Acta Mathematicae Applicatae Sinica, English Series

Vol. 21, No. 4 (2005) 571–580

Some Properties of Quasiconvex Functions on the

Heisenberg Group
Ming-bao Sun1,2 , Xiao-ping Yang2
1 Department of Applied Mathematics, Hunan Institute of Science and Technology, Yueyang 414000, China
(E-mail: sun mingbao@[Link])
2 Department of Applied Mathematics, Nanjing University of Science and Technology, Nanjing 210094, China
(E-mail: yangxp@[Link])

Abstract For the Heisenberg group, we introduce the concept of h-quasiconvex functions. We prove that the
notions of h-quasiconvex functions and h-convex set are equivalent and that h-quasiconvex functions are locally
bounded from above, and furthermore derive that h-convex functions are locally bounded, therefore it is locally
Lipschitz continuous by using recent results by Danielli-Garofalo-Nhieu. Finally we give estimates of the L∞
norm of the first derivatives of h-quasiconvex functions.

Keywords h-quasiconvex function, h-convex function, Heisenberg group

2000 MR Subject Classification 43A80, 26B25

1 Introduction

An important generalization of classical convex functions is the quasiconvex functions, which

have many interesting properties similar to the convex functions in Rn and play an important
role in mathematical programming (see [2, 4, 5, 7, 9]). Recently, some interesting properties
and notions of convex functions on the Heisenberg group have been investigated by Danielli-
Garofalo-Nhieu, and Lu-Manfredi-Stroffolini (see [3, 8]), the generalization of convex functions
to r-convex functions on the Heisenberg group is given in [12]. In this paper, motivated by
ideas in [2–5, 7–9, 12], we introduce the concept of h-quasiconvex functions in Heisenberg group
denoted by H, and give some interesting properties similar to h-convex functions on H. In
particular,we prove that the notions of h-quasiconvex functions and h-convex set are equivalent
and that h-quasiconvex functions are locally bounded from above, and furthermore derive that
h-convex functions are locally bounded, therefore it is locally Lipschitz continuous by using
recent results by Danielli-Garofalo-Nhieu. Finally we we give the estimates of the L∞ norm of
first derivatives of h-quasiconvex functions.
We begin by recalling some basic facts about the Heisenberg group H=(R3 , ·) (see [11]),
where · is the group operation given by

1 
p · q = x + x , y + y  , z + z  + (xy  − yx ) ,
2
Manuscript received February 4, 2004.
Supported in part by SF for Pure Research of Natural Sciences of the Education Department of Hunan Province
(No.2004c251), Natural Science Foundation of Hunan Province (No.05JJ30006) and National Natural Science
Foundation of China (No.10471063) and specialized Research Fund for Doctoral Program of Higher Education
of China.
572 M.B. Sun, X.P. Yang

where p = (x, y, z) and q = (x , y  , z  ). We also denote by e the group identity. H is a Lie group
with Lie algebra G generated by the left-invariant vector fields
∂ y ∂ ∂ x ∂
X1 = − , X2 = + .
∂x 2 ∂z ∂y 2 ∂z
∂
The only non-trivial Lie bracket is [X1 , X2 ] = ∂z = X3 . Thus H is a nilpotent Lie group of
step 2. For points p, q in H, let
Lp (q) = p · q
be the left-translations on H by an element p ∈ H on q. The group H has a family of dilations
that are group homomorphisms, parameterized by r > 0 and given by

δr (x, y, z) = (rx, ry, r2 z).

The horizontal tangent space at a point p = (x, y, z) is Th (p) =linear span {X1 (p), X2 (p)}. A
piecewise smooth curve t → γ(t) ∈ H is horizontal to γ  (t) ∈ Th (γ(t)) whenever γ  (t) exists.
Given two points p, q ∈ H, let

Γ(p, q) = {γ horizontal curves joining p and q}.

This set is never empty by Chow’s therem (see for example [1]). The Carnot-Carathèodory
distance is then defined by

dcc (p, q) = inf{length(γ) : γ ∈ Γ(p, q)}.

We compute the length of a tangent vector by considering the vectors {X1 , X2 } as an orthonomal
basis of a sub-riemannian metric defined on the distribution of horizontal tangent spaces Th (p).
The Carnot-Carathèodory ball of radius R centered at a point p is given by

B(p, R) = {q ∈ H : dCC (p, q) < R}.

The Carnot-Carathèodory gauge is given by

|p|CC = dCC (e, p).

The Carnot-Carathèodory distance, being constructed in terms of left-invariant vector fields, is

left-invariant and dilation invariant on H.
The exponential mapping exp : G → H, which takes the vector xX1 + yX2 + zX3 in the
Lie algebra G to the point (x, y, z) in the Lie group H, is a global diffeomorphism. We use
exponential coordinates of the first kind:

p = (x, y, z) = exp(xX1 + yX2 + zX3 ).

Given a point p ∈ H, the horizontal plane through p is given by

Hp = Lp (He ),

where He = {(x, y, 0)|(x, y) ∈ R2 } denotes the horizontal plan through e. Hence

Hp = Th (p) = p · He .

For a smooth function u : G → R, we consider

Du = (X1 u, X2 u, X3 u) ∈ R3 .
Some Properties of Quasiconvex Functions 573

the (full) gradient of u at point p. As a vector field, this is written as

Du = (X1 u)X1 + (X2 u)X2 + (X3 u)X3 .

The horizontal gradient of u at a point p is the projection of the gradient of u at p onto the
horizontal tangent space Th (p), and is given by

Dh u = (X1 u, X2 u) ∈ R2 ,

or as a vector field Dh u = (X1 u)X1 + (X2 u)X2 .

We will also use the Heisenberg gauge given by
1
|p|H = ((x2 + y 2 )2 + z 2 ) 4 .

Using a simple version of the ”ball-box” theorem in [1], we may infer the existence of a constant
C such that, for all p ∈ H, we have
1
|p|H ≤ |p|CC ≤ C|p|H .
C
The corresponding distance
dH (p, q) = |q −1 · p|H
is bi-Lipschitz equivalent to the Carnot-Carathèodory distance.
Let Ω ⊂ H be an open set. A function u : Ω → R is said to belong to the class Γ0,1 (Ω) if
there exists a positive constant L > 0 such that, for p, q ∈ Ω,

|u(p) − u(q)| ≤ LdH (p, q). (1.1)

We denote by dg the bi-invariant Haar measure on H obtained by pushing forward the Lebesgue
measure on G via the exponential map.
This paper, except for the introduction, is divided into three sections. In Section 2 we give
the definitions of h-quasiconvex functions and the main results in the Heisenberg group. By
using the results of Section 2 and 3, we will give the proof of main theorems in serves to set the
stage for Section 4.

2 Definitions and Main Results

In this section we introduce the definition of h-quasiconvexity which generalizes the notion of
h-convexity in the Heisenberg group H.
Let p = (x, y, z) and q = (x , y  , z  ) be two points in H and λ ∈ [0, 1], it is easy to infer that

1 1 1 
p · δλ (p−1 · q) = x + λ(x − x), y + λ(y  − y), z + λ(xy  − x y) + λ2 (z  − z + x y − xy  ) . (2.1)
2 2 2
If q ∈ Hp , i. e. p−1 · q ∈ He , then we get
1
z  − z + (x y − xy  ) = 0. (2.2)
2
From (2.1) and (2.2), we derive for any q ∈ Hp that

p · δλ (p−1 · q) = (1 − λ)p + λq. (2.3)

Note also that (2.2), (2.3) hold and q ∈ Hp if and only if p−1 · q ∈ He . We recall the following
notion of h-convexity in [8], also called weak H-convexity in [3], that preserves many interesting
properties of the classical convexity:
574 M.B. Sun, X.P. Yang

Definition 2.1. Let Ω ⊂ H be an open set. A function u : Ω → R is called h-convex if the

restriction of u to the segment [p, q] is a convex function for any two points p and q ∈ Ω such
that the segment [p, q] ⊂ Ω ∩ Hp .
Here [p, q] denotes the convex closure of the set {p, q} in the Euclidean sense.
Definition 2.2. A subset Ω of H is called h-convex if for any p ∈ Ω and every q ∈ Ω ∩ Hp
we have p · δλ (p−1 · q) ∈ Ω for every λ ∈ [0, 1].
Applying (2.2), (2.3) and the fact that a point q ∈ H belongs to Hp if and only if p−1 ·q ∈ He ,
it is easy to see that Definition 2.1 is equivalent to the following concept:
A function u : Ω → R is said to be h-convex if for any two points p and q ∈ Ω such that the
segment [p, q] ⊂ Ω ∩ Hp and λ ∈ [0, 1], we have
 
u p · δλ (p−1 · q) ≤ (1 − λ)u(p) + λu(q).

Recall the concept of quasiconvex function in Rn (see [2, 4]), a natural generalization of Defi-
nition 2.1 may be introduced in the following definition.
Definition 2.3. Let Ω ⊂ H be an open set. A function u : Ω → R is called h-quasiconvex
if the restriction of u to the segment [p, q] is a quasiconvex function for any two points p and
q ∈ Ω such that the segment [p, q] ⊂ Ω ∩ Hp .
Again applying (2.2), (2.3) and the fact that a point q ∈ H belongs to Hp if and only if
p−1 · q ∈ He , we can easily derive that the concept of h-quasiconvexity is equivalent to the
following notion:
A function u : Ω → R is called h-quasiconvex if for any two points p and q ∈ Ω such that
the segment [p, q] ⊂ Ω ∩ Hp and λ ∈ [0, 1], we have

u(p · δλ (p−1 · q)) ≤ max{u(p), u(q)}.

If Ω ⊂ H is an open set, it is easy to see that the following propositions are direct conse-
quences of Definition 2.3.
Proposition 2.4. Let {uα : Ω → R}α∈A be an arbitrary family of h-quasiconvex functions,
then sup uα is h-quasiconvex.
α∈A

Proposition 2.5. Let u : Ω → R be h-quasiconvex and f : R → R be non-decreasing, then

f ◦ u is h-quasiconvex.
Proposition 2.6. Let u : Ω → R be h-quasiconvex and p ∈ H, r > 0, then u◦Lp : Lp−1 (Ω) →
R and u ◦ δr : δ 1r (Ω) → R are h-quasiconvex.
Our main results are the following four theorems
Theorem 2.7. A function u : H → R is h-quasiconvex if and only if for any a ∈ R the level
sets Ωa = {p ∈ H|u(p) ≤ a} are h-convex.
Theorem 2.8. Let u : H → R be a h-quasiconvex function, then u is locally bounded from
above.
From Theorem 2.8, we also have
Theorem 2.9. Let u : H → R be a h-convex function, then u is locally bounded, and
therefore it is locally Lipschitz.
Theorem 2.10. Let Ω ⊂ H be an open set and u : Ω → R be a h-quasiconvex and continuous
function with u(q) ≤ u(p) for any fixed p ∈ Ω and every q ∈ Ω ∩ Hp . Let BR be ball such that
B3R ⊂ Ω. Then u is locally Lipschitz and we have the bound

Dh uL∞ (BR ) ≤ CuL∞ (B3R ) . (2.4)

Some Properties of Quasiconvex Functions 575

Here C is a constant independent of u and R.

3 Some Lemmas

To prove Theorems 2.8–2.10, we need some lemmas.

Lemma 3.1. Let p = q ∈ Rn and f : [p, q] → R be quasiconvex. Let M = max{f (p), f (q)},
then for every x ∈ [p, q] we have
f (x) ≤ M.

Proof. For any x ∈ [p, q], there exists a point λ ∈ [0, 1] such that x = (1 − λ)p + λq, hence we
have
f (x) = f ((1 − λ)p + λq) ≤ M
We will need the following Lemma 3.2, see Theorem1.5.1 in [10] or Theorem 11.3 in [3].
Lemma 3.2. Let p = q ∈ Rn , p = q  ∈ [p, q], x0 = 12 (p + q  ) and f : [p, q] → R be convex,
M = max{f (p), f (q)}, then one has

|f (y)| ≤ M + |f (x0 )| (3.1)

for every y ∈ [p , q  ].

It is easy to see that Theorem 11.6 in [3] implies the following result.
Lemma 3.3. An h-convex function u : H → R which is locally bounded is locally Lipschitz.
Lemma 3.4. Let Ω ⊂ H be a bounded open set. If for u ∈ Γ0,1 (Ω) and (1.1) holds, then
the distributional derivatives of u along the vector fields X1 , X2 belong to L∞ (Ω), and moreover
there exists C = C(H) > 0 such that

Xα uL∞ (Ω) ≤ CL α = 1, 2. (3.2)

Lemma 3.5. Let u ∈ L1,∞ (B(p0 , 3R)), with p0 ∈ H, R > 0, and B(p0 , 3R) ⊂ H, then
u can be modified on a set of dg-measure zero in B = B(p0 , R), so as to satisfy for every
p, q ∈ B(p0 , R) and for some C = C(H) > 0,

|u(p) − u(q)| ≤ CdH (p, q)uL1,∞ (B(p0 ,3R)) . (3.3)

Moreover, if u ∈ C ∞ (B(p0 , 3R)), then in the right hand side of the previous inequality one can
replace the term uL1,∞ (B(p0 ,3R)) with XuL∞ ((B(p0 ,3R)) .
For proof of Lemmas 3.4 and 3.5, the reader is referred to [6].

4 The Proofs of Theorems

Proof of Theorem 2.7. We first prove the necessity. Suppose that u : H → R is h-quasiconvex,
then for any p ∈ H and every q ∈ Hp ,
 
u p · δλ (p−1 · q) ≤ max{u(p), u(q)}.

Therefore for any a ∈ R, if p ∈ Ωa , and q ∈ Ωa ∩ Hp , we have

 
u p · δλ (p−1 · q) ≤ a.
576 M.B. Sun, X.P. Yang

Thus p · δλ (p−1 · q) ∈ Ωa , which establishes the first part of Theorem 2.7.

Now suppose that for any a ∈ R, the level set Ωa = {p ∈ H|u(p) ≤ a} is h-convex. We
will prove that u is h-quasiconvex. For any p, q ∈ H such that the segment [p, q] ⊂ Hp , let
a = max{u(p), u(q)}, then p ∈ Ωa and q ∈ Ωa ∩ Hp . Thus for every λ ∈ [0, 1], we have
p · δλ (p−1 · q) ∈ Ωa and therefore
 
u p · δλ (p−1 · q) ≤ max{u(p), u(q)}.

This proves that u is h-quasiconvex.

Proof of Theorem 2.8. The proof is divided into two parts
1. First we prove that u is bounded from above on any vertical segment.
Assume e+ −
1 = (1, 0, 0) and e1 = (−1, 0, 0), it is easy to see that two straight line segments

s   
s  
L+1 = 1, s,  − 1 ≤ s ≤ 1 and L− 1 = − 1, s, −  − 1 ≤ s ≤ 1
2 2
are respectively the horizontal segments of the horizontal plane He+ and He− . Using the
hypothesis and Definition 2.3, we have that u is quasiconvex on L+ −
1 and L1 . Applying Lemma
+ −
3.1 to L1 and L1 , we may infer that there exists a constant M1 > 0 such that for every
g ∈ L+ −
1 ∪ L1 ,
u(g) ≤ M1 . (4.1)
For every −1 ≤ t ≤ 1, we consider the point (0, 0, t) and the horizontal plane H(0,0,t) =
{(x, y, t)|(x, y) ∈ R2 }. Again from Definition 2.3, for any −1 ≤ t ≤ 1, we obtain that u is
quasiconvex provided it lies on the horizontal line segment L on H(0,0, 2t ) , which passes through
(0, 0, 2t ) and connects the points (1, t, 2t ) ∈ L+ t −
1 and (−1, −t, 2 ) ∈ L1 . Therefore, by (4.1) we
t
may deduce that (4.1) is valid for u on the vertical segment {(0, 0, 2 )| − 1 ≤ t ≤ 1}. From the
invariance of h-quasiconvexity by left translations and dilations, we get that u is bounded from
above on any vertical line segment and that bound may depend on the line segment.
2. Next, we prove that u is bounded from above on the 3-dimensional cube

L3 = {(x, y, z) ∈ H||x| ≤ 1, |y| ≤ 1, |z| ≤ 1}.

From the result proved in part 1, we have that u is respectively bounded from above on the
vertical segments
 − = {(−1, 1, t)| − 1 ≤ t ≤ 1},
L 1
 + = {(1, 1, t)| − 1 ≤ t ≤ 1},
L 1
 − = {(−1, −1, t)| − 1 ≤ t ≤ 1},
L 2
 + = {(1, −1, t)| − 1 ≤ t ≤ 1},
L 2

+ ∪ L
thus there exists a constant M2 > 0 such that for every g ∈ L − ∪ L
+ ∪ L
−,
1 1 2 2

u(g) ≤ M2 . (4.2)

Now we show that u is bounded from above on the face of L3

L2 = {(x, y, z) ∈ H||x| ≤ 1, y = 1, |z| ≤ 1}.

For any −1 ≤ t ≤ 1, by Definition 2.3, we obtain that u is quasiconvex provided it lies on

the horizontal line segment {(s, 0, t)| − 1 ≤ s ≤ 1} on H(0,0,t) = {(x, y, t)|(x, y) ∈ R2 }, which
passes through (0, 0, t). Therefore, from the invariance of h-quasiconvexity by left translations,
Some Properties of Quasiconvex Functions 577

we derive that horizontal line segment{(s, 1, t)| − 1 ≤ s ≤ 1} on H(0,0,t) , which is parallel to the
line segment {(s, 0, t)|−1 ≤ s ≤ 1} and connects the points (−1, 1, t) and (1, 1, t), is quasiconvex,
from (4.2) and applying Lemma 3.1 to the horizontal line segment {(s, 1, t)| − 1 ≤ s ≤ 1} for
|t| ≤ 1, we derive that
sup u(g) ≤ M2 .
g∈L2

This shows that u is bounded from above on L2 .

Similarly, from (4.2) and Lemma 3.1, we can derive that u is bounded from above on the
other faces of L3 which are parallel to the vertical axis. For any fixed (x, y, z) ∈ L3 , again from
Definition 2.3, Lemma 3.1 and (4.2), we derive that u is bounded from above provided it lies
on the horizontal line segment on the horizontal plane H(0,0,z) , which passes through (0, 0, z)
and (x, y, z) and intersects with the faces of cube L3 . Furthermore we have

u(x, y, z) ≤ M2 .

This show that u is bounded from above on L3 .

From the result proved in part 2, and the invariance of h-quasiconvexity by left translations
and dilations, we reach the conclusion of Theorem 2.8.
Proof of Theorem 2.9. According to Lemma 3.3, we only need to prove the first part of
Theorem 2.9. The proof is divided into three parts.
1. First we prove that u is bounded on any vertical segment. From Theorem 2.8, we get
that u is bounded from above on the vertical line segments
 + = {2, 0, t)| − 1 ≤ t ≤ 1} and L
L  − = {−2, 0, t)| − 1 ≤ t ≤ 1},
1 1

+ ∪ L
thus there exists M2 ≥ 0 such that for every g ∈ L −,
1 1

u(g) ≤ M2 . (4.3)

Consider again the segment

 = {(−2s, 0, s)| − 1 ≤ s ≤ 1}
L
on the horizontal plane H(0,1,0) = {(x, y, −x/2)|(x, y) ∈ R2 }. By Definition 2.1 we get that u
 Applying Lemma 3.2 to the segments L,
is convex on L.  we see that there exists a constant
M3 > 0 such that for every g ∈ L,
|u(g)| ≤ M3 . (4.4)
For −1/4 ≤ t ≤ 1/4, consider the horizontal segment {(s, 0, t)|−2 ≤ s ≤ 2} on the horizontal
plane H(0,0,t) = {(x, y, t)|(x, y) ∈ R2 }. Using Definition 2.1 we get that u is convex on such
segment, which intersects the segments L,  L  − in the points
 + and L
1 1

(−2t, 0, t), (2, 0, t) and (−2, 0, t),

respectively. Applying (4.3), (4.4) and Lemma 3.2 to the segments {(s, 0, t)| − 2 ≤ s ≤ 2} for
|t| ≤ 14 , we derive that u is bounded on the vertical segment

1  1
+ = 
L 1/4 , 0, t |t| ≤ ,
2 4
there exists a constant C(M2 , M3 ), which depends only on M2 and M3 , such that for every
g∈L + ,
1/4
|u(g)| ≤ C(M2 , M3 ).
578 M.B. Sun, X.P. Yang

From the invariance of H-convexity by left translations and dilations, we get that u is bounded
on any vertical line segments and the bound may depend on the line segment.
2. Now we show that u is bounded on the 2-dimensional square
L2 = {(x, y, z) ∈ H1 ||x| ≤ 1, y = 0, |z| ≤ 1}
From the result proved in part 1, we have that u is respectively bounded on the vertical segments
 − = {(−1, 0, t)| − 1 ≤ t ≤ 1}
L 1

L0 = {(0, 0, t)| − 1 ≤ t ≤ 1},
 + = {(1, 0, t)| − 1 ≤ t ≤ 1}.
L 1

+ ∪ L
There exists a constant M > 0 such that for every g ∈ L 0 ∪ L
− ,
1 1

|u(g)| ≤ M . (4.5)
For any −1 ≤ t ≤ 1, consider the horizontal segment {(s, 0, t)| − 1 ≤ s ≤ 1} on the plane
H(0,0,t) = {(x, y, t)|(x, y) ∈ R2 }. By Definition 2.1, we obtain that u is convex provided it
lies on the horizontal line segment {(s, 0, t)| − 1 ≤ s ≤ 1}, which passes through (0, 0, t) and
connects the points (−1, 0, t) and (1, 0, t). Therefore, from (4.5) and applying Lemma 3.2 to
the horizontal line segment {(s, 0, t)| − 1 ≤ s ≤ 1} for |t| ≤ 1, there exists a constant C(M )
which depends on M such that
sup |u(g)| ≤ C(M ).
g∈L2

This shows that u is bounded on L2 .

Similarly, we can deduce that u is bounded on the square
 2 = (x, y, z) ∈ H1 |x = 0, |y| ≤ 1, |z| ≤ 1 .
L
From the invariance of H-convexity by left translations and dilations, we get that u is bounded
on any squares which are parallel to the sequare L2 or L2.
3. Finally, we prove that u is bounded on the 3-dimensional cube
L3 = (x, y, z) ∈ H1 ||x| ≤ 1, |y| ≤ 1, |z| ≤ 1 .
By the results proved in Part 1 and 2, we obtain that u is bounded by some constant M on
the segment {(0, 0, z)| − 1 ≤ z ≤ 1} and the faces of L3 which are parallel to the vertical axis.
Moreover, for any fixed (x, y, z) ∈ L3 , we consider the horizontal segment S on the horizontal
plane H(0,0,t) = {(x, y, t)|(x, y) ∈ R2 }, which passes through (0, 0, z) and (x, y, z), and intersects
the faces of cube L3 . From Definition 2.1, we see that u is convex on such segment. Thus
applying Lemma 3.2 to such horizontal segment on the horizontal plane H(0,0,z) , we may infer
that u is bounded on such segment by a constant C(M ) dependent only on M . In addition we
have
|u(x, y, z)| ≤ C(M ).
This shows that u is bounded on L3 .
From the result proved in part 3 and the invariance of H-convexity by left translations and
dilations, we reach the conclusion of Theorem 2.9.
Proof of Theorem 2.10. Given a kernel K ∈ C0∞ (H), K ≥ 0, supp K ⊂B(e, 1), H K(g)dg = 1.
Consider the corresponding approximation to the identity {Kε = ε−4 K ◦ δ1/ε }ε>0 associated
with it. Define the convolution

uε = Kε  u(p) = Kε (h)u(h−1 · p)dh (4.6)

H
Some Properties of Quasiconvex Functions 579

for p ∈ Ωε = {p ∈ Ω|dist(p, ∂Ω) > ε}, then uε ∈ C ∞ (Ωε ). By hypothesis, u ∈ C(Ω), we then
have uε → u uniformly on compact subsets of Ω. Moreover, the function uε is h-quasiconvex in
Ωε . To see this note that for any p and q ∈ Ωε such that [p, q] ⊂ Ωε ∩ Hp and λ ∈ [0, 1], from
(4.6) and Definition 2.3 one has
    
uε p · δλ (p−1 · q) = Kε (h)u h−1 · p · δλ (p−1 · q) dh
H
  
= Kε (h)u (h−1 · p) · δλ (h−1 · p)−1 · (h−1 · q) dh
H

≤ Kε (h) max u(h−1 · p), u(h−1 · q) dh

H
= max{uε (p), uε (q)}. (4.7)

By the hypothesis u(q) ≤ u(p) for any fixed p ∈ Ωε and q ∈ Ωε ∩ Hp , we have

uε (q) ≤ uε (p). (4.8)

Again by the hypothesis and Definition 2.3, and note (4.7) and (4.8), for any fixed p ∈ Ωε
and [p, q] ∈ Ωε ∩ Hp we derive that for every λ > 0,

uε (p · δλ (p−1 · q)) − uε (p)

≤ 0. (4.9)
λ
Let p = (x, y, z), q = (x , y  , z  ) be two points in Ωε , we introduce the function ϕ : [0, 1] → R
defined by  
ϕ(λ) = uε p · δλ (p−1 · q) . (4.10)
From (2.1), we have

1
ϕ(λ) =uε x + λ(x − x), y + λ(y  − y), z + λ(xy  − x y)
2


2  1  1 
+ λ z − z + x y − xy . (4.11)
2 2
Clearly, ϕ(0) = uε (p), letting λ → 0 in (4.9), we get

ϕ (0) ≤ 0. (4.12)

Differentiating (4.11) with respect to λ, and applying

∂ y ∂ ∂ x ∂
X1 = − , X2 = + ,
∂x 2 ∂z ∂y 2 ∂z
we derive
ϕ (0) = X1 uε (p)(x − x) + X2 uε (p)(y  − y) = Dh uε (p), ζ , (4.13)
T  
where ζ = (x − x, y − y).
Now we fix a ball B(p0 , R) ⊂ B(p0 , 3R) ⊂ Ωε . For any fixed p ∈ B(p0 , R) such that
[p, q] ⊂ Ωε ∩ Hp and q = p, from (4.12) and (4.13), we get
 ζ 
Dh uε (p), ≤ 0. (4.14)
ζ
Passing to the supremum on all q ∈ ∂B(p, R) ∩ Hp in (4.14) for ε > 0 small enough, we derive

Dh uε (p) ≤ 0.
580 M.B. Sun, X.P. Yang

Since p ∈ B(p0 , R) is arbitrary, thus we have

Dh uε L∞ (B(p0 ,R)) ≤ 0. (4.15)

Using (4.15) and Lemma 3.5. we obtain for p, q ∈ B(p0 , R),

|uε (p) − uε (q)| ≤ Cuε L∞ (B(p0 ,3R)) dH (p, q), (4.16)

where C > 0 is an absolute constant. Taking ε → 0, we then have

|u(p) − u(q)| ≤ CuL∞ (B(p0 ,3R)) dH (p, q) (4.17)

for p, q ∈ B(p0 , R). Thus u is locally Lipschitz. From Lemma 3.4 and (4.17), we further deduce
that the weak derivatives X1 u, X2 u existdg-a.e. in Ω and belong to L∞ loc (Ω). We then further
obtain
Dh uL∞ (B(p0 ,R)) ≤ CuL∞ (B(p0 ,3R)) ,
where C is a constant independent of u and R. This completes the proof of Theorem 2.10.
Remark 4.1. All the results in this paper would generalize almost without changes to higher
dimensional Heisenberg groups.

References
[1] André, Bellaı̈che and Jean-Jacques Risler ed. Sub-Riemannian geometry. Progress in Mathematics, Vol.
144, Birkhauser, 1996
[2] Arrow, K.J., Enthoven, A.C. Quasi-concave programming. Economitrica, 29: 779–800 (1961)
[3] Danielli, D., Garofalo, N., Nhieu, D.M. Notions of convexity in Carnot groups. Comm. Anal. Geom., 11:
263–341 (2003)
[4] Fenchel, W. Convex cones, sets and functions. Princeton University, Princeton, New Jersey 1951
[5] Ferland, J.A. Matrix-theoretic criteria for the quasi-convexity of twice continuously differenciable functions.
Linear Alg, Appl., 38: 51–63 (1981)
[6] Garofalo, N., Nhieu, D.M. Lipschitz continuity, global smooth approximations and extension theorems for
Sobolev functions in Carnot-Carathèodory spaces. J. Anal. Math., 74: 67–97 (1998)
[7] Greenberg, H.J., Pierskalla, W.P. A review of quasi-convex function. OR, 19: 1553–1570 (1971)
[8] Lu, G.Z., Manfredi, J.J., Stroffolini, B. Convex functions on the Heisenberg group. Calc. Var. Partial
Differential Equations, 19: 1–22 (2003)
[9] Nikaido, H. On Von Neumann’s minimax theorem. Pacific J. Math., 4: 65–72 (1954)
[10] Schneider, R. Convex Bodies: The brunn-minkowski throry. Carmbridge Univ. Press, 1993
[11] Stein, E.M. Harmonic Analysis: Real varible methods, orthogonality and oscillatory integrals. Princeton
Univ. Press, 1993
[12] Sun, M.B., Yang, X.P. Inequalities of Hadamard type for r-convex functions in Carnots groups. Acta
Math. Appl. Sinica, 20: 123–132 (2004).