RACSAM (2019) 113:587–604

https://doi.org/10.1007/s13398-018-0500-5

ORIGINAL PAPER

Interior regularity for degenerate equations with drift on

homogeneous groups

Guangwei Du1 · Junqiang Han1 · Pengcheng Niu1

Received: 13 September 2016 / Accepted: 14 January 2018 / Published online: 22 January 2018
© Springer-Verlag Italia S.r.l., part of Springer Nature 2018

Abstract Let X 0 , X 1 , . . . , X m be left invariant real smooth vector fields on the homogeneous
group G = (R N , ◦, δλ ) satisfying the Hörmander condition. Assume that X 1 , . . . , X m (m ≤
N − 1) are homogeneous of degree one and X 0 is homogeneous of degree two with respect to
the family of dilations (δλ )λ>0 . Consider the second order degenerate equation with drift on G

m 
m
X i (ai j (x)X j u) + X 0 u = X j F j (x),
i, j=1 j=1

where the matrix {ai j (x)} of coefficients is symmetric positive definite on Rm belonging to
the VMO space defined on G. We prove a Sobolev type inequality for weak solutions to
the equation and then the higher L p estimates for the gradients of weak solutions. Interior
Morrey estimates and Hölder continuity for the weak solutions are derived by using the
representation formula of weak solutions and estimates of singular integrals.

Keywords Homogeneous group · Degenerate equation · Morrey estimates · Hölder

continuity

Mathematics Subject Classification 35R03 · 35B65 · 35C15

1 Introduction

Let {X 0 , X 1 , . . . , X m } be a system of real smooth vector fields in R N (m +1 ≤ N ), satisfying

the Hörmander condition [9]:

This work is supported by the National Natural Science Foundation of China (No. 11771354) and Natural
Science Basic Research Plan in Shaanxi Province of China (No. 2017JM5140).

B Junqiang Han
southhan@163.com
1 Department of Applied Mathematics, Northwestern Polytechnical University, Xi’an 710129, China
588 G. Du et al.

rank (Lie{X 0 , X 1 , . . . , X m }) = N .
We also assume that X 0 , X 1 , . . . , X m are left invariant with respect to the translations on the
Lie group G = (R N , ◦) and homogeneous with respect to the family of dilations (δλ )λ>0 on
R N . More precisely, X 1 , . . . , X m are homogeneous of degree one and X 0 is homogeneous
of degree two. Let Ω be a bounded domain in the homogeneous group G = (R N , ◦, δλ ).
The main purpose of this paper is to consider the interior regularity for weak solutions to the
second order degenerate equation with drift:

m 
m
L u := X i (ai j (x)X j u) + X 0 u = X j F j (x), (1.1)
i, j=1 j=1

where x ∈ Ω and F j ( j = 1, 2, . . . , m) are in suitable Morrey spaces. In the sequel, we

simply rewrite Eq. (1.1) as
div X (AXu) + X 0 u = div X (F) (1.2)
m
where X = (X 1 , . . . , X m ), A(x) = (ai j (x))m×m is a m ×m matrix, div X (F) = j=1 X j Fj ,
F = (F1 , . . . , Fm ). If A is the identity matrix, we write

m
L 0 := X 2j + X 0 . (1.3)
j=1

A function u is called a weak solution to (1.1) if u, X 1 u, . . . , X m u, X 0 u ∈ L 2loc (Ω) and

ˆ ˆ ˆ
AXu, X ϕd x − X 0 uϕd x = F, X ϕ (1.4)
Ω Ω Ω

for all ϕ ∈ C0∞ (Ω).Here ·, · denotes the scalar product in Rm .

We make the following assumptions:
(H) The coefficients ai j (x) = a ji (x) ∈ VMO (see Sect. 2) and there exists a constant
μ > 0 such that

m
μ−1 |ξ |2 ≤ ai j (x)ξi ξ j ≤ μ|ξ |2
i, j=1

for any x ∈ RN and ξ ∈ Rm ; moreover,

F j ∈ L 2Q/(Q−2),λ (Ω), j = 1, 2, . . . , m,
where 0 ≤ λ < Q, Q is the homogeneous dimension of G, L 2Q/(Q−2),λ (Ω) is the Morrey
space with respect to G (see
Sect. 2).
N
In (1.1), when X 0 = i, j=1 i j x i ∂ j − ∂t and X i = ∂i (i = 1, 2, . . . , m), we get a
b
Kolmogorov type equation of the form

m 
N 
m
Lu := ∂i (ai j (z)∂ j u) + bi j xi ∂ j u − ∂t u = ∂ j F j (z), (1.5)
i, j=1 i, j=1 j=1

where z ∈ R N +1 and (bi j ) N ×N is a suitable constant matrix. Equation (1.5) is the so-called
ultraparabolic equation and often appears in the kinetic theory of gases as well as in the
mathematical finance (see [15]). Clearly, if m = N , bi j = 0 and {ai j (x)} is symmetric
positive definite, then (1.5) is uniformly parabolic.
Interior regularity for degenerate equations with drift. . . 589

Equations like (1.5) with constant coefficients or Hölder continuous coefficients have been
extensively studied, see [11,13,16–18] and references therein. In these papers the existence
and estimates of the fundamental solution for the operator L, existence and uniqueness for
the Cauchy problem and some related regularity results were proved.
Since VMO spaces may contain some kind of discontinuous functions, ultraparabolic
equations in divergence or non-divergence form with VMO coefficients have attracted much
attention over the decades, see [3,14,19,20,24]. Manfredini and Polidoro [14] proved a priori
local L p estimates and a Hölder continuity result of the gradient of weak solutions to (1.5)
with VMO coefficients. Polidoro and Ragusa [20] derived the local Morrey estimates for
p
the gradient of weak solutions to (1.5) provided u ∈ L loc (Ω). In [3] and [19], the local L p
estimates and local Hölder continuity of strong solutions to the non-divergence ultraparabolic
equations were concluded.
Recently, regularity to degenerate equations in the homogeneous setting has been consid-
ered by several authors, see [2,4,5,10,21,22]. Bramanti and Brandolini [2] proved local L p
estimates for solutions to some non-divergence degenerate equations on the homogeneous
group. Cinti and Polidoro [5] proved an L ∞
loc bound of the weak solution u to the ultraparabolic
equation
m
∂t u = X i (ai j X j u) + X 0 u,
i, j=1

where the vector fields X 1 , . . . , X m and X 0 − ∂t are invariant with respect to a suitable
homogeneous Lie group and satisfy the Hörmander condition. By exploiting the representa-
tion formulas for solutions and estimates of singular integrals, Feng and Niu [6] proved that
p
if u is a weak solution to (1.2) and u ∈ L loc (Ω), then
 
X u L p (Ω ) ≤ c F L p (Ω ) + u L p (Ω ) , (1.6)
for any Ω ⊂⊂ Ω ⊂⊂ Ω and u is Hölder continuous. Global a priori estimates for
degenerate equations with constant coefficients on homogeneous group were derived by Hou
and Niu [10].
The main results of this paper are the following.

Theorem 1 Let Ω be a bounded domain in G and u a weak solution to (1.1), then for
any 0 ≤ λ < Q, p ∈ (1, 2∗ ], 2∗ = Q−2
2Q p,λ
, we have that u, X u ∈ L loc (Ω), and for every
Ω ⊂⊂ Ω ⊂⊂ Ω, there exists a constant c > 0 depending on Ω and Ω such that
 
u L p,λ (Ω ) ≤ c F L 2∗ ,λ (Ω) + u L 2 (Ω ) , (1.7)

and  
X u L p,λ (Ω ) ≤ c F L 2∗ ,λ (Ω) + u L 2 (Ω ) . (1.8)

Theorem 2 Let Ω be a bounded domain in G and u a weak solution to (1.1). Then for any
0 ≤ λ < Q and p ∈ (1, 2∗ ], p > Q − λ, and Ω ⊂⊂ Ω ⊂⊂ Ω, there exists a positive
constant c depending on Ω and Ω such that for any x, z ∈ Ω , x = z,
|u(x) − u(z)|  
Q λ
≤ c F L 2∗ ,λ (Ω) + u L 2 (Ω ) . (1.9)
1− +
z −1 ◦ x p p
p
Unlike [20] and [6], we do not assume u ∈ L loc (Ω). To obtain our results, a key component
in our proof is to establish a Sobolev type inequality for weak solutions to (1.1) by using the
idea in [5]. It is worth noting that in proving the Sobolev type inequality for weak solutions, the
590 G. Du et al.

technique of constructing a suitable test function in the classic parabolic case (i.e. X 0 = ∂t )
seems to be not useful because of the lacking of Sobolev embedding theorem with X 0 on the
homogeneous group. Our way is to represent the weak solution u in terms of the fundamental
solution of operator L 0 and then use some estimates in the Morrey spaces of singular integrals
to complete our proof.
The organization of the paper is as follows. In Sect. 2, we present some preliminaries and
auxiliary results. Section 3 is devoted to proofs of a Caccioppoli type inequality, a Sobolev
type inequality and the higher L p estimates of X u. In Sect. 4, inspired by the way of Polidoro
and Ragusa [20], Theorem 1 is proved. Based on Theorem 1 and the representation formula
for u, Theorem 2 is proved in Sect. 5.

2 Preliminaries

We describe first some notions on homogeneous groups. For more details, we refer to the
monograph [1].
Let ◦ be a given group law on R N and assume that the map (x, y) → y −1 ◦ x is smooth,
then R N together with this mapping forms a Lie group. If there exists an N -tuple of real
numbers σ = (σ1 , . . . , σ N ), with 1 ≤ σ1 ≤ . . . ≤ σ N , such that the dilation
δλ : δλ (x1 , . . . , x N ) → (λσ1 x1 , . . . , λσ N x N ), for every λ > 0
is an automorphism of the group, we say that the Lie group with this structure is a homoge-
neous group, denoted by G.
The homogeneous norm on G can be defined as follows. For any x ∈ G \ {0}, set
x = ρ ⇔ |δ1/ρ (x)| = 1,
where | · | denotes the Euclidean norm, and set 0 = 0. This norm enjoys the following
properties:
(i) δλ (x) = λx, for every x ∈ G and λ > 0;
(ii) the set {x ∈ G : x = 1} is the Euclidean unit sphere Σ N ;
(iii) the function x → x is smooth outside the origin;
(iv) there exists a constant c(G) ≥ 1 such that for every x, y ∈ G,
x ◦ y ≤ c(x + y) and x −1  ≤ cx.
In view of these properties, we can define the quasidistance d by
d(x, y) = y −1 ◦ x
and define the d-ball by
B(x, r ) = {y ∈ R N : d(x, y) < r }.
Note that B(0, r ) = δr (B(0, 1)). Moreover, it was proved (see [23, p. 619]) that the Lebesgue
measure in R N is the Haar measure of G and that
|B(x, r )| = |B(0, 1)|r Q , (2.1)
where Q = σ1 + · · · + σ N is socalled the homogeneous dimension of G. Throughout the
paper we always assume Q ≥ 3. By (2.1) the Lebesgue measure is a doubling measure with
respect to d, that is
|B(x, 2r )| ≤ c|B(x, r )|, x ∈ G, r > 0. (2.2)
Interior regularity for degenerate equations with drift. . . 591

A real function f on G is called homogeneous of degree α if for every x ∈ G and λ > 0,

it holds
f (δλ (x)) = λα f (x);
a differential operator Y is called homogeneous of degree β if
Y (ϕ(δλ (x))) = λβ (Y ϕ)(δλ (x))
for every test function ϕ and λ > 0. Note that if f is a homogeneous function of degree α and
Y is a differential operator homogeneous of degree β, then Y f is a homogeneous function
of degree α − β.
The convolution of two functions on G is defined by
ˆ ˆ
−1
( f ∗ g)(x) = f (x ◦ y )g(y)dy = g(y −1 ◦ x) f (y)dy,
G G
if the above integrals are well defined. Fixed ξ ∈ G, let τξ (x) = ξ ◦ x be the left translation
on G. A differential operator P is called left invariant if
P(ϕ ◦ τξ ) = (Pϕ) ◦ τξ
for any ξ ∈ G and every smooth function ϕ on G. It follows that for any left invariant
differential operator P,
P( f ∗ g) = f ∗ Pg
provided the integrals converge.
Now we define the Morrey spaces, BMO and VMO spaces associated with the structure
of G. Hereafter we shall denote by B any ball of G, by Br any ball of radius r and, for every
f ∈ L 1loc (G),
ˆ
1
fB = f (x)d x = f (x)d x.
B |B| B
p
Definition 1 Let Ω be a domain in G, p ∈ (1, ∞) and λ ∈ [0, Q]. We say that f ∈ L loc (Ω)
belongs to the Morrey space L p,λ (Ω) if
 ˆ 1
p
 f  L p,λ (Ω) = sup ρ −λ |u(y)| p dy < ∞.
x∈Ω, 0<ρ<diamΩ Ω∩Bρ (x)

Definition 2 We say that f ∈ L 1loc (G) belongs to BMO if

 f ∗ = sup | f (x) − f B |d x < ∞;

B B

f ∈ L 1loc (G) belongs to VMO if f ∈ BMO and

η f (r ) = sup | f (x) − f Bρ |d x → 0, r → 0.
ρ<r Bρ

The following observation is a simple consequence of the Hölder inequality (see e.g. [12]).

Lemma 1 If q1 ≤ q2 and λ1 ≤ q1
q2 λ2 + (1 − q1
q2 )Q, then L q2 ,λ2 (Ω) ⊂ L q1 ,λ1 (Ω), and
 f  L q1 ,λ1 (Ω) ≤ c  f  L q2 ,λ2 (Ω) .
592 G. Du et al.

Consider the “frozen” operator of L :


m
L0 = ai j (x0 )X i X j + X 0 . (2.3)
i, j=1

By a classical result of Folland [8, Theorem 2.1], we know that L0 has a fundamental solution
Γ (x0 ; ·) with pole at the origin. We always set
Γi = X i Γ and Γi j = X i X j Γ, i, j = 1, . . . , m.
The following properties hold (see [2,8]).

Lemma 2 For every x0 ∈ G,

(a) Γ (x0 ; ·) ∈ C ∞ (G) \ {0};
(b) Γ (x0 ; ·) is homogeneous of degree (2 − Q);
(c) Γi (x0 ; ·) and Γi j (x0 ; ·) is homogeneous of degree (1 − Q) and −Q, respectively;
(d) for every test function u and x ∈ G we have
ˆ
u(x) = (L0 u ∗ Γ (x0 ; ·))(x) = Γ (x0 ; y −1 ◦ x)L0 u(y)dy.
G

Furthermore, we also have the following pointwise estimates (see [6]): there exists a
positive constant C such that
C C
Γ (x0 ; y −1 ◦x) ≤ , Γ j (x0 ; y −1 ◦x) ≤ −1 , j = 1, . . . , m. (2.4)
y −1 ◦ x Q−2 y ◦ x Q−1

Remark Obviously, the above properties also hold for the fundamental solution Γ 0 (x; y) of
the operator L 0 defined in (1.3).
The following results about potential estimates due to Folland [8] play an important role
in our proof.

Lemma 3 Let K ∈ C(G \ {0}) be a homogeneous function of degree α − Q with α ∈ (0, Q)

and define an operator
ˆ
TK f (x) = K (y −1 ◦ x) f (y)dy
G
for f ∈ L p (G), p ∈ (1, ∞), then there exists a constant c = c(Q, p) such that
TK f  L q (G) ≤ c max |K (x)| f  L p (G) ,
x=1

1 1 α
for any q verifying = − .
q p Q
Introduce singular integrals
ˆ ˆ
Ti j f (x) = lim Γi j (x; y −1 ◦ x) f (y)dy := P.V. Γi j (x; y −1 ◦ x) f (y)dy
ε→0 y −1 ◦x≥ε G
(2.5)
and ˆ
Ci j [a, f ] (x) = P.V. Γi j (x; y −1 ◦ x)[a(y) − a(x)] f (y)dy, (2.6)
G
for f ∈ L p (G), i, j = 1, . . . , m.
Interior regularity for degenerate equations with drift. . . 593

Lemma 4 (see [7, Theorem 3.1]) If a ∈ BMO and f ∈ L p,λ (G), p > 1 and 0 ≤ λ < Q,
then Ti j f , Ci j [a, f ] ∈ L p,λ (G), i, j = 1, . . . , m, and there exists a constant c > 0 such
that
Ti j f  L p,λ (G) ≤ c f  L p,λ (G) ,
Ci j [a, f ] L p,λ (G) ≤ ca∗  f  L p,λ (G) .

Lemma 5 (see [7, Theorem 3.2]) Let a ∈ VMO, then for every ε > 0, there exists r > 0
depending on ε, p and λ, such that for every f ∈ L p,λ (Br ), p > 1 and λ ∈ [0, Q),
Ci j [a, f ] L p,λ (Br ) ≤ cε f  L p,λ (Br ) , i, j = 1, . . . , m.

By virtue of Lemma 3.6 in [7], we find that

ˆ ˆ
| f (y)|
|TK f (x)| ≤ |K (y −1 ◦ x)|| f (y)|dy ≤ c dy,
G G y −1 ◦ x Q−α
since K is a homogeneous function of degree α − Q. Then with a similar proof of Lemma
2.8 in [19], we have the following lemma.

Lemma 6 Let K ∈ C(G \ {0}) be a homogeneous function of degree α − Q, α ∈ (0, Q),

f ∈ L p1 ,μ1 (G) and p1 ∈ (1, ∞). If μ1 + αp1 < Q, then there exists a positive constant c
such that
TK f  L p2 ,μ2 (G) ≤ c  f  L p1 ,μ1 (G) ,
where p2 and μ2 satisfy
1 1 α μ1 Q p2
= − , μ2 = = μ1 .
p2 p1 Q Q − αp1 p1

Given two balls Bs and Br with Bs ⊂ Br and a function η ∈ C0∞ (G), let us write
Bs ≺ η ≺ Br to mean
0 ≤ η ≤ 1, η ≡ 1 on Bs and suppη ⊂ Br .
Existence of such smooth cut off functions on homogeneous groups was shown in [2]. Pick
a C ∞ function ϕ : [0, ∞) → [0, 1] such that
c
ϕ ≡ 1 on [0, s], ϕ ≡ 0 on [r, ∞), |ϕ | ≤ .
r −s
For any x0 ∈ Ω and r > 0 such that Br (x0 ) ⊂ Ω we let
η(y) = ϕ(y −1 ◦ x0 ), (2.7)
then Bs ≺ η ≺ Br and
c c
|X j η| ≤ , j = 1, 2, . . . m; |X 0 η| ≤ . (2.8)
r −s s(r − s)
If u is a solution of (1.1), we have
L (ηu) = divX (G) + g,

where
G = ηF + u AX η = (G 1 , . . . , G m ), g = AXu, X η − F, X η + u X 0 η. (2.9)
594 G. Du et al.

p,λ p,ν
Note that if F1 , . . . , Fm , u ∈ L loc (Ω) and X u ∈ L loc (Ω) with 1 < p < ∞, 0 ≤ ν ≤ λ < Q,
then G 1 , . . . , G m ∈ L p,λ (Br ) and g ∈ L p,ν (Br ), and there exists c = c(r, s) > 0 such that
 
G L p,λ (Br ) ≤ c u L p,λ (Br ) + F L p,λ (Br ) ,  (2.10)
g L p,ν (Br ) ≤ c u L p,ν (Br ) + X u L p,ν (Br ) + F L p,λ (Br ) ,


m 
m
where G L p,λ (Br ) = G j  L p,λ (Br ) , X u L p,ν (Br ) = X j u L p,ν (Br ) .
j=1 j=1

Lemma 7 [6] Let u be a weak solution of (1.1) and η, G 1 , . . . , G m , g are the functions
defined above. Then for j = 1, . . . , m,


m ˆ
X j (ηu)(x) = − P.V. Γ j h (x; y −1 ◦ x)[(ahk (x) − ahk (y))X k (ηu)(y) + G h (y)]dy
h,k=1 G
ˆ 
m
+ Γ j (x; y −1 ◦ x)g(y)dy + α j h (x)G h (x), (2.11)
G h=1

where α j h ’s are some uniformly bounded functions.

3 Higher L p estimates
p
Assume that u is a weak solution to (1.1). In this section we show u and X u ∈ L loc (Ω),
p ∈ (1, 2∗ ]. To do this, we first prove a Caccioppoli type inequality and a Sobolev type
inequality.

Lemma 8 (Caccioppoli type inequality) If u is a weak solution to (1.1), then for any B R ⊂⊂
Ω and γ ∈ (0, 1),
ˆ ˆ ˆ
c
|X u| d x ≤ 2
2
|u| d x + c
2
|F|2 d x. (3.1)
Bγ R R BR BR

Proof In the integral identity (1.4) we choose ϕ = η2 u, where η is a cutoff function with
Bγ R ≺ η ≺ B R . Then
ˆ ˆ ˆ
AXu, X uη2 d x = − 2 AXu, X ηuηd x + η2 u X 0 ud x
BR BR BR
ˆ ˆ (3.2)
+ F, X uη2 d x + 2 F, X ηuηd x.
BR BR

Using the identity

1
η2 u X 0 u = X 0 (u 2 η2 ) − u 2 ηX 0 η
2
and the divergence theorem, it follows
ˆ ˆ
η2 u X 0 ud x = − u 2 ηX 0 ηd x.
BR BR
Interior regularity for degenerate equations with drift. . . 595

The assumption (H) and Young inequality imply

ˆ
μ−1 η2 |X u|2 d x
BR
ˆ ˆ ˆ
≤2 |AXu, X η||uη|d x + u 2 |ηX 0 η|d x + |F, X u|η2 d x
BR BR BR
ˆ
+2 |F, X η||uη|d x
BR
ˆ ˆ ˆ ˆ
≤ cε η2 |X u|2 d x + cε u 2 |X η|2 d x + u 2 |ηX 0 η|d x + cε η2 |F|2 d x
BR BR BR BR
ˆ ˆ ˆ
cε
≤ cε η2 |X u|2 d x + 2 u 2 d x + cε |F|2 d x. (3.3)
BR R BR BR

Taking ε small enough such that μ−1 − cε > 0, our result follows. 


Since Γ 0 (x; y) = Γ 0 (y −1 ◦ x) is the fundamental solution of the operator L 0 , then the

following result is a direct consequence of Lemma 3.

Lemma 9 If we define that for any f ∈ L 2 (G) and x ∈ G,

ˆ ˆ
T ( f )(x) = Γ 0 (y −1 ◦ x) f (y)dy, T j ( f )(x) = Γ j0 (y −1 ◦ x) f (y)dy, j = 1, . . . , m,
G G

where Γ j0 = X j Γ 0 , then there exists a positive constant c = c(Q) such that

T ( f ) 2Q ≤ c  f  L 2 (G) , (3.4)
L Q−4 (G)
Tj ( f ) ∗
L 2 (G)
≤ c  f  L 2 (G) , j = 1, . . . , m. (3.5)

Before proving the Sobolev type inequality for weak solutions, we first make some obser-
vations. If u is a solution to (1.1), then u satisfies

L 0 u = L u + (L 0 − L )u = div X ( F̃), (3.6)


m
where F̃ j = F j + X j u − a jk X k u, j = 1, . . . , m. On the other hand, set
k=1

v(x) = η(x)u(x), (3.7)

where η satisfies
c c
0 ≤ η ≤ 1, η ≡ 1 on Bs , supp η ⊂ Br , |X η| ≤ and |X 0 η| ≤ . (3.8)
r −s s(r − s)
Then
L 0 v = u L 0 η + 2X u, X η + div X (η F̃) −  F̃, X η. (3.9)
p
Theorem 3 (Sobolev type inequality) Let u be a weak solution to (1.1), then u ∈ L loc (Ω),
p ∈ (1, 2∗ ], and for any 0 < r ≤ 1 such that Br ⊂⊂ Ω, we have
c  
u L 2∗ (Bs ) ≤ u L 2 (Br ) + X u L 2 (Br ) + F L 2 (Br ) , ∀ 0 < s < r. (3.10)
r −s
596 G. Du et al.

Proof Let Br ⊂⊂ Ω, 0 < s < r ≤ 1, v(x) = η(x)u(x) with η satisfying (3.8). For every
x ∈ Bs , we have by (3.9) and Lemma 2 that
ˆ
 
u(x) = Γ 0 (x; ·)L 0 v (y)dy
ˆG    
= Γ 0 (x; ·) u L 0 η + 2X u, X η −  F̃, X η − X Γ 0 (x; ·), η F̃ (y)dy
ˆG ˆ
 0   0 
= uΓ (x; ·)L 0 η (y)dy + 2 Γ (x; ·)X u, X η (y)dy
G G
ˆ   ˆ  
− Γ (x; ·) F̃, X η (y)dy −
0
X Γ 0 (x; ·), η F̃ (y)dy. (3.11)
G G

To the first term in the right hand side of (3.11), we use L 0 η = div X (X η) + X 0 η and then
integrate by parts to get
ˆ
 
uΓ 0 (x; ·)L 0 η (y)dy
G
ˆ
 
= −Γ 0 (x; ·)X η, X u − X Γ 0 (x; ·), u X η + uΓ 0 (x; ·)X 0 η (y)dy. (3.12)
G

Using F̃ = F + X u − AXu, the third and fourth term in the right hand side of (3.11)
becomes respectively
ˆ  
− Γ 0 (x; ·) F̃, X η (y)dy
G
ˆ
 
= −Γ 0 (x; ·)F, X η − Γ 0 (x; ·)X u, X η + Γ 0 (x; ·)AXu, X η (y)dy
G
(3.13)

and
ˆ  
− X Γ 0 (x; ·), η F̃ (y)dy
G
ˆ
 
=− X Γ 0 (x; ·), ηF + X Γ 0 (x; ·), ηX u − X Γ 0 (x; ·), η AXu (y)dy
ˆG
= − X Γ 0 (x; ·), ηF + X Γ 0 (x; ·), η(A − Im )X udy, (3.14)
G

where Im denotes the m × m identity matrix.

Taking (3.12)–(3.14) into (3.11), one gets
ˆ
 0 
u(x) = Γ (x; ·) (u X 0 η − F, X η + AXu, X η) (y)dy,
G
ˆ
 
+ −X Γ 0 (x; ·), u X η − X Γ 0 (x; ·), ηF + X Γ 0 (x; ·), η(A − Im )X u (y)dy
G
:= I1 (x) + I2 (x). (3.15)
Interior regularity for degenerate equations with drift. . . 597

From Lemma 9 and supp η ⊂ Br , we have

1
I1 (x) L 2∗ (Bs ) ≤ |Bs | Q I1 (x) 2Q
L Q−4 (Bs )
 
≤ cs u X 0 η L 2 (G) + F, X η L 2 (G) + AXu, X η L 2 (G)
c  
≤ u L 2 (Br ) + F L 2 (Br ) + X u L 2 (Br ) .
r −s
Similarly,
I2 (x) L 2∗ (Bs ) ≤ cu X η L 2 (G) + cηF L 2 (G) + cηX u L 2 (G)
c  
≤ u L 2 (Br ) + F L 2 (Br ) + X u L 2 (Br ) .
r −s
Thus (3.10) is proved. 


Theorem 4 Let u be a weak solution to (1.1), then for any p ∈ (1, 2∗ ], we have X u ∈
p
L loc (Ω). Furthermore, for any Br ⊂⊂ Ω (r ≤ 1) and 0 < s < r/2, we have
 
X u L p (Bs ) ≤ c u L 2 (Br ) + F L p (Br ) , (3.16)
where c = c(s, r ).

Proof Using (1.6), we have that for any p ∈ (1, 2∗ ] and 0 < s < r/2 (r ≤ 1),
 
X u L p (Bs ) ≤ c u L p (Br/2 ) + F L p (Br/2 ) . (3.17)

Noting (3.10) and (3.1), it implies

 
u L p (Br/2 ) ≤ c u L 2 (B3r/4 ) + X u L 2 (B3r/4 ) + F L 2 (B3r/4 )
 
≤ c u L 2 (B3r/4 ) + u L 2 (Br ) + F L 2 (Br ) + F L 2 (B3r/4 )
 
≤ c u L 2 (Br ) + F L p (Br ) , (3.18)
where c = c(s, r ). Now (3.16) follows by putting (3.18) into (3.17). 


4 Proof of Theorem 1

In order to prove Theorem 1, some preliminary results are needed.

p,ν
Lemma 10 Let u be a weak solution to (1.1). If u, X u and F ∈ L loc (Ω), 1 < p < ∞,
p,μ
0 ≤ ν < Q, then u ∈ L loc (Ω) for μ ∈ [0, Q) satisfying μ ≤ p + ν. Moreover, for every
Br ⊂⊂ Ω, s < r , there exists a positive constant c = c(s, r ) such that
 
u L p,μ (Bs ) ≤ c u L p,ν (Br ) + X u L p,ν (Br ) + F L p,ν (Br ) . (4.1)

Proof Let η be a cutoff function with Bs ≺ η ≺ Br , then by (3.15),

 
u(x) L p,μ (Bs ) ≤ c I1 (x) L p,μ (Bs ) + I2 (x) L p,μ (Bs ) .
Qp Qμ
, Q+2
To estimate I1 (x) L p,μ (Bs ) , we first claim by Lemma 6 that if f ∈ L Q+2 p p (G), then
ˆ
 0 
Γ (x; ·) f (y)dy ≤ c f  Qp , Qμ . (4.2)
G L p,μ (G) L Q+2 p Q+2 p (G)
598 G. Du et al.

Qp Qμ
, Q+2
Next, if f ∈ L p,ν (Br ), then we see by Lemma 1 that f ∈ L Q+2 p p (Br ) and
f Qp Qμ
, Q+2 p
≤ c  f  L p,ν (Br ) . (4.3)
L Q+2 p (Br )

It follows from (4.2), (4.3) and suppη ⊂ Br that

I1 (x) L p,μ (Bs ) ≤ AXu, X η − F, X η + u X 0 η Qp Qμ
,
L Q+2 p Q+2 p (G)

≤ c X u Qp , Qμ + F Qp , Qμ + u Qp , Qμ
L Q+2 p Q+2 p (Br ) L Q+2 p Q+2 p (Br ) L Q+2 p Q+2 p (Br )
 
≤ c X u L p,ν (Br ) + F L p,ν (Br ) + u L p,ν (Br ) . (4.4)
Now we estimate I2 (x) L p,μ (Bs ) . Since Γ j0 (y −1 ◦ x) is homogeneous of degree 1 − Q,
Qp Qμ
, Q+
one concludes from Lemma 6 that if f ∈ L Q+ p p (G), then
ˆ  
Γ j0 (x; ·) f (y)dy ≤ f Qp Qμ
, Q+ p
. (4.5)
G L p,μ (G) L Q+ p (G)

On the other hand, if f ∈ L p,ν (Br ), then we know by Lemma 1 and μ ≤ p + ν that
Qp
, Qμ
f ∈ L Q+ p Q+ p (Br ) and

f Qp ≤ c  f  L p,ν (Br ) . (4.6)

, Qμ
L Q+ p Q+ p (Br )

As a result of (4.5) and (4.6), we have

I2 (x) L p,μ (Bs ) ≤ c ηX u Qp , Qμ + ηF Qp , Qμ + u X η Qp , Qμ

L Q+ p Q+ p (G) L Q+ p Q+ p (G) L Q+ p Q+ p (G)

≤ c X u Qp Qμ + F Qp Qμ + u Qp Qμ
, , ,
L Q+ p Q+ p (Br ) L Q+ p Q+ p (Br ) L Q+ p Q+ p (Br )
 
≤ c X u L p,ν (Br ) + F L p,ν (Br ) + u L p,ν (Br ) . (4.7)
Then (4.1) follows by (4.4) and (4.7). 

Lemma 11 Let u be a weak solution to (1.1) and Br ⊂⊂ Ω. Suppose that F ∈ L p,λ (Br ),
u ∈ L p,μ (Br ) and X u ∈ L p,ν (Br ) with 1 < p < ∞ and 0 ≤ ν ≤ λ < Q, then there exists
p,μ
r0 > 0 such that X j u ∈ L loc (Br ), r ≤ r0 , μ = min{λ, ν + p}, j = 1, ..., m, and for any
0 < s < r,
 
X u L p,μ (Bs ) ≤ c u L p,μ (Br ) + X u L p,ν (Br ) + F L p,λ (Br ) , (4.8)
where c = c(r, s).
Proof If ν ≥ μ, (4.8) is obvious; the remaining task is to treat the case ν < μ.
We start to prove (4.8) assuming X u ∈ L p,μ (Br ).
Let v(x) = η(x)u(x), where η is the cutoff function satisfying Bs ≺ η ≺ Br . Setting
ˆ
T j g(x) = Γ j (x; y −1 ◦ x)g(y)dy,
G
where g is in (2.9) and using the notations defined in (2.5) and (2.6), the representation
formula (2.11) can be rewritten as
Interior regularity for degenerate equations with drift. . . 599


m 
m 
m
X j v(x) = C j h [ahk , X k v](x) − T j h (G h )(x) + T j g(x) + α j h G h (x). (4.9)
h,k=1 h=1 h=1

Since α j h ’s are uniformly bounded, it follows by using Lemma 4 that

⎛ ⎞

m 
m
X j v L p,μ (Br ) ≤ c ⎝ ahk ∗ X k v L p,μ (Br ) + G h  L p,μ (Br ) + T j g L p,μ (Br ) ⎠ .
h,k=1 h=1
(4.10)
Due to (4.5) and (4.6), we have
T j g L p,μ (Br ) ≤ cg Qp Qμ ≤ cg L p,ν (Br ) . (4.11)
,
L Q+ p Q+ p (Br )

Putting (4.11) into (4.10),

⎛ ⎞

m 
m
X j v L p,μ (Br ) ≤ c ⎝ ahk ∗ X k v L p,μ (Br ) + G h  L p,μ (Br ) + g L p,ν (Br ) ⎠ .
h,k=1 h=1

Making use of the VMO assumption on the coefficients ahk , we know from Lemma 5 that
there exists r0 > 0 such that for any r ≤ r0 ,
 
X j v L p,μ (Br ) ≤ c G L p,μ (Br ) + g L p,ν (Br ) . (4.12)
Since ν < μ ≤ λ and η is a cutoff function on Br , (4.8) immediately follows from (4.12)
and (2.10).
We next remove the assumption that X u ∈ L p,μ (Br ). Observe that the map
 m  m
T̃ : L p,ν (Br ) → L p,ν (Br )
defined by

m 
m
(T̃ U ) j (x) = C j h [ahk , Uk ](x) − T j h (G h )(x) + T j g(x)
h,k=1 h=1

m
+ α j h G h (x), j = 1, . . . , m,
h=1
 m
for every U ∈ L p,ν (Br ) , is a contraction. In fact, noting Lemma 5 and


m
 
(T̃ U ) j − (T̃ W ) j = C j h [ahk , Uk ] − C j h [ahk , Wk ]
L p,ν (Br )
h,k=1 L p,ν (Br )


m
= C j h [ahk , (Uk − Wk )]
h,k=1 L p,ν (Br )
≤ cεU − W  L p,ν (Br ) ,

we see that T̃ is a contraction by taking ε such that cε < 1. Hence we obtain from (4.9) that
X v is the unique fixed point of T̃ in (L p,ν (Br ))m . On the other hand, T̃ is also a contraction
in (L p,μ (Br ))m , then T̃ has a unique fixed point U ∗ ∈ (L p,μ (Br ))m . Since (L p,μ (Br ))m ⊂
(L p,ν (Br ))m , it means that X v must coincide with U ∗ . The proof is complete. 

600 G. Du et al.

Proof of Theorem 1 Fixed x0 ∈ Ω, let Br = Br (x0 ) ⊂⊂ Ω and p ∈ (1, 2∗ ]. We consider

the balls centered at x0 with radii ρ1 , ρ2 and ρ3 , where ρ3 < ρ2 < ρ1 < r/2.
By (3.16), we have
 
X u L p (Bρ1 ) ≤ c u L 2 (Br ) + F L p (Br ) . (4.13)

Combining (4.13) with (4.1) (letting ν = 0 and μ = min{ p, λ}), we deduce

 
u L p,μ (Bρ2 ) ≤ c u L p (Bρ1 ) + X u L p (Bρ1 ) + F L p (Bρ1 )
 
≤ c u L p (Bρ ) + u L 2 (Br ) + F L p (Br ) .
1

By virtue of (3.18),
 
u L p (Bρ ≤ c u L 2 (Br ) + F L p (Br ) ,
1)

and so  
u L p,μ (Bρ2 ) ≤ c u L 2 (Br ) + F L p (Br ) . (4.14)

Using Lemma 11 with ν = 0, (4.13) and (4.14), it follows

 
X u L p,μ (Bρ3 ) ≤ c u L p,μ (Bρ2 ) + X u L p (Bρ2 ) + F L p,λ (Bρ )
2
 
≤ c u L 2 (Br ) + F L p (Br ) + F L p,λ (Bρ )
2
 
≤ c u L 2 (Br ) + F L p,λ (Br ) . (4.15)

If λ ≤ p, then μ = min{ p, λ} = λ and the required conclusion is obtained, otherwise we

proceed as above, choosing ρ4 , ρ5 ∈ R such that ρ5 < ρ4 < ρ3 and using Lemmas 10 and
11 with ν = p. We have
 
u L p,μ1 (Bρ ) ≤ c u L 2 (Br ) + F L p,λ (Br ) (4.16)
5

and  
X u L p,μ1 (Bρ ) ≤ c u L 2 (Br ) + F L p,λ (Br ) , (4.17)
5

where μ1 = min{λ, 2 p}. If λ ≤ 2 p, then μ1 = min{2 p, λ} = λ, the proof is ended;

if λ > 2 p, we can continue to apply the previous technique. When one chooses ν = 0,
ν = p, ν = 2 p, . . . in using (4.1) and (4.8), since p is a constant, we surely arrive to
μ = min{λ, ν} = λ after finite steps and so Theorem 1 is proved. 


5 Proof of Theorem 2

To begin with, let us recall a mean value theorem for homogeneous functions which was
proved in [8, Proposition 1.15] (see also [2, Proposition 1]).

Lemma 12 Let f ∈ C 1 (G \ {0}) be homogeneous of degree κ < 1. Then there exist two
constants c = c(G, f ) > 0 and M = M(G) > 1 such that

| f (y) − f (x)| ≤ cx −1 ◦ yxκ−1 ,

for any x, y ∈ G such that x ≥ Mx −1 ◦ y.

Interior regularity for degenerate equations with drift. . . 601

Proof of Theorem 2 Let u be a weak solution to (1.1) and B2r ⊂⊂ Ω. Take v(x) = η(x)u(x),
where η is a cutoff function satisfying Bs ≺ η ≺ Br , 0 < s < r . From the representation
formula (3.15), we have that for any x, z ∈ Bs ,
ˆ 
  0  
|u(x) − u(z)| ≤  Γ (x; y) − Γ 0 (z; y) H (y)dy 
G
m ˆ 
  
 
+  Γ 0
(x; y) − Γ 0
(z; y) N (y)dy , (5.1)
 j j j 
j=1 G


m
where H = AXu, X η − F, X η + u X 0 η, N j = η a jk X k u − ηX j u − ηF j − u X j η.
k=1
We first consider one of the integrals in the last sum. It follows
ˆ   
 
 Γ 0
(x; y) − Γ 0
(z; y) N (y)dy 
 j j j 
G
ˆ
≤ |Γ j0 (x; y) − Γ j0 (z; y)||N j (y)|dy
y −1 ◦x≥Mx −1 ◦z
ˆ ˆ
+ |Γ j (x; y)||N j (y)|dy +
0
|Γ j0 (z; y)||N j (y)|dy
y −1 ◦x<Mx −1 ◦z y −1 ◦x<Mx −1 ◦z
:= L 1 + L 2 + L 3 . (5.2)

Since Γ j0 is homogeneous of degree (1 − Q), we have by Lemma 12 that

ˆ
x −1 ◦ z
L1 ≤ c |N j (y)|dy. (5.3)
y −1 ◦x≥Mx −1 ◦z y −1 ◦ x Q

Defining

E k = {y ∈ G : 2k−1 Mx −1 ◦ z ≤ y −1 ◦ x ≤ 2k Mx −1 ◦ z}, k = 1, 2, . . . ,

we have by the Hölder inequality that

∞ ˆ
 x −1 ◦ z
L1 ≤ c |N j (y)|dy
Ek y −1 ◦ x Q
k=1
∞ Q ˆ
2
≤c x −1 ◦ z |N j (y)|dy
2 Mx −1 ◦ z
k
B2k Mx −1 ◦z (x)
k=1
∞
2 Q   Q(1− 1 )
x −1 ◦ z 2k Mx −1 ◦ z
p
≤c N j  L p (B2k Mx −1 ◦z (x))
2k Mx −1 ◦ z
k=1
∞  − Q
2k Mx −1 ◦ z x −1 ◦ zN j  L p (B2k Mx −1 ◦z (x)) .
p
≤c
k=1

Noting
 λ
N j  L p (B2k Mx −1 ◦z (x)) ≤ 2k Mx −1 ◦ z N j  L p,λ (G) ,
p
602 G. Du et al.

it follows
∞ 
 − Q  λ
2k Mx −1 ◦ z x −1 ◦ z 2k Mx −1 ◦ z N j  L p,λ (G)
p p
L1 ≤ c
k=1
∞
 k
λ−Q p+λ−Q
≤c 2 p x −1 ◦ z p N j  L p,λ (G)
k=1
p+λ−Q
≤ cx −1 ◦ z p N j  L p,λ (G) , (5.4)

where we used the fact that the series in (5.4) is convergent from λ < Q.
To estimate L 2 , we denote
 
k = y ∈ G : 2−k Mx −1 ◦ z ≤ y −1 ◦ x ≤ 21−k Mx −1 ◦ z , k = 1, 2, . . .
E

and have from (2.4) that

ˆ
1
L2 ≤ |N j (y)|dy
y −1 ◦x<Mx −1 ◦z y −1 ◦ x Q−1
∞ ˆ
 1
≤c |N j (y)|dy
y −1 ◦ x Q−1
k
E
k=1
∞ Q−1 ˆ
2
≤c |N j (y)|dy
21−k Mx −1 ◦ z B21−k Mx −1 ◦z (x)
k=1
∞
 k
Q−( p+λ) p+λ−Q
≤ 2 p x −1 ◦ z p N j  L p,λ (G) .
k=1

The series is convergent by p + λ > Q and so

p+λ−Q
L 2 ≤ cx −1 ◦ z p N j  L p,λ (G) . (5.5)

For L 3 , we first note that if y −1 ◦ x < Mx −1 ◦ z, then y −1 ◦ z < c(y −1 ◦ x +
x −1◦ z) ≤ c(1 + M)x −1 ◦ z. So we can use the same argument to get
p+λ−Q
L 3 ≤ cx −1 ◦ z p N j  L p,λ (G) . (5.6)

Putting (5.4), (5.5) and (5.6) into (5.2), we obtain

ˆ   
  p+λ−Q
 Γ 0
(x; y) − Γ 0
(z; y) N (y)dy  ≤ cx −1 ◦ z p N j  L p,λ (G) . (5.7)
 j j j 
G

To the first integral in the right hand side of (5.1), we have

ˆ 
  0  
 Γ (x; y) − Γ (z; y) H (y)dy 
0

G
ˆ
≤ |Γ 0 (x; y) − Γ 0 (z; y)||H (y)|dy
y −1 ◦x≥Mx −1 ◦z
ˆ ˆ
+ |Γ 0 (x; y)||H (y)|dy + |Γ 0 (z; y)||H (y)|dy
y −1 ◦x<Mx −1 ◦z y −1 ◦x<Mx −1 ◦z
Interior regularity for degenerate equations with drift. . . 603

and use the similar proceed to obtaining (5.7) to show

ˆ 
  0   2q+μ−Q
 Γ (x; y) − Γ (z; y) H (y)dy  ≤ cx −1 ◦ z q H  L q,μ (G) .
0

G
pQ
Taking q = p+Q and μ = λq
p , we have
ˆ 
  0   p+λ−Q
 Γ (x; y) − Γ (z; y) H (y)dy  ≤ cx −1 ◦ z p H  L p,λ (G) .
0
(5.8)

G

Putting (5.7) and (5.8) into (5.1) yields

p+λ−Q p+λ−Q 
m
|u(x) − u(z)| ≤ cz −1 ◦ x p H  L p,λ (G) + cz −1 ◦ x p N j  L p,λ (G) . (5.9)
j=1

As a result of Theorem 1,
H  L p,λ (G) = AXu, X η − F, X η + u X 0 η L p,λ (G)
 
≤ c X u L p,λ (Br ) + F L p,λ (Br ) + u L p,λ (Br )
 
≤ c u L 2 (B2r ) + F L 2∗ ,λ (Ω)

and

m
N j  L p,λ (G) = η AXu − ηX u − ηF − u X η L p,λ (G)
k=1
 
≤ c X u L p,λ (Br ) + F L p,λ (Br ) + u L p,λ (Br )
 
≤ c u L 2 (B2r ) + F L 2∗ ,λ (Ω) .

Combining the above two inequalities with (5.9), one gets

p+λ−Q  
|u(x) − u(z)| ≤ cz −1 ◦ x p u L 2 (B2r ) + F L 2∗ ,λ (Ω) .

The proof of Theorem 2 is complete. 



References
1. Bonfiglioli, A., Lanconelli, E., Uguzzoni, F.: Stratified Lie Groups and Potential Theory for Their Sub-
Laplacians. Springer, New York (2007)
2. Bramanti, M., Brandolini, L.: L p estimates for nonvariational hypoelliptic operators with VMO coeffi-
cients. Trans. Am. Math. Soc. 352(2), 781–822 (2000)
3. Bramanti, M., Cerutti, M., Manfredini, M.: L p estimates for some ultraparabolic operators with discon-
tinuous coefficients. J. Math. Anal. Appl. 200(2), 332–354 (1996)
4. Bramanti, M., Zhu, M.: L p and Schauder estimates for nonvariational operators structured on Hörmander
vector fields with drift. Anal. PDE 6(8), 1793–1855 (2013)
5. Cinti, C., Polidoro, S.: Pointwise local estimates and Gaussian upper bounds for a class of uniformly
subelliptic ultraparabolic operators. J. Math. Anal. Appl. 338(2), 946–969 (2008)
6. Feng, X., Niu, P.: Interior regularity for degenerate elliptic equations with drift on homogeneous groups.
J. Lie Theory 23(3), 803–825 (2013)
7. Feng, X., Niu, P.: Local Sobolev-Morrey estimates for nondivergence operators with drift on homogeneous
groups. Rev. R. Acad. Cienc. Exactas Fís. Nat. Ser. A Mat. RACSAM 108(2), 683–709 (2014)
8. Folland, G.B.: Subelliptic estimates and function spaces on nilpotent Lie groups. Ark. Mat. 13, 161–207
(1975)
604 G. Du et al.

9. Hörmander, L.: Hypoelliptic second order differential equations. Acta Math. 119, 147–171 (1967)
10. Hou, Y., Niu, P.: Weighted Sobolev–Morrey estimates for hypoelliptic operators with drift on homoge-
neous groups. J. Math. Anal. Appl. 428(2), 1319–1338 (2015)
11. Lanconelli, E., Polidoro, S.: On a class of hypoelliptic evolution operators. Rend. Semin. Mat. Torino
52(1), 29–63 (1994)
12. Lu, G.: Embedding theorems on Campanato–Morrey spaces for vector fields of Hörmander type. Approx.
Theory Appl. 14(1), 69–80 (1998)
13. Manfredini, M.: The Dirichlet problem for a class of ultraparabolic equations. Adv. Differ. Equ. 2(5),
831–866 (1997)
14. Manfredini, M., Polidoro, S.: Interior regularity for weak solutions of ultraparabolic equations in diver-
gence form with discontinuous coefficients. Boll. Unione Mat. Ital. Sez. B, Artic. Ric. Mat. 1(8), 651–675
(1998)
15. Pascucci, A.: Kolmogorov equations in physics and in finance. Elliptic and Parabolic Problems. A Special
Tribute to the Work of Haim Brezis, pp. 353–364. Birkhäuser, Basel (2005)
16. Polidoro, S.: On a class of ultraparabolic operators of Kolmogorov–Fokker–Planck type. Matematiche
49(1), 53–105 (1994)
17. Polidoro, S.: Uniqueness and representation theorems for solutions of Kolmogorov–Fokker–Planck equa-
tions. Rend. Mat. Appl., VII. Ser 15(4), 535–560 (1995)
18. Polidoro, S.: A global lower bound for the fundamental solution of Kolmogorov–Fokker–Planck equations.
Arch. Ration. Mech. Anal. 137(4), 321–340 (1997)
19. Polidoro, S., Ragusa, M.A.: Sobolev–Morrey spaces related to an ultraparabolic equation. Manuscr. Math.
96(3), 371–392 (1998)
20. Polidoro, S., Ragusa, M.A.: Hölder regularity for solutions of ultraparabolic equations in divergence form.
Potential Anal. 14(4), 341–350 (2001)
21. Polidoro, S., Ragusa, M.A.: Harnack inequality for hypoelliptic ultraparabolic equations with a singular
lower order term. Rev. Mat. Iberoam. 24(3), 1011–1046 (2008)
22. Ragusa, M.A.: On weak solutions of ultraparabolic equations. Nonlinear Anal. 47(1), 503–511 (2001)
23. Stein, E.M.: Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscillatory Integrals. Prince-
ton University Press, Princeton, NJ (1993)
24. Zhu, M., Niu, P.: Interior W 1, p regularity and Hölder continuity of weak solutions to a class of divergence
Kolmogorov equations with discontinuous coefficients. Milan J. Math. 81(2), 317–346 (2013)