), which we can rewrite as
Using the latter two inequalities in (10.7) we conclude that the curvature-dimension
inequality CD( %, n) holds.
To prove the opposite direction, we assume that CD( %, n) holds and we want to
show that, given x ∈ M, for any v ∈ Tx M we have Ric(v, v) ≥ %|v| 2 . We can find
a sufficiently small neighborhood of x, Ux ⊂ M, and a function f ∈ C ∞ (Ux ) such
that ∇ f (x) = v and ∇2 f (x) = 0. By suitably modifying the function f outside Ux
we can assume that it is smooth in all of M. From the Bochner identity written in the
form (10.7), and the choice of f , we obtain
10.2 The main variational inequality. If we fix T > 0 and for a function
g ∈ C0∞ (M) we consider
v(x,t) = PT −t g(x),
then it should be clear that v solves the backward Cauchy problem
∆v + vt = 0 in M × (−∞,T ),
v(x,T ) = g(x), x ∈ M. (10.8)
In what follows we consider a fixed x ∈ M and a fixed T > 0. For f ∈ C0∞ (M),
f ≥ 0, we introduce the functionals
and
Φ(t) = Pt ((PT −t f )Γ(ln PT −t f )) (x), 0 ≤ t < T. (10.10)
At this point there is some delicate question having to do with the domain of the
functionals Ψ, Φ, but we will proceed formally without paying attention to them. For
the relevant technical details we refer the reader to [B86]. Notice that
and
2aγ 2
!
4aγ 4aγ
(aΦ) ≥ a + 2%a −
0 0
Φ+ ∆PT f − PT f . (10.16)
n n n
1
Γ2 (ln PT −t f ) ≥ (∆(ln PT −t f )) 2 + %Γ(ln PT −t f ).
n
Using the numerical inequality 2 βγ ≤ β 2 + γ 2 we obtain for every γ ∈ R,
(∆(ln PT −t f )) 2 ≥ 2γ∆(ln PT −t f ) − γ 2 .
Furthermore,
∆(PT −t f )
∆(ln PT −t f ) = − Γ(ln PT −t f ).
PT −t f
228 Nicola Garofalo
We infer that
!
2γ ∆(PT −t f ) 2γ
Γ2 (ln PT −t f ) ≥ + %− Γ(ln PT −t f ) − γ 2 .
n PT −t f n
nα 2
Γ(ln Pt f )(x) − ∂t (ln Pt f )(x) ≤ (10.18)
8t(α − 1)
%t
( )
n 2t
+% −1 − ∂t (ln Pt f )(x) .
2 α+1 α+1
Proof. Let x ∈ M and T > 0 be arbitrarily fixed. Consider the entropy functional
Φ(t) defined in (10.10). For any non-negative C 1 function a on the interval [0,T] we
choose the function γ in a such a way that the coefficient of Φ(t) in (10.16) vanishes,
i.e.,
4aγ
a0 − + 2%a = 0.
n
Solving for γ in this equation we obtain
n a0
!
γ= + 2% .
4 a
2 Hypoelliptic operators, etc. 229
In what follows it will prove useful to linearize the functional inequality (10.20) with
respect to the function a. With this objective in mind we write a = V 2 , obtaining
from (10.20),
Z T !
∆PT f (x)
Γ(ln PT f )(x) + 2% V 2 dt − 1 (10.21)
0 PT f (x)
Z T Z T !
n
≤ (V ) dt + %
0 2 2
V dt − % .
2
2 0 0
We now choose a = V 2 as
t α
a(t) = 1 − , α > 1.
T
Clearly, a ∈ C 1 ([0,T]), and a(0) = 1, a(T ) = 0. Noting that such a choice gives
α2
Z T Z T
T
V dt =
2
, (V 0 ) 2 dt = ,
0 α+1 0 4(α − 1)T
from inequality (10.21) we obtain
!
2%T ∆PT f (x)
Γ(ln PT f )(x) ≤ 1 −
α + 1 PT f (x)
%T nα 2
!
n
+ % −1 + .
2 α+1 8T (α − 1)
By the arbitrariness of T > 0 and the fact that ∆Pt f (x) = ∂t Pt f (x), we reach the
desired conclusion (10.18).
230 Nicola Garofalo
n %t
( 2t )
n
Γ(ln Pt f )(x) − ∂t (ln Pt f )(x) ≤ +% − 1 − ∂t (ln Pt f )(x) .
2t 2 3 3
(10.22)
n% t n| %| 2| %|t
( 2 )
n
Γ(uk ) − ∂t uk ≤ + + + ∂t uk .
2t 6 2 3
2| %|t n n%2 t n| %|
!
− 1+ ∂t uk ≤ −Γ(uk ) + + + . (10.24)
3 2t 6 2
2 Hypoelliptic operators, etc. 231
We now fix two points y, z ∈ M, and fix 0 < s < t < ∞. Since
1 n 3n| %|
−∂t uk (α(τ)) ≤ − Γ(uk (α(τ))) + τ + .
β(0) 2(t + T (s − t)) 4
232 Nicola Garofalo
We thus have
Z T Z T
Pt f k (y) t−s
log ≤ |∇uk (α(τ))|dτ − |∇uk (α(τ))| 2 dτ
Ps f k (z) 0 T β(0) 0
t−s T t − s 3n| %|
Z
n
+ τ dτ + T
T 0 2(t + T (s − t)) T 4
ε T
Z Z T
T t−s
≤ |∇uk (α(τ))| dτ +
2
− |∇uk (α(τ))| 2 dτ
2 0 2ε T β(0) 0
n t 3n| %|(t − s)
+ log + .
2 s 4
At this point we choose ε > 0 such that
ε t−s
= .
2 T β(0)
2| % |t
Recalling that β(0) = 1 + 3 , this gives
T2 2| %|t
!
Pt f k (y)
log ≤ 1+
Ps f k (z) 4(t − s) 3
n t 3n| %|(t − s)
+ log + .
2 s 4
If we now take the infimum on all competing T > 0, and we let k → ∞, we obtain
3n| %|r 2
r n !
2 2
p(x, x,r 0 ) ≤ p(x, x,r ) exp .
r0 4
Corollary 10.10 (Harnack inequality for the heat kernel). Let M be an n-dimen-
sional complete Riemannian with Ric(M) ≥ 0, and denote by p(x, y,t) the heat
kernel on it, i.e., the fundamental solution of the heat operator ∆ − ∂t on M × (0, ∞).
For every x, y, z ∈ M and every 0 < s < t < ∞ one has
d(y, z) 2
t n/2 !
p(x, y, s) ≤ p(x, z,t) exp .
s 4(t − s)
2 Hypoelliptic operators, etc. 233
the sphere Sn of constant sectional curvature κ when κ > 0,
Mκ = Rn when κ = 0,
the hyperbolic space Hn of constant sectional curvature κ
when κ < 0.
Furthermore, we have (see [Ch93, (3.25)])
Z r
Vκ (r) = σ n−1 Sκ (t) n−1 dt,
0
where
2π n/2
σ n−1 =
Γ(n/2)
is the area of the unit sphere in Rn , and
√
√1 sin
κ
κt, κ > 0,
Sκ (t) = t, κ = 0,
√1 sinh √ −κt, κ < 0.
−κ
234 Nicola Garofalo
V (x,r)
lim = 1,
r →0+ Vκ (r)
V (x,r)
r →
rn
is non-increasing. As a consequence, one has
Theorem 11.1 and Corollary 11.2 play an important role in studying the spectrum
of the Laplace–Beltrami on a manifold, Gaussian bounds on the heat kernel, isoperi-
metric theorems, etc. In this subsection we intend to illustrate how a weaker form of
Corollary 11.2 can be deduced from the entropy inequality in Theorem 10.4. This ap-
proach shows, in particular, that remarkably, (11.2) can be exclusively derived from
the Bochner identity on functions without direct use of the theory of Jacobi fields.
Recall the definition of the function V (x,r) = µ(B(x,r)), where now µ indicates the
Riemannian volume. Here is the main result that we intend to establish.
11.1.1 Preliminary reductions. From the semigroup property and the symmetry
of the heat kernel we have for any y ∈ M and t > 0,
Z
p(y, y, 2t) = p(y, z,t) 2 dµ(z).
M
√
Consider now a function h ∈ C0∞ (M) such that 0 ≤ h ≤ 1, h ≡ 1 on B(x, t/2), and
√
h ≡ 0 outside B(x, t). We thus have
Z
Pt h(y) = p(y, z,t)h(z)dµ(z)
M
Z 1/2 Z 1/2
2
≤ √ p(y, z,t) dµ(z) h(z) 2 dµ(z)
B(x, t) M
√
≤ p(y, y, 2t) 1/2
V (x, t) 1/2
.
p(x, x, 2r 2 ) V (x,r)
V (x, 2r) ≤ C(n) (11.7)
p(x, x, 4r 2 ) Pr 2 1 B(x,r ) (x) 2
V (x,r)
≤ C ∗ (n) ,
Pr 2 1 B(x,r ) (x) 2
11.1.2 The main estimate. From inequality (11.7) it is clear that we would be
finished with the proof of (11.2) if we could show that there exist universal constants
A ∈ (0, 1), and K > 0, such that one has for every x ∈ M and r > 0,
P Ar 2 1 B(x,r ) (x) ≥ K.
a(t) = τ + T − t,
n
γ(t) = − ,
4(τ + T − t)
where τ > 0 will later be optimized. Noting that we presently have a(0) = τ + T,
a(T ) = τ, and that
n 2aγ 2 n
a 0 ≡ −1, aγ ≡ − , − =− ,
4 n 8(τ + T − t)
from (10.16) in Theorem 10.4 we obtain
f ∈ Lipschitz(M) ∩ L ∞ (M), Γ( f ) ≤ 1.
n T
≥ −T∆PT e λ f − e λψ ln 1 + .
8 τ
The assumption Γ( f ) ≤ 1 implies
PT e λ f Γ( f ) ≤ PT e λ f = e λψ .
∂ψ λ
r !
n
− ≤ 1+ 2 −1 (11.11)
∂t 2 2λ t
! !
n 2 1
+ ln 1 + q = λG 2 ,
8λt 1 + 2λn2 t − 1 λ t
and so we have
Pt f ≤ ψ. (11.12)
Using (11.12), we infer
Z t !
1
Ps (λ f ) ≤ λψ(λ,t) + λ 2
G 2 dτ.
s λ τ
Letting s → 0+ we conclude that
Z t !
1
λ f ≤ λψ(λ,t) + λ 2 G dτ. (11.13)
0 λ2 τ
At this point consider the function
e λ f ≤ e−λr 1 B c + 1 B .
To estimate the first term in the right-hand side of the latter inequality, we use
(11.13) which gives
1 = e λ f (x) ≤ e λψ (λ, t )(x) eΦ(λ, t ) ,
where we have set Z t !
1
Φ(λ,t) = λ 2
G 2 dτ.
0 λ τ
2 Hypoelliptic operators, etc. 239
This gives
Pt (1 B )(x) ≥ e−Φ(λ, t ) − e−λr .
To make use of this estimate, we now choose λ = r1 , t = Ar 2 , obtaining
2
P Ar 2 (1 B )(x) ≥ e−Φ(1/r, Ar ) − e−1 .
We want to show that we can choose A > 0 sufficiently small, depending only on
n, and a K > 0 (we can in fact take K = (1 − e−1 )/2) such that
2
e−Φ(1/r, Ar ) − e−1 ≥ K, for every x ∈ M,r > 0. (11.14)
As noted above, a direct examination shows that G(t) = O(t 1/2 ) as t → +∞, and
therefore that G(t
t2
)
∈ L 1 (1, ∞). It is then clear that
Z ∞
G(t)
lim+ dt = 0,
A→0 A −1 t2
and therefore there exists A > 0 sufficiently small such that (11.14) holds with, say,
K = (1 − e−1 )/2 > 0.
where bi and σi j are smooth functions and the matrix (σi j (x))1≤i, j ≤n is symmetric
and positive semidefinite.
There is a natural gradient (or rather, a natural square of the length of a gradient)
canonically associated with L, and it is given by the quadratic functional Γ( f ) =
Γ( f , f ), where
1
Γ( f , g) = L( f g) − f Lg − gL f , f , g ∈ C ∞ (M).
(12.6)
2
The functional
1
Γ( f ) = L( f 2 ) − 2 f L f
(12.7)
2
is known as le carré du champ. Notice that (12.2) implies Γ(1) = 0. From (12.5) it
is easily recognized that in the above local chart U, the differential bilinear form Γ is
given by
n
X ∂f ∂f
Γ( f ) = σi j (x) .
i, j=1
∂x i ∂x j
This shows that Γ( f ) ≥ 0 on M and that Γ( f ) only involves differentiation of order
1. It is also worth noting that, by definition,
L( f 2 ) = 2 f L f + 2Γ( f ).
Free ebooks ==> www.Ebook777.com
L(ϕ ◦ f ) = ϕ 00 ◦ f Γ( f ) + ϕ 0 ◦ f L f ,
and therefore
Γ(ϕ ◦ f ) = (ϕ 0 ◦ f ) 2 Γ( f ).
If
n n
X ∂2 X ∂
L= σi j (x) + bi (x)
i, j=1
∂x i ∂x j i=1
∂x i
and
n n
X ∂2 X ∂
L= σ̃i j (x) + b̃i (x)
i, j=1
∂x i ∂x j i=1 ∂x i
12.1 Canonical distances on M. With the operator L, one can associate sev-
eral canonical distances. The most basic one hinges on the following definition,
which is based on Definitions 4.1 and 4.3 in Section 4.
Definition 12.1. A tangent vector v ∈ Tx M is called subunit for L at x if in a
local representation (12.5) of L at x, one has vi v j ≤ σi j (x) as matrices, where
v = i=1 vi ∂x∂ i . This amounts to having for every ξ ∈ Rn ,
Pn
n
X 2 n
X
vi ξ i ≤ σi j (x)ξ i ξ j . (12.8)
i=1 i, j=1
We emphasize that the notion of a subunit vector at x does not depend on the
local representation (12.5) of L.
Definition 12.2. A Lipschitz path γ : [0,T] → M is called subunit for L if γ 0 (t)
is subunit for L at γ(t) at every point of differentiability of γ, and therefore at a.e.
t ∈ [0,T].
For future use let us note one important consequence of (12.8). Suppose that
γ : [0,T] → M is a subunit curve for L. Given a C 1 function u on M define ϕ(t) =
u(γ(t)). If we fix a point t 0 ∈ [0,T] of differentiability for ϕ(t), then in a local chart
around γ(t 0 ) we have from (12.8),
s
n
∂u ∂u ∂u
X
|ϕ 0 (t)| = (γ(t))γi0 (t) ≤ σi j (γ(t)) (γ(t)) (γ(t)).
i=1 ∂x i ∂x i ∂x i
www.Ebook777.com
242 Nicola Garofalo
defines a true distance on M. This distance is canonically associated with the diffu-
sion operator L.
We next introduce another distance on M which is canonically associated with
the operator L. For a function g on M we let ||g||∞ = ess sup |g|. Given two points
M
x, y ∈ M we define
( )
% L (x, y) = sup | f (x) − f (y)| | f ∈ C0∞ (M), kΓ( f )k∞ ≤ 1 , x, y ∈ M. (12.11)
Let us notice the following trivial, yet useful, fact: for every f ∈ C ∞ (M) such that
||Γ( f )||∞ < ∞, we have
Proposition 12.4. Under Hypothesis 12.3 the function % L in (12.11) defines a metric
on M and for every x, y ∈ M we have
Henceforth, we assume
Hypothesis 12.5. The metric space (M, d L ) is complete.
2 Hypoelliptic operators, etc. 243
Γ Z ( f g, h) = f Γ Z (g, h) + gΓ Z ( f , h).
Γ Z ( f ) ≥ 0.
Besides Hypotheses 12.3 and 12.5 above, we will work with three general assump-
tions. The former two will be listed as Hypotheses 12.6 and 12.7 below, the third
one will be introduced in Definition 12.9 below.
Hypothesis 12.6. There exists an increasing sequence hk ∈ C0∞ (M) such that hk % 1
on M, and
||Γ(hk )||∞ + ||Γ Z (hk )||∞ → 0, as k → ∞.
We will also assume that the following commutation relation is satisfied.
Hypothesis 12.7. For any f ∈ C ∞ (M) one has
Γ( f , Γ Z ( f )) = Γ Z ( f , Γ( f )).
vertical directions. We also would like to mention at this point that the existence
of differential forms Γ, Γ Z satisfying Hypotheses 12.6 and 12.7 does not necessarily
imply that the sub-Riemannian structure has rank 2.
1
Γ2 ( f , g) = LΓ( f , g) − Γ( f , Lg) − Γ(g, L f ) ,
(12.13)
2
1
Γ2Z ( f , g) = LΓ Z ( f , g) − Γ Z ( f , Lg) − Γ Z (g, L f ) .
(12.14)
2
Observe that if Γ Z ≡ 0, then Γ2Z ≡ 0 as well. As for Γ and Γ Z , we will use the
notation
Γ2 ( f ) = Γ2 ( f , f ), Γ2Z ( f ) = Γ2Z ( f , f ).
1 κ
Γ2 ( f ) + νΓ2Z ( f ) ≥ (L f ) + %1 −
2
Γ( f ) + %2 Γ Z ( f ) (12.15)
d ν
holds for every f ∈ C ∞ (M) and every ν > 0.
The essential aspect of the theory developed in [BG09] and [BG15] is that, with
Hypotheses 12.3, 12.5, 12.6, and 12.7 in place, all the results are solely deduced from
the curvature-dimension inequality CD( %1 , %2 , κ, d) in (12.15). It is worth observing
explicitly that, as we have mentioned above, the Riemannian case corresponds to the
situation Γ Z = 0 and κ = 0 in (12.15). Hence in this case with L = ∆, and d = n =
dimension of M, our (12.15) is nothing but the Riemannian curvature-dimension
inequality CD( %1 , n) in (10.5) above. We also remark that, changing Γ Z into aΓ Z ,
where a > 0, changes the inequality CD( %1 , %2 , κ, d) into CD( %1 , a %2 , aκ, d). We
express this fact by saying that the quantity %κ2 is intrinsic.
2 Hypoelliptic operators, etc. 245
L = X 2 + Y 2. (12.16)
1
Γ( f ) = L( f 2 ) − 2 f L f = (X f ) 2 + (Y f ) 2 .
2
If we define
Γ Z ( f , g) = Z f Zg,
then from (i)–(iii) we easily verify that
Γ( f , Γ Z ( f )) = Γ Z ( f , Γ( f )).
We conclude that Hypothesis 12.7 is satisfied. It is not difficult to show that Hypoth-
esis 12.6 is also fulfilled.
246 Nicola Garofalo
[L, Z] = 0. (12.17)
1
Γ2Z ( f ) = L(Γ Z ( f )) − Γ Z ( f , L f ) = Z f [L, Z] f + (X Z f ) 2 + (Y Z f ) 2
2
= (X Z f ) 2 + (Y Z f ) 2 .
1
Γ2 ( f ) = L Γ( f ) − Γ( f , L f )
2
= %1 Γ( f ) + (X 2 f ) 2 + (Y X f ) 2 + (XY f ) 2 + (Y 2 f ) 2
+ 2Y f (X Z f ) − 2X f (Y Z f ).
We now notice that if we define the symmetrized Hessian of f with respect to the
horizontal distribution generated by X,Y in the following way:
X2 f 1
(XY f + Y X f )
!
∇H f = 1
2 2 ,
2 (XY f + Y X f ) Y2 f
then
1
(X 2 f ) 2 + (Y X f ) 2 + (XY f ) 2 + (Y 2 f ) 2 = ||∇2H f || 2 + Γ Z ( f ).
2
Substituting this information into the above formula we find that
1
Γ2 ( f ) = ||∇2H f || 2 + %1 Γ( f ) + Γ Z ( f ) + 2 Y f (X Z f ) − X f (Y Z f ) .
2
By the above expression for Γ2Z ( f ), using the Cauchy–Schwarz inequality, we ob-
tain, for every ν > 0,
1
|2Y f (X Z f ) − 2X f (Y Z f )| ≤ νΓ2Z ( f ) + Γ( f ).
ν
Similarly, one easily recognizes that
1
||∇2H f || 2 ≥ (L f ) 2 .
2
Combining these inequalities, we conclude that we have proved the following result.
2 Hypoelliptic operators, etc. 247
Proposition 12.10. For every %1 ∈ R the Lie group G( %1 ), with the sub-Laplacian
L in (12.16), satisfies the curvature-dimension inequality CD( %1 , 12 , 1, 2). Precisely,
for every f ∈ C ∞ (G( %1 )) and any ν > 0 one has
!
1 1 1
Γ2 ( f ) + νΓ2Z ( f ) ≥ (L f ) 2 + %1 − Γ( f ) + Γ Z ( f ).
2 ν 2
Proposition 12.10 provides basic motivation for Definition 12.9. It is also impor-
tant to observe at this point that the Lie group G( %1 ) can be endowed with a natural
CR structure. In fact, denoting by H the subbundle of TG( %1 ) generated by the
vector fields X and Y , the endomorphism J of H, defined by
J (Y ) = X, J (X ) = −Y,
12.5 The model spaces. In this final subsection we consider three model spaces
which correspond to the Riemannian space forms of constant sectional curvature:
the unit sphere, flat Rn , and hyperbolic space. These three model spaces correspond
respectively to the cases %1 = 1, %1 = 0, and %1 = −1 of the Sasakian three manifolds
discussed in Section 12.4. For more general situations the reader should see the
papers [BG09, BG15], and also the subsequent lecture notes of Fabrice Baudoin in
this volume.
Example 12.11 (%1 = 1). The Lie group SU(2) is the group of 2 × 2, complex,
unitary matrices of determinant 1. Its Lie algebra su(2) consists of 2 × 2, complex,
skew-hermitian matrices with trace 0. A basis of su(2) is formed by the following
matrices X = 2i σ1 , Y = 2i σ2 , Z = 2i σ3 , where σk , k = 1, 2, 3, are the Pauli matrices:
! ! !
1 0 1 1 0 i 1 i 0
X= , Y= , Z= .
2 −1 0 2 i 0 2 0 −i
248 Nicola Garofalo
and thus %1 = 1.
Example 12.12 (%1 = 0). The Heisenberg group H is the group of 3 × 3 matrices
1 x z
*. 0 1 y +/ , x, y, z ∈ R.
, 0 0 1 -
The Lie algebra of H is spanned by the matrices
0 1 0 0 0 0 0 0 1
X= 0
*
. 0 0 +/ , Y = *. 0 0 1 +/ , Z = *. 0 0 0 +/ ,
, 0 0 0 - , 0 0 0 - , 0 0 0 -
for which the following commutation relations hold:
Acknowledgments
The author is deeply grateful to the organizers of the semester for the gracious invita-
tion: Andrei Agrachev, Davide Barilari, Ugo Boscain, Yacine Chitour, Frédéric Jean,
Ludovic Rifford, Mario Sigalotti. He also gratefully acknowledges the most gracious
hospitality and the financial support of the Institut Henri Poincaré, the CNRS, and
the ERC StG 2009 “GeCoMethods”, contract number 239748 (PI: Ugo Boscain).
Bibliography 249
Bibliography
[AL14] A. Agrachev and P. Lee, Generalized Ricci curvature bounds on three-dimensional contact
sub-Riemannian manifolds, Math. Ann. 360 (2014), no. 1-2, 209–253. Zbl 06357943.
[A76] T. Aubin, Problèmes isopérimétriques et espaces de Sobolev (French), J. Differential Ge-
ometry 11 (1976), no. 4, 573–598. Zbl 0371.46011.
[B86] D. Bakry, Un critère de non-explosion pour certaines diffusions sur une variété riemanni-
enne complète, C.R. Acad. Sci. Paris Sér. I Math. 303 (1986), no. 1, 23–26. Zbl 0589.60069.
[B94] D. Bakry, L’hypercontractivité et son utilisation en théorie des semigroupes, Ecole d’Eté de
Probabilites de St-Flour, Lecture Notes in Math, (1994), 1–114. Zbl 0856.47026.
[B06] D. Bakry, Functional inequalities for Markov semigroups, Probability measures on groups:
Recent directions and trends, 91–147, Tata Inst. Fund. Res., Mumbai, 2006. Zbl 1148.60057.
[BBBQ08] D. Bakry, F. Baudoin, M. Bonnefont, and B. Qian, Subelliptic Li-Yau estimations on
three dimensional model spaces, Potential theory and stochastics in Albac, 1-10, Theta Ser.
Adv. Math., 11, Theta, Bucharest, 2009. Zbl 1199.35051.
[BE83] D. Bakry and M. Émery, Diffusions hypercontractives, Sémin. de probabilités XIX, Univ.
Strasbourg, Springer, 1983. Zbl 0561.60080.
[BGL14] D. Bakry, I. Gentil, and M. Ledoux, Analysis and geometry of Markov diffusion oper-
ators, Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathe-
matical Sciences], 348. Springer, Cham, 2014. xx+552 pp. Zbl 06175511.
[BL06] D. Bakry and M. Ledoux, A logarithmic Sobolev form of the Li-Yau parabolic inequality,
Revista Mat. Iberoamericana 22 (2006), 683–702.
[Ba04] F. Baudoin, An introduction to the geometry of stochastic flows, Imperial College Press,
London, 2004. x+140 pp. Zbl 1116.58024.
[BG09] F. Baudoin and N. Garofalo, Generalized Bochner formulas and Ricci lower bounds for
sub-Riemannian manifolds of rank two, ArXiv, April 2009. arXiv:0904.1623.
[BG11] F. Baudoin and N. Garofalo, Perelman’s entropy and doubling property on Riemannian
manifolds, J. Geom. Anal. 21 (2011), no. 4, 1119–1131. Zbl 1238.58024.
[BG13] F. Baudoin and N. Garofalo, A note on the boundedness of Riesz transform for some subel-
liptic operators, Int. Math. Res. Not. IMRN 2013, no. 2, 398–421. Zbl 1316.58029.
[BG15] F. Baudoin and N. Garofalo, Curvature-dimension inequalities and Ricci lower bounds for
sub-Riemannian manifolds with transverse symmetries, J. European Math. Soc. (JEMS), to
appear. arXiv:1101.3590.
[BG16] F. Baudoin and N. Garofalo, Curvature dimension inequalities in Riemannian and sub-
Riemannian geometry, book in preparation.
[BG14] F. Baudoin, M. Bonnefont, and N. Garofalo, A sub-Riemannian curvature-dimension in-
equality, volume doubling property and the Poincaré inequality, Math. Ann. 358 (2014), no.
3-4, 833–860. Zbl 1287.53025.
250 Nicola Garofalo
[BBGM14] F. Baudoin, M. Bonnefont, N. Garofalo, and I.H. Munive, Volume and distance com-
parison theorems for sub-Riemannian manifolds, J. Funct. Anal. 267 (2014), no. 7, 2005–
2027. Zbl 1303.53042.
[Be96] A. Bellaïche, The tangent space in sub-Riemannian geometry, in Sub-Riemannian geome-
try, 1–78, Progr. Math., 144, Birkhäuser, Basel, 1996. Zbl 0862.53031.
[BLU07] A. Bonfiglioli, E. Lanconelli, and F. Uguzzoni, Stratified Lie groups and potential the-
ory for their sub-Laplacians, Springer Monographs in Mathematics. Springer, Berlin, 2007.
xxvi+800 pp. Zbl 1128.43001.
[BL12] A. Bonfiglioli and E. Lanconelli, Lie groups related to Hörmander operators and
Kolmogorov-Fokker-Planck equations, Comm. Pure Appl. Anal. 11 (2012), no. 5, 1587–
1614. Zbl 1266.35096.
[BU04] A. Bonfiglioli and F. Uguzzoni, Nonlinear Liouville theorems for some critical problems
on H-type groups, J. Funct. Anal. 207 (2004), no. 1, 161–215. Zbl 1045.35018.
[Bu70] H. Busemann, Recent synthetic differential geometry, Ergebnisse der Mathematik und ihrer
Grenzgebiete, Band 54. Springer, New York-Berlin, 1970. vii+110 pp. Zbl 0194.53701.
[C37] R. Caccioppoli, Sui teoremi di esistenza di Riemann, Ann. Sc. N. Sup. Pisa 7 (1938), 177–
187. Zbl 0018.40401.
[Ca76] A.P. Calderón, Inequalities for the maximal function relative to a metric, Studia Math. 57
(1976), no. 3, 297–306. Zbl 0341.44007.
[CY92] H.D. Cao and S.T. Yau, Gradient estimates, Harnack inequalities and estimates for heat
kernels of the sum of squares of vector fields, Mathematische Zeitschrift, 211 (1992), 485–
504. Zbl 0808.58037.
[CDG94] L. Capogna, D. Danielli, and N. Garofalo, The geometric Sobolev embedding for vec-
tor fields and the isoperimetric inequality, Comm. Anal. Geom. 2 (1994), no. 2, 203–215.
Zbl 0864.46018.
[CDG96] L. Capogna, D. Danielli, and N. Garofalo, Capacitary estimates and the local behavior
of solutions of nonlinear subelliptic equations, Amer. J. Math. 118 (1996), no. 6, 1153–1196.
Zbl 0878.35020.
[CDG97] L. Capogna, D. Danielli, and N. Garofalo, Subelliptic mollifiers and a basic pointwise
estimate of Poincaré type, Math. Z. 226 (1997), no. 1, 147–154. Zbl 0893.35023.
[CDPT07] L. Capogna, D. Danielli, S.D. Pauls, and J. Tyson, An introduction to the Heisen-
berg group and the sub-Riemannian isoperimetric problem, Progress in Mathematics, 259.
Birkhäuser, Basel, 2007. xvi+223 pp. Zbl 1138.53003.
[CKS87] E. Carlen, S. Kusuoka, and D. Stroock, Upper bounds for symmetric Markov transi-
tion functions, Ann. Inst. H. Poincaré Probab. Statist. 23 (1987), no. 2, suppl., 245–287.
Zbl 0634.60066.
[Ch93] I. Chavel, Riemannian geometry: A modern introduction, Cambridge tracts in mathematics,
Vol. 108, Cambrige Univ. Press, 1993. Zbl 0810.53001.
[Che99] J. Cheeger, Differentiability of Lipschitz functions on metric measure spaces, Geom.
Funct. Anal. 9 (1999), 428–517. Zbl 0942.58018.
Free ebooks ==> www.Ebook777.com
Bibliography 251
[C39] W.L. Chow, Über Systeme von linearen partiellen Differentialgleichungen erster Ordnung,
Math. Ann., 117 (1939), 98–105. Zbl 0022.02304.
[CLN06] B. Chow, P. Lu, and L. Ni, Hamilton’s Ricci flow, Amer. Math. Soc., Grad. Studies in
Math., vol. 77, 2006. Zbl 1118.53001.
[Ci38/1] G. Cimmino, Sulle equazioni lineari alle derivate parziali del secondo ordine di tipo
ellittico sopra una superficie chiusa, Ann. Sc. N. Sup. Pisa 7 (1938), 73–96. Zbl 0018.02501.
[Ci38/2] G. Cimmino, Nuovo tipo di condizione al contorno e nuovo metodo di trattazione per
il problema generalizzato di Dirichlet, Rend. Circ. Mat. Palermo 61 (1938), 177–221.
Zbl 0019.26301.
[CGL93] G. Citti, N. Garofalo, and E. Lanconelli, Harnack’s inequality for sum of squares of
vector fields plus a potential, Amer. J. Math., 3, 115 (1993), 699–734. Zbl 0795.35018.
[CW71] R.R. Coifman and G. Weiss, Analyse harmonique non-commutative sur certains espaces
homogènes. (French) Étude de certaines intégrales singulières, Lecture Notes in Mathemat-
ics, Vol. 242. Springer, Berlin-New York, 1971. v+160 pp. Zbl 0224.43006.
[CM97] T.H. Colding and W.P. Minicozzi, II, Harmonic functions on manifolds, Ann. of Math. (2)
146 (1997), no. 3, 725–747. Zbl 0928.53030.
[CGr90] L. Corwin and F.P. Greenleaf, Representations of nilpotent Lie groups and their applica-
tions, Part I: basic theory and examples, Cambridge Studies in Advanced Mathematics 18,
Cambridge University Press, Cambridge (1990). Zbl 0704.22007.
[CDKR91] M. Cowling, A.H. Dooley, A. Korányi, and F. Ricci, H-type groups and Iwasawa
decompositions, Adv. in Math., 87 (1991), 1–41. Zbl 0761.22010.
[Cy81] J. Cygan, Subadditivity of homogeneous norms on certain nilpotent Lie groups, Proc. Amer.
Math. Soc. 83 (1981), no. 1, 69–70. Zbl 0475.43010.
[DGN98] D. Danielli, N. Garofalo, and D.M. Nhieu, Trace inequalities for Carnot-Carathéodory
spaces and applications, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 27 (1998), no. 2, 195–252.
Zbl 0938.46036.
[D89] E.B. Davies, Heat kernels and spectral theory, Cambridge Tracts in Mathematics, 92. Cam-
bridge University Press, Cambridge, 1989. Zbl 0699.35006.
[DG57] E. De Giorgi, Sulla differenziabilità e l’analiticità delle estremali degli integrali multi-
pli regolari (Italian), Mem. Accad. Sci. Torino Cl. Sci. Fis. Mat. Nat. (3) 3 (1957), 25–43.
Zbl 0084.31901.
[DG54] E. De Giorgi, Su una teoria generale della misura (r − 1)-dimensionale in uno spazio ad
r dimensioni, Ann. Mat. Pura Appl. (4) 36 (1954), 191–213. Zbl 0055.28504.
[DG61] E. De Giorgi, Frontiere orientate di misura minima, Seminario di Matematica della Scuola
Normale Superiore di Pisa, 1960-61 Editrice Tecnico Scientifica, Pisa, 1961.
[DT06] S. Dragomir and G. Tomassini, Differential geometry and analysis on CR manifolds,
Birkhäuser, Vol. 246, 2006. Zbl 1099.32008.
[FaS86] E.B. Fabes and D. Stroock, A new proof of Moser’s parabolic Harnack inequality using the
old ideas of Nash, Arch. Rational Mech. Anal. 96 (1986), no. 4, 327–338. Zbl 0652.35052.
www.Ebook777.com
252 Nicola Garofalo
[GM14] N. Garofalo and A. Mondino, Li-Yau and Harnack type inequalities in RCD∗ (K, N ) metric
measure spaces, Nonlinear Anal. 95 (2014), 721–734. Zbl 1286.58016.
[GN96] N. Garofalo and D.-M. Nhieu, Isoperimetric and Sobolev inequalities for Carnot-
Carathéodory spaces and the existence of minimal surfaces, Comm. Pure Appl. Math. 49
(1996), no. 10, 1081–1144. Zbl 0880.35032.
[GN98] N. Garofalo and D.-M. Nhieu, Lipschitz continuity, global smooth approximations and
extension theorems for Sobolev functions in Carnot-Carathéodory spaces, J. Anal. Math. 74
(1998), 67–97. Zbl 0906.46026.
[GV01] N. Garofalo and D. Vassilev, Symmetry properties of positive entire solutions of Yamabe-
type equations on groups of Heisenberg type, Duke Math. J. 106 (2001), no. 3, 411–448.
Zbl 1012.35014.
[GW79] R.E. Greene and H. Wu, Function theory on manifolds which possess a pole, Lecture
Notes in Math., vol. 699, Springer, 1979. Zbl 0414.53043.
[Gri91] A.A. Grigor’yan, The heat equation on noncompact Riemannian manifolds, (Russian)
Mat. Sb. 182 (1991), no. 1, 55–87; translation in Math. USSR-Sb. 72 (1992), no. 1, 47–77.
Zbl 0776.58035.
[Gri09] A. Grigor’yan, Heat kernels and function theory on metric measure spaces, Amer. Math.
Soc., Studies in Advanced Math., S.-T. Yau, Series Ed., 2009. Zbl 1048.58021.
[G96] M. Gromov, Carnot-Carathéodory spaces from within, in Sub-Riemannian geometry, 1–78,
Progr. Math., 144, Birkhäuser, Basel, 1996. Zbl 0864.53025.
[GSC10] P. Gyrya and L. Saloff-Coste, Neumann and Dirichlet heat kernels in inner uniform do-
mains, (English, French summary) Astérisque No. 336 (2011), viii+144 pp. Zbl 1222.58001.
[Ha] P. Hajlasz, Sobolev spaces on metric-measure spaces. Heat kernels and analysis on mani-
folds, graphs, and metric spaces, (Paris, 2002), 173–218, Contemp. Math., 338, Amer. Math.
Soc., Providence, RI, 2003. Zbl 1048.46033.
[HK] P. Hajlasz and P. Koskela, Sobolev met Poincaré, Mem. Amer. Math. Soc. 145 (2000), no.
688, x+101 pp. Zbl 0954.46022.
[Hei] J. Heinonen, Lectures on analysis on metric spaces, Universitext. Springer, New York, 2001.
x+140 pp . Zbl 0985.46008.
[HH97] J. Heinonen and I. Holopainen, Quasiregular maps on Carnot groups, J. Geom. Anal. 7
(1997), no. 1, 109–148. Zbl 0905.30018.
[HeK98] J. Heinonen and P. Koskela, Quasiconformal maps in metric spaces with controlled ge-
ometry, Acta Math. 181 (1998), 1–61. Zbl 0915.30018.
[Hel] B. Helffer, Conditions nécessaires d’hypoanalyticité pour des opérateurs invariants à gauche
homogènes sur un groupe nilpotent gradué, (French) [Necessary conditions of hypoanalyt-
icity for homogeneous left-invariant operators on a graded nilpotent group], J. Differential
Equations 44 (1982), no. 3, 460–481. Zbl 0458.35019.
[Ho67] L. Hörmander, Hypoelliptic second-order differential equations, Acta Math., 119 (1967),
147–171. Zbl 0156.10701.
254 Nicola Garofalo
[S93] E.M. Stein, Harmonic analysis: Real variable methods, orthogonality and oscillatory inte-
grals, Princeton Univ. Press, (1993). Zbl 0821.42001.
[SW71] E.M. Stein and G. Weiss, Introduction to Fourier analysis on Euclidean spaces, Princeton
Univ. Press, (1971). Zbl 0232.42007.
[St83] R. Strichartz, Analysis of the Laplacian on the complete Riemannian manifold, Journal
Func. Anal., 52, 1, 48–79 (1983). Zbl 0515.58037.
[St86] R. Strichartz, Sub-Riemannian geometry, Journ. Diff. Geom., 24 (1986), 221–263.
Zbl 0609.53021.
[St89] R. Strichartz, corrections to “Sub-Riemannian geometry” [Journ. Diff. Geom., 24 (1986),
221-263], 30 (2) (1989), 595–596. Zbl 0609.53021.
[Stu94] K.T. Sturm, Analysis on local Dirichlet spaces. I. Recurrence, conservativeness and L p -
Liouville properties, J. reine angew. Math., 456 (1994), 173–196. Zbl 0806.53041.
[Stu95] K.T. Sturm, Analysis on local Dirichlet spaces. II. Upper Gaussian estimates for the
fundamental solutions of parabolic equations, Osaka J. Math., 32 (1995), no. 2, 275–312.
Zbl 0854.35015.
[Stu96] K.T. Sturm, Analysis on local Dirichlet spaces. III. The parabolic Harnack inequality, J.
Math. Pures Appl. (9) 75 (1996), no. 3, 273–297. Zbl 0854.35016.
[Stu06] K.T. Sturm, On the geometry of metric measure spaces. I, Acta Math. 196 (2006), no. 1,
65–131. Zbl 1105.53035.
[Stu06] K.T. Sturm, On the geometry of metric measure spaces. II, Acta Math. 196 (2006), no. 1,
133–177. Zbl 1106.53032.
[T76] G. Talenti, Best constant in Sobolev inequality, Ann. Mat. Pura Appl. (4) 110 (1976), 353–
372. Zbl 0353.46018.
[V87] N.Th. Varopoulos, Fonctions harmoniques sur les groupes de Lie, (French) [Harmonic
functions on Lie groups] C.R. Acad. Sci. Paris Sér. I Math. 304 (1987), no. 17, 519–521.
Zbl 0614.22002.
[VSC] N. Varopoulos, L. Saloff-Coste, and T. Coulhon, Analysis and geometry on groups, Cam-
bridge Tracts in Mathematics, 100. Cambridge University Press, Cambridge, 1992. xii+156
pp. ISBN: 0-521-35382-3. Zbl 1179.22009.
[Va06] D. Vassilev, Existence of solutions and regularity near the characteristic boundary for
sub-Laplacian equations on Carnot groups, Pacific J. Math. 227 (2006), no. 2, 361–397.
Zbl 1171.35388.
[W40] H. Weyl, The method of orthogonal projection in potential theory, Duke Math. J. 7 (1940),
411–444. Zbl 0026.02001.
[Y75] S.T. Yau, Harmonic functions on complete Riemannian manifolds, Comm. Pure Appl. Math.
28 (1975), 201–228. Zbl 0291.31002.
[Y76] S.T. Yau, Some function-theoretic properties of complete Riemannian manifold and their
applications to geometry, Indiana Univ. Math. J. 25 (1976), no. 7, 659–670. Zbl 0335.53041.
Bibliography 257
[Y78] S.T. Yau, On the heat kernel of a complete Riemannian manifold, J. Math. Pures Appl., ser.
9, 57 (1978), 191–201. Zbl 0405.35025.
[Z89] W.P. Ziemer, Weakly differentiable functions, Springer, 1989. Zbl 0692.46022.
Chapter 3
Sub-Laplacians and hypoelliptic operators on
totally geodesic Riemannian foliations
Fabrice Baudoin1
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
2 Riemannian foliations and their Laplacians . . . . . . . . . . . . . . . . . . . . 261
3 Horizontal Laplacians and heat kernels on model spaces . . . . . . . . . . . . . 268
4 Transverse Weitzenböck formulas . . . . . . . . . . . . . . . . . . . . . . . . . 282
5 The horizontal heat semigroup . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
6 The horizontal Bonnet–Myers theorem . . . . . . . . . . . . . . . . . . . . . . 302
7 Riemannian foliations and hypocoercivity . . . . . . . . . . . . . . . . . . . . 308
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
1 Introduction
It is a fact that many interesting hypoelliptic diffusion operators may be studied by in-
troducing a well-chosen Riemannian foliation. In particular, several sub-Laplacians
on sub-Riemannians manifolds often appear as horizontal Laplacians of a foliation
and several of the Kolmogorov-type hypoelliptic diffusion operators which are used
in the theory of kinetic equations appear as the sum of the vertical Laplacian of a
foliation and of a first-order term.
The goal of the present notes is to survey some geometric analysis tools to study
this kind of diffusion operator. We specially would like to stress the importance
of subelliptic Bochner-type identities in this framework and show how they can be
used to deduce a variety of results ranging from topological information on a sub-
Riemannian manifold to hypocoercive estimates and convergence to equilibrium for
1fbaudoin@purdue.edu
Department of Mathematics, Purdue University, 150 N. University Street, West Lafayette, IN 47907-2067,
United States.
260 Fabrice Baudoin
2.1 Riemannian submersions. Let (M, g) and (B, j) be smooth and connected
Riemannian manifolds.
Definition 2.1. A smooth surjective map π : (M, g) → (B, j) is called a Riemannian
submersion if its derivative maps Tx π : Tx M → Tπ(x) B are orthogonal projections,
i.e., for every x ∈ M, the map Tx π(Tx π) ∗ : Tp(x) B → Tp(x) B is the identity.
Example 2.2 (Warped products). Let (M1 , g1 ) and (M2 , g2 ) be Riemannian man-
ifolds and f be a smooth and positive function on M1 . Then the first projection
(M1 × M2 , g1 ⊕ f g2 ) → (M1 , g1 ) is a Riemannian submersion.
Example 2.3 (Quotient by an isometric action). Let (M, g) be a Riemannian man-
ifold and G be a closed subgroup of the isometry group of (M, g). Assume that
the projection map π from M to the quotient space M/G is a smooth submersion.
Then there exists a unique Riemannian metric j on M/G such that π is a Riemannian
submersion.
If π is a Riemannian submersion and b ∈ B, the set π −1 ({b}) is called a fiber.
For x ∈ M, V x = Ker(Tx π) is called the vertical space at x. The orthogonal
complement of H x shall be denoted H x and will be referred to as the horizontal
space at x. We have an orthogonal decomposition
Tx M = H x ⊕ V x
g = gH ⊕ gV .
all the situations we will consider the horizontal distribution is everywhere bracket
generating in the sense that for every x ∈ M, Lie(H)(x) = Tx M. In that case it is
natural to study the sub-Riemannian geometry of the triple (M, H, gH ). As we will
see, many interesting examples of sub-Riemannian structures arise in this framework
and this is really the situation which is interesting for us.
We shall mainly be interested in submersion with totally geodesic fibers.
Example 2.6 (Principal bundle). Let M be a principal bundle over B with fiber F and
structure group G. Then, given a Riemannian metric j on B, a G-invariant metric k
on F and a G connection form ϑ, there exists a unique Riemannian metric g on M
such that the bundle projection map π : M → B is a Riemannian submersion with
totally geodesic fibers isometric to (F, k) and such that the horizontal distribution of
ϑ is the orthogonal complement of the vertical distribution. We refer to [51, page
78] for a proof. In the case of the tangent bundle of a Riemannian manifold, the
construction yields the Sasaki metric on the tangent bundle.
As we will see, for a Riemannian submersion with totally geodesic fibers, all the
fibers are isometric. The argument, due to Hermann [38] relies on the notion of a
basic vector field that we now introduce.
Let π : (M, g) → (B, j) be a Riemannian submersion. A vector field X ∈
Γ∞ (TM) is said to be projectable if there exists a smooth vector field X on B such
that for every x ∈ M, Tx π(X (x)) = X (π(x)). In that case, we say that X and X are
π-related.
If X is a smooth vector field on B, then there exists a unique basic vector field X
on M which is π-related to X. This vector is called the lift of X.
Notice that if X is a basic vector field and Z is a vertical vector field, then
Tx π([X, Z](x)) = 0 and thus [X, Z] is a vertical vector field. The following result is
due to Hermann [38].
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 263
Proposition 2.8. The submersion π has totally geodesic fibers if and only if the flow
generated by any basic vector field induces an isometry between the fibers.
(L X g)(Z1 , Z2 ) = −hX, D Z 1 Z2 + D Z 2 Z1 i
= −2hX, D Z 1 Z2 i.
Thus the flow generated by any basic vector field induces an isometry between the
fibers if and only if D Z 1 Z2 is always vertical, which is equivalent to the fact that the
fibers are totally geodesic submanifolds.
and
m
X
∆V = − Zi∗ Zi ,
i=1
264 Fabrice Baudoin
where (·)H denotes the horizontal part of the vector. In a similar way we obviously
have
Xm Xn m
X
∆V = Zi2 − (D X i X i )V − (D Z i Zi )V .
i=1 i=1 i=1
We can observe that the Laplace–Beltrami operator ∆ of M can be written
∆ = ∆H + ∆V .
It is worth noting that, in general, ∆H is not the lift of the Laplace–Beltrami operator
∆B on B. Indeed, let us denote by X 1 , . . . , X n the vector fields on B which are
π-related to X1 , . . . , X n . We have
n n
2
X X
∆B = Xi − DX i X i .
i=1 i=1
As a consequence,
This condition is equivalent to the fact that the flow generated by X induces an isom-
etry between the fibers, and so from Hermann’s Proposition 2.8 this is equivalent to
the fact that the fibers are totally geodesic.
Theorem 2.10. The Riemannian submersion π has totally geodesic fibers if and only
if any basic vector field X commutes with the vertical Laplacian ∆V . In particular,
if π has totally geodesic fibers, then for every f ∈ C ∞ (M),
∆H ∆V f = ∆V ∆H f .
Proof. Assume that the submersion is totally geodesic. Let X be a basic vector field
and ξ t be the flow it generates. Since ξ induces an isometry between the fibers, we
have
ξ t∗ (∆V ) = ∆V .
Differentiating at t = 0 yields [X, ∆V ] = 0.
Conversely, assume that for every basic field X, [X, ∆V ] = 0. Let X1 , . . . , X n be a
local orthonormal frame of basic vector fields and Z1 , . . . , Z m be a local orthonormal
frame of the vertical distribution. The second-order part of the operator [X, ∆V ] must
be zero. Given the expression of ∆V , this implies
m
X
[X, Zi ]Zi = 0.
i=1
266 Fabrice Baudoin
So X leaves the symbol of ∆V invariant which is the metric on the vertical distri-
bution. This implies that the flow generated by X induces isometries between the
fibers.
Finally, as we have seen, if the submersion is totally geodesic then in a local basic
orthonormal frame
Xn n
X
∆H = X i2 − (D X i X i )H .
i=1 i=1
Since the vectors (D X i X i )H are basic, from the previous result ∆H commutes with
∆V .
The triple (M, ϑ, g) is called a contact Riemannian manifold. We see then that the
Reeb foliation is totally geodesic with bundle-like metric if and only if the Reeb
vector field T is a Killing field, i.e.,
LT g = 0.
In that case (M, ϑ, g) is called a K-contact Riemannian manifold.
Example 2.14 (Sub-Riemannian manifolds with transverse symmetries). The con-
cept of sub-Riemannian manifold with transverse symmetries was introduced in [14].
Let M be a smooth, connected manifold with dimension n + m. We assume that M
is equipped with a bracket-generating distribution H of dimension n and a fiberwise
inner product gH on that distribution. It is said that M is a sub-Riemannian mani-
fold with transverse symmetries if there exists an m- dimensional Lie algebra V of
sub-Riemannian Killing vector fields such that for every x ∈ M,
Tx M = H (x) ⊕ V (x),
where
V (x) = {Z (x), Z ∈ V (x)}.
The choice of an inner product gV on the Lie algebra V naturally endows M with a
Riemannian metric that makes the decomposition Tx M = H (x) ⊕ V (x) orthogonal:
g = gH ⊕ gV .
(x 1 , y1 , z1 ) ? (x 2 , y2 , z2 ) = (x 1 + x 2 , y1 + y2 , z1 + z2 + hx 1 , y2 iRn − hx 2 , y1 iRn ).
[X i ,Yj ] = 2δ i j Z, [X i , Z] = [Yi , Z] = 0.
The map
H2n+1 → R2n ,
π:
(x, y, z) → (x, y)
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 269
is then a Riemannian submersion with totally geodesic fibers. The horizontal Lapla-
cian is the left invariant operator
n
X
∆H = (X i2 + Yi2 )
i=1
n n
∂2 ∂2 ∂ ∂ ∂ 2 ∂
2
X X !
= + + 2 x i − y i + (k xk 2
+ k yk )
i=1
∂x 2i ∂yi2 i=1
∂yi ∂x i ∂z ∂z 2
∂2
∆V = .
∂z 2
The horizontal distribution
∆H ( f ◦ %) = (∆H f ) ◦ %.
From this invariance property in order to study the heat kernel and fundamental
solution of ∆H at 0 it suffices to study the heat kernel and the fundamental solution
of ∆H at 0.
We denote by pt (r, z) the heat kernel at 0 of ∆H . It was first computed explicitly
by Gaveau [33], building on previous works by Paul Lévy.
Proposition 3.1. For r ≥ 0 and z ∈ R,
!n
λ
Z
1 iλz
e−(λr /2) coth(2λt ) dλ.
2
pt (r, z) = e
(2π) n+1 R sinh(2λt)
∂
Proof. Since ∂z commutes with ∆H , the idea is to use a Fourier transform in z. We
see then that Z
1
pt (r, z) = ei λ z Φt (r, λ)dλ,
2π R
270 Fabrice Baudoin
∂2 2n − 1 ∂
Lλ = + − λ 2r 2 .
∂r 2 r ∂r
L λ e λr /2 f = e λr /2 (2nλ + G λ ) f ,
2 2
where
∂2 2n − 1 ∂
!
G λ = 2 + 2λr + .
∂r r ∂r
The operator G λ turns out to be the radial part of the Ornstein–Uhlenbeck operator
∆R2n + 2λhx, ∇R2n i whose heat kernel at 0 is a Gaussian density with mean 0 and
1
variance 2λ (e4λt − 1). This means that the heat kernel at 0 of G λ is given by
!n
1 2λ 2 /(e 4λ t −1)
qt (r) = e−λr .
(2π) n e4λt − 1
We conclude that
!n
e2nλt 2λ 2 /2 2 /(e 4λ t −1)
Φt (r, z, λ) = e−λr e−λr .
(2π) n e4λt − 1
Though it does not seem very explicit, this representation of the heat kernel has
many applications and can be used to get very sharp estimates and small-time asymp-
totics (see [21] and [41, 42]).
Free ebooks ==> www.Ebook777.com
3.2 The Hopf fibration. The second simplest and geometrically relevant ex-
ample is given by the celebrated Hopf fibration. The horizontal heat kernel was first
computed in [19]; we follow that, but simplify it, since the CR structure of the sphere
is not relevant for us.
Let us consider the odd-dimensional unit sphere
The generator of this action shall be denoted by T. We thus have for every f ∈
C ∞ (S2n+1 ),
d
T f (z) = f (eiϑ z) |ϑ=0 ,
dϑ
so that
n+1
∂ ∂
X !
T =i zj − zj .
j=1
∂z j ∂z j
The quotient space S2n+1 /U(1) is the projective complex space CPn and the projec-
tion map π : S2n+1 → CPn is a Riemannian submersion with totally geodesic fibers
isometric to U(1). The fibration
∂
coordinates, it is clear that T = ∂ϑ and that the vertical Laplacian is
∂2
∆V = .
∂ϑ2
2We will call north pole the point with complex coordinates z 1 = 0, . . . , z n+1 = 1.
www.Ebook777.com
272 Fabrice Baudoin
Our goal is now to compute the horizontal Laplacian ∆H . This operator is invariant
by the action on the variables (w1 , . . . , wn ) of the group of isometries of CPn that
fix the north pole of S2n+1 (this group is SU(n)). Therefore the heat kernel at the
north pole depends only on the variables (r, ϑ) and can be computed through the
heat kernel of the radial part ∆H of ∆H .
Proposition 3.3. Consider the submersion
∂
∆H = ∆ − ∆V = ∆ − .
∂ϑ2
In our parametrization of S2n+1 we have
Therefore if δ1 denotes the Riemannian distance based at the north pole, we have
cos δ1 = cos r cos ϑ and if δ2 denotes the Riemannian distance based at the point
with real coordinates (0, . . . , 0, 1) then we have cos δ2 = cos r sin ϑ. The formula
for the Laplace–Beltrami operator acting on functions depending on the Riemannian
distance based at a point is well known and we deduce from it that ∆ acts on functions
depending only on δ1 , δ2 as
∂2 ∂ ∂2 ∂
+ 2n cot δ1 + 2 + 2n cot δ2 .
∂δ1
2 ∂δ1 ∂δ2 ∂δ2
∂2 ∂ 1 ∂2
+ ((2n − 1) cot r − tan r) + .
∂r 2 ∂r cos2 r ∂ϑ2
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 273
∂2 ∂ 1 ∂2 ∂
∆H = + ((2n − 1) cot r − tan r) + − .
∂r 2 ∂r cos r ∂ϑ
2 2 ∂ϑ2
We can observe that ∆H is symmetric with respect to the measure
2π n
dµ = (sin r) 2n−1 cos r dr dϑ,
Γ(n)
where the normalization is chosen in such a way that
Z π Z π/2
2π n+1
d µ = µ(S2n+1 ) = .
−π 0 Γ(n + 1)
As mentioned above, the heat kernel at the north pole of ∆H depends only on
(r, ϑ), that is, p weiϑ cos r, eiϑ cos r = pt (r, ϑ), where pt is the heat kernel at 0
of ∆H .
Proposition 3.4. For t > 0, r ∈ [0, π/2), ϑ ∈ [−π, π],
Γ(n)
pt (r, ϑ) =
2π n+1
+∞ X+∞
m + |k | + n − 1 −λ m, k t+ikϑ
X !
n−1, |k |
× (2m + |k | + n) e (cos r) |k | Pm (cos 2r),
k=−∞ m=0
n − 1
n−1, |k | (−1) m dm
Pm (x) = ((1 − x) n−1+m (1 + x) |k |+m )
2m m!(1 − x) n−1 (1 + x) |k | dx m
is a Jacobi polynomial.
Proof. Similarly to the Heisenberg group case, we observe that ∆H commutes with
∂
∂ϑ , so the idea is to expand pt (r, ϑ) as a Fourier series in ϑ. We can write
+∞
1 X ikϑ
pt (r, ϑ) = e φk (t,r),
2π k=−∞
∂φk ∂ 2 φk ∂φk
= + ((2n − 1) cot r − tan r) − k 2 tan2 r φk .
∂t ∂r 2 ∂r
274 Fabrice Baudoin
Note that as a by-product of the previous result we obtain that the L 2 spectrum
of −∆H is given by
Sp(−∆H ) = {4m(m + k + n) + 2kn, k ∈ N, m ∈ N} . (3.2)
We can give another representation of the heat kernel pt (r, ϑ) which is easier to
∂
handle analytically. The key idea is to observe that since ∆ and ∂ϑ commute, we
formally have
2 /∂ϑ 2 )
e t∆H = e−t (∂ e t∆ . (3.3)
This gives a way to express the horizontal heat kernel in terms of the Riemannian
one. Let us recall that the Riemannian heat kernel on the sphere S2n+1 is given by
+∞
Γ(n) X
qt (cos δ) = (m + n)e−m(m+2n)t Cm
n
(cos δ), (3.4)
2π n+1 m=0
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 275
where V (t, δ) = √ 1
P −(δ−2k π) 2 /4t is a theta function.
4πt k ∈Z e
Using the commutation (3.3) and the formula cos δ = cos r cos ϑ, we then infer
the following proposition which is easy to prove (see [19] for the details).
Proposition 3.5. For t > 0, r ∈ [0, π/2), ϑ ∈ [−π, π],
Z +∞
1
e−(y+iϑ) /4t qt (cos r cosh y)dy.
2
pt (r, ϑ) = √ (3.6)
4πt −∞
Applications of this formula are given in [19]. We can, in particular, deduce
from it small asymptotics of the kernel when t → 0. Interestingly, these small-time
asymptotics allow the sub-Riemannian distance to be computed explicitly. For a
study of the distance and related geodesics, we refer to [26] and [44].
There is an isometric group action of the Lie group SU(2) on S4n+3 which is
given by
g · (q1 , . . . , qn+1 ) = (gq1 , . . . , gqn+1 ).
The three generators of this action are given by
n+1
∂f ∂f ∂f ∂f
!
d Iϑ
X
f (e q) |ϑ=0 = −x i + ti − zi + yi ,
dϑ i=1
∂t i ∂x i ∂yi ∂z i
n+1
∂f ∂f ∂f ∂f
!
d X
f (e Jϑ q) |ϑ=0 = −yi + zi + ti − xi ,
dϑ i=1
∂t i ∂x i ∂yi ∂z i
and
n+1
∂f ∂f ∂f ∂f
!
d Kϑ
X
f (e q) |ϑ=0 = −z i − yi + xi + ti .
dϑ i=1
∂t i ∂x i ∂yi ∂z i
The quotient space S4n+3 /SU(2) is the projective quaternionic space HPn and the
projection map π : S4n+3 → HPn is a Riemannian submersion with totally geodesic
fibers isometric to SU(2). The fibration
Proposition 3.6. Let us denote by % the submersion from S4n+3 to [0, π/2) × [0, π)
such that
% (cos r)e I ϑ1 +Jϑ2 +K ϑ3 w, (cos r)e I ϑ1 +Jϑ2 +K ϑ3 = (r, η) ,
qP q
where r = arctan n
j=1 |w j | 2 and η = ϑ21 + ϑ22 + ϑ23 . Then for every smooth
function f : [0, π/2) × [0, π) → R,
∆H ( f ◦ %) = (∆H f ) ◦ %, ∆V ( f ◦ %) = (∆V f ) ◦ %,
where
∂2 ∂ ∂2 ∂
!
∆H = 2 + ((4n − 1) cot r − 3 tan r) + tan2 r + 2 cot η
∂r ∂r ∂η 2 ∂η
and
∂2 ∂
∆V = + 2 cot η .
∂η 2 ∂η
Proof. The formula for ∆V is clear because SU(2) is isometric to the sphere S3 . For
the horizontal Laplacian, the proof follows the same lines as the case of the classical
Hopf fibration. Let δ1 be the distance based at the point (0, 1) ∈ Hn × H, δ2 be the
distance based at the point (0, I) ∈ Hn × H, δ3 be the distance based at the point
(0, J) ∈ Hn × H, and δ4 be the distance based at the point (0, K ) ∈ Hn × H.
The Laplace–Beltrami operator ∆ acts on functions depending only on δ1 , δ2 , δ3 ,
δ4 as
4
* ∂ + (4n + 2) cot δ i ∂ + .
X 2
i=1 ,
∂δ2i ∂δ i -
Observing now that
where
k + m + 2n
!
Γ(2n)
α k, m = 2n+2 (2k + m + 2n + 1)(m + 1) ,
2π 2n − 1
λ k, m = 4 [k (k + 2n + m + 1) + nm] ,
and
(−1) k dk
Pk2n−1, m+1 (x) = (1 − x) 2n−1+k (1 + x) m+1+k
2k k!(1 − x) 2n−1 (1 + x) m+1 dx k
is a Jacobi polynomial.
Proof. The idea is to expand the subelliptic kernel in spherical harmonics as follows:
+∞
X sin(m + 1)η
pt (r, η) = φ m (t,r)
m=0
sin η
sin(m+1)η ∂2 ∂
where sin η is the eigenfunction of ∆˜ SU (2) = ∂η 2
+ 2 cot η ∂η which is associated
∂p t
to the eigenvalue −m(m + 2). To determine φ m , we use ∂t = L̃pt and find that
∂φ m ∂ 2 φm ∂φ m
= + ((4n − 1) cot r − 3 tan r) − m(m + 2) tan2 r φ m .
∂t ∂r 2 ∂r
Let φ m (t,r) = e−4nmt (cos r) m ϕ m (t,r); then ϕ m (t,r) satisfies the equation
∂ϕ m ∂ 2 ϕm ∂ϕ m
= + [(4n − 1) cot r − (2m + 3) tan r] .
∂t ∂r 2 ∂r
We now change the variable and denote by ϕ m (t,r) = gm (t, cos 2r); then we have
that gm (t, x) satisfies the equation
∂gm ∂ 2 gm ∂gm
= 4(1 − x 2 ) + 4[(m + 2 − 2n) − (2n + m + 2)x] .
∂t ∂x 2 ∂x
∂ 2 ∂
We define Ψm = (1 − x 2 ) ∂x 2 + [(m + 2 − 2n) − (2n + m + 2)x] ∂x , and find that
∂gm
= 4Ψm (gm ).
∂t
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 279
The equation
Ψm (gm ) + k (k + 2n + m + 1)gm = 0
(−1) k dk
Pk2n−1, m+1 (x) = (1 − x) 2n−1+k
(1 + x) m+1+k
.
2k k!(1 − x) 2n−1 (1 + x) m+1 dx k
+∞ X
∞
X sin(m + 1)η
pt (r, η) = α k, m e−4[k (k+2n+m+1)+nm]t (cos r) m Pk2n−1, m+1 (cos 2r)
m=0 k=0
sin η
+∞ X
+∞
X sin(m + 1)η 2n−1, m+1
f (r, η) = bk, m Pk (cos 2r) · (cos r) m
m=0 k=0
sin η
+∞ X
X +∞
f (0, 0) = bk, m (m + 1)Pk2n−1, m+1 (1),
m=0 k=0
2n−1+k
and we observe that Pk2n−1, m+1 (1) = k . The measure dµ is given in cylindri-
cal coordinates by
8π 2n+1
dµr = (sin r) 4n−1 (cos r) 3 (sin η) 2 dr dη.
Γ(2n)
280 Fabrice Baudoin
Moreover, since
Z πZ π
2
pt (r, η) f (−r, −η)dµr
0 0
2n+2 +∞ X
+∞ Z π
4π X 2
= α k, m bk, m e −λ k, m t * (cos r) 2m+3 |Pk2n−1, m+1 | 2 (sin r) 4n−1 dr +
Γ(2n) m=0 k=0 , 0 -
+∞ X+∞
2π 2n+2 X α k, m bk, m e−λ m, k t Γ(k + 2n)Γ(k + m + 2)
= ,
Γ(2n) m=0 k=0
2k + m + 2n + 1 Γ(k + 1)Γ(k + 2n + m + 1)
∂2 ∂ ∂2
If we define ∆SL(2) = ∂η 2
+ 2 coth η ∂η , then from the fact that ∆V = ∂η 2
+
∂
2 cot η ∂η , it is not hard to see that
We are thus left with the computation of e t∆R3 . The operator ∆R3 is the radial part of
the Laplacian ∆R3 ; thus after a routine computation, for x ∈ R3 ,
Z +∞ ηr
1 r
e−(r +η )/(4t ) f (r)dr.
2 2
e t∆R3
f (η) = √ sinh
πt 0 η 2t
As a consequence, we get the integral representation of pt .
Proposition 3.10. For t > 0, r ∈ [0, π/2), η ∈ [0, π],
Z +∞ sinh y sin ηy
e−t 2t 2 2
pt (r, η) = √ e−(y −η )/(4t ) qt (cos r cosh y)dy.
πt 0 sin η
We refer to [20] for applications of this formula to the computation of the small-
time asymptotics of the kernel and, as a by-product, of the sub-Riemannian distance.
282 Fabrice Baudoin
4.1 The Bott connection. Let M be a smooth, connected manifold with dimen-
sion n + m. We assume that M is equipped with a Riemannian foliation F with
bundle-like metric g and totally geodesic m-dimensional leaves.
As usual, the subbundle V formed by vectors tangent to the leaves will be referred
to as the set of vertical directions and the subbundle H which is normal to V will be
referred to as the set of horizontal directions. The metric g can be split as
g = gH ⊕ gV .
horizontal and the vertical bundle may not be parallel. More adapted to the geometry
of the foliation is the Bott connection that we now define. It is an easy exercise to
check that there exists a unique affine connection ∇ such that
• for U,V ∈ Γ∞ (V ), ∇U V ∈ Γ∞ (V );
(D X Y )H , X,Y ∈ Γ∞ (H ),
[X,Y ]H , Γ∞ (V ),Y ∈ Γ∞ (H),
X∈
∇X Y =
[X,Y ]V , X∈ Γ∞ (H),Y ∈ Γ∞ (V ),
(D X Y )V , ∈ Γ∞ (V ),
X,Y
where the subscript H (resp. V ) denotes the projection on H (resp. V ). Observe that
for horizontal vector fields X,Y the torsion T (X,Y ) is given by
T (X,Y ) = −[X,Y ]V .
Example 4.1. Let (M, ϑ, g) be a K-contact Riemannian manifold. The Bott connec-
tion coincides with the Tanno connection that was introduced in [47] and which is
the unique connection that satisfies
1. ∇ϑ = 0;
2. ∇T = 0;
3. ∇g = 0;
We now introduce some tensors and definitions that will play an important role
in the sequel.
For Z ∈ Γ∞ (TM), there is a unique skew-symmetric endomorphism JZ : H x →
H x such that for all horizontal vector fields X and Y ,
where R is the curvature of the connection form. By using the second Bianchi iden-
tity, the Yang–Mills condition is equivalent to the fact that the Ricci tensor of the
connection form is a Codazzi tensor, that is, for any vector fields X,Y, Z in Γ∞ (H ),
kη kε2 = kη kH
2
+ εkη kV2 .
By using the duality given by the metric g, (1, 1) tensors can also be seen as
linear maps on the cotangent bundle T ∗ M. More precisely, if A is a (1, 1) tensor,
we will still denote by A the fiberwise linear map on the cotangent bundle which is
defined as the g-adjoint of the dual map of A. The same convention will be used for
any (r, s) tensor.
We define the horizontal Ricci curvature Ric H as the fiberwise symmetric linear
map on one-forms such that for all smooth functions f , g,
We observe that TVε is skew-symmetric for the metric gε so that ∇−T ε is a gε -metric
connection.
If η is a one-form, we define the horizontal gradient of η in a local frame as the
(0, 2) tensor
Xn
∇H η = ∇ Xi η ⊗ ϑi .
i=1
ε ∗ ε 1 1
ε = −(∇H − TH ) (∇H − TH ) − J2 + δHT − Ric H ,
ε ε
where the adjoint is understood with respect to the metric gε . It is easily seen that,
in a local horizontal frame,
n
ε ∗ ε
(∇ X i − T εX i ) 2 − (∇∇ X i X i − T∇ε X
X
−(∇H − TH ) (∇H − TH )= X i ). (4.2)
i
i=1
ε ∗ ε 1
ε = −(∇H − TH ) (∇H − TH ) − J2 − Ric H .
ε
As a consequence, in the Yang–Mills case the operator ε is seen to be symmetric
for the metric gε .
The following theorem that was proved in [18] is the main result of the section.
Theorem 4.6. For every f ∈ C ∞ (M), we have
d∆H f = ε d f .
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 287
Proof. We only sketch the proof and refer to [18] for the details. If Z1 , . . . , Z m is a
local vertical frame of the leaves, we define
m
X
J(η) = − JZ ` (ι Z ` dη V ),
`=1
where η V is the projection of η to the vertical cotangent bundle. It does not depend
on the choice of the frame and therefore defines a globally defined tensor. Also, let
us consider the map T : Γ∞ (∧2T ∗ M) → Γ∞ (T ∗ M) which is given in a local coframe
ϑ i ∈ Γ∞ (H∗ ), νk ∈ Γ∞ (V ∗ ),
T (ϑ i ∧ ϑ j ) = −γi`j ν` , T (ϑ i ∧ νk ) = T (νk ∧ ν` ) = 0.
ε ∗ ε 2 1 1
−(∇H − TH ) (∇H − TH ) = ∆H + 2J − T ◦ d + δHT ∗ − δHT + J2 .
ε ε ε
Thus, we just need to prove that if ∞ is the operator defined on one-forms by
∞ = ∆H + 2J − Ric H + δHT ∗ ,
The idea is now to multiply this by any g ∈ C0∞ (M) and integrate over M. For that,
observe that
Z Z
ε ∗ ε ε ε
gh(∇H − TH ) (∇H − TH )η, ηiε dµ = h(∇H − TH )η, (∇H − TH )(gη)iε dµ.
M M
We have now
ε ε
(∇H − TH )(gη) = g(∇H − TH )(η) + η ⊗ ∇H g
and
Z Z
ε
h(∇H − TH )η, η ⊗ ∇H giε dµ = h∇H η, η ⊗ ∇H giε dµ
M M
Z
1
= h∇H g, ∇H kη k 2 iε dµ.
2 M
ε ∗ ε ε 1
h(∇H − TH ) (∇H − TH )η, ηiε = k∇H η − TH η kε2 − ∆H kη kε2 .
2
ε 1 ε
k∇H η − TH η kε2 = k∇#H η kε2 − TrH (Jη2 ) + εk∇H η − TH η kV2 ;
4
thus by the Cauchy–Schwarz inequality we have
1 1 2 1
∆H kη kε2 − hε η, ηiε ≥ TrH ∇#H η − TrH (Jη2 ) + hRic H (η), ηiH (4.3)
2 n 4
1
− hδHT (η), ηiV + hJ2 (η), ηiH .
ε
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 289
hRic H (η 1 ), η 1 iH ≥ %1 kη 1 kH
2
, −hJ2 η 1 , η 1 iH ≤ κkη 1 kH
2
,
1
− TrH (Jη22 ) ≥ %2 kη 2 kV2 ,
4
for some %1 ∈ R, κ, %2 > 0. The third assumption can be thought as a uniform
bracket-generating condition of the horizontal distribution H and from Hörmander’s
theorem, it implies that the horizontal Laplacian ∆H is a subelliptic diffusion opera-
tor. We insist that for the following results to be true, the positivity of %2 is required.
We introduce the following operators defined for f , g ∈ C ∞ (M):
1
Γ( f , g) = (∆H ( f g) − g∆H f − f ∆H g) = h∇H f , ∇H giH ,
2
ΓV ( f , g) = h∇V f , ∇V giV ,
hRic H (η 1 ), η 1 iH ≥ %1 kη 1 kH
2
, −hJ2 η 1 , η 1 iH ≤ κkη 1 kH
2
,
1
− TrH (Jη22 ) ≥ %2 kη 2 kV2
4
immediately yields the expected result. The intertwining Γ( f , ΓV ( f )) = ΓV ( f , Γ( f ))
is proved in Theorem 2.9.
4.4 Sharp lower bound for the first eigenvalue of the horizontal Lapla-
cian. In this section, as a second application of the transverse Weitzenböck formula
proved in the previous chapter, we obtain a sharp lower for the first non-zero eigen-
value of the horizontal Laplacian.
Let M be a compact, smooth, connected manifold with dimension n + m. We
assume that M is equipped with a Riemannian foliation F with bundle-like metric g
and totally geodesic m-dimensional leaves. We also assume that M is of Yang–Mills
type.
We prove the following result that was first obtained in [17] in a less general
setting. Let us point out that this bound may not be obtained as a consequence of the
generalized curvature-dimension inequality only.
Tr(Jη∗ Jη ) ≥ %2 kη kV2 ,
with %1 , %2 > 0 and κ ≥ 0. Then the first eigenvalue λ 1 of the horizontal Laplacian
−∆H satisfies
%1
λ1 ≥ .
1− 1
n + 3κ
%2
• Consider now the quaternionic Hopf fibration SU(2) → S4d+3 → HPd . The
sub-Laplacian ∆H is then the lift of the Laplace–Beltrami operator on HPd
and in that case, λ 1 = 4d (see 3.9). For this example, %1 = 4(d + 2), κ = 3,
%2 = 4d. Thus the bound of Theorem 4.9 is still sharp in this example.
We also mention that it has even been proved in [17] that for some Riemannian
foliations the equality λ 1 = %1 /(1 − n1 + 3κ
% 2 ) actually implies that the foliation is
equivalent to the classical or the quaternionic Hopf fibration.
Proof. As for the classical Lichnerowicz estimate on Riemannian manifolds, the
idea is to integrate on the manifold the Bochner–Weitzenböck equality in Theorem
4.7 but some tricks are needed. Let f ∈ C ∞ (M). Let us first observe that
Z Z
− hε d f , d f iε dµ = − hd∆H f , d f iε dµ
M
ZM Z
=− hd∆H f , d f iH dµ − ε hd∆H f , d f iV dµ.
M M
Thus, by integrating the Bochner–Weitzenböck equality in Theorem 4.7, we ob-
tain
Z Z Z
ε
(∆H f ) 2 dµ − ε hd(∆H f ), d f iV dµ ≥ k∇H d f − TH d f kε2 dµ (4.4)
M M M
κ
Z
+ %1 − kd f kH2
dµ.
ε M
We now compute
Z Z Z
ε ε ε
k∇H d f − TH d f kε dµ =
2
k∇H d f − TH d f kH dµ + ε
2
k∇H d f − TH d f kV2 dµ
M M M
Z Z
ε
= k∇H d f − TH d f kH dµ + ε
2
k∇H d f kV2 dµ
M M
Z Z
ε ε
− 2ε h∇H d f , TH (d f )iV dµ + ε kTH d f kV2 dµ.
M M
(4.5)
Using the definition of ε
TH together with the Yang–Mills assumption, we see that
Z Z
ε 1
h∇H d f , TH (d f )iV dµ = Tr(J∇∗ V f J∇V f )dµ. (4.6)
M ε M
By using (4.6), the trick is now to write
Z Z Z
ε 3 ε 1 ε
h∇H d f , TH (d f )iV dµ = h∇H d f , TH (d f )iV dµ − h∇H d f , TH (d f )iV dµ
M 2 M 2 M
Z Z
3 ε 1
= h∇H d f , TH (d f )iV dµ − Tr(J∇∗ V f J∇V f )dµ.
2 M 4ε M
www.Ebook777.com
292 Fabrice Baudoin
We thus deduce
Z Z !Z
n−1 9κ
(∆H f ) dµ − ε
2
hd(∆H f ), d f iV dµ ≥ %1 − 2
kd f kH dµ
n M M 4ε M
Z
3
+ %2 kd f kV2 dµ.
4 M
Z Z Z
Γ( f , h f )dµ = −
2
f ∆H (h f )dµ = −
2
(∆∗H f )(h2 f )dµ
M M M
Z
= −λ f 2 h2 dµ ≤ 0.
M
Since
Γ( f , h2 f ) = h2 Γ( f , f ) + 2 f hΓ( f , h),
we deduce that Z Z
h Γ( f )dµ + 2
2
hΓ( f , h)dµ ≤ 0.
M M
294 Fabrice Baudoin
• P0 = Id;
• Pt+s = Pt Ps , s,t ≥ 0;
• kPt f k2 ≤ k f k2 ;
where p(x, y,t) > 0 is the so-called heat kernel associated to Pt . Such a function is
smooth and it is symmetric, i.e.,
For a more analytic view of (Pt )t ≥0 , we recall that it can be seen as the unique
solution of a parabolic Cauchy problem in L p (M, µ), 1 < p < +∞.
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 295
ε ∗ ε 1
ε = −(∇H − TH ) (∇H − TH ) − J2 − Ric H ,
ε
then for any smooth function f ,
d∆H f = ε d f
From our assumptions, the symmetric tensor − ε1 J2 − Ric H is bounded from above,
thus by choosing λ big enough, we have
Z Z
ε
hn k∇H η − TH η kε + 2
2 2
hn hη, ∇∇H h n ηiε ≤ 0.
M M
ε η k 2 = 0 which implies ∇ η −
By letting n → ∞, we easily deduce that k∇H η − TH ε H
ε
TH η = 0. If we come back to the equation ε η = λη and the expression of ε , we
see that it implies !
1 2
− J − Ric H (η) = λη.
ε
Our choice of λ then forces η = 0.
Since ε is essentially self-adjoint, it admits a unique self-adjoint extension
which generates, thanks to the spectral theorem, a semigroup Qεt = e tε . We re-
call that Pt = e t∆H is the semigroup generated by ∆H . We have the following
commutation property.
Lemma 5.4. If f ∈ C0∞ (M), then for every t ≥ 0,
dPt f = Qεt d f .
dL = ε d,
where α is any smooth one-form. By using Gronwall’s lemma, we have for every
t ≥ 0,
kτtε α(X t )kε ≤ e (K +(κ/ε))t /2 kα(X t )kε .
By the Feynman–Kac formula, we have for every smooth and compactly supported
one-form,
Q t /2 η(x) = E x (τt η(X t )1t <e ) .
Since dPt = Qεt d, it follows easily that
It is well known that this type of gradient bound implies the stochastic complete-
ness of Pt . More precisely, adapting an argument of Bakry [2] yields the following
result.
Theorem 5.6. For t ≥ 0, one has Pt 1 = 1.
Proof. Let f , g ∈ C0∞ (M), we have
∂
Z Z tZ !
(Pt f − f )g dµ = Ps f g dµ ds
M 0 M ∂s
Z tZ
= (∆H Ps f ) g dµ ds
0 M
Z tZ
=− Γ(Ps f , g)dµ ds.
0 M
5.4 Li–Yau estimates. We show in this section how to obtain the Li–Yau esti-
mate which is a crucial ingredient to prove the Bonnet–Myers theorem.
Henceforth, we will indicate Cb∞ (M) = C ∞ (M) ∩ L ∞ (M). A key lemma is the
following.
Lemma 5.8. Let f ∈ Cb∞ (M), f > 0, and T > 0, and consider the functions
∂φ1
∆H φ1 + = 2(PT −t f )Γ2 (ln PT −t f )
∂t
and
∂φ2
∆H φ2 + = 2(PT −t f )Γ2V (ln PT −t f ).
∂t
Proof. This is direct computation without a trick. Let us just point out that the
formula
∂φ2
∆H φ2 + = 2(PT −t f )Γ2V (ln PT −t f )
∂t
uses the fact that Γ(g, ΓV (g)) = ΓV (g, Γ(g)) and thus that the foliation is totally
geodesic.
We now show how to prove the Li–Yau estimates for the horizontal semigroup.
The method we use is adapted from [7].
Theorem 5.9. Let α > 2. For f ∈ C0∞ (M), f ≥ 0, f , 0, the following inequality
holds for t > 0:
ακ 2%1 ∆H Pt f n%21
!
2%2 V
Γ(ln Pt f ) + tΓ (ln Pt f ) ≤ 1 + − t + t
α (α − 1) %2 α Pt f 2α
! n(α − 1) 2 1 + ακ 2
%1 n ακ (α−1) % 2
− 1+ + .
2 (α − 1) %2 8(α − 2)t
b0 + 2% a = 0.
2
2
a 0 + 2% a − 2κ a − 4aγ = 0,
1
b n
b0 + 2% a = 0,
2
Observe that if Ric H ≥ 0, then we can take %1 = 0 and the estimate simplifies to
2%2 V
Γ(ln Pt f ) + tΓ (ln Pt f )
α
2
ακ
∆H Pt f n(α − 1) 1 + (α−1) %2
2
ακ
!
≤ 1+ + .
(α − 1) %2 Pt f 8(α − 2)t
By adapting a classical method of Li and Yau [43] and integrating this last in-
equality on sub-Riemannian geodesics leads to a parabolic Harnack inequality (de-
tails are in [14]). For α > 2, we denote
ακ
n(α − 1) 2 1 + (α−1) % 2
Dα = . (5.6)
4(α − 2)
The minimal value of Dα is difficult to compute, depends on κ, %2 , and does not seem
relevant because the constants we get are anyhow not optimal. We just point out that
the choice α = 3 turns out to simplify many computations and is actually optimal
when κ = 4%2 .
Corollary 5.10. Let us assume that Ric H ≥ 0. Let f ∈ L ∞ (M), f ≥ 0, and consider
u(x,t) = Pt f (x). For every (x, s), (y,t) ∈ M × (0, ∞) with s < t, one has, with Dα
as in (5.6),
t D α /2 Dα d(x, y) 2
!
u(x, s) ≤ u(y,t) exp .
s n 4(t − s)
Here d(x, y) is the sub-Riemannian distance between x and y.
Since the work by Li and Yau (see [43]), it is classical and not difficult to prove
that a parabolic Harnack inequality implies a Gaussian upper bound on the heat
kernel. With the curvature-dimension inequality in hand, it is actually also possible,
but much more difficult, to prove a lower bound. The final result proved in [11] is as
follows.
302 Fabrice Baudoin
Theorem 5.11. Let us assume that Ric H ≥ 0; then for any 0 < ε < 1 there exists a
constant C(ε) = C(n, κ, %2 , ε) > 0, which tends to ∞ as ε → 0+ , such that for every
x, y ∈ M and t > 0 one has
is not sharp. This is because the method we use, that comes from joint work with
Garofalo [14], is an adaptation of the energy-entropy inequality methods developed
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 303
by Bakry in [3]. Even in the Riemannian case, Bakry’s methods are known to lead to
non-sharp constants. An analytical method that leads to sharp diameter constants is
based on sharp Sobolev inequalities (see [6]), however as of today, it is still an open
question to prove those sharp Sobolev inequalities.
|I f (g)| ≤ kgk1 .
For x, y ∈ Ω, we define
and assume that d is a distance everywhere finite that induces the topology of Ω.
Theorem 6.2. Assume that for every f ∈ L 2 (Ω, µ) and t ≥ 0,
1
kPt f k∞ ≤ k f k2 ,
(1 − e−αt ) D/2
with α, D > 0. Then Ω is compact and its diameter for the distance d satisfies
r
2D
diam(Ω) ≤ 2π .
α
Proof. Since
1
kPt f k∞ ≤ k f k2 ,
(1 − e−αt ) D/2
304 Fabrice Baudoin
from Davies’ theorem ([27, Theorem 2.2.3]), for f ∈ L 2 (Ω) such that Ω f 2 dµ = 1,
R
we obtain
Z Z
f 2 ln f 2 dµ ≤ 2t Γ( f )dµ − D ln 1 − e−αt , t > 0.
Ω Ω
By minimizing the right-hand side of the above inequality over t, we get that for
f ∈ L 2 (M) such that M f 2 dµ = 1,
R
Z Z !
2 2
f ln f dµ ≤ Φ Γ( f )dµ ,
M M
where " ! ! !#
2 2 2 2
Φ(x) = D 1+ x ln 1 + x − x ln x .
αD αD αD αD
The function Φ enjoys the following properties:
• Φ0 (x)/x 1/2 and Φ(x)/x 3/2 are integrable on (0, ∞);
• Φ is concave;
R +∞ R +∞ Φ0 (x) +∞ √
• 21 0 Φ(x) dx = 0 = −2 xΦ00 (x)dx < +∞.
R
√ dx
x 3/2 x 0
We can therefore apply [3, Theorem 5.4] to deduce that the diameter of M is finite
and Z +∞
√ 00
diam(Ω) ≤ −2 xΦ (x)dx.
0
Since Φ00 (x) = − x (2x+α
2D
D) , a routine calculation shows
Z +∞ r
√ 00 2D
−2 xΦ (x)dx = 2π .
0 α
6.2 Proof of the compactness theorem. We now turn to the proof of Theo-
rem 6.1. Let M be a smooth, connected manifold with dimension n + m. As usual,
we assume that M is equipped with a Riemannian foliation with bundle-like com-
plete metric g and totally geodesic m-dimensional leaves. We also assume that the
horizontal distribution is of Yang–Mills type and that for any smooth horizontal one-
form η ∈ Γ∞ (H∗ ),
D E
hRic H (η), ηiH ≥ %1 kη kH
2
, −J2 (η), η ≤ κkη kH
2
,
H
and that for any vertical η ∈ Γ∞ (V ∗ ),
1
Tr(Jη∗ Jη ) ≥ %2 kη kV2 ,
4
with %1 , %2 > 0 and κ ≥ 0.
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 305
b0 + 2%2 a = 0
and
a2
a 0 + 2%1 a − 2κ = 0.
b
With this choice, we get
∂φ
∆H φ +
≥ 0.
∂t
From the parabolic comparison theorem, Theorem 5.7, we deduce
κ + %2 V κ + %2
!
−2% 1 % 2 t /(κ+% 2 ) V
Γ(Pt f ) + Γ (Pt f ) ≤ e Pt (Γ( f )) + Pt (Γ ( f )) .
%1 %1
306 Fabrice Baudoin
∂
Z Z tZ !
(Pt f − f )g dµ = Ps f g dµ ds
M 0 M ∂s
Z tZ
= (∆H Ps f ) g dµ ds
0 M
Z tZ
=− Γ(Ps f , g)dµ ds.
0 M
By means of the previous bound and the Cauchy–Schwarz inequality we find that
Z
(Pt f − f )g dµ (6.1)
M
κ + %2 V
Z t ! r Z
−% 1 % 2 s/(κ+% 2 )
≤ e ds kΓ( f )k∞ + kΓ ( f )k∞ Γ(g) 1/2 dµ.
0 %1 M
and thus Z
1
P∞ f = f dµ.
µ(M) M
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 307
Finally, using the Cauchy–Schwarz inequality, we find that for x ∈ M, f ∈ L 2 (M, µ),
s,t, τ ≥ 0,
Thus, we have Z
1
Pt f (x) →t→+∞ f dµ.
µ(M) M
Since µ(M) < +∞, we can assume µ(M) = 1. The second lemma we need is a
uniform bound on the heat kernel of ∆H , from which we will immediately deduce
Theorem 6.1 by using Theorem 6.2.
where
n β−1 κ
! !
Dβ = 1+ β−1 .
4 β−2 %2
Proof. We fix T > 0 and consider functions a, b : [0,T] → R ≥0 and γ : [0,T] → R
such that
a2 4aγ
0
+ = 0,
a 2%1 a − 2κ −
b n
b + 2%2 a = 0,
0
a(T ) = b(T ) = 0.
Let us choose
b(t) = (e−αt − e−αT ) β , 0 ≤ t ≤ T,
2% 1 % 2
with β > 2, α = β (% 2 +κ) , and a, γ such that
a2 4aγ
0
+ = 0,
a 2%1 a − 2κ −
b n
b0 + 2% a = 0.
2
The optimal β does not lead to a nice formula. The value β = 3 yields the bound
s
√ %2 + κ
!
3κ
diam M ≤ 2 3π 1+ n.
%1 %2 2%2
L = ∆V + Y,
L = ∆v − v · ∇v + ∇ x V · ∇v − v · ∇ x ,
∂h
= ∆v h − v · ∇v h + ∇V · ∇v h − v · ∇ x h, (x, v) ∈ R2n (7.1)
∂t
is then known as the kinetic Fokker–Planck equation with confinement potential V .
It has been extensively studied due its importance in mathematical physics. We refer
for instance to [29, 37, 50, 56]. This equation is the Kolmogorov–Fokker–Planck
equation associated to the stochastic differential system
dx t = vt dt,
dvt = −vt dt − ∇V (x t )dt + dBt ,
where (Bt )t ≥0 is a Brownian motion in Rn .
310 Fabrice Baudoin
L = ∆V + Y,
and assume in this section that the Riemannian foliation is totally geodesic with a
bundle-like metric. Our first task will be to prove a Bochner-type inequality for L. If
f ∈ C ∞ (M), we denote
1
T2 ( f ) = L(k∇ f k 2 ) − 2h∇ f , ∇L f i ,
2
where ∇ is the whole Riemannian gradient. We denote by RicV the Ricci curvature
of the leaves and we denote by DY the tensor defined by DY (U,V ) = hDU Y,V i
where D is the Levi-Civita connection.
We have then the following Bochner inequality.
Theorem 7.2. For every f ∈ C ∞ (M),
T2 ( f ) ≥ (RicV − DY )(∇ f , ∇ f ).
where
1
Γ2H ( f ) = ∆V (k∇H f k 2 ) − 2h∇H f , ∇H ∆V f i ,
2
1
Γ2V ( f ) = ∆V (k∇V f k 2 ) − 2h∇V f , ∇V ∆V f i ,
2
and
1
Γ2Y ( f ) = Y (k∇ f k 2 ) − 2h∇ f , ∇Y f i .
2
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 311
Γ2H ( f ) = k∇V ∇H f k 2 .
So we have Γ2H ( f ) ≥ 0.
Since ∆V is the Laplace–Beltrami operator on the leaves, from the usual Bochner
formula we have
Γ2V ( f ) = k∇2V f k 2 + RicV (∇ f , ∇ f ).
Thus we have
Γ2V ( f ) ≥ RicV (∇ f , ∇ f ).
Finally, we see that
1 1
Y k∇ f k 2 = DY k∇ f k 2 = h∇ f , DY ∇ f i
2 2
and
h∇ f , ∇Y f i = h∇ f , ∇hY, ∇ f ii = DY (∇ f , ∇ f ) + h∇ f , DY ∇ f i.
A difficulty that arises when studying Kolmogorov-type operators is that, in gen-
eral, they are not symmetric with respect to any measure. As a consequence, we
cannot use functional analysis and the spectral theory of self-adjoint operators to de-
fine the semigroup generated by L. A typical assumption to ensure that L generates
a well-behaved semigroup is the existence of a nice Lyapounov function.
So, in the sequel, we will assume that there exists a function W such that W ≥ 1,
k∇W k ≤ CW , LW ≤ CW for some constant C > 0 and {W ≤ m} is compact for
every m. This condition is actually not too restrictive and may be checked in concrete
situations. If M is compact, it is obviously satisfied. A non-compact situation where
it is satisfied is the following: Assume that M is non-compact and that any two points
of M can be joined by a unique geodesic. Also assume that the Riemannian foliation
comes from a Riemannian submersion π : M → B and that the Ricci curvature of the
leaves RicV is bounded from below by a negative constant −K. If x ∈ B, we denote
L x = π −1 ({x}). Any geodesic γ : [0, L] → B in the base space can be lifted into a
geodesic in M. For x ∈ Lγ(0) , denote by τγ (x) the endpoint of the unique horizontal
lift of γ starting from x. Since the leaves are assumed to be totally geodesic, the map
τγ induces an isometry between Lγ(0) and Lγ(L) . Now fix a base point x 0 ∈ M and
for x ∈ Lπ(x0 ) define %V (x) = d(x 0 , x). If x < Lπ(x0 ) , then consider γ : [0, L] → B
to be the unique geodesic between π(x 0 ) and π(x) and define %V (x) = d(τγ (x 0 ), x).
Consider also the function %H (x) = d(π(x 0 ), π(x)) and finally define
We have
Φn0 (s) = Psn (L n k∇Pt−s
n n
f k 2 − 2h∇L n Pt−s f , ∇Pt−s
n
f i).
Now, observe that by assumption, and denoting by K − the negative part of K,
n
L n k∇Pt−s n
f k 2 − 2h∇L n Pt−s f , ∇Pt−s
n
fi
= h2n T2 (Pt−s
n
f , Pt−s
n n
f ) − 4hn LPt−s f h∇hn , ∇Pt−s
n
fi
n
≥ −2K h2n k∇Pt−s n
f k 2 − 4hn LPt−s f h∇hn , ∇Pt−s
n
fi
n
≥ −2K h2n k∇Pt−s n
f k 2 − 4Pt−s L n f h∇ ln hn , ∇Pt−s
n
fi
n n
≥ −2K h2n k∇Pt−s f k 2 − 4kL f k∞ k∇ ln hn kk∇Pt−s fk
≥ −(2K − + 2)k∇Pt−s
n
f k 2 − 2kL f k∞
2
k∇ ln hn k 2 .
where C now depends on f and t, but still does not depend on n. Integrating this
inequality from 0 to t yields a bound of the type
k∇Ptn f k ≤ C,
where C depends on f and t. This bounds holds uniformly on the set {W ≤ n}.
We now pick any x, y ∈ M, f ∈ C0∞ (M), and n big enough so that x, y ∈ {W ≤ n}
and Supp( f ) ⊂ {W ≤ n}. We have from the previous inequality
We therefore reach the important conclusion that Pt transforms C0∞ (M) into a subset
of the set of smooth and Lipschitz functions. With this conclusion in hand, we can
now run the usual Bakry–Émery machinery.
Let f ∈ C0∞ (M), and T > 0, and consider the function
We have
∂φ
Lφ + = 2T2 (PT −t f , PT −t f ) ≥ −2K φ.
∂t
Since we know that φ is bounded, we can use a parabolic comparison principle sim-
ilar to the one in Proposition 5.7 to conclude, thanks to Gronwall’s inequality, that
This inequality is then easily extended to any bounded and Lipschitz function f .
Under the same assumptions, we can actually get slightly stronger bounds:
Theorem 7.4. Let us assume that for some K ∈ R,
RicV − DY ≥ −K;
p
then for every non-negative function f ∈ C ∞ (M) such that f is bounded and
Lipschitz, we have, for t ≥ 0,
we have
∂φ
Lφ + = 2(PT −t f ) T2 (ln PT −t f , ln PT −t f ),
∂t
where we use the fact that since the foliation is totally geodesic we have for every
smooth g,
h∇H g, ∇H k∇V gk 2 i = h∇V g, ∇V k∇H gk 2 i.
The proof then follows the same lines as the proof of Theorem 7.3.
3 Sub-Laplacians and hypoelliptic operators on totally geodesic Riemannian foliations 315
RicV − DY ≥ %,
and that there exists a probability measure µ on M such that for every x ∈ M and
bounded f , Z
lim Pt f (x) = f dµ.
t→+∞ M
Then, µ satisfies the log-Sobolev inequality
Z "Z Z ! Z !#
1
2
f k∇ ln f k dµ ≥ f ln f dµ − f dµ ln f dµ .
M 2% M M M
Proof. Let g ∈ C0∞ (M), g ≥ 0 and define f = g + ε where ε > 0. Since µ needs to
be an invariant measure for L, we have
Z +∞
∂
Z Z Z Z
f ln f dµ − f dµ ln f dµ = − (Pt f )(ln Pt f )dµ dt
∂t
M M M
Z0 +∞ Z M
=− (LPt f )(ln Pt f )dµ dt
0 M
Z +∞ Z
k∇V Pt f k 2
= dµ dt
0 M Pt f
Z +∞ Z
= Pt f k∇V ln Pt f k 2 dµ dt
Z0 +∞ M Z
−2%t
≤ e dt k f ∇ ln f k 2 dµ
0 M
Z
1
≤ f k∇ ln f k 2 dµ.
2% M
p
We then extend the inequality to any non-negative f such that f is bounded and
Lipschitz.
We now study the converse question which is to understand how a functional
inequality satisfied by an invariant measure implies the convergence to equilibrium
of the semigroup.
The easiest convergence to deal with is L 2 convergence and, as is well known, it
is connected to the Poincaré inequality.
316 Fabrice Baudoin
Theorem 7.6. Assume that there exist two constants %1 ≥ 0, %2 > 0 such that for
every X ∈ Γ∞ (TM),
Assume moreover that the operator L admits an invariant probability measure µ that
satisfies the Poincaré inequality
Z Z Z ! 2
k∇ f k 2 dµ ≥ κ f 2 dµ − f dµ .
M M M
Z Z
( %1 + %2 ) (Pt f ) dµ +
2
k∇Pt f k 2 dµ
M M
Z Z !
≤ e−λt ( %1 + %2 ) f 2 dµ + k∇ f k 2 dµ ,
M M
2% 2 κ
where λ = κ+% 1 +% 2 .
By repeating the arguments of the proof of the previous theorem, we get the differ-
ential inequality
Z s
Ψ(s) − Ψ(0) ≥ 2%2 Pu (k∇Pt−u f k 2 )du.
0
% 1 +% 2
Now define ε = κ+% 1 +% 2 . We have, from the assumed Poincaré inequality,
Z Z
ε k∇Pt−u f k 2 dµ ≥ εκ (Pt−u f ) 2 dµ.
M M
Assume moreover that the operator L admits an invariant probability measure µ that
satisfies the log-Sobolev inequality
Z "Z Z ! Z !#
f k∇ ln f k dµ ≥ κ
2
f ln f dµ − f dµ ln f dµ .
M M M M
∞
p
RThen for every positive and bounded f ∈ C (M), such that k∇ f k is bounded and
M
f dµ = 1,
Z Z
2( %1 + %2 ) Pt f ln Pt f dµ + Pt f k∇ ln Pt f k 2 dµ
M M
Z Z !
≤e −λt
2( %1 + %2 ) f ln f dµ + f k∇ ln f k dµ ,
2
M M
2% 2 κ
where λ = κ+2(% 1 +% 2 ) .
7.3 The kinetic Fokker–Planck equation. In this section we study the ki-
netic Fokker–Planck equation which is an important example of an equation to which
our methods apply.
Let V : Rn → R be a smooth function. The kinetic Fokker–Planck equation with
confinement potential V is the parabolic partial differential equation
∂h
= ∆v h − v · ∇v h + ∇ x V · ∇v h − v · ∇ x h, (x, v) ∈ R2n . (7.2)
∂t
The operator
L = ∆v − v · ∇v + ∇ x V · ∇v − v · ∇ x
is a Kolmogorov-type operator. The foliation on R2n which is relevant here is not
the trivial one. We endow R2n with the translation-invariant metric that makes
∂ ∂ ∂
( )
2 + , ,1 ≤ i ≤ n
∂x i ∂vi ∂vi
an orthonormal basis at any point. We then consider the foliations with leaves
{(x, v), v ∈ Rn }. It is obviously totally geodesic.
318 Fabrice Baudoin
It is readily checked that L is not symmetric with respect to µ. The operator L is the
generator of a strongly continuous sub-Markov semigroup (Pt )t ≥0 . If we assume
that the Hessian ∇2V is bounded, which we do in the sequel, then Pt is Markovian.
Observe that since ∇V is Lipschitz, the function W (x, v) = 1 + k xk 2 + kvk 2 is
such that, for some constant C > 0, LW ≤ CW and k∇W k ≤ CW .
The quadratic form T2 is easy to compute in this case and we then obtain the
following result that was first obtained in [9]:
Proposition 7.8. For every 0 < η < 21 , there exists K (η) ≥ − 12 such that for every
f ∈ C ∞ (R2n ),
T2 ( f , f ) ≥ −K (η)k∇V f k 2 + η k∇H f k 2 .
The previous proposition shows that Theorems 7.6 and 7.7 thus apply to the ki-
netic Fokker–Planck operator. We mention that the entropic of the semigroup under
the assumption that the invariant measure satisfies a log-Sobolev inequality was first
established by Villani (see [50, Theorem 35]) but the rate of convergence given by
Theorems 7.6 and 7.7 is more explicit.
Bibliography
[1] A. Agrachev, U. Boscain, J.P. Gauthier & F. Rossi, The intrinsic hypoelliptic Laplacian and
its heat kernel on unimodular Lie groups. J. Funct. Anal. 256 (2009), no. 8, 2621–2655.
Zbl 1165.58012.
[2] D. Bakry, Un critère de non-explosion pour certaines diffusions sur une variété riemannienne
complète. (French. English summary.) [A non-explosion criterium for some diffusions on a
complete Riemannian manifold] C. R. Acad. Sci. Paris Sér. I Math. 303 (1986), no. 1, 23–26.
Zbl 0589.60069.
[3] D. Bakry, L’hypercontractivité et son utilisation en théorie des semigroupes. Ecole d’Eté de
Probabilites de St-Flour, Lecture Notes in Math, (1994). Zbl 0856.47026.
[4] D. Bakry, F. Baudoin, M. Bonnefont, B. Qian, Subelliptic Li-Yau estimates on three dimen-
sional model spaces. Potential Theory and Stochastics in Albac, Aurel Cornea Memorial
Volume (2009). Zbl 1199.35051.
[5] D. Bakry & M. Émery, Diffusions hypercontractives. Sémin. de probabilités XIX, Univ. Stras-
bourg, Springer, 1983. Zbl 0561.60080.
[6] D. Bakry & M. Ledoux, Sobolev inequalities and Myers diameter theorem for an abstract
Markov generator. Duke Math. J. 85 (1996), no. 1, 253–270. Zbl 0870.60071.
Bibliography 319
[7] D. Bakry & M. Ledoux, A logarithmic Sobolev form of the Li-Yau parabolic inequality. Rev.
Mat. Iberoam. 22 (2006), no. 2, 683–702. Zbl 1116.58024.
[8] F. Baudoin, Stochastic analysis on sub-Riemannian manifolds with transverse symmetries.
(2014), To appear in Annals of Probability, Arxiv preprint. arXiv:1402.4490.
[9] F. Baudoin, Bakry-Émery meet Villani. (2013), Arxiv preprint. arXiv:1308.4938.
[10] F. Baudoin & M. Bonnefont, Log-Sobolev inequalities for subelliptic operators satisfying
a generalized curvature dimension inequality. Journal of Functional Analysis 262 (2012),
2646–2676. Zbl 1254.58004.
[11] F. Baudoin, M. Bonnefont & N. Garofalo, A sub-Riemannian curvature-dimension inequality,
volume doubling property and the Poincaré inequality. Math. Ann. 358 (2014), no. 3-4, 833–
860. Zbl 1287.53025.
[12] F. Baudoin, M. Bonnefont, I. Munive & N. Garofalo, Volume and distance comparison theo-
rems for sub-Riemannian manifolds. Journal of Functional Analysis 267 (2014), no. 7, 2005–
2027, Arxiv preprint. arXiv:1211.0221. Zbl 1303.53042.
[13] F. Baudoin & N. Garofalo, Generalized Bochner formulas and Ricci lower bounds for sub-
Riemannian manifolds of rank two. arXiv:0904.1623.
[14] F. Baudoin & N. Garofalo, Curvature-dimension inequalities and Ricci lower bounds for sub-
Riemannian manifolds with transverse symmetries. In minor revision for the Journal of the
EMS, Arxiv preprint. arXiv:1101.3590.
[15] F. Baudoin & N. Garofalo, A note on the boundedness of Riesz transform for some subelliptic
operators. Int. Math. Res. Not. IMRN 2013 (2013), no. 2, 398–421. Zbl 1316.58029.
[16] F. Baudoin & B. Kim, Sobolev, Poincaré and isoperimetric inequalities for subelliptic diffu-
sion operators satisfying a generalized curvature dimension inequality. Revista Matematica
Iberoamericana 30 (2014), 1, 109–131. Zbl 1287.53026.
[17] F. Baudoin & B. Kim, The Lichnerowicz–Obata theorem on sub-Riemannian manifolds with
transverse symmetries. J. Geom. Anal. 26 (2016) no. 1, 156–170. Zbl 06551697.
[18] F. Baudoin, B. Kim & J. Wang, Transverse Weitzenböck formulas and curvature dimension
inequalities on Riemannian foliations with totally geodesic leaves. (2014). arXiv:1408.0548.
[19] F. Baudoin & J. Wang, The subelliptic heat kernel on the CR sphere. Math. Z. 275 (2013),
no. 1-2, 135–150. Zbl 1277.32037.
[20] F. Baudoin & J. Wang, The subelliptic heat kernels of the quaternionic Hopf fibration. Poten-
tial Analysis (2014). Zbl 1304.35234.
[21] Beals, R., Gaveau, B., Greiner, P.C. Hamilton-Jacobi theory and the heat kernel on Heisen-
berg groups. J. Math. Pures Appl. 79, 7 (2000) 633–689. Zbl 0959.35035.
[22] L. Bérard-Bergery, J.P. Bourguignon, Laplacians and Riemannian submersions with totally
geodesic fibres. Illinois J. Math. 26 (1982), no. 2, 181–200. Zbl 0483.58021.
[23] A. Besse, Einstein manifolds. Reprint of the 1987 edition. Classics in Mathematics. Springer,
Berlin, 2008. xii+516 pp. Zbl 1147.53001.
320 Fabrice Baudoin
[24] M. Bonnefont, Functional inequalities for subelliptic heat kernels. PhD dissertation, Paul
Sabatier University, Toulouse, 2009.
[25] M. Bonnefont, The subelliptic heat kernel on SL(2, R) and on its universal covering
SL(2,
K R): Integral representations and some functional inequalities. Potential analysis 36
(2012), no. 2, 275–300. Zbl 1236.43006.
[26] D.C. Chang, I. Markina, A. Vasil’ev, Hopf fibration: geodesics and distances. J. Geom. Phys.
61 (2011), no. 6, 986–1000. Zbl 1217.53036.
[27] E.B. Davies, Heat kernels and spectral theory. Cambridge Tracts in Mathematics, 92. Cam-
bridge University Press, Cambridge, 1989. Zbl 0699.35006.
[28] S. Dragomir & G. Tomassini, Differential geometry and analysis on CR manifolds. Birk-
häuser, Vol. 246, 2006. Zbl 1099.32008.
[29] J-P. Eckmann, M. Hairer, Spectral properties of hypoelliptic operators. Communications in
Mathematical Physics 235 (2003), Issue 2, 233–253. Zbl 1040.35016.
[30] K.D. Elworthy, Decompositions of diffusion operators and related couplings. Preprint, 2014.
[31] R. Escobales, Riemannian submersions with totally geodesic fibers. J. Differential Geom. 10
(1975), 253–276. Zbl 0301.53024.
[32] M. Fukushima, Dirichlet forms and Markov processes. Amsterdam-Oxford-New York, North
Holland, 1980. Zbl 0422.31007.
[33] B. Gaveau, Principe de moindre action, propagation de la chaleur et estimées sous elliptiques
sur certains groupes nilpotents. Acta Math. 139 (1977), no. 1-2, 95–153. Zbl 0366.22010.
[34] D. Geller, The Laplacian and the Kohn Laplacian for the sphere. J. Differential Geometry 15
(1980), 417–435. Zbl 0507.58049.
[35] E. Grong, A. Thalmaier, Curvature-dimension inequalities on sub-Riemannian manifolds ob-
tained from Riemannian foliations, Part I. arXiv:1408.6873.
[36] E. Grong, A. Thalmaier, Curvature-dimension inequalities on sub-Riemannian manifolds ob-
tained from Riemannian foliations, Part II. arXiv:1408.6872.
[37] B. Helffer, F. Nier, Hypoelliptic estimates and spectral theory for Fokker-Planck opera-
tors and Witten Laplacians. Lecture Notes in Mathematics, 1862. Springer, Berlin, (2005).
Zbl 1072.35006.
[38] R. Hermann, A sufficient condition that a mapping of Riemannian manifolds be a fibre bundle.
Proc. Amer. Math. Soc. 11 (1960), 236–242. Zbl 0112.13701.
[39] R. Hladky, Connections and curvature in sub-Riemannian geometry. Houston J. Math. 38
(2012), no. 4, 1107–1134. Zbl 1264.53040.
[40] B. Kim, Poincaré inequality and the uniqueness of solutions for the heat equation associated
with subelliptic diffusion operators. arXiv:1305.0508.
[41] H.-Q. Li, Estimation optimale du gradient du semi-groupe de la chaleur sur le groupe de
Heisenberg. J. Funct. Anal. 236 (2) (2006), 369–394. Zbl 1106.22009.
Bibliography 321
[42] H.-Q. Li, Estimations asymptotiques du noyau de la chaleur sur les groupes de Heisenberg.
C. R. Acad. Sci. Ser. I Math. (2007) 497–502. Zbl 1114.22007.
[43] P. Li & S.T. Yau, On the parabolic kernel of the Schrödinger operator. Acta Math. 156
(1986), 153–201. Zbl 0611.58045.
[44] M.G. Molina & I. Markina, Sub-Riemannian geodesics and heat operator on odd dimensional
spheres. Anal. Math. Phys. 2 (2012), no. 2, 123–147. Zbl 1279.53029.
[45] M. Reed & B. Simon, Methods of modern mathematical physics. Functional analysis. Sec-
ond edition. Academic Press, [Harcourt Brace Jovanovich, Publishers], New York, 1980.
Zbl 0459.46001.
[46] R. Strichartz, Analysis of the Laplacian on the complete Riemannian manifold. J. Func. Anal.
52 (1983), 1, 48–79. Zbl 0515.58037.
[47] S. Tanno, Variation problems on contact Riemannian manifolds. Trans. Amer. Math. Soc.
314 (1989), 1, 349–379. Zbl 0677.53043.
[48] M.E. Taylor, Partial differential equations. II. Applied Mathematical Sciences 116, Springer,
New York (1996). Zbl 1206.35003.
[49] P. Tondeur, Foliations on Riemannian manifolds. Universitext. Springer, New York, 1988.
xii+247 pp. Zbl 0643.53024.
[50] C. Villani, Hypocoercivity. Mem. Amer. Math. Soc. 202 (2009), no. 950. Zbl 1197.35004.
[51] J. Vilms, Totally geodesic maps. Journal of Differential Geometry, vol. 4, (1970), 73–79.
Zbl 0194.52901.
[52] F-Y Wang, Generalized curvature condition for subelliptic diffusion processes. http://arxiv.
org/pdf/1202.0778v2.
[53] F-Y Wang, Analysis for diffusion processes on Riemannian manifolds. Advanced Se-
ries on Statistical Science and Applied Probability, Vol. 18. World Scientific, (2014).
Zbl 1296.58001.
[54] J. Wang, Sub-Riemannian heat kernels on model spaces and curvature-dimension inequalities
on contact manifolds. PhD dissertation, Purdue University, (2014).
[55] J. Wang, The subelliptic heat kernel on the CR hyperbolic spaces. Preprint, 2014.
[56] L. Wu, Large and moderate deviations and exponential convergence for stochastic damping
Hamiltonian systems. Stochastic Process. Appl. 91 (2001), no. 2, 205–238. Zbl 1047.60059.
Index
area formula, 26, 46–52, 56, 79 distance d ∞ , 9∗
divergence theorem, 93
Baker–Campbell–Hausdorff formula,
doubling condition, 189, 233–234
133, 138
ball-box theorem, 205 Engel group, 37∗ , 91, 107, 139∗
basic vector field, 262∗ , 263 Euclidean group, 4∗ , 105
Bishop–Gromov theorem, 234 exponential coordinates, 3∗ , 135∗
blow-up theorem, 90 finite rank condition, 125∗
Bochner’s identity, 219 Folland–Stein embedding theorem, 156
Bott connection, 282∗ fractional integration operator, 198∗
bounded variation, 179∗
BV -function, 32∗ G-perimeter, 33∗ , 34, 46, 83–107, 181∗
G-regular hypersurface, 43∗ , 45
Caccioppoli set (or G-Caccioppoli set), gauge, 132, 144, 175
33∗ , 34, 92 generalized curvature-dimension
Carnot group, 3∗ , 131∗ , 136 inequality, 244∗
Carnot–Caratheódory distance, 8∗ , 162∗ generating system of vector fields, 3∗
carré du champ, 240 group constants, 136
characteristic point, 44 group law structure, 5
of a function, 97∗ group of Heisenberg type, 141–143,
of a regular surface, 95 146–147, 149–150
of a set with finite Euclidean
H-regular surface, 54∗ , 55, 72, 75, 76
perimeter, 87∗
harmonic distribution, 203∗
Chow–Rashevsky theorem, 164
harmonic function, 134∗
Cohn-Vossen theorem, 171
Harnack inequality, 230
complementary subgroup, 35∗ , 38–41
Hausdorff dimension, 13∗ , 14, 17, 49, 50
cone, 57∗
Hausdorff measure, 12∗ , 16–19, 48–51
connection, 218∗
heat kernel, 294∗ , 301
contact manifold, 267∗ , 283
heat semigroup, 292
corkscrew condition, 163, 191
Heisenberg group, 4∗ , 52, 72, 96, 106,
CR Yamabe problem, 154∗
137∗ , 268∗
curvature-dimension inequality, 224∗ ,
H-linear map, 21∗ , 23, 26
289∗
homogeneous dimension, 10∗ , 17, 26,
cut locus, 220
31, 34, 45, 49, 50, 128, 132∗ ,
cut point, 220
190
dilation, 4∗ , 131∗ homogeneous norm, 7∗
Dirichlet form, 303∗ Hopf fibration, 271∗ , 290
324 Index
In the last twenty years, sub-Riemannian geometry has emerged as an independent Volume I
research domain, with extremely rich motivations and ramifications in several parts of
pure and applied mathematics, such as geometric analysis, geometric measure theory,
stochastic calculus and evolution equations together with applications in mechanics,
Davide Barilari
The aim of the lectures collected here is to present sub-Riemannian structures for the Ugo Boscain
use of both researchers and graduate students.
Mario Sigalotti
Editors
ISBN 978-3-03719-162-0
www.ems-ph.org
www.Ebook777.com
Barilari et al. Vol. I | Rotis Sans | Pantone 287, Pantone 116 | 170 x 240 mm | RB: 16 ?? mm