J. Math. Anal. Appl.

368 (2010) 677–691

Contents lists available at ScienceDirect

Journal of Mathematical Analysis and

Applications
www.elsevier.com/locate/jmaa

Rough singular integrals supported on submanifolds ✩

Yanping Chen a , Yong Ding b,∗ , Honghai Liu c
a
Department of Mathematics and Mechanics, Applied Science School, University of Science and Technology Beijing, Beijing 100083, PR China
b
School of Mathematical Sciences, Laboratory of Mathematics and Complex Systems (BNU), Ministry of Education of China, Beijing Normal University,
Beijing 100875, PR China
c
School of Mathematical Sciences, Beijing Normal University, Beijing 100875, PR China

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, the authors study the boundedness of the singular integral operators
Received 2 September 2009 associated to submanifolds on Triebel–Lizorkin spaces and Besov spaces. Moreover, the
Available online 24 February 2010 authors also establish an lq -valued inequality for the maximal operators associated to
Submitted by M. Milman
submanifolds in this paper.
Keywords:
© 2010 Elsevier Inc. All rights reserved.
Rough singular integrals
Triebel–Lizorkin spaces
Maximal operators
Vector-valued norm inequality

1. Introduction

In 1986, E.M. Stein [12] studied the L p boundedness of the singular integral operator along polynomials T P defined by

 
T P f (x) = p.v. f x − P ( y ) K ( y ) dy , (1.1)
Rn

where x ∈ Rd , P = ( P 1 , P 2 , . . . , P d ) is a mapping from Rn into Rd with P j being polynomials in y ∈ Rn and K is the

Calderón–Zygmund standard kernel, which means that K is homogeneous of degree −n, smooth away from the origin and
has vanishing mean-valued.

Theorem A. (See [12] or see also [11].) T P is bounded on L p (Rd ) for 1 < p < ∞. The bounds of T P may depend on deg( P j ), but it is
independent of the coefficients of P j .

In 1997, Fan and Pan [6] proved that the smoothness condition assumed on the kernel K is not necessary for the L p
boundedness of the operator T P . To state the results of Fan and Pan, let us recall some definitions. A measurable function b
defined on R+ := (0, ∞) is said to be in  for some 1    ∞, if b satisfies

R
1  
sup b(t ) dt < ∞. (1.2)
R >0 R
0

✩
The research was supported by NSF of China (Grants: 10931001, 10901017) and SRFDP of China (Grant: 20090003110018).
* Corresponding author.
E-mail addresses: yanpingch@126.com (Y. Chen), dingy@bnu.edu.cn (Y. Ding), hhl@mail.bnu.edu.cn (H. Liu).

0022-247X/$ – see front matter © 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.jmaa.2010.02.021
678 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691

For b ∈ 1 , replacing K ( y ) by b(| y |)Ω( y )| y |−n in (1.1), then the rough singular integral operator along polynomials is
defined by

b(| y |)Ω( y )  
TΩ, P f (x) = p.v. f x − P ( y ) dy , (1.3)
| y|n
Rn

where Ω is a homogeneous function of degree zero on Rn with Ω ∈ L 1 (Sn−1 ) and satisfies


   
Ω y  d σ y  = 0. (1.4)
Sn−1

Theorem B. (See [6].) Suppose that Ω ∈ H 1 (Sn−1 ) and b ∈  (R+ ) for some  > 1 (see Section 2 for the definition of the Hardy
space H 1 (Sn−1 )). Then for | 1p − 12 | < min{ 12 , 1 }, there exists a constant C > 0 such that for every f ∈ L p (Rd ),

TΩ, P f L p (Rd )  C Ω H 1 (Sn−1 )  f L p (Rd )

where C = C n,d,b, p ,deg( P j ) is independent of the coefficients of P j .

On the other hand, it is well known that the Triebel–Lizorkin spaces and Besov spaces contain many important function
spaces, such as Lebesgue spaces, Hardy spaces, Sobolev spaces and Lipschitz spaces. The homogeneous Triebel–Lizorkin
p ,q p ,q
space Ḟ α (Rd ) and homogeneous Besov space Ḃ α (Rd ) are defined as following. For 0 < p , q  ∞ ( p = ∞) and α ∈ R,
p ,q d
Ḟ α (R ) is defined by
 
1/q  
p ,q      − j αq 
Ḟ α Rd
=  d 
f ∈ S R :  f  Ḟ p,q (Rd ) =  2 |Ψ j ∗ f |q  <∞
α 
j ∈Z L p (Rd )

p ,q
and Ḃ α (Rd ) is defined by
 
1/q 
p ,q    
f ∈ S  Rd :  f  Ḃ p,q (Rd ) = 2− j αq Ψ j ∗ f 
q
Ḃ α Rd = α L p (Rd )
<∞ ,
j ∈Z

j (ξ ) = φ(2 j ξ ) for j ∈ Z and φ ∈ C c∞ (Rd ) satisfies the

where S  (Rd ) denotes the tempered distribution class on Rd , Ψ
conditions:

(i) 0  φ(x)  1;
(ii) supp(φ) ⊂ {x: 1/2  |x|  2};
(iii) φ(x) > c > 0 if 3/5  |x|  5/3.

p ,q p ,q
See [9,10,15,16] for more properties of Ḟ α (Rd ) and Ḃ α (Rd ).
Recently, the previous two authors of this paper gave the boundedness of the rough singular integral operator T Ω with
its kernel Ω ∈ H 1 (Sn−1 ) on the Triebel–Lizorkin spaces and Besov spaces in [3]. See also [1] and [2] for the boundedness
of the rough singular integrals on the Triebel–Lizorkin spaces. The main purpose of this paper is to study the mapping
properties of TΩ, P on Triebel–Lizorkin spaces and Besov spaces. Our main results are as follows:

Theorem 1.1. Suppose Ω ∈ H 1 (Sn−1 ) and b ∈  (R+ ) for some  > 1. Then for | 1p − 12 | < min{ 12 , 1 } and | 1q − 12 | < min{ 12 , 1 },
p ,q
α ∈ R, there exists a constant C > 0 such that for f ∈ Ḟ α (Rd ),
TΩ, P f  Ḟ αp,q (Rd )  C Ω H 1 (Sn−1 )  f  Ḟ αp,q (Rd ) ,

where C = C n,d,b, p ,q,α ,deg( P j ) is independent of the coefficients of P j .

Theorem 1.2. Suppose Ω ∈ H 1 (Sn−1 ) and b ∈  (R+ ) for some  > 1. Then for | 1p − 12 | < min{ 12 , 1 }, 1 < q < ∞, α ∈ R, there
p ,q
exists a constant C > 0 such that for f ∈ Ḟ α (R ), d

TΩ, P f  Ḃ αp,q (Rd )  C Ω H 1 (Sn−1 )  f  Ḃ αp,q (Rd ) ,

where C = C n,d,b, p ,q,α ,deg( P j ) is independent of the coefficients of P j .

 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691 679

p ,2 p ,2
Remark 1.3. Note that Ḟ 0 (Rd ) = Ḃ 0 (Rd ) = L p (Rd ) for 1 < p < ∞, so, Theorems 1.1 and 1.2 are an extension of Theorem B.

In 1986, Stein [12] gave the L p boundedness of the maximal operator along polynomials MP , which is defined by

1   
MP f (x) = sup  f x − P ( y )  dy ,
n
r >0 r
| y |r

where x ∈ Rd , P = ( P 1 , P 2 , . . . , P d ), P j are real-valued polynomials of y ∈ Rn .

To prove our main theorems, we need to set the following lq -valued L p boundedness of the maximal operator MP :

Theorem 1.4. For 1 < p , q < ∞, there is a constant C p ,q such that


1q  
1q 
      
 |MP f i |q  
 C p ,q  | fi|q  for { f i } ∈ L p lq Rd .
  
i ∈Z L p (Rd ) i ∈Z L p (Rd )

Furthermore, the bound on the operator MP can be taken to independent of the coefficients of P j .

The above theorem is an extension of the famous result on the lq -valued L p boundedness of the Hardy–Littlewood
maximal operator obtained by C. Fefferman and E.M. Stein in [7]. Hence, Theorem 1.4 has its own interest.

p ,q p ,q
Remark 1.5. Denote by F α and B α the inhomogeneous Triebel–Lizorkin space and Besov spaces, respectively. Then the
following properties are well known (see [9] or [10], for example):

p ,q p ,q
(a) F α ∼ Ḟ α ∩ L p and  f  F αp,q ∼  f  Ḟ αp,q +  f  L p (α > 0);
p ,q p ,q
(b) B α ∼ Ḃ α ∩ L p and  f  B p,q ∼  f  Ḃ p,q +  f  L p (α > 0).
α α

Hence, by the properties (a) and (b) above and Theorems 1.1, 1.2 and Theorem B, we get the following conclusion
immediately:

Corollary 1.6. Under the same conditions of Theorems 1.1 and 1.2 (for α > 0), the rough operator TΩ, P is also bounded on the
p ,q p ,q
inhomogeneous Triebel–Lizorkin space F α and Besov spaces B α , respectively.

This paper is organized as follows. First, accepting Theorem 1.4, we prove the main results of this paper, i.e. Theorems 1.1
and 1.2, which will be arranged in Section 2. Then, in Section 3, we give the proof of lq -valued norm inequality for the max-
imal operator MP , i.e. Theorem 1.4. In the last section, we show an lq -valued analytic interpolation theorem (Lemma 3.1)
briefly, which will be used in the proof of Theorem 1.4.
Throughout this paper the letter “C ” will stand for a positive constant which is independent of the essential variables
and not necessarily the same one in each occurrence. As usual, | E | denotes the Lebesgue measure of a measurable set E
in Rn and for p  1, p  = p /( p − 1) denotes the dual exponent of p. For matrix A, A t is the transpose of A. Let J is an
invertible transformation, J −1 denote the inverse of J .

2. Proofs of Theorems 1.1 and 1.2

First let us recall the definitions of the Hardy space and H 1 -atoms on the unit sphere Sn−1 . The Hardy space H 1 (Sn−1 )
is defined by
    
              
H 1 n −1
S = ω∈L S 1 n −1  
 ω H 1 (Sn−1 ) =  sup  ω y P r (·) y dσ y  <∞ ,
0r <1 L 1 (Sn−1 )
Sn−1

where P r y  (x ) denotes the Possion kernel on Sn−1 defined by

  1 − r2
P r y  x = , 0  r < 1, and x , y  ∈ Sn−1 .
|r y  − x |n
By [4], if Ω ∈ H 1 (Sn−1 ) satisfying (1.4), then
∞

Ω= λ ja j , (2.1)
j =1
∞
where j =1 |λ j |  C Ω H 1 (Sn−1 ) and each a j is a regular H 1 (Sn−1 ) atom.
680 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691

Definition 2.1. A function a on Sn−1 is called to be a regular atom if there exist ζ ∈ Sn−1 and ρ ∈ (0, 2] such that

(i) supp(a) ⊂ Sn−1 ∩ B (ζ, ρ ), where B (ζ, ρ ) = { y ∈ Rn : | y − ζ | < ρ };

(ii) a L ∞  ρ −n+1 ;

(iii) Sn−1 a( y ) dσ ( y ) = 0.

For s ∈ {1, . . . , d}, define the projection operator πsd on Rd by

πsd (ξ ) = (ξ1 , . . . , ξs ),
. . , ξd ) ∈ Rd . Let ς ∈ C 1 (R+ ), 0  ς (t )  1, supp(ς ) ⊂ {2−l C  t  2l C } for some positive inte-
where ξ = (ξ1 , . . . , ξs , ξs+1 , .
j ∈Z ς j (t ) = 1, where ς j (t ) = ς (2 t ) for j ∈ Z. Set ψ j (ξ ) = ς j (|πs ξ |) for j ∈ Z. Then we have
2 jl d
ger l. Moreover, ς satisfies
the following conclusion:

Lemma 2.2. Define operators S j by ( S j g )∧ (ξ ) = ψ j (ξ ) ĝ (ξ ). Then for 1 < p , q < ∞ and { f k } ∈ L p (lq )(Rd ), we have
 
q/2
1/q  
1/q 
   
 | S f | 2   C | f |q  .
 j k   k 
k∈Z j ∈Z L p (Rd ) k∈Z L p (Rd )

Proof. We can easily find that

  d ∨  
ψ j πs · (x) = 2− jsl Ψ 2− jl x1 , . . . , 2− jl xs ⊗ δRd−s = Ψ j ⊗ δRd−s ,

and Ψ ∈ S (Rs ), where δRs denote the Dirac delta on Rs . Denote x = (x0 , x1 ), where x0 = (x1 , . . . , xs ), x1 = (xs+1 , . . . , xd ).
Then, we have
 
q/2
1/q  p    

   2 q/2 p /q 0 1
 |S j fk| 2  =  (Ψ j ⊗ δRd−s ) ∗ f k (x) dx dx
  p
k∈Z j ∈Z L (Rd ) k∈Z j ∈Z
Rd−s Rs
   

    q/2 p /q 0 1
=  Ψ j ∗ f k ·, x1 x0 2 dx dx
k∈Z j ∈Z
Rd−s Rs
  

  0 1 q p /q 0 1
C  fk x , x  dx dx
k∈Z
Rd−s Rs

1/q  p
 
= C
 | f k |q 
 p . 2
k∈Z L (Rd )

Let us now turn to the proof of Theorem 1.1.

Proof of Theorem 1.1. Let a be a regular atom. Without loss of generality, we may assume that 0 < ρ < 1/4 and ζ = e1 :=
(1, 0, . . . , 0) in Definition 2.1. Write

  a( y  )  
B ( f )(x) = p.v. b | y| f x − P ( y ) dy .
| y |n
Rn

By (2.1), it suffices to show

 B f  Ḟ αp,q (Rd )  C  f  Ḟ αp,q (Rd ) , (2.2)

where C is independent of a and the coefficients of P j .

For k ∈ Z, we define D k = {x ∈ Rn : 2k  |x| < 2k+1 }. For a suitable mapping Γ : Rn → Rd , we define the measures
{σk,Γ }k∈Z on Rd by
 
    a(x )
f dσk,Γ = f Γ (x) b |x| dx.
|x|n
Rd Dk
 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691 681

It is obvious that

    a( y  )
f ∗ σk,P (x) = f x − P ( y) b | y| dy
| y |n
Dk

and

B ( f )(x) = f ∗ σk,P (x).
k∈Z

Below we give some notations, which are same as in [6]. Let deg(P ) = max{deg( P j ) | 1  j  d}. For l ∈ N, Al denote the
class of polynomials of l variables with real coefficients and V l denote the space of real-valued homogeneous polynomials
of degree l on Rn . Then there are integers 0 < l1 < l2 < · · · < lm  deg(P ), and polynomials Q uj ∈ V lu ⊂ An , R j ∈ A1 with
deg( R j )  deg(P ) for 1  u  m, 1  j  d such that

m
 
P (x) = Qu (x) + R |x| ,
u =1

where Q (x) = ( Q 1u (x), Q 2u (x), . . . , Q du (x)), R(t ) = ( R 1 (t ), R 2 (t ), . . . , R d (t )). For j = 1, 2, . . . , d, denote

u

Q uj (x) = b u j β xβ .
|β|=lu

For l ∈ N and α ∈ (N ∪ {0})n with |α | = l, we choose ηl,α (·) ∈ An−1 such that
 α 
x − ηl,α (x̃)  C ρ 4l(n−1)

for x ∈ Sn−1 ∩ B (e1 , ρ ). For each u ∈ {1, . . . , m}, j ∈ {1, . . . , d}, we define quj ∈ An−1 by

quj (x̃) = b u j β ηlu ,β (x̃)
|β|=lu

and set qu (x̃) = (q1u (x̃), q2u (x̃), . . . , qdu (x̃)).

For each fixed u ∈ {1, . . . , m}, there are positive integers ν (u ), 0 < hu,1 < · · · < hu,ν (u) , and polynomials { W ujη | j = 1,
. . . , d; η = 1, . . . , ν (u )} ⊂ An−1 such that

(i) for j ∈ {1, . . . , d},

η ∈ {1, . . . , ν (u )}, W ujη (·) is homogeneous of degree hu,η ;
(ii) for each η ∈ {1, . . . , ν (u )}, there exists at least one j ∈ {1, . . . , d} such that W ujη = 0;
ν ( u )
(iii) for each j ∈ {1, . . . , d}, there is a v uj ∈ R such that quj (x̃) = η=1 W j η (x̃) + v j . For u ∈ {1, . . . , m} and
u u
η ∈ {1, . . . , ν (u )},
we define R (x) and W
u u ,η
(x̃) by
   
Ru (x) = R |x| + |x|lh v h1 , . . . , v dh + Qh (x)
u hm 1hu −1

and
 
W u ,η (x̃) = W 1uη (x̃), . . . , W duη (x̃) .
u
Let M (u ) = h=1 [
ν (h) + 1] for 1  u  m, M (0) = 0, and define Γ0 , . . . , ΓM (m) by


x̃
Γ M (u −1)+θ (x) = Ru (x) + |x|lu W u ,κ
|x|
1κ θ

for 1  u  m, 0  θ  M (u ) − M (u − 1), and Γ M (m) (x) = P (x). Let d(u ) = dim( V lu ) for each u and write
  n     
β ∈ N ∪ {0}  |β| = lu = β(u , 1), . . . , β u , d(u ) .
Hence we can write

d (u )
Q uj (x) = bu js xβ(u ,s)
s =1
682 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691

where bu js = b u j β(u ,s) . Denote by d(u , η) the number of distinct elements in {Θ ∈ (N ∪ {0})n−1 | |Θ| = h u ,η }. For 1  u  m,
1  η  ν (u ) and 1  j  d, write
     
Θ  |Θ| = hu ,η = Θ(u , η, 1), . . . , Θ u , η, d(u , η)
and
d(u ,η)

W ujη (x̃) = w u , j ,η,s x̃Θ(u ,η,s) .
s =1

Define Λ1 , . . . , Λ M (m) ∈ N by

d(u , θ), if 1  θ < M (u ) − M (u − 1), 1  u  m;
Λ M (u −1)+θ =
d(u ), if θ = M (u ) − M (u − 1), 1  u  m.

Following that, we define linear transformations L i : Rd → RΛi for 1  i  M (m) by

⎧
⎨ ( d w u , j ,θ,s ξ j , . . . , d w u , j ,θ,d(u ,θ ) ξ j ), if 1  θ < M (u ) − M (u − 1);
j =1 j =1
L M (u −1)+θ ξ =
⎩ ( d b ξ j , . . . , d b if θ = M (u ) − M (u − 1).
j =1 u j1 j =1 u jd(u ) ξ j ),

For η = 1, . . . , M (m), we let

⎧  
⎪ 1
⎪
⎨ l(η) = lu , δ(η) = hu ,θ , γ (η) = , if η ∈ M (u − 1), M (u ) ;
4h u ,θ lu 
⎪
⎪ 1
⎩ l(η) = lu , δ(η) = 4lu (n − 1), γ (η) = , if η = M (u ).
8lu 
With the notations above, the following estimates are well known (see (7.32) and (7.34) in [6]):
   
σ̂k,Γ (ξ )  C 2kl(η) ρ δ(η) | L η ξ | −γ (η) , (2.3)
η
 
σ̂k,Γ (ξ ) − σ̂k,Γ (ξ )  C 2kl(η) ρ δ(η) | L η ξ |. (2.4)
η η −1

For η ∈ {1, . . . , M (m)}, s(η) = rank( L η ), there are two nonsingular linear transformations H η : Rs(η) → Rs(η) and
G η : Rd → Rd such that
   
 H η π d G η ξ   | L η ξ |  Λη  H η π d G η ξ  for ξ ∈ Rd .
s (η ) s (η )

Fix a function φ ∈ C 0∞ (R) such that φ(t ) = 1 for |t |  1/2 and φ(t ) = 0 for |t |  1. Let ϕ (t ) = φ(t 2 ) and define the measures
{τk,η } by
  
τ̂k,η (ξ ) = σ̂k,Γη (ξ ) ϕ 2kl( j) ρ δ( j) H s( j) πsd( j) G s( j) ξ 
η< j  M (m)
  
− σ̂k,Γη−1 (ξ ) ϕ 2kl( j) ρ δ( j) H s( j) πsd( j) G s( j) ξ 
η−1< j  M (m)

for k ∈ Z and 1  η  M (m) where we use the convention j ∈∅ a j = 1. Then, for ξ ∈ R , k ∈ Z, and 1  η  M (m), we can
d

get the following estimate from (2.3) and (2.4) (see also (7.39) in [6])
      
τ̂k,η (ξ )  C min 2kl(η) ρ δ(η) Λ−1 | L η ξ |, 2kl(η) ρ δ(η) Λ−1 | L η ξ | −1 γ (η) . (2.5)
η η
M (m)
Notice that Γ M (m) (x) = P (x), we find that σk,P = η=1 τk,η and

M
(m)
M (m)
B f (x) = τk,η ∗ f (x) = τk,η ∗ f (x).
k∈Z η =1 η=1 k∈Z

For fixed η ∈ {1, 2, . . . , M (m)}, let B η f (x) = k∈Z τk,η ∗ f (x). Thus, to get (2.2), we just need to prove
 
 
 τk,η ∗ f  =  B η f  Ḟ αp,q (Rd )  C η  f  Ḟ αp,q (Rd ) . (2.6)
 p ,q
k∈Z Ḟ α (Rd )
 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691 683

Letς ∈ C 1 (R+ ), 0  ς (t )  1, supp(ς ) ⊂ {2−l(η) ρ −δ(η) Λη  t  2l(η) ρ −δ(η) Λη }. Moreover, ς satisfies j∈Z ς 2j (t ) = 1,
where ς j (t ) = ς (2 jl(η) t ). Denote ψ j (ξ ) = ς j (|πsd(η) ξ |), and define operators S j by ( S j f )∧ (ξ ) = ψ j (ξ ) f̂ (ξ ). Then we have the

decomposition f = j ∈Z S j S j f and get




j
τk,η ∗ f = τk,η ∗ S j +k S j +k f = S j +k (τk,η ∗ S j +k f ) =: Bη f . (2.7)
k∈Z k∈Z j ∈Z j ∈Z k∈Z j ∈Z

When L η = πsd(η) , by (2.5) we get

         
τ̂k,η (ξ )  C min 2kl(η) ρ δ(η) Λ−1 π d ξ , 2kl(η) ρ δ(η) Λ−1 π d ξ  −1 γ (η) . (2.8)
η s (η ) η s (η )

By (2.7) and Plancherel’s theorem, we have the following L 2 (Rd ) estimate:


 j 2    
Bη f  2 C τ̂k,η (ξ )2  f̂ (ξ )2 dξ. (2.9)
L (Rd )
k∈Z −( j +k+1)l(η) −δ(η)
2 ρ Λη |πsd(η) ξ |2−( j +k−1)l(η) ρ −δ(η) Λη

Thus, if j  −1, then by (2.8) and (2.9) we have


 j 2  
Bη f  2  C 22( j +1)l(η)γ (η)  f̂ (ξ )2 dξ.
L (Rd )
Rd

Similarly, if j  1, we have

 j 2  
Bη f  2  C 22(1− j )l(η)γ (η)  f̂ (ξ )2 dξ.
L (Rd )
Rd

If j = 0, since |τ̂k,η (ξ )|  C η , we can easily get that

 0 2
B f  2  C  f 2L 2 (Rd ) .
η L (Rd )

p ,2
Notice that Ḟ 0 = L p , then we have

 j   j 
Bη f  2,2 =  B η f L 2 (Rd )  C 2−| j |l(η)γ (η)  f  Ḟ 2,2 (Rd ) . (2.10)
Ḟ 0 (Rd ) 0

Following that, we will prove that, for | 1p − 12 | < min{ 12 , 1 }, | 1q − 12 | < min{ 12 , 1 }, α ∈ R, j ∈ Z,

 j 
Bη f  p ,q  C  f  Ḟ αp,q (Rd ) . (2.11)
Ḟ α (Rd )

To do this, it suffices to show that, if there exists a constant C η , independent of j and ρ , such that


1  
1q 
  j q q      
  B η gi    Cη | g |q  for { g i } ∈ L p lq Rd (2.12)
   i 
i ∈Z L p (Rd ) i ∈Z L p (Rd )

holds, then we will get (2.11). In fact, by (2.12), we have


1q 
 j   −i α q  j q 
Bη f  =  2  Ψ ∗ B f 
Ḟ α
p ,q
(Rd )  i η 
i ∈Z L p (Rd )


1 
  j  −i α q q 

=  Bη 2 Ψi ∗ f  

i ∈Z L p (Rd )
684 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691


1 
  −i α q q 

 Cη 2 Ψi ∗ f  

i ∈Z L p (Rd )

= C η  f  Ḟ α
p ,q
(Rd ) .

This is just (2.11). To get (2.12), we need to give the following result:

Proposition 2.3. If | 1p − 12 | < min{ 12 , 1 }, | 1q − 12 | < min{ 12 , 1 } and { gk, j } ∈ L p (lq (l2 ))(Rd ). Then there is a constant C such that

 
2q
1q   
q2
1q 
   
 | τ ∗ g | 2   C | g | 2  .
 k,η k, j   k, j 
j ∈Z k∈Z L p (Rd ) j ∈Z k∈Z L p (Rd )

Proof. The main ideal of the proof is taken from [6]. It is obvious that  (R+ ) ⊆ 2 (R+ ) when   2. Thus, we may
assume that
2 2
1 <  < 2, 2<p< and 2<q< .
2− 2−

For 1  ι  M (m), let Φ ι (ξ 0 ) be a radial function in S (Rs(ι) ), where ξ 0 ∈ Rs(ι) . Define J ι and X ι by
   
J ι f (x) = f G ts(ι) H ts(ι) ⊗ idRd−s(ι) x
and
  
X ι f (x) = J ι−1 Φkι ⊗ δRd−s(ι) ∗ J ι f (x),

where Φkι (x0 ) = [2kl(ι) ρ δ(ι) ]−s(ι) Φ ι (2−kl(ι) ρ −δ(ι) x0 ), x0 ∈ Rs(ι) , G s(ι) and H s(ι) are proper nonsingular linear transformations
on Rd and Rs(ι) , respectively. Notice that, for η  ι  M (m), we have
   
 X ι f (x)  C ι J −1 ◦ ( M (ι) ⊗ id
ι Rd−s(ι) ) ◦ J ι ( f )(x),

where M (ι) denotes the Hardy–Littlewood maximal operator on Rs(ι) . Hence we have
 
q
1   
q
1 
   2 q p   −1 2 2 q  p
  X ι ( gk, j )(·)2  
 C  J ι ◦ ( M (ι) ⊗ idRd−s(ι) ) ◦ J ι ( gk, j )(·) 
  p  p
j ∈Z k∈Z L (Rd ) j ∈Z k∈Z L (Rd )

   
q
p
    0 2 2 q 0 1
 C | J ι|  M (ι) J ι gk, j ·, x1 x  dx dx
j ∈Z k∈Z
Rd−s(ι) Rs(ι)
   
q
p
  0 1 2 2 q 0 1
 Cι| J ι|  J ι gk , j x , x  dx dx
j ∈Z k∈Z
Rd−s(ι) Rs(ι)
 
q
1 
  2 2 q  p

 Cι  gk, j (·)  . (2.13)
 p
j ∈Z k∈Z L (Rd )

On the other hand, by the definitions of X ι , we have

τk,η ∗ f (x) = σk,η ∗ X η+1 ◦ X η+2 · · · X M (m) f (x) − σk,η−1 ∗ X η ◦ X η+1 · · · X M (m) f (x). (2.14)
Thus, by (2.14) and using the estimate (2.13) repeatedly, to prove Proposition 2.3 it remains to show the following estimate
holds for 1  η  M (m):
 
q2
1q   
2q
1q 
   
 | σ ∗ g | 2   Cη | g | 2  . (2.15)
 k,η k, j   k, j 
j ∈Z k∈Z L p (Rd ) j ∈Z k∈Z L p (Rd )

p  q 
Suppose { f j } ∈ L ( 2 ) (l( 2 ) )(Rd ) with { f j } ( 2 ) ( 2 )
p q  1. Then (see (7.7) in [6])
L (l )(Rd )
 
   
σk,η ∗ gk, j (x)2 f j (x) dx  C  gk, j (x)2 ( M̃ η f j )(x) dx, (2.16)
Rd Rd
 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691 685

where
  
   |a( y  )|   2− 
( M̃ η f )(x) = sup f x + Γη ( y ) b | y|  dy .
k∈Z | y |n
2k | y |<2k+1

Notice that |b(·)|2− ∈   (R+ ), by Hölder’s inequality, we have

2−

  
   |a( y  )|   2− 
sup f x + Γη ( y ) b | y |  dy 

k∈Z | y |n
2k | y |<2k+1
 
           
C a y  sup 1  f x + Γη t y  b |t | 2− dt dσ y 
r >0 r
Sn−1 |t |<r
  
2(−1)
          
C a y  sup 1  f x + Γη t y   2(−1) dt dσ y  . (2.17)
r >0 r
Sn−1 |t |<r

For fixed y  ∈ Sn−1 , Γη (t y  ) is a polynomial of t and deg(Γη (t y  ))  deg(P ). Applying Theorem 1.4 and (2.17), for

2(−1)
< u , v < ∞, we have


1v  
1v 
   
 | M̃ f | v   C u,v  | f | v  . (2.18)
 η j   j 
j ∈Z L u (Rd ) j ∈Z L u (Rd )

By (2.16) and (2.18) we obtain

 
2q
1q 2
 
 | σ ∗ g | 2 
 k,η k, j  p
j ∈Z k∈Z L (Rd )

  
= sup σk,η ∗ gk, j (x)2 f j (x) dx
{ f j } ( ) ( )
p q 1 j ∈Z k∈Z
L 2 (l 2 ) Rd
 
q2
1q 2 
1v 
   
C sup  | gk , j | 2   | M̃ η f j | v 
  p  
{ f j } ( ) ( )
p q 1 j ∈Z k∈Z L (Rd ) j ∈Z L u (Rd )
L 2 (l 2 )

 
2q
1q 2
 

 C | gk , j | 2  ,
 p
j ∈Z k∈Z L (Rd )

where we take v = ( 2 ) and u = ( 2 ) . Then, Proposition 2.3 is proved.

q p
2

Let us continue the proof of Theorem 1.1. By Lemma 2.2 and Proposition 2.3, for { g i } ∈ L p (lq )(Rd ), we have


1 
  j q q 
 Bη fi  
 
i ∈Z L p (Rd )
 
 
= sup  τk,η ∗ S j+k f i (x) S j+k hi (x) dx
∗

{hi } ( ) ( )
p q 1 i ∈Z k∈Z
L 2 (l 2 ) Rd
 
q2
1q   
q
1 
    ∗ 2 2 q 
 Cη
 | τk,η ∗ S f
j +k i | 2 



 S hi 
j +k

 
i ∈Z k∈Z L p (Rd ) i ∈Z k∈Z L p (Rd )


1q 
 
 Cη
 | f i |q 
 .
i ∈Z L p (Rd )
686 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691

We therefore get (2.12). Then, by applying an interpolation theorem (see [8] or [10]) between (2.10) and (2.11), it is clear
that for α ∈ R, | 1p − 12 | < min{ 12 , 1 }, | 1q − 12 | < min{ 12 , 1 },
 j 
Bη f  p ,q  C η 2−ε| j |  f  Ḟ αp,q (Rd ) .
Ḟ α (Rd )

Thus, (2.6) follows from this, i.e.,

 
Bη f  p ,q  C η  f  Ḟ αp,q (Rd ) .
Ḟ α (Rd )

Hence, we complete the proof of Theorem 1.1. 2

Proof of Theorem 1.2. The proof is simple. In fact, by Theorem B, for | 1p − 12 | < min{ 12 , 1 }, 1 < q < ∞, α ∈ R, we have

1q
2−i αq Ψi ∗ TΩ, P f  p
q
TΩ, P f  Ḃ αp,q (Rd ) =
i ∈Z

1
  −i α q q
=  TΩ, P 2 Ψi ∗ f 
p
i ∈Z

1q
−i α q q
C 2 Ψi ∗ f p
i ∈Z
= C  f  Ḃ αp,q (Rd ) .
This is just the conclusion of Theorem 1.2. 2

3. Proof of Theorem 1.4: lq -valued inequality for maximal operator

We first prove Theorem 1.4 for the case where P j are all monomials (i.e., Proposition 3.2 below). We give some notations.
Let D = max{deg( P j )}. For α = (α1 , . . . , αn ) with α j ∈ Z+ ∪ {0}, denote |α | = |α1 | + · · · + |αn |. Denote N = {α : 1  |α |  D }
and
    
Λ = α 1 , α 2 , . . . , α N  1  α j   D for 1  j  N .
Moreover, R N = {(xα ) = (xα 1 , . . . , xα N ) | α j ∈ Λ} is the N-dimension Euclid space, whose coordinates are labeled by the
multi-indices α . Then we define the maximal operator Mp corresponding to the polynomial map
   
p : Rn → R N , p(t ) = t α 1|α | D := t 1α1 · · · tnαn 1|α | D
by
  
1   
Mp f (x) = sup n  f x − p(t ) dt .
r r >0
|t |r

By Proposition 2 in [11, p. 477], Mp can be extended as an operator of type ( p , p ) for 1 < p  ∞.

In the proof of Proposition 3.2, we need an lq -valued analytic interpolation conclusion (Lemma 3.1 below). Before stating
this result, let us give some notations. Denote by S the closed strip {a  Re( z)  b} ⊂ C, {U z }z∈ S is a family of uniformly
bounded linear operators on L 2 (Rn ), i.e., there is an M > 0 such that
 z
U   M, ∀z ∈ S .
L2 →L2
Moreover, the function

z → U z ( f )(x) g (x) dx
Rn

is continuous on S and analytic in the interior of S, whenever f , g ∈ L 2 (Rn ).

Lemma 3.1. With the above assumptions, if for 1 < p i , qi < ∞ (i = 0, 1), {U z } satisfies the following conditions:


1 
q1 
  z q0 q0   0 
(i)  U f j   
 M0 | f j|q0  when Re( z) = a;
  
j ∈Z L p0 j ∈Z L p0
 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691 687


1 
q1 
  z q1 q1   1 
(ii)  U f j   
 M1 | f j|q1  when Re( z) = b.
  
j ∈Z L p1 j ∈Z L p1

Then we have

1  
1q 
  a(1−θ )+θ b q q   
 U f    M 01−θ M 1θ  | f |q  , (3.1)
 j   j 
j ∈Z Lp j ∈Z Lp

1 1−θ 1 1−θ
where 0 < θ < 1, q
= q0
+ θ
q1
and p
= p0
+ p1
θ
.

We should point out that Lemma 3.1 is only a consequence of the Banach-valued interpolation theorem of analytic
families of operators, which was given by Cwikel and Janson in [5]. However, we also present a direct simple proof of
Lemma 3.1 in the last section.

Proposition 3.2. For 1 < p , q < ∞, there is a constant C p ,q such that for all { f i } ∈ L p (lq )(R N ),

1q  
1q 
   
 |Mp f i |q  
 C p ,q  | fi|q  .
  
i ∈Z L p (R N ) i ∈Z L p (R N )

Proof. For δ > 0, consider a nonisotropic dilations of R N given by

 1 
x → δ ◦ x = δ |α | xα 1 , . . . , δ |α | xα N ,
N

where x = (xα ) ∈ R N . The corresponding norm on R N is defined by

 j
|

ρ (x) = max |xα j |1/|α , α j ∈ Λ.
1 j  N

With the above definitions, it is easy to check that

ρ (δ ◦ x) = δ ρ (x) for all δ > 0, x ∈ RN

and

δ ◦ p(t ) = p(δt ) for all δ > 0, t ∈ Rn ,

where δt = (δt 1 , . . . , δtn ) is the usual dilation on Rn . Fix a smooth nonnegative function η on Rn with η(t ) = 1 for |t |  1,
and η(t ) = 0 for |t | > 2. Let

   
A j f (x) = 2nj f x − p(t ) η 2 j t dt .
Rn

Define measure dμ on R N by
 
 
f dμ = f p(t ) η(t ) dt .
RN Rn

Similarly, measure dμ j is defined by

 
 
f dμ j = f 2− j ◦ p(t ) η(t ) dt .
RN Rn

For each complex z, we denote by ν z the distribution on RN with Fourier transform

 
νz (ξ ) = d
z/2
μ(ξ ) · 1 + |ξ |2 ,
and let νzj (ξ ) = νz (2− j ◦ ξ ). We then define A zj ( f ) = f ∗ ν zj and N z f (x) = sup j | A zj f (x)|. Observe that

N 0 ( f ) = sup A j ( f )  Mp ( f ) if f  0. (3.2)
j

By (30) in [11, p. 480], we have

 z 
N f   A z  f L 2 (RN ) for Re( z) < 1/ D , (3.3)
L 2 (R N )
688 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691

and the bound A z grow at most polynomially in (1 + | Im( z)|). Then, it is obvious that

 
1/2 
  z 2 1/2   
 N f i    Az | f | 2  for Re( z) < 1/ D . (3.4)
   i 
i ∈Z L2 i ∈Z L2

On the other hand, by (31) in [11, p. 480] and (33) in [11, p. 481]1 respectively, the following estimates hold:
 z 
N f   A z  f L p (RN ) for 1 < p  ∞ and Re( z) < 0 (3.5)
L p (R N )

and

 
supν zj (x − y ) − ν zj (x) dx  A z for Re( z) < 0. (3.6)
j
ρ (x)2ρ ( y )
By (3.5), (3.6) and the theory of lq -valued extension of nonlinear operators (see [10]), we have

1  
1q 
  z q q   
 N f i   
 Az | fi|q  for 1 < p  ∞ and Re( z) < 0. (3.7)
  
i ∈Z Lp i ∈Z Lp

Now, let (x, i ) → j (x, i ) be an arbitrary integer-valued function on R N × Z and define the analytic family of operators U z by
2
U z ( g )(x) = e z · A zj(x,i ) ( g )(x).
2
The factor e z compensates for the growth of the bound A z . Thus, by (3.3) we know that U z  L 2 → L 2 is bounded uniformly
in z for Re( z) < 1/ D. In view of (3.4) and (3.7), for 1 < p , q < ∞, we have

 
1/2 
  z 2 1/2   
 U f i    A | f | 2  if Re( z) = σ0 < 1/ D ,
   i 
i ∈Z L2 i ∈Z L2

and

1  
1q 
  z q q   
 U f i    A | f |q  if Re( z) = σ1 < 0,
   i 
i ∈Z Lp i ∈Z Lp

where the bound A is independent of z and the choice of function j (x, i ).

Now, for all 1 < p , q < ∞, we can choose θ , p 1 and q1 with 0 < θ < 1 and 1 < p 1 , q1 < ∞, so that
1 θ 1−θ 1 θ 1−θ
= + , = + .
p 2 p1 q 2 q1
1
For θ chosen above, take σ0 = 2D
, then there exists a σ1 < 0 satisfying
0 = θ σ0 + (1 − θ)σ1 .
Applying Lemma 3.1, we have

1  
1q 
  0 q q   
 U f i    A | f |q  , 1 < p , q < ∞, (3.8)
   i  p
i ∈Z Lp i ∈Z L

where A is independent of the choice of j (x, i ). Thus, by (3.2) and (3.8), we get

1q  
1q  
1q 
     
 |Mp f i |q  
 sup | A j f i |q  
 A | fi|q  , 1 < p , q < ∞.
    p
i ∈Z Lp i ∈Z
j Lp i ∈Z L

We therefore complete the proof of Proposition 3.2. 2

To get Theorem 1.4 by using Proposition 3.2, we need to introduce some auxiliary operators. For a finite measure dμ
on R N , the operator T is defined by a convolution with dμ, i.e., T ( f ) = f ∗ dμ. Suppose L is a linear mapping from R N
to Rd , then the operator T L is defined by

 
T L g (x) = g x − L ( z) dμ( z),
RN

1
Where Re( z) > 0 in (33) of [11, p. 481] should be Re( z) < 0.
 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691 689

where x ∈ Rd . By the same way, for a family of finite measures {dμ j } on R N , we can define operators family { T j } and { T Lj }
related to {dμ j }.
Using a similar idea of proving lemma in [11, p. 484], we may get the following results, here we omit the details.

Lemma 3.3. Suppose L, T and T j as above, 1 < p , q < ∞.

(a) Then the norm of operator T L acting on L p (lq )(Rd ) does not exceed the norm of T acting on L p (lq )(R N ).
(b) Assume additionally that the dμ j are nonnegative and set
   
T ∗ g (x) = sup T j g (x), and T ∗L g (x) = sup T Lj g (x).
j j

If T ∗ is bounded on L (l )(R ), then T ∗ is also bounded on L p (lq )(Rd ), and its norm does not exceed that of T ∗ .
p q N L

Finally, we complete the proof of Theorem 1.4 by applying Lemma 3.1, Proposition 3.2 and Lemma 3.3, which idea follows
Stein–Wainger’s way (see [11] or [13]).

Proof of Theorem 1.4. For r > 0, the measure dμr on R N is defined by

 
1  
f (x) dμr (x) = f p(t ) dt .
rn
RN |t |r

Let T r ( f ) = f ∗ dμr , then Mp ( f ) = supr >0 | T r ( f )|.

We may assume that the P j vanish at the origin, otherwise, a translation just needed. Write

N
i
P j ( y) = a jα y α = a jαi y α , αi ∈ Λ
1|α | D i =1

and define the linear mapping L : R N → Rd by

N
L (x) j = z j = a j α i xi ,
i =1

where x = (x1 , . . . , x N ) ∈ R N , z = ( z1 , . . . , zd ) ∈ Rd . Then dμrL is given by

  
1    1  
f (x) dμrL (x) = f L p(t ) dt = f P (t ) dt .
rn rn
Rd |t |r |t |r

Let = f ∗ dμ MP ( f ) =
T rL ( f ) L
r, supr >0 | T rL ( f )|.
For 1 < p , q < ∞, by Lemma 3.3 and Proposition 3.2, restricting to a countable dense set of r for the interval 0 < r < R,
and letting R → ∞, then Theorem 1.4 is established. 2

4. Proof of Lemma 3.1: the lq -valued analytic interpolation

Denote
  

m
 k
k q n
S= χ A k (x)c j : m ∈ N, c j j ∈Z
∈ l , Ak ⊂ R , | Ak | < ∞, Ak ∩ A i = ∅ for k = i ,
k =1 j ∈Z

then S is a dense subset of L p (lq )(Rd ) (see [10] or [16]). Thus, by a simple limiting argument, we need only to show (3.1)
for functions in S. For 0 < θ < 1, 1q = 1q−θ + qθ and 1p = 1p−θ + pθ , let
0 1 0 1


q( q1 − q1 )(z−θ ) 
p ( p1 − p1 )( z−θ )
|C j (x)| 1 0 {C j (x)}lq 1 0
C j ( z, x) = C j (x) ,
{C j (x)}lq {C j (·)}L p (lq )
where z ∈ { z ∈ C: 0  Re( z)  1} and

m
C j (x) = χ Ak (x)ckj , {C j } j ∈Z ∈ S.
k =1

It is obvious C j (θ, x) = C j (x). By a simple calculation, we have

690 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691

       
 C j (1 + it , ·)  =  C j (·) L p (lq ) and  C j (it , ·)  L p0 (lq0 ) =  C j (·)  L p (lq ) . (4.1)
L p 1 (lq1 )

On the other hand, denote

  

h
 l
 q
S = blj χ B l (x) : h ∈ N, b j j ∈Z
n
∈ l , B l ⊂ R , | B l | < ∞, B l ∩ B i = ∅ for l = i ,
l =1 j ∈Z
  1−θ 1−θ
then S is a dense subset of L p (lq )(Rn ). Note that, for 0 < θ < 1, 1
q
= q0
+ θ
q1
and 1
p
= p 0
+ θ
p 1
, if denote


q ( 1
− 1
)( z−θ ) 
p ( 1
− 1
)( z−θ )
| E j (x)| q q { E j (x)}lq p p
E j ( z, x) = E j (x)
1 0 1 0
,
{ E j (x)}lq { E j (·)}L p (lq )
h
where z ∈ { z ∈ C: 0  Re( z)  1} and E j (x) = l
χ Bl (x), then we also have
l=1 b j
       
 E j (it , ·)  p  q =  E j (·)    and  E j (1 + it , ·)  p q =  E j (·)  L p (lq ) . (4.2)
L 0 (l 0 ) L p (lq ) 1 1 L (l )

Without loss of generality, we may assume that {C j (·)} L p (lq )  1. Thus, to get (3.1), it suffices to show
  
 
sup  U a(1−θ )+bθ
(C j )(x) E j (x) dx  M 01−θ M 1θ . (4.3)

{ E j }∈S j ∈ZRn
{ E j (·)} p  q 1
L (l )

Define a function F ( z) on { z ∈ C: 0  Re( z)  1} by

  
F ( z) = U a(1−z)+bz C j ( z, ·) (x) E j ( z, x) dx.
j ∈ZRn

With our assumption on {C j } and { E j }, it is easy to check that F ( z) is continuous in the strip { z ∈ C: 0  Re( z)  1} and
analytic in its interior. Let z = u + it, 0  u  1, 1r = 1q−u + qu , by Hölder’s inequality, we have
0 1

h 
m
  
1 1
   k r ( q − q )(u −θ )+r  l r  ( qq − qq )(u −θ )+r  r  r
 F ( z)  C c  q1 q0 b  1 0  C p ,q,m,h,θ .
j j
k =1 l =1 j ∈Z j ∈Z

So, F ( z) is bounded on strip { z ∈ C: 0  Re( z)  1}. Using Hölder’s inequality again and (4.1), (4.2), we have
     
 F (it )  M 0  C j (it , ·)   E j (it , ·)  p  q  M0.
L p 0 (lq0 ) L 0 (l 0 )

and
     
 F (1 + it )  M 1  C j (1 + it , ·)   E j (1 + it , ·)  p  q  M1.
L p 1 (lq1 ) L 1 (l 1 )

Thus, | F (θ)|  M 01−θ M 1θ by applying Phragmen–Lindelöf three lines theorem (see [14]). We therefore get (4.3) and Lemma 3.1
follows.

Acknowledgments

The authors would like to express their deep gratitude to the referee for his/her valuable comments and suggestions. In particular, the authors greatly
appreciate the referee for pointing out them Cwikel and Janson’s early paper [5], so that the authors learn that Lemma 3.1 is indeed a consequence of the
theorem on interpolation of analytic families of operators obtained in [5].

References

[1] J. Chen, D. Fan, Y. Ying, Singular integral operators on function spaces, J. Math. Anal. Appl. 276 (2002) 691–708.
[2] J. Chen, C. Zhang, Boundedness of rough singular integral on the Triebel–Lizorkin spaces, J. Math. Anal. Appl. 337 (2008) 1048–1052.
[3] Y. Chen, Y. Ding, Rough singular integrals on Triebel–Lizorkin space and Besov space, J. Math. Anal. Appl. 347 (2008) 493–501.
[4] L. Colzani, Hardy spaces on spheres, PhD thesis, Washington University, St. Louis, 1982.
[5] M. Cwikel, S. Janson, Interpolation of analytic families of operators, Studia Math. 79 (1984) 61–71.
[6] D. Fan, Y. Pan, Singular integral operators with rough kernels supported by subvarieties, Amer. J. Math. 119 (1997) 799–839.
[7] C. Fefferman, E.M. Stein, Some maximal inequalities, Amer. J. Math. 93 (1971) 107–115.
[8] M. Frazier, B. Jawerth, A discrete transform and decompositions of distribution spaces, J. Funct. Anal. 93 (1990) 34–170.
[9] M. Frazier, B. Jawerth, G. Weiss, Littlewood–Paley Theory and the Study of Function Spaces, CBMS Reg. Conf. Ser., vol. 79, Amer. Math. Soc., Providence,
RI, 1991.
[10] L. Grafakos, Classical and Modern Fourier Analysis, Prentice Hall, Upper Saddle River, NJ, 2003.
[11] E.M. Stein, Harmonic Analysis: Real-Variable Methods, Orthogonality and Oscillatory Integrals, Princeton Univ. Press, Princeton, NJ, 1993.
 Y. Chen et al. / J. Math. Anal. Appl. 368 (2010) 677–691 691

[12] E.M. Stein, Problems in harmonic analysis related to curvature and oscillatory integrals, in: Proceedings of the International Congress of Mathematicians,
Berkeley, 1986, pp. 196–221.
[13] E.M. Stein, S. Wainger, Problems in harmonic analysis related to curvature, Bull. Amer. Math. Soc. 84 (1978) 1239–1295.
[14] E.M. Stein, G. Weiss, Introduction to Fourier Analysis on Euclidean Spaces, Princeton Univ. Press, Princeton, NJ, 1971.
[15] H. Triebel, Theory of Function Spaces, Monogr. Math., vol. 78, Birkhäuser Verlag, Basel, 1983.
[16] H. Triebel, Interpolation Theory, Function Spaces and Differential Operators, second ed., Johann Ambrosius Barth Verlag, Heidelberg, Leipzig, 1995.