Startseite Geometric characterization of generalized Hajłasz-Sobolev embedding domains
Artikel Open Access

Geometric characterization of generalized Hajłasz-Sobolev embedding domains

  • Ziwei Li , Dachun Yang und Wen Yuan EMAIL logo
Veröffentlicht/Copyright: 27. März 2025

Abstract

In this article, the authors study the embedding properties of Hajłasz-Sobolev spaces with generalized smoothness on Euclidean domains, whose regularity is described via a smoothness weight function ϕ : [ 0 , ) [ 0 , ) . Given any bounded domain with the slice property, the authors prove that it is a generalized Hajłasz-Sobolev embedding domain if and only if it is a generalized F -weak cigar domain, where F is a modulus of continuity related to the weight function ϕ of the generalized Hajłasz-Sobolev spaces under consideration. Comparing with the classical Hajłasz-Soblev spaces, one main difficulty in dealing with generalized Hajłasz-Sobolev spaces lies in that both its smoothness weight function ϕ and the related modulus of continuity F have no explicit expression. To overcome this, the authors introduce and use some key indices to accurately describe the increasing or the decreasing behavior of both ϕ and F . Besides the classical Hajłasz-Sobolev spaces, this result can be applied to many other nontrivial spaces such as Hajłasz-Sobolev spaces with logarithmic smoothness and is of wide generality.

1 Introduction

Throughout this article, let R n be the Euclidean space with dimension n 2 and Ω R n a domain, that is, a connected open set in R n . Assume that X ( Ω ) and Y ( Ω ) are two function spaces defined on Ω , equipped respectively with the quasi-norms X ( Ω ) and Y ( Ω ) . Recall that X ( Ω ) is said to be continuously embedded into Y ( Ω ) , denoted by X ( Ω ) Y ( Ω ) , if there exists a positive constant C such that, for any f X ( Ω ) , one has f Y ( Ω ) and f Y ( Ω ) C f X ( Ω ) ; moreover, Ω is said to support an embedding from X ( Ω ) to Y ( Ω ) if X ( Ω ) is continuously embedded into Y ( Ω ) . The study on embedding properties of various function spaces is a topic with a long history and plays a key role in many branches of mathematics [1,17,2830,5557,62,67].

It is known that the embedding property of function spaces on domains strongly depends not only on the analytical character of function spaces themselves but also on the geometric character of the underlying domain under consideration. There already exist a lot of works on the geometric characterizations of domains supporting various embeddings of function spaces. For instance, for any α ( 0 , 1 ] , Gehring and Martio [22] studied the α -Lipschitz class Lip α and its local version loc Lip α on domains and showed that a domain Ω supports the embedding loc Lip α ( Ω ) Lip α ( Ω ) if and only if it is a weak α -cigar domain. Here, and thereafter, a domain Ω is called a weak α -cigar domain if, for any x 1 , x 2 Ω , there exists a rectifiable curve Γ Ω joining x 1 and x 2 such that

(1.1) Γ [ dist ( z , Ω ) ] α 1 d z C x 1 x 2 α ,

where C is a positive constant independent of both x 1 and x 2 and dist ( z , Ω ) denotes the distance between z and the boundary Ω of Ω . This result was extended by Lappalainen [47] to the generalized Lipschitz spaces Lip h and loc Lip h related to a modulus of continuity, h , which is a continuous increasing function on [ 0 , ) whose derivative is decreasing on ( 0 , ) and which satisfies h ( 0 ) = 0 . It was proved in [47] that a domain Ω supports the embedding loc Lip h ( Ω ) Lip h ( Ω ) if and only if it is a weak h -cigar domain, which is defined via replacing [ dist ( z , Ω ) ] α 1 and x 1 x 2 α in (1.1), respectively, by h ( dist ( z , Ω ) ) dist ( z , Ω ) and h ( x 1 x 2 ) . Concerning Sobolev spaces, Buckley and Koskela [4] showed that, for any given p ( n , ) and α = ( p n ) ( p 1 ) , a local weak α -cigar (resp. bounded weak α -cigar) domain always supports an embedding W 1 , p ( Ω ) C 0 , 1 n p ( Ω ¯ ) [resp. W ˙ 1 , p ( Ω ) C ˙ 0 , 1 n p ( Ω ¯ ) ] and, if in addition the underlying domain has a slice property, then the converse implication also holds.

To reflect the geometric character of a set Ω in R n , the extendability of functions from the given set to the entire space is also an effective tool. Recall that Ω is called an X-extension set if there exist a positive constant C and a continuous linear operator Ω : X ( Ω ) X ( R n ) such that, for any f X ( Ω ) , Ω f Ω = f and Ω f X ( R n ) C f X ( Ω ) . Let

X ( R n ) Ω { h Ω : h X ( R n ) }

and, for any f X ( R n ) Ω ,

(1.2) f X ( R n ) Ω inf { h X ( R n ) : h Ω = f , h X ( R n ) } .

If Ω is an X -extension set, an elementary consequence is that

(1.3) X ( Ω ) = X ( R n ) Ω

with equivalent quasi-norms. In particular, if a domain Ω is an X -extension set, we call it X-extension domain. There exists a close connection between extension and embedding domains. In 1998, Koskela [42] proved that, when p ( n , ) , the Sobolev embedding theorem on the domain Ω implies the W 1 , p -extension property on Ω , that is, the existence of the bounded linear extension operator of W 1 , p ( Ω ) into W 1 , p ( R n ) ; moreover, a weak ( p n ) ( p 1 ) -cigar domain has the W 1 , q -extension property for any q ( p , ) . This result was further extended by Shvartsman [59]. Zhou [67] showed that a domain Ω R n with n 2 is a W s , p -extension or a W s , p -embedding domain for all/some ( s , p ) ( 0 , 1 ) × ( 0 , ) if and only if it is a uniform domain. For further related works, we also refer the reader to, for instance, [12,28,29,43,46,60,67].

In particular, there also exist some related works on Hajłasz-Sobolev type spaces M ˙ s , p [27,32,45,65] and their local versions M ˙ ball s , p [44,66]. Recall that the space M ˙ ball s , p with s = 1 was used in [44] to characterize Hardy-Sobolev spaces. Hajłasz et al. [29] proved that an Euclidean domain is an M 1 , p -extension domain with p ( 1 , ) if and only if it is a plump domain. In [66], Zhou showed that, when n 2 and α ( 0 , 1 ) , a bounded weak α -cigar domain in R n is an M ˙ ball s , p -extension domain and also an M ˙ ball s , p -embedding domain for any s ( α , 1 ] and p = ( n α ) ( s α ) ; conversely, if s ( 0 , 1 ] and p ( n s , ) , then a bounded M ˙ ball s , p -extension domain or a bounded M ˙ ball s , p -embedding domain having the slice property is a weak α -cigar domain. For more related results, we refer the reader to previous studies [2,25,31,3640] and the references therein.

On the other hand, the study of function spaces with generalized smoothness can be traced back to the mid-1970s [24,35] and has found various applications in harmonic analysis, probability theory and stochastic processes [18,19], fractal analysis and the related spectral theory [63, Chapters 18–23], and interpolations and both embedding of function spaces and entropy numbers [48,54]. As a special case, function spaces with logarithmic smoothness have received an increasing interest in the recent two decades [59,1316,26,33]. Very recently, the Hajłasz-Sobolev spaces with generalized smoothness, M ˙ ϕ , p and M ϕ , p , whose regularity is described via a general function ϕ : [ 0 , ) [ 0 , ) , were introduced and studied in [49,50]. Recall that, for any domain Ω R n , any p ( 0 , ] , and certain function ϕ : [ 0 , ) [ 0 , ) , the homogeneous ϕ -Hajłasz-Sobolev space M ˙ ϕ , p ( Ω ) is defined as the collection of all measurable and almost everywhere finite functions u on Ω such that there exists a nonnegative function g L p ( Ω ) satisfying that, for almost every x , y Ω ,

(1.4) u ( x ) u ( y ) ϕ ( x y ) [ g ( x ) + g ( y ) ] ;

moreover, the inhomogeneous ϕ -Hajłasz-Sobolev space M ϕ , p ( Ω ) is defined by setting

M ϕ , p ( Ω ) M ˙ ϕ , p ( Ω ) L p ( Ω ) .

If the inequality (1.4) holds only for almost every x , y Ω satisfying

x y 1 2 min { dist ( x , Ω ) , dist ( y , Ω ) } ,

then the related local homogeneous and local inhomogeneous ϕ -Hajłasz-Sobolev spaces are denoted, respectively, by M ˙ ball ϕ , p ( Ω ) and M ball ϕ , p ( Ω ) ; see Section 2 for more details.

Motivated by the aforementioned works, the main purpose of this article is to find the geometric character of the underlying domain Ω supporting the Sobolev-type embedding properties of the above Hajłasz-Sobolev spaces with generalized smoothness. To this end, throughout this article, for any function ϕ : [ 0 , ) [ 0 , ) and p ( 0 , ) , we define ϕ ( p ) : ( 0 , ) ( 0 , ) by setting, for any t ( 0 , ) ,

(1.5) ϕ ( p ) ( t ) t n p ϕ ( t ) ;

if lim t 0 + ϕ ( p ) ( t ) [ 0 , ] exists, then we extend ϕ ( p ) to [ 0 , ) by setting ϕ ( p ) ( 0 ) lim t 0 + ϕ ( p ) ( t ) . Here, and thereafter, t 0 + means that t ( 0 , ) and t 0 . Define ϕ ( ) ϕ . To state the main result of this article, we first present the following concept of the M ˙ ball ϕ , p -embedding and the M ball ϕ , p -embedding domains.

Definition 1.1

Let ϕ : [ 0 , ) [ 0 , ) and p ( 0 , ] be such that ϕ ( p ) in (1.5) is a well-defined almost increasing function on [ 0 , ) satisfying ϕ ( p ) ( 0 ) = 0 . A domain Ω R n is called an M ˙ ball ϕ , p (resp. M ball ϕ , p )-embedding domain if it supports the embedding

(1.6) M ˙ ball ϕ , p ( Ω ) M ˙ ϕ ( p ) , ( Ω ) [ resp. M ball ϕ , p ( Ω ) M ϕ ( p ) , ( Ω ) ] .

Clearly, if s ( 0 , 1 ] and ϕ ( t ) t s for any t [ 0 , ) , then the embedding in (1.6) becomes the Sobolev embedding

M ˙ ball s , p ( Ω ) M ˙ s n p , ( Ω ) resp. M ball s , p ( Ω ) M s n p , ( Ω ) ,

and the M ˙ ball ϕ , p (resp. M ball ϕ , p )-embedding domains are precisely the M ˙ ball s , p (resp. M ball s , p )-embedding domains studied by Zhou [66]. It is worth mentioning that, ignoring a set with zero measure, the Hajłasz-Sobolev spaces, M ˙ s , and M ˙ ball s , with s ( 0 , 1 ] , are exactly the s -Lipschitz class Lip s and the locally s -Lipschitz class loc Lip s , respectively. In this sense, the Hajłasz-Sobolev spaces can be regarded as a generalization of Lipschitz function spaces. We also recall that, if p = , then, in the almost everywhere sense, M ˙ ball ϕ , ( Ω ) = loc Lip ϕ ( Ω ) and M ˙ ϕ , ( Ω ) = Lip ϕ ( Ω ) , where the spaces loc Lip ϕ ( Ω ) and Lip ϕ ( Ω ) were originally considered by Lappalainen [47].

The main result of this article reads as follows, which gives an intrinsic geometric characterization of bounded M ˙ ball ϕ , p -embedding domains.

Theorem 1.2

Suppose that Ω R n is a bounded domain with the slice property. Let ϕ : [ 0 , ) [ 0 , ) be a modulus of continuity with α ϕ ( 0 , 1 ) and let p ( n [ log 2 α ϕ ] , ) be such that the function F [ ϕ ( p ) ] p p 1 satisfies, for any given L ( 0 , ) ,

liminf r 0 + sup t ( 0 , L ] r F ( t ) F ( r t ) = 0

and that the function F * F ( n ) , as in (1.5) with ϕ replaced by F, satisfies

inf y x : x , y ( 0 , ) , F * ( x ) > F * ( y ) > 0 .

Then Ω is an M ˙ ball ϕ , p -embedding domain if and only if it is a weak F-cigar domain.

The concepts of the number α ϕ , the slice property of Ω , and the weak F -cigar domain are presented, respectively, in (2.10) and Definitions 2.5 and 3.10.

Remark 1.3

Clearly, Theorem 1.2 generalizes the corresponding results in [66] for classical fractional Hajłasz-Sobolev spaces (namely in the case where ϕ ( t ) t s for any t [ 0 , ) with s ( 0 , 1 ] ). There also exists a series of other nontrivial functions ϕ such that Theorem 1.2 holds. Below are some examples.

  1. Let Ω 1 be a bounded domain with diam Ω < e 1 and, for some given ε ( 0 , 1 ] and any t [ 0 , diam Ω ] , let ϕ 1 ( t ) [ ln ( 1 t ) ] ε . According to the definition of the F -cigar domain (Definition 3.10), the value of F on ( diam Ω , ) is none of our concern. Thus, we choose the function ϕ as any extension of ϕ 1 to [ 0 , ) such that α ϕ + ( 0 , 1 ) and β ϕ + ( 0 , 2 ] . That is, Theorem 1.2 can be applied to some Hajłasz-Sobolev spaces with logarithmic smoothness.

  2. Let Ω 2 be a bounded domain with diam Ω = π 4 and ϕ 2 ( t ) 2 sin t for any t [ 0 , π 4 ] . By the same reason as in (i), we choose the function ϕ as any extension of ϕ 2 to [ 0 , ) such that α ϕ + ( 0 , 1 ) and β ϕ + ( 0 , 2 ] .

  3. We also refer to [50, Examples 3.12 and 3.13] for two more examples.

Although Theorem 1.2 is on the embedding domain of generalized Hajłasz-Sobolev spaces, we also have some discussions on the related extension domains. Indeed, we prove in Theorem 4.10 that M ball ϕ , p -embedding domains are M ball ϕ , q -extension domains for any q > p , so are their homogeneous versions. Thus, the boundedness together with the weak F -cigar property becomes a sufficient condition for a domain to be an M ball ϕ , q -extension domain; see Theorem 5.2 for more details. We also point out that, due to the variety of the generalized smoothness, the result of this article is of wide generality and can be applied to many nontrivial spaces, including some Hajłasz-Sobolev spaces with logarithmic smoothness.

The remainder of this article is organized as follows.

In Section 2, we state some basic notation and concepts used in this article, including some classical geometric concepts on domains such as uniform domains, cigar domains, and the slice property, as well as some basic assumptions on the function ϕ used to describe the regularity of the spaces under consideration. We also present the definitions of ϕ -Hajłasz-Sobolev spaces and some of their basic properties.

In Section 3, we first introduce the concept of the modulus of continuity, which generalizes those introduced by Lappalainen in [47] in the sense that it contains more elements such as almost increasing but not increasing functions. Then we introduce the (local) weak F -cigar domain related to a given modulus of continuity, F . These (local) weak F -cigar domains are a generalization of the classical weak α -cigar domains, and we present a series of typical examples of these generalized cigar domains to illustrate their generality and flexibility. Furthermore, inspired by Shvartsman [59], we study the weak self-improving property of these weak F -cigar domains in Theorem 3.13, which provides an essential tool to study the sufficiency of Theorem 1.2. Different from the classical weak α -cigar domains, to establish the weak self-improving property of weak F -cigar domains, one main obstacle is that the function F has no specific expression in general, which brings some essential difficulties in the calculation of some integrations and summations and also in finding the related inverse functions. In addition, as the class of modulus of continuity containing more functions, compared with those strictly increasing or decreasing functions, the implicit positive constant that occurs in the definition of almost increasing or decreasing functions sometimes needs to be taken into consideration. To overcome these difficulties, we introduce some classes (Definitions 3.5 and 3.7 ) and indices [(2.10) and (2.11)] to accurately describe the increasing or the decreasing speed of a given modulus of continuity, which reflect some key and intrinsic characters of the modulus of continuity under consideration.

In Section 4, we discuss the relation between embedding and extension properties of ϕ -Hajłasz-Sobolev spaces and show that M ball ϕ , p -embedding domains are M ball ϕ , q -extension domains for any q > p (Theorem 4.10). Both the Whitney ball-covering and the related reflected quasi-balls play a key role in the construction of the desired extension mappings.

Section 5 is devoted to the proof of Theorem 1.2. We prove the sufficiency, that is, a weak F -cigar domain is an M ˙ ball ϕ , p -embedding domain, in Theorems 5.2 and 5.4. This, together with Theorem 4.10, further implies that a weak F -cigar domain is also a ϕ -Hajłasz-Sobolev extension domain. The proof of the sufficiency of Theorem 1.2 does not need the slice property, but makes full use of the weak self-improving property of the weak F -cigar domains (Theorem 3.13) and the Besicovitch covering lemma, which is used to confirm the existence of a sequence of balls with bounded overlap. The necessity is proved in Theorem 5.5. It should be pointed out that, in the proof of the necessity, the choice of both an appropriate test function u M ˙ ball ϕ , p and its Hajłasz gradient function g is the key point, which is mainly based on the fact, ensured by the slice property (Definition 2.5), that diam S i dist ( x i , Ω ) . We also find that, if ϕ satisfies some stronger assumption, with the same F , the bounded F -cigar domain also becomes a sufficient condition for M ˙ ball ϕ , p ˜ -extension domains for any p ˜ ( p * , ) with some p * [ n ( log 2 α ϕ ) , p ) . Indeed, in both sides of the sufficiency and the necessity, we obtain more results than Theorem 1.2.

2 Preliminaries

We begin this section with some notational conventions and then recall some classical geometrical concepts and the Hajłasz-Sobolev spaces with generalized smoothness.

2.1 Basic notation and concepts

Let Z be the collection of all integers, N the collection of all positive integers, and Z + N { 0 } . Throughout this article, we denote by C some positive constant, which is independent of the main parameters involved but might be different from line to line, and we denote by C ( c 1 , c 2 , ) some positive constant depending on the parameters c 1 , c 2 , …. We use the symbols A B to denote A C B and A B to denote A B A .

For any x , y R n , let x y be the Euclidean distance between x and y , and x y ¯ be the line segment connecting x and y . For any ball B B ( x , r ) R n and λ ( 0 , ) , let λ B B ( x , λ r ) . For any given set E R n , we denote by E ¯ its closure in R n , diam E sup { x y : x , y E } its diameter, E its Lebesgue measure, and 1 E its characteristic function. For any function f on R n , the support supp ( f ) of f is defined by setting

supp ( f ) { x R n : f ( x ) 0 } ¯ .

For any two nonempty subsets E and E 1 of R n and x R n , let

dist ( E 1 , E ) inf { x 1 x 2 : x 1 E 1 , x 2 E } = dist ( E , E 1 ) ,

and dist ( x , E ) dist ( { x } , E ) .

For any given domain Ω R n , let Ω be its boundary, L 0 ( Ω ) be the collection of all measurable functions on Ω , which are finite almost everywhere, and L loc 1 ( Ω ) be the collection of all locally integrable functions on Ω . Moreover, for any given function f L loc 1 ( Ω ) and any measurable set E R n with E [ 0 , ) , the integral average of f over E is defined by setting

f E Ω E Ω f ( x ) d x 1 E Ω E Ω f ( x ) d x , E Ω ( 0 , ) ; 0 , E Ω = 0 .

In particular, when Ω = R n , we write f E f E Ω . For any L ( 0 , ) , a function f on Ω is said to be L-Lipschitz if

f ( x ) f ( y ) L x y , x , y Ω .

For any curve Γ Ω , we use leng Γ to denote its length and d z to denote its arc differential element. Throughout this article, we always assume that Γ is a rectifiable curve in Ω and is identified with the function

Γ : [ 0 , leng Γ ] Ω .

For any two points x , y Ω , a rectifiable curve joining x and y in Ω means a rectifiable curve, which is identified with the function Γ : [ 0 , leng Γ ] Ω satisfying Γ ( 0 ) = x and Γ ( leng Γ ) = y and is denoted by Γ x y .

We often use the following elementary inequality: if q ( 0 , 1 ] , then

(2.1) i Z a i q i Z a i q , { a i } i Z C .

2.2 Some classical geometric concepts

We now recall some geometric concepts.

Definition 2.1

  1. A domain Ω R n is called a John domain if there exist positive constants c 1 and c 2 , with c 1 c 2 , and a point x 0 Ω , called a John center, such that, for any x Ω , there is a rectifiable curve Γ Ω joining both x and x 0 and satisfying

    L leng ( Γ ) c 1 and c 2 L 1 s 1 dist ( Γ ( s ) , Ω ) , s [ 0 , L ] .

    Here, Γ ( 0 ) = x and Γ ( L ) = x 0 .

  2. A domain Ω R n is called a uniform domain if there exists a positive constant C 0 [ 1 , ) such that, for any x , y Ω , there is a rectifiable curve Γ Ω joining x and y , which satisfies L leng ( Γ ) C 0 x y and

    (2.2) t ( 0 , L ) B ( Γ ( t ) , C 0 1 min { t , L t } ) Ω .

    The smallest positive constant playing the role of C 0 is called the uniformity constant with respect to Ω .

Remark 2.2

Uniform domains were introduced based on John domains by Martio and Sarvas [52]. It is known that bounded uniform domains are John domains and convex John domains are uniform domains; see [2123,51,64] for some more details on these domains.

Definition 2.3

Let α ( 0 , 1 ] and β ( 0 , α ] . A domain Ω R n is called a weak ( α , β ) -cigar domain if there exists a positive constant C such that, for any x 1 , x 2 Ω , there exists a rectifiable curve Γ Ω joining both x 1 and x 2 and satisfying

(2.3) Γ [ dist ( z , Ω ) ] α 1 d z C x 1 x 2 β .

In particular, Ω is called a weak α -cigar domain if β = α .

Remark 2.4

  1. The concept of weak ( α , β ) -cigar domains was introduced in [4]. By [4, Proposition 2.4], a bounded weak ( α , β ) -cigar domain is a bounded weak ( α ˜ , β ˜ ) -cigar domain if and only if α ˜ [ α , 1 ) and β ˜ α ˜ β α .

  2. The weak α -cigar domain is precisely the Lip α -extension domain in [22] and, especially, when the domain is bounded, it coincides with the α -subhyperbolic domain studied in [59]. Recall that the α -subhyperbolic domain is defined as in (2.3) via replacing the arbitrary x 1 , x 2 Ω by x 1 , x 2 Ω with x 1 x 2 smaller than some given positive constant.

  3. By [22, Theorem 2.24], we find that, for any α ( 0 , 1 ] , the class of weak α -cigar domains contains all uniform domains. Furthermore, the example given in [22, Example 2.26(c)] implies that the set of uniform domains is a proper subset of the class of weak α -cigar domains.

The following slice property was introduced by Buckley and Koskela [4].

Definition 2.5

Let C S ( 1 , ) . A domain Ω R n is said to have the C S -slice property if, for any pair of points x , y Ω , there exist a rectifiable curve Γ joining x and y and a collection { S i } i = 0 N , with N N , of pairwise disjoint open subsets of Ω such that

  1. x S 0 , y S N , and, for any i { 1 , , N 1 } , x and y lie in the different components of Ω \ S i ¯ ;

  2. if γ Ω is a curve containing both x and y , then, for any i { 1 , , N 1 } , diam ( S i ) C S leng ( γ S i ) ;

  3. for any z Γ , B ( z , C S 1 dist ( z , Ω ) ) i = 0 N S i ¯ ;

  4. for any i { 0 , , N } , there exists x i Γ i Γ S i such that x 0 x , x N y , B ( x i , C S 1 dist ( x i , Ω ) ) S i , and, for any z Γ i , diam S i C S dist ( z , Ω ) .

The constant C S is called the slice constant. Furthermore, Ω is said to have the slice property if it has the C S -slice property for some positive constant C S ( 1 , ) .

Remark 2.6

As pointed out in [4, p. 890], both (iii) and (iv) of Definition 2.5 imply that, for any domain Ω having the slice property, one can choose a rectifiable curve Γ as an appropriate polygonal path such that, for any i { 0 , , N } ,

(2.4) Γ S i [ dist ( z , Ω ) ] 1 d z 1 ,

where the implicit positive constant only depends on the slice constant C S .

Definition 2.7

A measurable set E is said to be regular if there exist positive constants C ( 0 , 1 ] and R E ( 0 , ) such that, for any x E ¯ and r ( 0 , R E ) , B ( x , r ) E C B ( x , r ) .

Remark 2.8

  1. The condition in Definition 2.7 is usually called the measure density condition.

  2. Regular domains are also called plump domains [66]. Obviously, domains having cusps are not regular. We also refer the reader to [34] for some examples of regular sets in R n .

  3. We point out that changing “any x E ¯ and r ( 0 , R E ) ” into “almost every x E ¯ and r ( 0 , R E ) ” in Definition 2.7 leads to an equivalent definition of regular sets. To show this, assume that the inequality in Definition 2.7 holds for almost every x E ¯ and r ( 0 , R E ) . Then, for any given ( x ˜ , r ˜ ) E ¯ × ( 0 , R E ) , there exists a pair ( x , r ) E ¯ × ( 0 , R E ) such that 1 2 r ˜ r r ˜ , B ( x , r ) B ( x ˜ , r ˜ ) , and the inequality with the positive constant C therein in Definition 2.7 holds for this ( x , r ) . This further implies that

    B ( x ˜ , r ˜ ) E B ( x , r ) E C B ( x , r ) 2 n C B ( x ˜ , r ˜ ) ,

    and hence, the measurable set E is regular.

  4. The value of R E in Definition 2.7 is not essential. To prove this, we let R 1 < R 2 and show that, if a measurable set E is regular with R E = R 1 , then, for any x E ¯ and r ( 0 , R 2 ) , B ( x , r ) E B ( x , r ) . Indeed, for any x E ¯ and r ( 0 , R 2 ) , if r ( 0 , R 1 ) , there is nothing to prove; if r [ R 1 , R 2 ) , then there exists a ball B ( y , r ˜ ) B ( x , r ) satisfying R 1 2 < r ˜ < R 1 and, by the assumption, one has

    B ( x , r ) E B ( y , r ˜ ) E B ( y , r ˜ ) r ˜ r n B ( x , r ) R 1 2 R 2 n B ( x , r ) ,

    which competes the proof of the aforementioned claim.

2.3 The ϕ -Hajłasz-Sobolev spaces

We now recall the main function spaces under consideration. To this end, we begin with some basic concepts and properties on their generalized smoothness.

Definition 2.9

A function f : [ 0 , ) [ 0 , ) is said to be almost increasing (resp. decreasing) if there exists a positive constant C [ 1 , ) such that, for any t 1 , t 2 [ 0 , ) with t 1 t 2 (resp. t 1 t 2 ), f ( t 1 ) C f ( t 2 ) . The least positive constant playing the role of C is called the increasing (resp. decreasing) constant related to f and is denoted by C f , (resp. C f , ).

This concept has been widely used [41, Section 2.2.1]. Some basic properties of almost increasing or decreasing functions are listed in the following two propositions. Since the proof of the first one is fundamental, we omit the details, and hence, we only show the second one. In what follows, when we prove a theorem (or the like), in its proof, we always use the same symbols as in the statement itself of that theorem (or the like).

Proposition 2.10

Let f 1 and f 2 be two functions on [ 0 , ) such that f 1 f 2 is almost decreasing. If f 1 is almost increasing, then f 2 is almost increasing; if f 2 is almost decreasing, then f 1 is almost decreasing.

Proposition 2.11

Let F : [ 0 , ) [ 0 , ) be an almost decreasing function. Then, for any closed set I and any continuous function f : I [ 0 , ) ,

(2.5) min t I F ( f ( t ) ) F ( max t I f ( t ) ) C F , min t I F ( f ( t ) )

and

(2.6) C F , 1 max t I F ( f ( t ) ) F ( min t I f ( t ) ) max t I F ( f ( t ) ) ,

where C F , is the decreasing constant related to F as shown in Definition 2.9 with f replaced by F.

Proof

Since, for any closed set I and any s I , we have

min t I f ( t ) f ( s ) max t I f ( t ) ,

then, from the almost decreasing property of F , we deduce that

(2.7) F ( max t I f ( t ) ) C F , min t I F ( f ( t ) ) and C F , 1 max t I F ( f ( t ) ) F ( min t I f ( t ) ) .

By using the assumptions that I is closed and f is continuous, we conclude that there exist t 1 , t 2 I such that

f ( t 1 ) = max t I f ( t ) and f ( t 2 ) = min t I f ( t ) ,

and hence,

min t I F ( f ( t ) ) F ( f ( t 1 ) ) = F ( max t I f ( t ) )

and

max t I F ( f ( t ) ) F ( f ( t 2 ) ) = F ( min t I f ( t ) ) ,

which, combined with (2.7), further imply (2.5) and (2.6), respectively. This then finishes the proof of Proposition 2.11.□

Recall that a sequence { σ j } j Z of positive real numbers is called an admissible sequence if there exist two positive constants d 0 and d 1 such that, for any j Z ,

(2.8) d 0 σ j σ j + 1 d 1 σ j ;

see, for instance, [19]. A continuous measurable function ϕ : [ 0 , ) [ 0 , ) is said to be of admissible growth if { ϕ ( 2 j ) } j Z is an admissible sequence and, for any k Z and t [ 2 k , 2 k + 1 ) , ϕ ( t ) ϕ ( 2 k ) with the positive equivalence constants independent of both k and t [50].

We now recall the class [10], which consists of all continuous functions ϕ : [ 0 , ) [ 0 , ) such that, for any s [ 0 , ) ,

ϕ ¯ ( s ) sup t [ 0 , ) ϕ ( s t ) ϕ ( t ) < .

We also consider a larger class 0 , which is defined to be the collection of all continuous functions ϕ : [ 0 , ) [ 0 , ) such that, for any s [ 0 , ) ,

(2.9) ϕ ¯ ( s ) sup t ( 0 , ) ϕ ( s t ) ϕ ( t ) < .

Obviously, the only difference between ϕ ¯ and ϕ ¯ is the range over which the supremum on t is taken.

Remark 2.12

Clearly, if ϕ is of admissible growth, then ϕ 0 . Indeed, for any given s [ 0 , ) and any t ( 0 , ) , there always exist two integers k and j such that t [ 2 k , 2 k + 1 ) and s [ 2 j , 2 j + 1 ) , and, furthermore,

ϕ ( s t ) ϕ ( t ) ϕ ( 2 k + j ) ϕ ( 2 k ) max { d 0 j , d 1 j } ,

where d 0 and d 1 are precisely the positive constants in (2.8) and the implicit positive constants are independent of both s and t . Thus, for any given s [ 0 , ) , ϕ ¯ ( s ) < , which further implies that ϕ 0 .

For any given function ϕ 0 , let

(2.10) α ϕ max { α ϕ , α ϕ + } and β ϕ max { β ϕ , β ϕ + } ,

where

α ϕ limsup k ϕ ( 2 k ) ϕ ( 2 k + 1 ) , α ϕ + limsup k + ϕ ( 2 k ) ϕ ( 2 k + 1 )

and

(2.11) β ϕ limsup k ϕ ( 2 k + 1 ) ϕ ( 2 k ) , β ϕ + limsup k + ϕ ( 2 k + 1 ) ϕ ( 2 k ) .

Remark 2.13

Let ϕ : [ 0 , ) [ 0 , ) be a continuous function of admissible growth and ϕ ( p ) defined in (1.5). In particular, for any t ( 0 , ) , let

ϕ * ( t ) ϕ ( n ) ( t ) = t 1 ϕ ( t ) .

  1. By an argument similar to that used in the proofs of [50, (2.5) and (2.6)] and [49, (9) and (10)], we find that, if α ϕ ( 0 , 1 ) , then ϕ is almost increasing, ϕ ( 0 ) = 0 , lim t ϕ ( t ) = , and there exists a positive constant C 1 such that, for any j Z and t [ 0 , ) ,

    (2.12) i = j ϕ ( 2 i t ) C 1 ϕ ( 2 j t ) .

    Notice that, for any p ( n [ log 2 α ϕ ] , ) , α ϕ ( p ) = 2 n p α ϕ ( 0 , 1 ) . Thus, ϕ ( p ) is a well-defined almost increasing function on [ 0 , ) satisfying

    ϕ ( p ) ( 0 ) lim t 0 + ϕ ( p ) ( t ) = 0 .

    Similarly, if β ϕ ( 0 , 2 ) , then ϕ * is a well-defined almost decreasing function on [ 0 , ) satisfying ϕ * ( 0 ) = and there exists a positive constant C 2 such that, for any j Z and t [ 0 , ) ,

    i = j ϕ * ( 2 i t ) C 2 ϕ * ( 2 j t ) .

  2. If ϕ * is almost decreasing on ( 0 , ) , then β ϕ [ 0 , 2 ] . Indeed, β ϕ 0 is obvious. To show β ϕ 2 , we argue by contradiction. If β ϕ > 2 , then there exists ε ( 0 , 1 ) such that β ϕ > 2 + ε > 2 . Without loss of generality, we may assume that β ϕ + > 2 + ε > 2 . Thus, by (2.11), we conclude that there exists an increasing subsequence { k j } j N of Z such that k j as j and

    lim j ϕ ( 2 k j + 1 ) ϕ ( 2 k j ) 2 + ε ,

    which further implies that

    lim j ϕ * ( 2 k j + 1 ) ϕ * ( 2 k j ) ( 2 + ε ) 2 > 1 .

    By combining this and Definition 2.9, we find that ϕ * cannot be an almost decreasing function, which leads to a contradiction. Thus, β ϕ [ 0 , 2 ] .

  3. Similarly, if ϕ is almost increasing on ( 0 , ) , then α ϕ [ 0 , 1 ] . Furthermore, by the observation that 1 α ϕ β ϕ , we conclude that, when ϕ is almost increasing and ϕ * is almost decreasing, it holds that α ϕ [ 1 2 , 1 ] and β ϕ [ 1 , 2 ] .

In what follows, for simplicity, we always denote by the symbol A the class of all continuous and almost increasing functions ϕ : [ 0 , ) [ 0 , ) with ϕ ( 0 ) = 0 such that { ϕ ( 2 j ) } j Z is admissible.

Definition 2.14

Let ϕ A and Ω R n be a domain.

  1. For any given u L 0 ( Ω ) , let D ϕ ( u ) be the collection of all nonnegative measurable functions g on Ω satisfying that there exists a set E R n with E = 0 such that, for any x , y Ω \ E ,

    (2.13) u ( x ) u ( y ) ϕ ( x y ) [ g ( x ) + g ( y ) ] .

  2. Let θ ( 0 , ) . For any given u L 0 ( Ω ) , let D ball ϕ , θ ( u ) be the collection of all nonnegative measurable functions g on Ω satisfying that there exists a set E R n with E = 0 such that, for any x , y Ω \ E with

    (2.14) x y θ min { dist ( x , Ω ) , dist ( y , Ω ) } ,

    (2.13) holds. Especially, simply write D ball ϕ ( u ) D ball ϕ , 1 2 ( u ) .

Definition 2.15

Let ϕ A , Ω R n be a domain, and p ( 0 , ] .

  1. The homogeneous ϕ -Hajłasz-Sobolev space M ˙ ϕ , p ( Ω ) is defined to be the set of all u L 0 ( Ω ) such that

    u M ˙ ϕ , p ( Ω ) inf g D ϕ ( u ) g L p ( Ω ) < .

  2. The homogeneous ball ϕ -Hajłasz-Sobolev space M ˙ ball ϕ , p ( Ω ) is defined to be the set of all u L 0 ( Ω ) such that

    u M ˙ ball ϕ , p ( Ω ) inf g D ball ϕ ( u ) g L p ( Ω ) < .

  3. The inhomogeneous ϕ -Hajłasz-Sobolev spaces M ϕ , p ( Ω ) is defined as M ˙ ϕ , p ( Ω ) L p ( Ω ) equipped with the quasi-norm

    M ϕ , p ( Ω ) L p ( Ω ) + M ˙ ϕ , p ( Ω ) .

  4. The inhomogeneous ball ϕ -Hajłasz-Sobolev spaces M ball ϕ , p ( Ω ) is defined as M ˙ ball ϕ , p ( Ω ) L p ( Ω ) equipped with the quasi-norm

    M ball ϕ , p ( Ω ) L p ( Ω ) + M ˙ ball ϕ , p ( Ω ) .

Remark 2.16

  1. When s ( 0 , 1 ] and ϕ ( t ) t s for any t [ 0 , ) , M ˙ ball ϕ , p ( Ω ) [resp. M ball ϕ , p ( Ω ) ] is precisely the space M ˙ ball s , p ( Ω ) [resp. M ball s , p ( Ω ) ] studied in [44] when s = 1 and in [66] when s ( 0 , 1 ) .

  2. There exists a positive constant C such that, for any u M ˙ ϕ , ( Ω ) , there exists g D ϕ ( u ) L ( Ω ) such that, for almost every x , y Ω ,

    u ( x ) u ( y ) ϕ ( x y ) [ g ( x ) + g ( y ) ] 2 ϕ ( x y ) g L ( Ω ) C ϕ ( x y ) u M ˙ ϕ , ( Ω ) .

We now recall a Poincaré-type inequality related to the above spaces, which is precisely [49, (16)] in the Euclidean setting.

Lemma 2.17

Let ϕ A with α ϕ ( 0 , 1 ) and let p ( 0 , n [ log 2 α ϕ ] ) . For any given ε ( 0 , log 2 α ϕ ) , there exists a positive constant C such that, for any ball B R n , any u M ˙ ϕ , p ( 2 B ) , and any g D ϕ ( u ) ,

inf c R B u ( y ) c n p n ε p d y n ε p n p C ϕ ( diam B ) { 2 B [ g ( y ) ] p d y } 1 p ,

where the positive constant C is independent of B, u, and g.

Recall that the classical Hardy-Littlewood maximal operator is defined by setting, for any f L loc 1 ( R n ) and x R n ,

( f ) ( x ) sup r ( 0 , ) B ( x , r ) f ( y ) d y .

Below, for any f L loc 1 ( Ω ) , we still write the Hardy-Littlewood maximal function of the nature 0-extension of f from Ω to R n as ( f ) , that is,

( f ) ( x ) sup r ( 0 , ) B ( x , r ) f ˜ ( y ) d y ,

where f ˜ L loc 1 ( R n ) satisfies f ˜ Ω = f and f ˜ R n \ Ω = 0 .

The following proposition gives the fact that, for any θ ( 0 , 1 ) , using D ball ϕ , θ ( u ) instead of D ball ϕ ( u ) still defines the same ϕ -Hajłasz-Sobolev space M ˙ ball ϕ , p ( Ω ) , which generalizes both [44, p. 736] and [66, Lemma 2.1].

Proposition 2.18

Let ϕ A with α ϕ ( 0 , 1 ) , p ( n ( n log 2 α ϕ ) , ] , and u L 0 ( Ω ) . Then, for any given θ 1 , θ 2 ( 0 , 1 ) , there exists a positive constant C = C ( θ 1 , θ 2 ) such that

C 1 inf g D ball ϕ , θ 1 ( u ) g L p ( Ω ) inf g D ball ϕ , θ 2 ( u ) g L p ( Ω ) C inf g D ball ϕ , θ 1 ( u ) g L p ( Ω ) .

Proof

Due to the symmetry, without loss of generality, we may assume that θ 1 < θ 2 . By its definition and θ ( 0 , 1 ) , we find that D ball ϕ , θ ( u ) is decreasing, and hence,

inf g D ball ϕ , θ 1 ( u ) g L p ( Ω ) inf g D ball ϕ , θ 2 ( u ) g L p ( Ω ) .

To prove inf g D ball ϕ , θ 2 ( u ) g L p ( Ω ) inf g D ball ϕ , θ 1 ( u ) g L p ( Ω ) , we let g D ball ϕ , θ 1 ( u ) , k 1 , k 2 Z with k 1 k 2 be such that θ 1 [ 2 k 1 1 , 2 k 1 ) and θ 2 [ 2 k 2 1 , 2 k 2 ) , and m 3 2 k 1 k 2 + 2 .

For any x , y Ω with x y θ 2 min { dist ( x , Ω ) , dist ( y , Ω ) } , we find that the line segment x y ¯ Ω . Let { B k B ( z k , r k ) } k Z be a sequence of balls with

z k x , r k 2 m + k x y 2 m if k m , z k ( m k ) x + ( m + k ) y 2 m , r k x y 2 m if m + 1 k m 1 , z k y , r k 2 m k x y 2 m if k m .

Then, by m 12 , we obtain, for any k Z , B k 3 B k + 1 , B k + 1 3 B k , 6 B k Ω , and

diam 6 B k 2 k 2 k 1 1 x y < θ 1 θ 2 x y θ 1 min { dist ( x , Ω ) , dist ( y , Ω ) } ,

which further implies that, for any k Z and almost every x , y 6 B k ,

(2.15) u ( x ) u ( y ) ϕ ( x y ) [ g ( x ) + g ( y ) ] .

Notice that, by ϕ A and (2.12) with t = x y , we have

k Z ϕ ( r k ) ( 2 m 1 ) ϕ x y 2 m + 2 k = m ϕ 2 m + k x y 2 m ϕ x y 2 m ϕ ( x y ) .

Thus, using the Lebesgue differentiation theorem, (2.15), and Lemma 2.17 with B replaced by 6 B k , we conclude that, for almost every x , y Ω with x y θ 2 min { dist ( x , Ω ) , dist ( y , Ω ) } ,

(2.16) u ( x ) u ( y ) k Z u B k u B k + 1 k Z inf c R 3 B k u ( z ) c d z k Z ϕ ( r k ) 6 B k [ g ( z ) ] n n + ε d z n + ε n ϕ ( x y ) g n n + ε ( x ) n + ε n + g n n + ε ( y ) n + ε n ,

where ε ( 0 , log 2 α ϕ ) is such that p > n ( n + ε ) , denotes the Hardy-Littlewood maximal operator, and the implicit positive constants depend only on ϕ , n , p , θ 1 , and θ 2 . This further implies that [ ( g n ( n + ε ) ) ] ( n + ε ) n Ω is a positive constant multiple of an element in D ball ϕ , θ 2 ( u ) . From this and the L p ( n + ε ) n ( R n ) -boundedness of the Hardy-Littlewood maximal operator , it follows that

inf g D ball ϕ , θ 2 ( u ) g L p ( Ω ) inf g D ball ϕ , θ 1 ( u ) g L p ( Ω ) .

This finishes the proof of Proposition 2.18.□

Remark 2.19

For any θ ( 0 , 1 ) , by choosing θ 1 θ and θ 2 1 2 in Proposition 2.18, we find that

u M ˙ ball ϕ , p ( Ω ) inf g D ball ϕ , θ ( u ) g L p ( Ω ) ,

where the positive equivalence constants depend only on θ .

Moreover, the space M ˙ ball ϕ , p ( Ω ) can also be equivalently defined via replacing the collection D ball ϕ by D ˜ ball ϕ , 1 4 , where D ˜ ball ϕ , 1 4 is defined the same as D ball ϕ via replacing (2.14) by

(2.17) x y 1 4 dist ( x , Ω ) .

Indeed, if (2.14) holds with θ 1 4 , then (2.17) holds, which means D ball ϕ , 1 4 D ˜ ball ϕ , 1 4 . Thus, by Proposition 2.18, we have

inf g D ˜ ball ϕ , 1 4 g L p ( Ω ) u M ˙ ball ϕ , p ( Ω ) .

Conversely, if (2.17) holds, then

3 4 dist ( x , Ω ) dist ( y , Ω ) 5 4 dist ( x , Ω ) ,

and hence

x y min 1 4 dist ( x , Ω ) , 1 3 dist ( y , Ω ) 1 3 min { dist ( x , Ω ) , dist ( y , Ω ) } .

By this, we find that D ˜ ball ϕ , 1 4 D ball ϕ , 1 3 , which, combined with Proposition 2.18, further implies that

u M ˙ ball ϕ , p ( Ω ) inf g D ˜ ball ϕ , 1 4 g L p ( Ω ) .

Thus, M ˙ ball ϕ , p ( Ω ) coincides with the space defined as M ˙ ball ϕ , p ( Ω ) via replacing D ball ϕ by D ˜ ball ϕ , 1 4 , which completes the proof of the above claim.

Obviously, by the definitions of both M ˙ ball ϕ , p ( Ω ) and M ball ϕ , p ( Ω ) , one has M ˙ ϕ , p ( Ω ) M ˙ ball ϕ , p ( Ω ) and M ϕ , p ( Ω ) M ball ϕ , p ( Ω ) , but the converse direction, in general, is not true. For instance, taking Ω B ( ( 0 , 0 ) , 1 ) \ { ( x , 0 ) : 0 x < 1 } R 2 and f ( r , θ ) θ r for any r ( 0 , ) and θ ( 0 , 2 π ) , one has f M ball ϕ , ( Ω ) but f M ϕ , ( Ω ) . Observe here that this Ω is not a uniform domain. However, inspired by [44, Theorem 19], the following proposition shows that one obtains the coincidences M ˙ ball ϕ , p ( Ω ) = M ˙ ϕ , p ( Ω ) and M ball ϕ , p ( Ω ) = M ϕ , p ( Ω ) if the underlying domain Ω satisfies an additional uniformity condition.

Proposition 2.20

Let Ω be a uniform domain, ϕ A with α ϕ ( 0 , 1 ) , and p ( n ( n log 2 α ϕ ) , ] . Then M ˙ ball ϕ , p ( Ω ) = M ˙ ϕ , p ( Ω ) and M ball ϕ , p ( Ω ) = M ϕ , p ( Ω ) .

Proof

By similarity, we only consider the homogeneous case. Since M ˙ ϕ , p ( Ω ) M ˙ ball ϕ , p ( Ω ) is trivial, we only need to prove M ˙ ball ϕ , p ( Ω ) M ˙ ϕ , p ( Ω ) .

Let x , y Ω , C 0 be the uniformity constant in Definition 2.1(ii), and Γ a rectifiable curve joining x and y such that (2.2) holds.

By using the uniformity assumption of Ω in Definition 2.1(ii), we choose a sequence { B k } k Z of balls in the following way. First, let B 0 B ( Γ ( L 2 ) , L ( 48 C 0 ) ) and 0 { B 0 } . Next, observe that, for any given k N , along the curve Γ ( [ 2 k 1 L , 2 k L ] ) , there exist 24 C 0 + 2 balls, centered on Γ ( [ 2 k 1 L , 2 k L ] ) , with different centers but the same radius ( 2 k L ) ( 48 C 0 ) to cover the curve Γ ( [ 2 k 1 L , 2 k L ] ) . We use k to denote the collection of all these 24 C 0 + 1 balls together with the ball B ( Γ ( 2 k 1 L ) , ( 2 k L ) ( 48 C 0 ) ) . Similarly, for any negative integer k , by a symmetrical construction, we find a collection k of 24 C 0 + 2 balls, centered on the curve Γ ( [ ( 1 2 k ) L , ( 1 2 k 1 ) L ] ) with different centers but the same radius ( 2 k L ) ( 48 C 0 ) , such that their union covers Γ ( [ ( 1 2 k ) L , ( 1 2 k 1 ) L ] ) . Clearly, the union of all balls in k Z k covers the curve Γ ( ( 0 , L ) ) . Finally, we rearrange all these balls in k Z k and rewrite this new sequence of balls as { B k B ( z k , r k ) } k Z such that, for any k Z , 12 B k Ω , B k B k + 1 , r k 2 r k + 1 2 r k , and

lim k dist ( x , B k ) = 0 = lim k dist ( y , B k ) .

Since r k L C 0 x y for any k Z , applying the symmetry of the construction of this sequence of balls, ϕ A , and (2.12) with t = x y , we find that

(2.18) k Z ϕ ( r k ) = ϕ ( r 0 ) + 2 k = 1 ϕ ( r k ) ϕ ( L [ 48 C 0 ] ) + 2 ( 24 C 0 + 2 ) k = 1 ϕ ( [ 2 k L ] [ 48 C 0 ] ) ϕ ( x y ) .

Let u M ˙ ball ϕ , p ( Ω ) and g D ball ϕ ( u ) be such that g L p ( Ω ) u M ˙ ball ϕ , p ( Ω ) . Observe that, for any k Z and x k , y k 6 B k ,

x k y k < 12 r k 1 2 min { dist ( x k , Ω ) , dist ( y k , Ω ) } .

By applying this, the properties of { B k } k Z , (2.18), Lemma 2.17 with B replaced by 6 B k , and an estimation similar to that used in (2.16), we conclude that [ ( g n ( n + ε ) ) ] ( n + ε ) n Ω is a positive constant multiple of an element in D ϕ ( u ) , which, together with the L p ( n + ε ) n ( R n ) -boundedness of the Hardy-Littlewood maximal operator and the choice of g , further implies M ˙ ball ϕ , p ( Ω ) M ˙ ϕ , p ( Ω ) . This finishes the proof of Proposition 2.20.□

Remark 2.21

  1. When ϕ ( t ) t for any t [ 0 , ) , Proposition 2.20 in this case is precisely [44, Theorem 19], which is on the particular first order Hajłasz-Sobolev space. In this sense, Proposition 2.20 generalizes [44, Theorem 19]. On the basis of some ideas from the proof of [44, Theorem 19], we also describe the specific construction of the sequence of balls in detail to keep the completeness of the whole proof of Proposition 2.20.

  2. In Proposition 2.20, if Ω is further assumed to be bounded, the assumption α ϕ ( 0 , 1 ) can be reduced to α ϕ ( 0 , 1 ) . This is because, for any x , y Ω , x y has a uniform upper bound, and hence, α ϕ ( 0 , 1 ) is sufficient to ensure the validity of (2.18).

3 Generalized weak F -cigar domains and their weak self-improving property

In this section, we generalize the classical concept of weak α -cigar domains (also named as Lip α -extension domains) via replacing the scalar parameter α by a function parameter F and study the weak self-improving property of these generalized weak F -cigar domains.

We begin with some notational convention. For any given function F : [ 0 , ) [ 0 , ) , define F * : ( 0 , ) ( 0 , ) by setting, for any t ( 0 , ) ,

(3.1) F * ( t ) t 1 F ( t ) .

Moreover, if the limit lim t 0 + F * ( t ) [ 0 , ] exists, we extend F * to [ 0 , ) by setting F * ( 0 ) lim t 0 + F * ( t ) . Also, for this given F and any given rectifiable curve Γ Ω , define

(3.2) F Γ * ( t ) F * ( dist ( Γ ( t ) , Ω ) ) , t [ 0 , leng Γ ] .

Observe that, for any t [ 0 , leng Γ ] , dist ( Γ ( t ) , Ω ) > 0 .

3.1 Generalized weak F -cigar domains

We begin this subsection with the following definition.

Definition 3.1

A continuous function F : [ 0 , ) [ 0 , ) is called a modulus of continuity if it satisfies the following conditions:

  1. F ( 0 ) = 0 and, for any t ( 0 , ) , F ( t ) > 0 ;

  2. F is almost increasing on [ 0 , ) ;

  3. F * , defined in (3.1), is almost decreasing on ( 0 , ) .

Remark 3.2

  1. Recall that the modulus of continuity h introduced by Lappalainen in [47] was originally defined as a continuous increasing function h : [ 0 , ) [ 0 , ) satisfying that h ( 0 ) = 0 , h ( t ) > 0 for any t ( 0 , ) , and the derivative of h exists and is decreasing on ( 0 , ) . Compared with these h , the restrictions of F in Definition 3.1 are relaxed, and the class of moduli of continuity defined via Definition 3.1 can contain more functions, for instance, some nondifferentiable functions or some almost increasing but not increasing functions. To show this, we only need to notice that any increasing (resp. decreasing) function is almost increasing (resp. almost decreasing) and, if a function F is differentiable with its derivative decreasing on ( 0 , ) , then F * is also decreasing on ( 0 , ) .

  2. A modulus of continuity in Definition 3.1 belongs to both A and 0 . Indeed, let F be a modulus of continuity in Definition 3.1 and F ¯ in (2.9) with ϕ replaced by F . Obviously, F ¯ ( 0 ) = 0 < and F ¯ ( 1 ) = 1 < . For any s ( 0 , 1 ) , by s t < t for any t ( 0 , ) and by the almost increasing property of F , we obtain F ¯ ( s ) C F , < ; similarly, for any s ( 1 , ) , by the almost decreasing property of F * , we obtain

    F ¯ ( s ) = sup t ( 0 , ) s F * ( s t ) F * ( t ) C F * , s < ,

    where C F , and C F * , are precisely the positive constants as in Definition 2.9 with f replaced, respectively, by F and F * . Thus, F 0 . Furthermore, by taking s = 1 2 and s = 2 in the aforementioned argument, we conclude that { F ( 2 j ) } j Z is admissible, which further implies that F A .

  3. Let F : [ 0 , ) [ 0 , ) be of admissible growth with α F ( 0 , 1 ) and β F ( 0 , 2 ) , where α F and β F are the same as in (2.10) with ϕ replaced by F . Then F is a modulus of continuity.

  4. By an argument similar to that used in the proof of [47, Theorem 2.24], we find that every modulus of continuity F defines a metric in Ω : for any x , y Ω ,

    (3.3) d F Ω ( x , y ) inf Γ = Γ x y Ω Γ F * ( dist ( z , Ω ) ) d z ,

    where the infimum is taken over all rectifiable curves joining x and y in Ω and F * is the same as in (3.1).

We have the following properties of the modulus of continuity.

Proposition 3.3

Let F : [ 0 , ) [ 0 , ) be a modulus of continuity as in Definition 3.1, F * the same as in (3.1), and C F * , and C F , the same positive constants as in Definition 2.9 with f replaced, respectively, by F * and F. Then, for any given c ( 1 , ) , there exist positive constants C 1 and C 2 , depending only on C F * , , C F , , and c, such that, for any x , y ( 0 , ) with x c y , F ( x ) C 1 F ( y ) a n d F * ( x ) C 2 F * ( y ) .

Proof

Let x , y [ 0 , ) with x c y . By the almost increasing property of F , the almost decreasing property of F * , and c > 1 , we have

F ( x ) y F ( c y ) y F * ( c y ) F * ( y ) and F * ( x ) F * ( c y ) F ( c y ) y F * ( y ) .

This finishes the proof of Proposition 3.3.□

Proposition 3.4

Let F : [ 0 , ) [ 0 , ) be a modulus of continuity as in Definition 3.1 and F * the same as in (3.1). Then, for any c ( 0 , ) and x ( 0 , ) , F ( c x ) F ¯ ( c ) F ( x ) and F * ( c x ) c 1 F ¯ ( c ) F * ( x ) , where F ¯ ( c ) is the same as in (2.9) with ϕ F .

Definition 3.5

A function g : ( 0 , ) ( 0 , ) is said to belong to the class Δ , denoted by g Δ , if

(3.4) C g Δ inf y x : x , y ( 0 , ) , g ( x ) > g ( y ) > 0 .

Remark 3.6

  1. Define F 1 and F 2 by setting, for any t ( 0 , ) , F 1 ( t ) t s with s ( , 1 ) and F 2 ( t ) ln ( 1 + t ) . Then both F 1 * and F 2 * , defined as in (3.1), belong to Δ .

  2. Any strictly decreasing function g : ( 0 , ) ( 0 , ) belongs to Δ with C g Δ = 1 .

  3. There exist functions belonging to Δ , which are not strictly decreasing. For instance, let F : ( 0 , ) ( 0 , ) be a continuous function and N 1 , N 2 ( 0 , ) with N 1 < N 2 be two positive constants such that F is strictly decreasing on ( 0 , ) \ [ N 1 , N 2 ] and, for any t 0 [ N 1 , N 2 ] ,

    lim t F ( t ) < F ( t 0 ) < lim t 0 + F ( t ) .

    Then F still belongs to Δ .

We now introduce another class of functions on [ 0 , ) as follows.

Definition 3.7

Let p [ 0 , ) . A function F : [ 0 , ) [ 0 , ) is said to belong to the class Λ p if, for any given L ( 0 , ) ,

liminf r 0 + sup t ( 0 , L ] r p F ( t ) F ( r t ) = 0 .

Remark 3.8

  1. Recall that an Orlicz function Φ : [ 0 , ) [ 0 , ) is said to be of upper type p for some p [ 0 , ) if there exists a positive constant C such that, for any s [ 1 , ) and t [ 0 , ) , Φ ( s t ) C s p Φ ( t ) . If an Orlicz function Φ is of upper type p 1 [ 0 , ) , then, for any p 2 ( p 1 , ) , Φ Λ p 2 .

  2. Indeed, for any function Φ of upper type p 1 , there exists a positive constant C such that, for any s ( 0 , 1 ] ,

    sup t ( 0 , ) s p 1 Φ ( t ) Φ ( s t ) C .

    Thus, by p 1 < p 2 , we conclude that

    liminf s 0 + sup t ( 0 , L ] s p 2 Φ ( t ) Φ ( s t ) liminf s 0 + s p 2 p 1 sup s ( 0 , 1 ] sup t ( 0 , ) s p 1 Φ ( t ) Φ ( s t ) C liminf s 0 + s p 2 p 1 = 0 ,

    which further implies that Φ Λ p 2 .

  3. Let p [ 0 , ) and F : [ 0 , ) [ 0 , ) be a function such that lim t 0 + F ( t ) t p [ 0 , ] exists. Define

    F p ( t ) F ( t ) t p if t [ 0 , ) , lim t 0 + F ( t ) t p if t = 0 .

    Notice that (2.9) further implies that

    sup t ( 0 , L ] s p F ( t ) F ( s t ) = sup t ( 0 , L ] F p ( t ) F p ( s t ) F p ¯ ( s ) , s ( 0 , ) ,

    where F p ¯ is defined as in (2.9) with ϕ replaced by F p . Then, if F also satisfies the condition liminf s 0 + F p ¯ ( s ) = 0 , then

    liminf s 0 + sup t ( 0 , L ] s p F ( t ) F ( s t ) = 0 ,

    which means that F Λ p .

Proposition 3.9

Let F : [ 0 , ) [ 0 , ) be a continuous function of admissible growth and F * the same as in (3.1).

  1. If β F ( 0 , 2 ) , then F * Δ .

  2. Let θ [ 0 , 1 ] . If β F ( 0 , 2 θ ) , then F Λ θ .

Proof

To prove (i), we argue by contradiction. Let C F * Δ be the same constant as in (3.4) with g replaced by F * . Obviously, C F * Δ 0 . If C F * Δ = 0 , then, for any j N , there exists x j , y j ( 0 , ) such that

(3.5) x j > 2 j y j and F * ( x j ) > F * ( y j ) .

By using the admissible growth property of F , without loss of generality, we may assume that y j = 2 k j for some k j Z . By β F ( 0 , 2 ) and Remark 2.13(i), we find that F * is almost decreasing; moreover, there exists δ 2 ( 0 , ) such that β F + δ 2 < 2 and

F ( 2 k j + j ) ( β F + δ 2 ) j F ( 2 k j ) , j N .

Thus,

F * ( 2 j y j ) [ ( β F + δ 2 ) 2 ] j F * ( y j ) , j N ,

with the implicit positive constant independent of both k and j . From this, (3.5), and the almost decreasing property of F * , we infer that

F * ( x j ) > F * ( y j ) β F + δ 2 2 j F * ( 2 j y j ) β F + δ 2 2 j F * ( x j )

with the implicit positive constants independent of j . Then, by using β F + δ 2 < 2 and choosing j large enough, we obtain a contradiction to (3.5). This further implies that C F * Δ > 0 , and hence, F * Δ .

Now, we show (ii). To this end, choose both δ ˜ 2 ( 0 , ) such that β F + δ ˜ 2 < 2 θ and a sequence { r j 2 j } j N . Then, by (2.11), we obtain

liminf r 0 + sup t ( 0 , L ] r θ F ( t ) F ( r t ) lim j sup t ( 0 , L ] 2 j θ F ( t ) F ( 2 j t ) lim j β F + δ ˜ 2 2 θ j = 0 ,

which means that F Λ θ . This finishes the proof of (ii), and hence, Proposition 3.9.□

Now, we introduce the following concept of weak F -cigar domains, which generalizes the concept of weak cigar domains from [4, p. 885].

Definition 3.10

Let F : [ 0 , ) [ 0 , ) be a modulus of continuity and d F Ω ( x , y ) defined in (3.3). A domain Ω R n is called a locally weak F-cigar domain if there exist two constants R F Ω ( 0 , ) and C F Ω [ 1 , ) such that, for any x , y Ω satisfying F ( x y ) F ( R F Ω ) ,

(3.6) d F Ω ( x , y ) C F Ω F ( x y ) .

A domain Ω is called a weak F-cigar domain if, for any x , y Ω , (3.6) holds.

In the remainder of this article, for any given (locally) weak F -cigar domain Ω , we always denote by C F Ω the smallest positive constant and R F Ω the largest positive constant in Definition 3.10. We also denote by Cig ( F ) the collection of all weak F -cigar domains. When F ( t ) t a , t [ 0 , ) , with a ( 0 , 1 ] , we denote Cig ( F ) by Cig ( a ) .

Obviously, any weak F -cigar domain is a local weak F -cigar domain. Moreover, if a bounded domain Ω with the diameter R is a locally weak F -cigar domain with F ( R F Ω ) [ C F , F ( R ) , ) , then it is a weak F -cigar domain, where C F , is the same positive constant as in Definition 2.9 with f replaced by F .

Recall that the class of weak α -cigar domains is strictly increasing on α ( 0 , 1 ] ; see [4, Proposition 2.4] and also [47, Corollary 4.10]. The following proposition proves that the generalized weak F -cigar domain keeps a similar property. Since its proof is similar to that of [47, Theorem 4.6], we omit the details.

Proposition 3.11

Let F and G be two moduli of continuity such that G F is almost increasing. Then Cig ( F ) Cig ( G ) .

Remark 3.12

(i) Let α ( 0 , 1 ] . When F ( t ) t α for any t [ 0 , ) , the generalized weak F -cigar domain is precisely the classical weak α -cigar domain in [4] and also the Lip α -extension domain in [22, Theorem 2.2]. Furthermore, by an argument similar to that used in the proof of [47, Theorem 3.6], we find that, for any given modulus of continuity F , a domain is a weak F -cigar domain if and only if it is a Lip F -extension domain in the sense of [47, Definition 3.1].

(ii) Recall that, when F ( t ) t a , t [ 0 , ) with a ( 0 , 1 ] , all uniform domains are weak F -cigar domains [22, Theorem 2.24].

There also exist some other functions F such that uniform domains are weak F -cigar domains. For instance, if s ( 0 , 1 ] and

F 1 ( t ) t s ln 1 t if t ( 0 , e 2 ] , t s + e 2 s if t ( e 2 , ) ,

then, by Proposition 3.11, we find that all uniform domains are weak F 1 -cigar domains.

Furthermore, by an argument similar to those used in the proofs of [47, Theorems 4.12 and 4.17], we conclude that, for any modulus of continuity F , if there exists a positive constant K (resp. exist two positive constants K and t K ) such that, for any t ( 0 , ) (resp.  t ( 0 , t K ] ),

(3.7) 0 t F * ( s ) d s K F ( t ) ,

where F * is the same as in (3.1), then all uniform (resp. bounded uniform) domains Ω are weak F -cigar domains with C F Ω depending on F , K , and the uniformity constant C (resp. on F , K , t K , the uniformity constant C , and the diameter of Ω ).

In addition, there also exist weak F -cigar domains but not uniform; see [22, Example 2.26] and [47, Lemma 4.28].

(iii) Proposition 3.11 not only generalizes the classical result that Cig ( a ) Cig ( b ) with a , b ( 0 , 1 ] and a < b but also asserts that, besides power functions, there exist some other functions F such that Cig ( a ) Cig ( F ) Cig ( b ) ; for instance, if, for any t [ 0 , ) , F ( t ) t a log ( 1 + t ) , then

a ( 0 , a ] Cig ( a ) Cig ( F ) a ( a , 1 ] Cig ( a ) .

(iv) The class of weak F -cigar domains strictly contain all weak a -cigar domains with a ( 0 , 1 ) . To show this, we only need to prove that there exist domains containing outward-directed cusps with the angle zero that are weak F -cigar domains for suitable functions F because, due to [47, Remark 4.35], such domains are not weak a -cigar domains for any a ( 0 , 1 ) . Indeed, inspired by [47, Example 4.32], we consider the dimension n = 2 , and let

Ω { ( x , y ) R 2 : 0 < x < e 2 , y < x 2 }

and, for any given ε ( 0 , 1 ] ,

(3.8) F ( t ) t ln 1 t ε if t ( 0 , e 2 ] , t + 2 ε 1 e 2 if t ( e 2 , ) .

In this case, F * ( t ) = [ ln ( 1 t ) ] ε for any t ( 0 , e 2 ] . It has been showed in [47, Example 4.32] that, for any a ( 0 , 1 ) , Ω Cig ( a ) . Next, we prove Ω Cig ( F ) . To this end, it suffices to show that, for any z 1 ( x 1 , y 1 ) , z 2 ( x 2 , y 2 ) Ω , (3.6) holds. Without loss of generality, we may assume that x 1 x 2 . To prove this, we consider two cases of the positions of z 1 and z 2 .

When z 1 and z 2 are both in a uniform subdomain of Ω , we let K 2 and t K e 2 . By the Hölder inequality, we have, for any t ( 0 , t K ] ,

0 t F * ( s ) d s 0 t ln 1 s d s ε t 1 ε ( t t ln t ) ε t 1 ε 2 t ln 1 t ε = 2 F ( t ) ,

which means that F satisfies (3.7). From this and [47, Theorem 4.12], we infer that (3.6) holds in this case.

When z 1 and z 2 are not in a uniform subdomain of Ω , z 1 must be close to ( 0 , 0 ) and x 1 x 2 must be large enough. Inspired by [47, Example 4.32], we choose the curve Γ connecting z 1 and z 2 as the particular poly-line

Γ 1 Γ 2 Γ 3 z 1 ( x 1 , 0 ) ¯ ( x 1 , 0 ) ( x 2 , 0 ) ¯ ( x 2 , 0 ) z 2 ¯ ,

that is, the union of three line segments Γ 1 , Γ 2 , and Γ 3 , which connect z 1 and ( x 1 , 0 ) , ( x 1 , 0 ) and ( x 2 , 0 ) , and ( x 2 , 0 ) and z 2 , respectively. By the definition of Ω , one has Γ Ω . Since both Γ 1 and Γ 3 are contained in some uniform subdomain of Ω , then (3.6) holds, respectively, for both z 1 and ( x 1 , 0 ) and for both ( x 2 , 0 ) and z 2 and, therefore, by Proposition 3.4, we have

z 1 ( x 1 , 0 ) ¯ F * ( dist ( z , Ω ) ) d z F ( y 1 ) F ( z 1 z 2 )

and

( x 2 , 0 ) z 2 ¯ F * ( dist ( z , Ω ) ) d z F ( y 2 ) F ( z 1 z 2 ) .

In the following, it suffices to show that

( x 1 , 0 ) ( x 2 , 0 ) ¯ F * ( dist ( z , Ω ) ) d z F ( z 1 z 2 ) ,

where F * is the same as in (3.1). From the geometric observation that, for any x ( 0 , 1 e 2 ) , dist ( ( x , 0 ) , Ω ) x 2 2 , the Hölder inequality, and the almost increasing property of F with x 1 x 2 z 1 z 2 , we deduce that

( x 1 , 0 ) ( x 2 , 0 ) ¯ F * ( dist ( z , Ω ) ) d z x 1 x 2 ln 2 x 2 ε d x ( x 2 x 1 ) + x 1 x 2 ln 1 x ε d x ( x 2 x 1 ) + x 1 x 2 ln 1 x d x ε ( x 2 x 1 ) 1 ε ( x 2 x 1 ) + [ x 2 x 1 + ( x 1 ln x 1 x 2 ln x 2 ) ] ε ( x 2 x 1 ) 1 ε ( x 2 x 1 ) + ( x 2 x 1 ) ln 1 x 2 x 1 ε ( x 2 x 1 ) 1 ε F ( x 1 x 2 ) F ( z 1 z 2 ) ,

which proves (3.6) in this case. This finishes the proof of the aforementioned claim.

Let F be the same as in (3.8). We also mention that F * Δ and, for any p ( 1 ε , 1 ] , F Λ p . Thus, except all weak a -cigar domains for a ( 0 , 1 ) , there exists at least one weak F -cigar domain such that F is in both Δ and Λ p . Indeed, on the one hand, F * is strictly decreasing, which, together with Remark 3.6(i), further implies that F Δ . On the other hand, by an ordinary calculation, β F = 2 1 ε , which, combined with Proposition 3.9(ii), further implies that, for any p ( 1 ε , 1 ] , F Λ p .

(v) There also exist domains that are not weak F -cigar domain for any modulus of continuity F . Indeed, by [47, Example 3.5], we find that the domain

Ω B ( ( 0 , 0 ) , 1 ) \ { ( x , 0 ) : 0 x < 1 } R 2

is not a Lip F -extension domain and hence, by (i) of this remark, not a weak F -cigar domain for any modulus of continuity F .

3.2 Weak self-improving property

It is known that the class of weak α -cigar domains is strictly contained in the intersection of all sets of weak α -cigar domains with α ( α , 1 ] (see both [4, Proposition 2.4] and [47, Example 6.7]). However, as pointed out by Shvartsman [59], there exists no answer whether the set of weak α -cigar domains can be obtained by the union of the sets of weak τ -cigar domains with τ strictly smaller than α . Instead, Shvartsman established and used the weak self-improving property of weak α -cigar domains, which, roughly speaking, states that there exists a rectifiable curve Γ such that, for any τ smaller than α and larger than some α * ( 0 , α ) , (2.3) holds on some subcurve of Γ , which can be arbitrarily close to Γ [59, Theorem 1.5]. The following theorem, which is on the weak self-improving property for generalized weak F -cigar domains, is a generalization of the study by Shvartsman [59, Theorem 1.5].

Theorem 3.13

Let F be a modulus of continuity with lim t F ( t ) = , F Λ 1 , and F * Δ and let Ω be a local weak F-cigar domain in R n . Then there exist constants q * ( 1 , ) , θ ( 0 , ) , and C C ( F , C F Ω ) [ 1 , ) , where C F Ω is the same as in Definition 3.10, such that, for any given ε ( 0 , ) and x , y Ω with x y θ , there exist both a rectifiable curve Γ Ω joining x and y and a subset Γ ˜ Γ consisting of a finite number of arcs satisfying that

  1. (3.9) Γ F * ( dist ( z , Ω ) ) d z C F ( x y ) and F ( leng Γ ) C F ( x y ) ;

  2. (3.10) leng ( Γ \ Γ ˜ ) < ε

    and, for any ball B centered in Γ ˜ with the radius at most x y ,

    (3.11) diam B C leng ( B Γ ˜ ) ;

  3. for any modulus of continuity G : [ 0 , ) [ 0 , ) with G * ( F * ) q * almost increasing on ( 0 , ) and G F q * almost decreasing on ( 0 , ) ,

    (3.12) Γ ˜ G * ( dist ( z , Ω ) ) d z C G ( x y ) .

Inspired by the argument in [59, pp. 2222–2223], the proof of Theorem 3.13 relies on the following three main steps. First, find a family of m -adic intervals of the level at most k (both m and k are chosen later) by the standard iterative procedure. Second, show that the function F Γ * ( t ) F * ( dist ( Γ ( t ) , Ω ) ) , t [ 0 , leng Γ ] , is a Muckenhoupt A 1 -weight with respect to the aforementioned family. Finally, apply the technique of parametrization by arc length and the construction of the family of intervals obtained in the first step to find the expected curve and its sub-curve, and then complete the proof of Theorem 3.13.

We begin with the following lemma, which, in the case where the two points x and y are far away enough from the boundary of Ω , implies the conclusion of Theorem 3.13. Throughout this subsection, we always assume that F is a modulus of continuity.

Lemma 3.14

Let F be a modulus of continuity and x , y Ω such that

max { dist ( x , Ω ) , dist ( y , Ω ) } > 2 x y .

Then the line segment x y ¯ Ω and, moreover,

x y ¯ F * ( dist ( z , Ω ) ) d z C F * , F ( x y ) ,

where F * is the same as in (3.1) and C F * , the same positive constant as in Definition 2.9 with f replaced by F * .

Proof

Without loss of generality, we may assume that dist ( x , Ω ) > 2 x y . By this, we have

x y ¯ B ( x , 2 x y ) B ( x , dist ( x , Ω ) ) Ω .

Since, for any z x y ¯ , x z x y , then, from the assumption dist ( x , Ω ) > 2 x y , we infer that

dist ( z , Ω ) dist ( x , Ω ) x z dist ( x , Ω ) x y > x y ,

which, combined with the almost decreasing property of F * (due to both the fact that F is a modulus of continuity and Definition 3.1) as well as its definition, further implies that

x y ¯ F * ( dist ( z , Ω ) ) d z C F * , x y ¯ F * ( x y ) d z = C F * , F ( x y ) .

This finishes the proof of Lemma 3.14.□

We also need the following two auxiliary lemmas, which generalize [59, Lemmas 2.2 and 2.7], respectively.

Lemma 3.15

Let F be a modulus of continuity, x , y Ω , Γ x y Ω be a rectifiable curve joining x and y, and L leng ( Γ x y ) . If there exists a constant λ [ 1 , ) such that

(3.13) Γ x y F * ( dist ( z , Ω ) ) d z λ F * ( L ) ,

then

  1. for any given c ( 0 , ) , there exist z 0 Γ x y and a positive constant C 1 , depending only on λ , c , and F * , such that

    dist ( z 0 , Ω ) = max { dist ( z , Ω ) : z Γ x y }

    and

    F * ( dist ( z 0 , Ω ) c ) C 1 F * ( L ) ;

  2. there exists a positive constant C 2 , depending only on both λ and F * , such that

    Γ x y F * ( dist ( z , Ω ) ) d z C 2 inf z Γ x y F * ( dist ( z , Ω ) ) = C 2 min z Γ x y F * ( dist ( z , Ω ) ) ,

    where F * is the same as in (3.1).

Proof

By Remark 3.2(ii), we find that F 0 .

We first prove (i). The existence of z 0 is obvious. Then, by the definition of z 0 , Proposition 3.4 with c therein replaced by 1 c , Proposition 2.11, and (3.13), we conclude that

F * dist ( z 0 , Ω ) c = Γ x y F * max z Γ x y { dist ( z , Ω ) } c d z c F ¯ ( 1 c ) Γ x y F * ( max z Γ x y { dist ( z , Ω ) } ) d z Γ x y min z Γ x y F * ( dist ( z , Ω ) ) d z Γ x y F * ( dist ( z , Ω ) ) d z F * ( L ) ,

where F ¯ ( 1 c ) is the same as in (2.9) with ϕ replaced by F . This shows (i).

Now, we prove (ii). Since, for any z 1 , z 2 Γ x y ,

dist ( z 1 , Ω ) dist ( z 2 , Ω ) + z 1 z 2 dist ( z 2 , Ω ) + L ,

it follows that

max z Γ x y dist ( z , Ω ) min z Γ x y dist ( z , Ω ) + L ,

which further implies that, whenever max z Γ x y dist ( z , Ω ) > 2 L ,

(3.14) 1 2 max z Γ x y dist ( z , Ω ) min z Γ x y dist ( z , Ω ) .

Using both (2.5) with the almost decreasing function F , the continuous function f , and the closed set I therein replaced, respectively, by F * , dist ( , Ω ) 2 , and Γ x y , and Proposition 3.4 with c therein replaced by 1/2, we find that

(3.15) F * 1 2 max z Γ x y { dist ( z , Ω ) } min z Γ x y F * 1 2 dist ( z , Ω ) 2 F ¯ 1 2 min z Γ x y { F * ( dist ( z , Ω ) ) }

with F ¯ ( 1 2 ) in (2.9) with ϕ replaced by F and the implicit positive constants depending only on λ and C F * , , where C F * , is the same positive constant as in Definition 2.9 with f replaced by F * . By (2.6), we also conclude that

(3.16) max z Γ x y { F * ( dist ( z , Ω ) ) } F * ( min z Γ x y { dist ( z , Ω ) } ) .

Therefore, when max z Γ x y dist ( z , Ω ) 2 L , applying (3.13), the almost decreasing property of F * with L 1 2 max z Γ x y dist ( z , Ω ) , and (3.15), we find that

Γ x y F * ( dist ( z , Ω ) ) d z λ F * 1 2 max z Γ x y { dist ( z , Ω ) } λ min z Γ x y { F * ( dist ( z , Ω ) ) } ,

while, when max z Γ dist ( z , Ω ) > 2 L , from (3.16), the almost decreasing property of F * with (3.14), (3.15), and λ [ 1 , ) , we infer that

Γ x y F * ( dist ( z , Ω ) ) d z Γ x y max z Γ x y { F * ( dist ( z , Ω ) ) } d z F * ( min z Γ x y { dist ( z , Ω ) } ) F * 1 2 max z Γ x y { dist ( z , Ω ) } λ min z Γ x y { F * ( dist ( z , Ω ) ) } .

This finishes the proof of (ii), and hence, Lemma 3.15.□

Lemma 3.16

Let F be the same as in Definition 3.10, Ω be a locally weak F-cigar domain, and x , y Ω such that F ( x y ) F ( R F Ω ) and

(3.17) 2 x y max { dist ( x , Ω ) , dist ( y , Ω ) } ,

where R F Ω is the same as in Definition 3.10. Let δ ( 0 , d F Ω ( x , y ) ] with d F Ω ( x , y ) in (3.3), F * be the same as in (3.1), Γ x y Ω a rectifiable curve joining x and y such that

(3.18) Γ x y F * ( dist ( z , Ω ) ) d z < d F Ω ( x , y ) + δ ,

and L leng ( Γ x y ) . Then

  1. (3.19) Γ x y F * ( dist ( z , Ω ) ) d z 2 C F Ω F ( x y ) ,

    where C F Ω [ 1 , ) is the same as in Definition 3.10, and, furthermore, there exists a constant C 3 [ 1 , ) , depending only on both F and C F Ω , such that F ( L ) C 3 F ( x y ) ;

  2. there exists a constant C 4 [ 1 , ) , depending only on both F and C F Ω , such that, for any u , v Γ x y with F ( u v ) F ( R F Ω ) and F ( leng Γ u v ) δ ,

    (3.20) Γ u v F * ( dist ( z , Ω ) ) d z C 4 F * ( leng Γ u v ) ,

    where Γ u v denotes the sub-curve of Γ x y joining u and v.

Proof

From Remark 3.2(ii), it follows that F 0 .

We first show (i). By using (3.18), the assumptions of x , y , and δ , and (3.6), we conclude that (3.19) holds.

To prove the remainder of (i), without loss of generality, we may assume that L > 2 x y ; otherwise, by the almost increasing property of F , there exists nothing else to show. From this assumption and (3.17), it follows that

(3.21) max t [ 0 , L ] { dist ( Γ ( t ) , Ω ) } max t [ 0 , L ] { x Γ ( t ) } + dist ( x , Ω ) L + 2 x y < 2 L .

Thus, by the definition of F * , the almost decreasing property of F * with (3.21), Proposition 3.4 with c replaced by 1/2, (2.5), and (3.19), we conclude that

F ( L ) = Γ x y F * ( L ) d z Γ x y F * 1 2 max t [ 0 , L ] { dist ( Γ ( t ) , Ω ) } d z 2 F ¯ ( 1 2 ) Γ x y F * ( max t [ 0 , L ] { dist ( Γ ( t ) , Ω ) } ) d z Γ x y min t [ 0 , L ] F * ( { dist ( Γ ( t ) , Ω ) } ) d z Γ x y F * ( dist ( Γ ( t ) , Ω ) ) d z F ( x y ) ,

where the implicit positive constants depend on both F and C F Ω .

Next, we prove (ii). We first claim that, for any u , v Γ x y ,

(3.22) Γ u v F * ( dist ( z , Ω ) ) d z < d F Ω ( u , v ) + δ .

We show this by contradiction. Assume that there exist u 0 , v 0 Γ x y such that

Γ u 0 v 0 F * ( dist ( z , Ω ) ) d z d F Ω ( u 0 , v 0 ) + δ .

By this, (3.3), and the triangle inequality for the metric d F Ω , we find that

Γ x y F * ( dist ( z , Ω ) ) d z = Γ x u 0 F * ( dist ( z , Ω ) ) d z + Γ u 0 v 0 + Γ v 0 y d F Ω ( x , u 0 ) + [ d F Ω ( u 0 , v 0 ) + δ ] + d F Ω ( v 0 , y ) d F Ω ( x , y ) + δ ,

which contradicts (3.18). This proves the aforementioned claim.

Since, for any u , v Γ x y , u v leng Γ u v , then, from (3.22), (3.6) with x and y therein replaced, respectively, by u and v , and the almost increasing property of F with u v leng Γ u v , we deduce that, for any u , v Γ x y with both F ( u v ) F ( R F Ω ) and F ( leng Γ u v ) δ ,

Γ u v F * ( dist ( z , Ω ) ) d z < C F Ω F ( u v ) + δ F ( u v ) + F ( leng Γ u v ) F ( leng Γ u v ) ,

where the implicit positive constant depends on both F and C F Ω . Thus, (3.20) holds. This finishes the proof of (ii), and hence, Lemma 3.16.□

Let L ( 0 , ) and I 0 [ 0 , L ] . For any k Z + and m N , define

(3.23) S k , m j = 0 k j , m ,

where, for any m N and j Z + with j k ,

j , m L m j i , L m j ( i + 1 ) : i { 0 , , m j 1 } .

Let S S k , m . Recall that a function f , defined on U { I : I S } , is called an m-adic Muckenhoupt A 1 -weight on U with respect to S if there exists a constant C f [ 1 , ) such that, for any t U ,

(3.24) sup I S , I t I f ( u ) d u C f f ( t ) ,

where the supremum is taken over all elements of the given family S , which contains t ; see [53] for more information on this concept.

Definition 3.17

A family S of intervals is said to be nonoverlapping if, for any I , J S with I J ,

( I \ I ) ( J \ J ) = .

The following lemma is precisely [59, Corollary 2.10].

Lemma 3.18

Let k Z + , m N , S S k , m be a family of intervals in (3.23) with some given L ( 0 , ) , f an m -adic Muckenhoupt A 1 -weight with respect to this family S, and

(3.25) q * 1 2 1 + log m log ( m C f 1 [ m 1 ] ) ,

where C f [ 1 , ) is the same positive constant as in (3.24). Then, for any q [ 1 , q * ] , there exists a positive constant C, depending on both m and C f , such that, for any nonoverlapping subfamily S 0 S ,

1 L I S 0 I f ( t ) d t q I 1 q C 1 L 0 L f ( t ) d t .

In the remainder of the proof of Theorem 3.13, as mentioned earlier, after finding a family of m -adic intervals of the level at most k as the first step and proving F Γ * to be a Muckenhoupt A 1 -weight with respect to this family as the second step, we then use Lemma 3.18 and the characteristic of this family to show (3.12), which is the main part of the following proof of Theorem 3.13.

Proof of Theorem 3.13

Notice that, for any x , y Ω with

max { dist ( x , Ω ) , dist ( y , Ω ) } > 2 x y ,

by setting Γ x y ¯ Γ ˜ and using Lemma 3.14 twice with F therein replaced, respectively, by F and G , we conclude that (3.9)–(3.12) hold. Therefore, without loss of generality, we may assume that

max { dist ( x , Ω ) , dist ( y , Ω ) } 2 x y .

Let m be any given positive integer and k be some positive integer, which is determined later; let

(3.26) θ C 3 1 C F , 1 F ( R F Ω )

and x , y Ω with F ( x y ) θ ; also, let

(3.27) δ min { d F Ω ( x , y ) , C F , 1 F ( m k x y ) } ,

where C 3 , C F , [ 1 , ) , and R F Ω are the positive constants, respectively, in Lemma 3.16(i), Definition 2.9 with f replaced by F , and Definition 3.10 and where d F Ω ( x , y ) is defined in (3.3). Then, by (3.3), we find that there exists a rectifiable curve Γ Ω joining x and y such that

Γ F * ( dist ( z , Ω ) ) d z < d F Ω ( x , y ) + δ .

From this, (3.27), F ( x y ) θ , (3.26), and Lemma 3.16(i), it follows that

(3.28) Γ F * ( dist ( z , Ω ) ) d z 2 C F Ω F ( x y )

and

(3.29) F ( L ) C 3 F ( x y ) C F , 1 F ( R F Ω ) ,

where L leng ( Γ ) .

For k and m as mentioned earlier, define

S k , m j = 0 k j , m j = 0 k L i m j , L ( i + 1 ) m j : i { 0 , , m j 1 } .

Below, we assume that the curve Γ is parameterized by arc length via the function Γ : [ 0 , L ] Ω with Γ ( 0 ) = x and Γ ( L ) = y and, for simplicity, for any j { 0 , , k } and i { 0 , , m j 1 } , we let u j , i Γ ( m j L i ) and Γ j , i Γ ( [ m j L i , m j L ( i + 1 ) ] ) .

By the definition of S k , m , we find that, for any j { 0 , , k } and I j , m , I = m j L and hence, for any I S k , m ,

I m k L m k x y .

Thus, by this, the almost increasing property of F , and (3.27), we obtain

F ( I ) C F , 1 F ( m k x y ) δ ,

which further implies that, for any j { 0 , , k } and i { 0 , , m j 1 } ,

(3.30) F ( leng ( Γ j , i ) ) = F ( m j L ) = F ( I ) δ .

In addition, by both the almost increasing property of F with Γ ( m j L i ) Γ ( m j L [ i + 1 ] ) L and (3.29), we also have, for any j { 0 , , k } and i { 0 , 1 , , m j 1 } ,

(3.31) F ( u j , i u j , i + 1 ) = F ( Γ ( m j L i ) Γ ( m j L [ i + 1 ] ) ) C F , F ( L ) F ( R F Ω ) .

By combining (3.30) and (3.31) and applying Lemma 3.16(ii) with u , v , and Γ u v therein replaced, respectively, by u j , i , u j , i + 1 , and Γ j , i , we conclude that

(3.32) Γ j , i F * ( dist ( z , Ω ) ) d z C 4 F * ( [ m j L i , m j L ( i + 1 ) ] ) = C 4 F * ( m j L ) ,

where C 4 [ 1 , ) is the same constant as in Lemma 3.16(ii). Observe that (3.32) is precisely (3.13) with Γ x y , L , and λ therein replaced, respectively, by Γ j , i , m j L , and C 4 . Thus, by Lemma 3.15(ii), we conclude that there exists a positive constant C 2 , depending on both C 4 and F , such that, for any j { 0 , , k } and i { 0 , , m j 1 } ,

Γ j , i F * ( dist ( z , Ω ) ) d z C 2 min z Γ j , i F * ( dist ( z , Ω ) ) ,

which, together with the arbitrariness of both i and j , further implies that, for any I S k , m ,

Γ ( I ) F * ( dist ( z , Ω ) ) d z C 2 min z Γ ( I ) F * ( dist ( z , Ω ) ) ,

and hence,

I F Γ * ( t ) d t C 2 min t I F Γ * ( t ) .

From this, we infer that, for any t [ 0 , L ] ,

sup I S k , m , I t I F Γ * ( t ) d t C 2 F Γ * ( t ) .

Thus, F Γ * is an m -adic Muckenhoupt A 1 -weight with respect to the family S k , m . Here, and thereafter, F Γ * is defined in (3.2).

Next, we prove that, for any I j , m and j { 0 , , k 1 } , there exist t I I and a subinterval I ˜ j + 1 , m of I with I ˜ = I m such that

(3.33) max t I ˜ F Γ * ( t ) F Γ * ( t I ) min t I ˜ F Γ * ( t ) ,

where the implicit positive constants are independent of Γ , k , m , and I .

Indeed, on the one hand, for any given j { 0 , , k 1 } and any given interval I j , m , we denote by z I Γ ( I ) the point such that

dist ( z I , Ω ) = max { dist ( z , Ω ) : z Γ ( I ) } ,

by t I I the point such that z I = Γ ( t I ) , and I ˜ j + 1 , m be the subinterval of I such that t I I ˜ and I ˜ = I m . By using (3.32), Lemma 3.15(i) with Γ x y , L , λ , and c therein replaced, respectively, by Γ j , i , m j L , C 4 , and 2 C F * Δ , and the arbitrariness of both i and j , we find that there exists a positive constant C 1 [ 1 , ) such that, for any j { 0 , , k 1 } and any interval I j , m ,

(3.34) F * C F * Δ dist ( z I , Ω ) 2 C 1 F * ( I ) ,

where C F * Δ is the same positive constant in Definition 3.5 with g replaced by F * and C 1 depends only on C F * Δ , C 4 , and F .

On the other hand, the assumption F Λ 1 , together with Definition 3.7 with r 1 m , enables us to choose m Z + large enough such that

(3.35) sup t ( 0 , L ] F ( t ) m F ( t m ) = sup t ( 0 , L ] F * ( t ) F * ( t m ) < 1 C F * , C 1 .

Notice that, for any j N , m j L ( 0 , L ] . Thus, for any j N , by choosing t m j L in (3.35), we obtain

(3.36) F * ( m j 1 L ) F * ( m j L ) > C F * , C 1 .

Recall that z I Γ ( I ˜ ) . Then, for any I S k 1 , m and t I ˜ , we have

dist ( z I , Ω ) dist ( Γ ( t ) , Ω ) z I Γ ( t ) leng ( Γ ( I ˜ ) ) = I ˜ = I m ,

which, combined with the almost decreasing property of F * , (3.36) with m j L = I , and (3.34), further implies that, for any t I ˜ ,

F * ( dist ( z I , Ω ) dist ( z ( t ) , Ω ) ) C F * , 1 F * ( I m ) > C 1 F * ( I ) F * ( C F * Δ dist ( z I , Ω ) 2 ) .

Thus, by the arbitrariness of t I ˜ , we obtain

F * ( max t I ˜ dist ( Γ ( t ) , Ω ) min t I ˜ dist ( Γ ( t ) , Ω ) ) > F * 1 2 C F * Δ max t I ˜ dist ( Γ ( t ) , Ω ) .

By using this and F * Δ , we conclude that

max t I ˜ dist ( Γ ( t ) , Ω ) min t I ˜ dist ( Γ ( t ) , Ω ) 1 2 max t I ˜ dist ( Γ ( t ) , Ω ) ,

which further implies that

1 2 dist ( z I , Ω ) = 1 2 max t I ˜ dist ( Γ ( t ) , Ω ) min t I ˜ dist ( Γ ( t ) , Ω ) max t I ˜ dist ( Γ ( t ) , Ω ) .

From this, the almost decreasing property of F * , Proposition 2.11, Proposition 3.4 with c replaced by 1/2, and the definition of z I , it follows that, for any I S k 1 , m , there exists a subinterval I ˜ of I in S k , m such that

max t I ˜ F * ( dist ( z ( t ) , Ω ) ) F * ( min t I ˜ dist ( z ( t ) , Ω ) ) F * 1 2 dist ( z I , Ω ) F * ( dist ( z I , Ω ) ) min t I ˜ F * ( dist ( z ( t ) , Ω ) ) ,

which shows (3.33).

Following the proof of [59, Theorem 1.5], we define I 0 [ 0 , L ] , A 1 { I 0 ˜ } , U 1 I 0 ˜ , and

D 1 { I 1 , m : ( I \ I ) [ 0 , L ] \ I 0 ˜ } .

Here, I 0 ˜ is precisely the same subinterval as in (3.33) with j 0 . Then I 0 ˜ 1 , m with I 0 ˜ = I 0 m . Next, if k 2 , then, for any j { 2 , , k } , define

A j { I ˜ j , m : I D j 1 } , U j U j 1 ( I A j I ) ,

and

D j { I j , m : ( I \ I ) [ 0 , L ] \ U j } .

Write U U k and S 0 j = 1 k A j . By this construction, we find that S 0 is a subfamily of S k , m consisting of nonoverlapping intervals (Definition 3.17) and, furthermore, by (3.33), we have, for any I S 0 ,

(3.37) max t I F Γ * ( t ) min t I F Γ * ( t ) ,

where the implicit positive constant depends only on F .

Let q * ( 1 , ) be the same as in (3.25) with m in (3.36) and C f therein replaced by C F Γ * , where C F Γ * is the same positive constant as in (3.24) with f replaced by F Γ * . Notice that U = I S 0 I . Thus, from (3.37), Lemma 3.18 with f replaced by F Γ * , and (3.28), we deduce that

(3.38) 1 L U [ F Γ * ( t ) ] q * d t = 1 L I S 0 I I [ F Γ * ( t ) ] q * d t 1 L I S 0 I [ max t I F Γ * ( t ) ] q * 1 L I S 0 I min t I F Γ * ( t ) q * 1 L I S 0 I I F Γ * ( t ) d t q * 1 L 0 L F Γ * ( t ) d t q * L q * [ F ( x y ) ] q * ,

where the implicit positive constants depend on F , C F , Ω , m , and q * .

By the assumption that max { dist ( x , Ω ) , dist ( y , Ω ) } 2 x y , we conclude that, for any z Γ ,

dist ( z , Ω ) dist ( x , Ω ) + x z 2 x y + L 3 L .

Using this, the almost increasing property of G * [ ( F * ) q * ] , F , G 0 [see Remark 3.2(ii)], (3.38), and the almost decreasing property of G ( F q * ) with x y L , we find that

(3.39) U G * ( dist ( z , Ω ) ) d z = U G * ( dist ( z , Ω ) ) [ F * ( dist ( z , Ω ) ) ] q * [ F * ( dist ( Γ ( t ) , Ω ) ) ] q * d t G * ( L ) [ F * ( L ) ] q * U [ F Γ * ( t ) ] q * d t L 1 q * G * ( L ) [ F * ( L ) ] q * [ F ( x y ) ] q * L 1 q * G * ( L ) [ F * ( L ) ] q * [ F ( L ) ] q * G ( L ) G ( x y ) G ( x y ) .

Since, by the construction of U , we have [ 0 , L ] \ U = ( m 1 m ) k L , then, from (3.29), we infer that

(3.40) F m m 1 k [ 0 , L ] \ U = F ( L ) C 3 F ( x y ) .

By using lim t F ( t ) = , we conclude that there exists N ( 0 , ) such that F ( N ) > C F , C 3 F ( x y ) . For such an N and any given ε ( 0 , ) , by choosing k N large enough such that [ ( m 1 ) m ] k N < ε , we conclude that, if [ 0 , L ] \ U ε , then, by the almost increasing property of F , (3.40), and the choice of N , it holds that

F ( N ) C F , F m m 1 k [ 0 , L ] \ U C F , C 3 F ( x y ) < F ( N ) ,

which leads to a contradiction. Thus, we obtain

(3.41) [ 0 , L ] \ U < ε .

By letting Γ ˜ Γ U , we find that (3.9) follows from both (3.28) and (3.29), (3.10) follows from (3.41), and (3.12) follows from (3.39).

Finally, since the proof of the remaining inequality (3.11) depends only on the construction of both the family S 0 and the set U , which has been provided in the proof of [59, Theorem 1.5], we omit the details. This finishes the proof of Theorem 3.13.□

Example 3.19

Let q * ( 1 , ) be the same as in Theorem 3.13.

  1. Let a ( 0 , 1 ) and, for any t [ 0 , ) , define F ( t ) t a . Obviously, F Λ 1 and F * Δ . In addition, for any given b [ 1 q * ( 1 a ) , a ] and any t [ 0 , ) , let G ( t ) t b . Since

    b 1 q * ( a 1 ) 0 and b a q * < b a 0 ,

    we deduce that G * ( t ) [ F * ( t ) ] q * = t b 1 q * ( a 1 ) is increasing and G ( t ) [ F ( t ) ] q * = t b a q * is decreasing. Thus, the function G satisfies all the assumptions of Theorem 3.13(iii). Under all these assumptions, Theorem 3.13 reduces back to [59, Theorem 1.5] with α , α * , and τ therein replaced, respectively, by a , 1 q * ( 1 a ) , and b .

  2. Let F : [ 0 , ) [ 0 , ) be a continuous function of admissible growth with both α F ( 0 , 1 ) and β F ( 0 , 2 ) , where α F and β F are the same as in (2.10) with ϕ replaced by F . By Remark 3.2(iii), F is a modulus of continuity and, by Proposition 3.9, F Λ 1 and F * Δ . Furthermore, we have 1 2 < β F 2 < 1 α F β F < 2 . From this, we infer that q * ( log 2 β F 1 ) < q * log 2 α F . Thus, there always exist continuous functions G : [ 0 , ) [ 0 , ) of admissible growth with both α G ( 0 , 1 ) and β G ( 0 , 2 ) such that 1 log 2 α G > q * ( log 2 β F 1 ) and

    log 2 β G < 1 q * log 2 α F .

    For such a function G , we have α H 1 = 2 1 q * α G ( β F ) q * ( 0 , 1 ) and β H 2 = β G ( α F ) q * ( 0 , 2 ) , which further implies that H 1 is almost increasing and H 2 is almost decreasing, where H 1 G * ( F * ) q * and H 2 G F q * . Thus, G satisfies all the assumptions of Theorem 3.13(iii).

4 Relations between ϕ -Hajłasz-Sobolev embedding and extension domains

In this section, we discuss some embedding properties of ϕ -Hajłasz-Sobolev spaces. On the one hand, we prove that M ball ϕ , p -extension domains are M ball ϕ , p -embedding domains; see Remark 4.2. On the other hand, we show that M ball ϕ , p -embedding domains are M ball ϕ , q -extension domains for any q > p ; see Theorem 4.10 for more details. The Whitney ball-covering and the related reflected quasi-balls play a key role in the construction of the desired extension mappings.

Let p ( 0 , ] and ϕ A . Recall that ϕ ( p ) is defined in (1.5). A basic embedding result for ϕ -Hajłasz-Sobolev spaces on R n reads as follows.

Proposition 4.1

Let ϕ be a modulus of continuity with α ϕ ( 0 , 1 ) and let p ( n ( log 2 α ϕ ) , ] . Then

  1. M ˙ ϕ , p ( R n ) is embedded into M ˙ ϕ ( p ) , ( R n ) ;

  2. M ϕ , p ( R n ) is embedded into M ϕ ( p ) , ( R n ) .

Proof

Notice that, when p = , one has ϕ ( ) = ϕ . Thus, there is nothing to prove for both (i) and (ii) in this case.

Now, we let p < . By p > n ( log 2 α ϕ ) and Remark 2.13(i), we find that ϕ ( p ) is a well-defined almost increasing function on [ 0 , ) with ϕ ( p ) ( 0 ) = 0 . From the assumption of ϕ and Remark 2.13(ii), we deduce that log 2 α ϕ ( 0 , 1 ] , and hence, p > n 2 .

To show (i), let u M ˙ ϕ , p ( R n ) and g D ϕ ( u ) with g L p ( R n ) u M ˙ ϕ , p ( R n ) . For any Lebesgue points x , y of u , by g D ϕ ( u ) , the almost increasing property of ϕ with the observation that z 1 z 2 2 k + 2 x y for any z 1 B ( x , 2 k + 1 x y ) and z 2 B ( x , 2 k x y ) , the Hölder inequality with p > 1 , and (2.12) with α ϕ ( p ) = 2 n p α ϕ ( 0 , 1 ) , we have

(4.1) u ( x ) u B ( y , 2 x y ) k = 0 u B ( x , 2 k x y ) u B ( x , 2 k + 1 x y ) + u B ( x , 2 x y ) u B ( y , 2 x y ) k = 0 B ( x , 2 k x y ) B ( x , 2 k + 1 x y ) u ( z 1 ) u ( z 2 ) d z 1 d z 2 k = 0 ϕ ( 2 k + 2 x y ) B ( x , 2 k + 1 x y ) g ( z ) d z k = 0 ϕ ( p ) ( 2 k + 2 x y ) B ( x , 2 k + 1 x y ) [ g ( z ) ] p d z 1 p ϕ ( p ) ( x y ) g L p ( R n ) ;

similarly, we also have

(4.2) u ( y ) u B ( y , 2 x y ) ϕ ( p ) ( x y ) g L p ( R n ) .

Thus, by combining (4.1) and (4.2), we conclude that

u ( x ) u ( y ) u ( x ) u B ( y , 2 x y ) + u B ( y , 2 x y ) u ( y ) ϕ ( p ) ( x y ) g L p ( R n ) ,

which further implies u M ˙ ϕ ( p ) , ( R n ) g L p ( R n ) u M ˙ ϕ , p ( R n ) . This finishes the proof of (i).

Next, we prove (ii). Let u M ϕ , p ( R n ) , g D ϕ ( u ) with g L p ( R n ) u M ˙ ϕ , p ( R n ) , and N be any given positive constant. Thus, by the Hölder inequality with p > 1 , (4.2) with y and 2 x y replaced, respectively, by x and N , and g L p ( R n ) u M ˙ ϕ , p ( R n ) , we find that, for almost every x R n ,

u ( x ) u ( x ) u B ( x , N ) + u B ( x , N ) u ( x ) u B ( x , N ) + B ( x , N ) u ( z ) p d z 1 p ϕ ( p ) ( N ) g L p ( R n ) + N n p u L p ( R n ) ϕ ( p ) ( N ) + N n p u M ϕ , p ( R n ) ,

which means that u L ( R n ) u M ϕ , p ( R n ) . This, together with the conclusion of (i), further implies that u M ϕ ( p ) , ( R n ) u M ϕ , p ( R n ) , which completes the proof of (ii), and hence, Proposition 4.1.□

Remark 4.2

Recall that the M ˙ ball ϕ , p -embedding domain is defined in Definition 1.1 and the extension domain is defined via (1.3). We claim that, if Ω is an M ˙ ball ϕ , p -extension domain, then Ω is also an M ˙ ball ϕ , p -embedding domain. Indeed, by using Proposition 4.1 with the same assumptions of both ϕ and p as therein, we conclude that

M ˙ ball ϕ , p ( R n ) = M ˙ ϕ , p ( R n ) M ˙ ϕ ( p ) , ( R n ) ,

which means that R n is an M ˙ ball ϕ , p -embedding domain. From this and (1.2), we further infer that M ˙ ball ϕ , p ( R n ) Ω M ˙ ϕ ( p ) , ( R n ) Ω . Thus, if Ω is an M ˙ ball ϕ , p -extension domain, we then have

M ˙ ball ϕ , p ( Ω ) = M ˙ ball ϕ , p ( R n ) Ω M ˙ ϕ ( p ) , ( R n ) Ω M ˙ ϕ ( p ) , ( Ω ) ,

where the last embedding naturally holds by (1.2). This means that Ω is an M ˙ ball ϕ , p -embedding domain, which shows the aforementioned claim. The aforementioned claim is also true for the inhomogeneous case.

To characterize the extension and the embedding domains of M ϕ , p spaces, we first state several technical lemmas. To prove them, we borrow some ideas from the proof in the study by Zhou [66, Theorem 4.1].

Lemma 4.3

Let ϕ be a modulus of continuity and p [ n ( log 2 α ϕ ) , ] such that ϕ ( p ) in (1.5) is a well-defined almost increasing function on [ 0 , ) satisfying ϕ ( p ) ( 0 ) = 0 . Assume also that Ω supports the embedding from M ball ϕ , p ( Ω ) to M ˙ ϕ ( p ) , ( Ω ) . Then there exist positive constants λ and C such that, for any u M ˙ ball ϕ , p ( Ω ) and almost every x , y Ω with x y λ ,

(4.3) u ( x ) u ( y ) C ϕ ( p ) ( x y ) u M ˙ ball ϕ , p ( Ω ) .

Proof

Let u M ˙ ball ϕ , p ( Ω ) and g D ball ϕ ( u ) with g L p ( Ω ) u M ˙ ball ϕ , p ( Ω ) . Notice that, for almost every x Ω , u ( x ) < . Fix λ 1 ( 0 , ) and let x , y Ω be such that x y < λ 1 and u ( x ) u ( y ) < . Without loss of generality, we may assume that, for any z Ω , 0 u ( y ) u ( z ) u ( x ) ; otherwise we can consider the replacements

u ˜ + u ˜ 1 { z Ω : u ˜ ( z ) 0 } and u ˜ u ˜ 1 { z Ω : u ˜ ( z ) 0 } ,

where

u ˜ u ( x ) 1 { z Ω : u ( z ) > u ( x ) } + u ( y ) 1 { z Ω : u ( z ) < u ( y ) } + u 1 { z Ω : u ( y ) u ( z ) u ( x ) } .

Here, we find that, for any z Ω ,

0 u ˜ + ( y ) u ˜ + ( z ) u ˜ + ( x ) , 0 u ˜ ( x ) u ˜ ( z ) u ˜ ( y ) ,

u ( x ) u ( y ) u ˜ + ( x ) u ˜ + ( y ) + u ˜ ( x ) u ˜ ( y ) ,

and

u ˜ + M ˙ ball ϕ , p ( Ω ) + u ˜ M ˙ ball ϕ , p ( Ω ) 2 u M ˙ ball ϕ , p ( Ω ) .

This means that, if (4.3) holds for both u ˜ + and u ˜ , then (4.3) also holds for u .

Now, we introduce an auxiliary function, which is supported in some ball. Let φ be a λ 1 1 -Lipschitz function with supp φ B ( x , 2 λ 1 ) , 0 φ 1 , and φ 1 B ( x , λ 1 ) 1 . Define v [ u u ( y ) ] φ and

g ˜ { g + [ ϕ ( λ 1 ) ] 1 u ( x ) u ( y ) } 1 B ( x , 4 λ 1 ) Ω .

Then, by φ ( x ) = 1 and v ( y ) = 0 , we have

(4.4) v ( x ) v ( y ) = u ( x ) u ( y ) .

Next, we show that g ˜ is a positive harmless constant multiple of an element in D ball ϕ ( v ) . For almost every z 1 , z 2 Ω with z 1 z 2 min { dist ( z 1 , Ω ) , dist ( z 2 , Ω ) } 2 , we argue it by considering four cases.

Case 1. When z 1 , z 2 B ( x , 4 λ 1 ) Ω , then, by the definition of v , the λ 1 1 -Lipschitz continuity of φ , g D ball ϕ ( u ) , the almost decreasing property of ϕ * with z 1 z 2 8 λ 1 , 0 φ 1 , and Proposition 3.3 with F * , x , y , and c therein replaced, respectively, by ϕ * , 8 λ 1 , λ 1 , and 8, we have

(4.5) v ( z 1 ) v ( z 2 ) u ( z 1 ) u ( y ) φ ( z 1 ) φ ( z 2 ) + φ ( z 2 ) u ( z 1 ) u ( z 2 ) λ 1 1 z 1 z 2 u ( x ) u ( y ) + φ L ( R n ) ϕ ( z 1 z 2 ) [ g ( z 1 ) + g ( z 2 ) ] λ 1 1 ϕ ( z 1 z 2 ) ϕ * ( z 1 z 2 ) u ( x ) u ( y ) + ϕ ( z 1 z 2 ) [ g ( z 1 ) + g ( z 2 ) ] ϕ ( z 1 z 2 ) u ( x ) u ( y ) λ 1 ϕ * ( 8 λ 1 ) + g ( z 1 ) + g ( z 2 ) ϕ ( z 1 z 2 ) u ( x ) u ( y ) ϕ ( λ 1 ) + g ( z 1 ) + g ( z 2 ) ϕ ( z 1 z 2 ) [ g ˜ ( z 1 ) + g ˜ ( z 2 ) ] ,

where the implicit positive constants depend only on ϕ .

Case 2. When z 1 B ( x , 2 λ 1 ) Ω and z 2 Ω \ B ( x , 4 λ 1 ) , from supp v B ( x , 2 λ 1 ) , 0 φ 1 , the assumption of u , and the almost increasing property of ϕ with z 1 z 2 > 2 λ 1 > λ 1 , we deduce that

(4.6) v ( z 1 ) v ( z 2 ) = v ( z 1 ) u ( z 1 ) u ( y ) u ( x ) u ( y ) = ϕ ( z 1 z 2 ) [ ϕ ( z 1 z 2 ) ] 1 u ( x ) u ( y ) ϕ ( z 1 z 2 ) [ ϕ ( λ 1 ) ] 1 u ( x ) u ( y ) ϕ ( z 1 z 2 ) [ g ˜ ( z 1 ) + g ˜ ( z 2 ) ] .

Case 3. When z 2 B ( x , 2 λ 1 ) Ω and z 1 Ω \ B ( x , 4 λ 1 ) , by the symmetry, we find that the same estimate as (4.6) holds.

Case 4. In all other cases, one has v ( z 1 ) = 0 = v ( z 2 ) and hence v ( z 1 ) v ( z 2 ) = 0 .

Altogether, we conclude that g ˜ is a positive harmless constant multiple of an element in D ball ϕ ( v ) . From this, the definitions of both v and g ˜ , the choice of g , and supp v B ( x , 2 λ 1 ) , we infer that

(4.7) v M ball ϕ , p ( Ω ) g ˜ L p ( Ω ) + v L p ( Ω ) g L p ( Ω ) + [ ϕ ( λ 1 ) ] 1 u ( x ) u ( y ) 1 B ( x , 4 λ 1 ) L p ( R n ) + u ( x ) u ( y ) 1 B ( x , 2 λ 1 ) L p ( R n ) u M ˙ ball ϕ , p ( Ω ) + { 1 + [ ϕ ( λ 1 ) ] 1 } λ 1 n p u ( x ) u ( y ) ,

where the implicit positive constant is independent of λ 1 , and hence, v M ball ϕ , p ( Ω ) . By the assumption of Ω , we conclude that v M ˙ ϕ ( p ) , ( Ω ) and v M ˙ ϕ ( p ) , ( Ω ) v M ball ϕ , p ( Ω ) . From this, Remark 2.16(ii) with u and ϕ therein replaced, respectively, by v and ϕ ( p ) , and (4.7), we deduce that

(4.8) u ( x ) u ( y ) = v ( x ) v ( y ) ϕ ( p ) ( x y ) v M ˙ ϕ ( p ) , ( Ω ) ϕ ( p ) ( x y ) v M ball ϕ , p ( Ω ) ϕ ( p ) ( x y ) u M ˙ ball ϕ , p ( Ω ) + λ 1 n p + [ ϕ ( p ) ( λ 1 ) ] 1 ϕ ( p ) ( x y ) u ( x ) u ( y ) ,

where the implicit positive constant in the last quantity, denoted by C 0 , is independent of λ 1 .

By ϕ ( p ) ( 0 ) = 0 , we are able to choose λ ( 0 , λ 1 ] small enough such that

ϕ ( p ) ( λ ) 1 2 C ϕ ( p ) , C 0 { λ 1 n p + [ ϕ ( p ) ( λ 1 ) ] 1 } ,

where C ϕ ( p ) , is the same positive constant as in Definition 2.9 with f replaced by ϕ ( p ) . Then, from the almost increasing property of ϕ ( p ) , we infer that, for almost every x , y Ω with x y λ ,

ϕ ( p ) ( x y ) C ϕ ( p ) , ϕ ( p ) ( λ ) 1 2 C 0 { λ 1 n p + [ ϕ ( p ) ( λ 1 ) ] 1 } ,

and, furthermore, from (4.8), it follows that

u ( x ) u ( y ) C 0 ϕ ( p ) ( x y ) u M ˙ ball ϕ , p ( Ω ) + 1 2 u ( x ) u ( y ) ,

which further implies (4.3). This finishes the proof of Lemma 4.3.□

Lemma 4.4

Let all the assumptions be as in Lemma 4.3. Then Ω is regular.

Proof

Let λ be the same positive constant as in Lemma 4.3. For any given x 0 Ω ¯ and r ( 0 , λ ) , define u and g , respectively, by setting, for any z Ω , u ( z ) r 1 dist ( z , R n \ B ( x 0 , r ) ) and g ( z ) [ ϕ ( r ) ] 1 1 2 B ( x 0 , r ) ¯ Ω ( z ) . Then we find that 0 u 1 , u ( x 0 ) = 1 , u R n \ B ( x 0 , r ) 0 , and, for any z 1 , z 2 Ω , u ( z 1 ) u ( z 2 ) r 1 z 1 z 2 . Thus, by these facts, the almost decreasing property of ϕ * , the almost increasing property of ϕ , and an argument similar to that used in the estimations of both (4.5) and (4.6) with v , g ˜ , and λ 1 therein replaced, respectively, by u , g , and r , we conclude that g is a positive harmless constant multiple of an element in D ball ϕ ( u ) , and hence,

(4.9) u M ˙ ball ϕ , p ( Ω ) g L p ( R n ) [ ϕ ( r ) ] 1 2 B ( x 0 , r ) ¯ Ω 1 p

with the implicit positive constant independent of both x 0 and r . Then, applying Lemma 4.3 with x replaced by x 0 and (4.9), we conclude that, for almost every r ( 0 , λ ) and almost every y B ( x 0 , r ) ,

1 = u ( x 0 ) u ( y ) ϕ ( p ) ( x 0 y ) u M ˙ ball ϕ , p ( Ω ) ϕ ( p ) ( r ) [ ϕ ( r ) ] 1 2 B ( x 0 , r ) ¯ Ω 1 p r n p 2 B ( x 0 , r ) Ω 1 p .

Thus, 2 B ( x 0 , r ) Ω ( 2 r ) n , which, combined with Definition 2.7 with R Ω 2 λ and Remark 2.8(iii), further implies the regularity of Ω . This finishes the proof of Lemma 4.4.□

Remark 4.5

Let ϕ A and p [ n ( log 2 α ϕ ) , ] be such that ϕ ( p ) in (1.5) is a well-defined almost increasing function on [ 0 , ) satisfying ϕ ( p ) ( 0 ) = 0 . If Ω is a bounded M ˙ ball ϕ , p -embedding domain, then, for any u M ˙ ball ϕ , p ( Ω ) , (4.3) naturally holds with λ diam Ω . Furthermore, if ϕ is also a modulus of continuity, then, by the proof of Lemma 4.4, we conclude that Ω is a regular domain with the positive constant R Ω , as in Definition 2.7 with E replaced by Ω , that is precisely equal to diam Ω .

We now have the following conclusion on the relations between M ˙ ball ϕ , p -embedding domains and M ball ϕ , p -embedding domains.

Proposition 4.6

Let ϕ be a modulus of continuity and p [ n ( log 2 α ϕ ) , ] such that ϕ ( p ) in (1.5) is a well-defined almost increasing function on [ 0 , ) satisfying ϕ ( p ) ( 0 ) = 0 .

  1. If Ω is an M ˙ ball ϕ , p -embedding domain, then Ω supports the embedding from M ball ϕ , p ( Ω ) to M ˙ ϕ ( p ) , ( Ω ) .

  2. The domain Ω is an M ball ϕ , p -embedding domain if and only if Ω supports the embedding from M ball ϕ , p ( Ω ) to M ˙ ϕ ( p ) , ( Ω ) .

  3. If Ω is bounded, then Ω is an M ˙ ball ϕ , p -embedding domain if and only if it is an M ball ϕ , p -embedding domain.

Proof

If Ω is an M ˙ ball ϕ , p -embedding domain or an M ball ϕ , p -embedding domain, that is, M ˙ ball ϕ , p ( Ω ) M ˙ ϕ ( p ) , ( Ω ) or M ball ϕ , p ( Ω ) M ϕ ( p ) , ( Ω ) , it is then obvious that Ω supports the embedding from M ball ϕ , p ( Ω ) to M ˙ ϕ ( p ) , ( Ω ) , which proves both (i) and the necessity of (ii).

Next, we show the sufficiency of (ii). Let M ball ϕ , p ( Ω ) M ˙ ϕ ( p ) , ( Ω ) and u M ball ϕ , p ( Ω ) . To prove that Ω is an M ball ϕ , p -embedding domain, that is, M ball ϕ , p ( Ω ) M ϕ ( p ) , ( Ω ) , it suffices to show

(4.10) u L ( Ω ) u M ball ϕ , p ( Ω ) .

By both the assumption of ϕ and Remark 2.13(ii), we find that log 2 α ϕ ( 0 , 1 ] , and hence p > n 2 . Let λ be the same positive constant as in Lemma 4.3. Using the Hölder inequality with p > 1 , Lemmas 4.3 and 4.4, the almost increasing property of ϕ ( p ) on [ 0 , λ ] , and Remark 3.2(ii), we conclude that, for almost every x Ω ,

u ( x ) u ( x ) u B ( x , λ 2 ) Ω + u B ( x , λ 2 ) Ω B ( x , λ 2 ) Ω u ( x ) u ( z ) d z + B ( x , λ 2 ) Ω u ( z ) p d z 1 p ϕ ( p ) ( λ ) u M ˙ ball ϕ , p ( Ω ) + λ n p u L p ( Ω ) u M ball ϕ , p ( Ω ) ,

where the implicit positive constant in the last quantity only depends on ϕ , λ , n , p , and the constant related to the regularity of Ω as in Definition 2.7 with E replaced by Ω . Thus, (4.10) holds, which proves the sufficiency of (ii).

Finally, we show (iii). We first prove its necessity. If Ω is an M ˙ ball ϕ , p -embedding domain, then, by both (i) and (ii), we find that Ω is an M ball ϕ , p -embedding domain. [Since both (i) and (ii) hold for all domains, the bounded condition of Ω is not needed for the necessity.]

Conversely, to show the sufficiency, we let Ω be a bounded M ball ϕ , p -embedding domain. Then (ii) implies that Ω supports the embedding from M ball ϕ , p ( Ω ) to M ˙ ϕ ( p ) , ( Ω ) . To prove that Ω is an M ˙ ball ϕ , p -embedding domain, let u M ˙ ball ϕ , p ( Ω ) and λ be the same as in Lemma 4.3. Without loss of generality, we may assume that λ < diam Ω . For almost every x , y Ω , if x y λ , then (4.3) holds. Now, we assume that x y > λ . Since Ω is bounded, it can be covered by a finite number of balls with radius λ 2 centered in Ω , denoted by J { B i B ( x i , λ 2 ) : x i Ω } i = 1 N , where N is a positive integer depending only on both Ω and λ . Then there obviously exist balls { B i j : B i j J , j { 1 , , N ( x , y ) } } such that x B i 0 , y B i N ( x , y ) , and, for any j { 1 , , N ( x , y ) 1 } , B i j B i j + 1 . Here, N ( x , y ) N . Write z 0 x and z N ( x , y ) + 1 y , and, for any j { 1 , , N ( x , y ) } , choose z j B i j 1 B i j Ω . Then, for any j { 1 , , N ( x , y ) } , z j z j + 1 2 λ 2 = λ . From this, Lemma 4.3 with x and y therein replaced, respectively, by z j and z j + 1 , N ( x , y ) N , and the almost increasing property of ϕ ( p ) with z j z j + 1 λ < x y , it follows that

u ( x ) u ( y ) j = 0 N ( x , y ) u ( z j ) u ( z j + 1 ) ( N + 1 ) ϕ ( p ) ( x y ) u M ˙ ball ϕ , p ( Ω ) .

Thus, h u M ˙ ball ϕ , p ( Ω ) 1 Ω is a positive harmless constant multiple of an element of D ϕ ( p ) ( u ) , which, combined with h L ( Ω ) = u M ˙ ball ϕ , p ( Ω ) , further implies that u M ˙ ϕ ( p ) , ( Ω ) u M ˙ ball ϕ , p ( Ω ) , and hence, Ω is an M ˙ ball ϕ , p -embedding domain. This finishes the proof of Proposition 4.6.□

Let R ( 0 , ] and u L loc 1 ( Ω ) . We continue to use the same symbol R Ω ( u ) as in [66] to denote the restricted maximal function, which is defined by setting, for any x Ω ,

(4.11) R Ω ( u ) ( x ) sup r ( 0 , R ) B ( x , r ) Ω u ( y ) d y .

When Ω = R n , we denote R Ω simply by R ; when R = , we denote R Ω simply by Ω ; furthermore, when Ω = R n and R = , R Ω = is precisely the classical Hardy-Littlewood maximal operator .

Remark 4.7

Recall that a measure μ in R n is said to be doubling if there exists a positive constant C μ [ 1 , ) such that, for any x R n and r ( 0 , ) , μ ( B ( x , 2 r ) ) C μ μ ( B ( x , r ) ) . If Ω is a regular domain, then the measure μ given by setting, for any measurable set E R n , μ ( E ) E Ω , is doubling. Indeed, by Definition 2.7, one has, for any x R n and r ( 0 , ) ,

μ ( B ( x , 2 r ) ) = B ( x , 2 r ) Ω B ( x , 2 r ) B ( x , r ) B ( x , r ) Ω μ ( B ( x , r ) ) .

Then, by the Hardy-Littlewood maximal function theorem for doubling measures [11], we find that, for any p ( 1 , ] , Ω is bounded on L p ( Ω ) .

We have the following technical lemma.

Lemma 4.8

Let ϕ be a modulus of continuity with α ϕ ( 0 , 1 ) , p ( n ( log 2 α ϕ ) , ] , and Ω be a domain supporting the embedding from M ball ϕ , p ( Ω ) to M ˙ ϕ ( p ) , ( Ω ) , where ϕ ( p ) is the same as in (1.5). Then there exist positive constants λ , N , and C = C ( ϕ , p , Ω , λ , N ) , which all depend only on ϕ , p , and Ω , such that, for any q [ p , ] , u M ˙ ball ϕ , q ( Ω ) , and g D ball ϕ ( u ) L q ( Ω ) and for almost every x , y Ω with x y λ ,

(4.12) u ( x ) u ( y ) C ϕ ( x y ) [ N x y Ω ( g p ) ] 1 p ( x ) + [ N x y Ω ( g p ) ] 1 p ( y ) ,

where, when p = , [ N x y Ω ( g p ) ] 1 p ( x ) and [ N x y Ω ( g p ) ] 1 p ( y ) denote the essential suprema of g, respectively, on B ( x , N x y ) Ω and B ( y , N x y ) Ω .

Proof

Let λ be the same positive constant as in Lemma 4.3. By u L 0 ( Ω ) , we find that, for almost every x , y Ω , u ( x ) u ( y ) < . Now, we take x , y Ω satisfying that both x y λ and u ( x ) u ( y ) < .

Similarly to the proof of Lemma 4.3, without loss of generality, we may assume that, for any z Ω , 0 u ( y ) u ( z ) u ( x ) . Let N 0 [ 1 , ) and φ be an N 0 1 x y 1 -Lipschitz function such that supp φ B ( x , 2 N 0 x y ) , 0 φ 1 , and φ B ( x , N 0 x y ) 1 . Define v [ u u ( y ) ] φ and

g ˜ { g + [ ϕ ( N 0 x y ) ] 1 u ( x ) u ( y ) } 1 B ( x , 4 N 0 x y ) Ω .

By an argument similar to that used in the proof of Lemma 4.3 with λ 1 therein replaced by N 0 x y , we conclude that g ˜ is a positive harmless constant multiple of an element in D ball ϕ ( v ) , which, together with the boundedness of the support of g ˜ , the Hölder inequality with q [ p , ] , and g L q ( Ω ) , further implies that

v M ˙ ball ϕ , p ( Ω ) g ˜ L p ( Ω ) g L p ( B ( x , 4 N 0 x y ) Ω ) + [ ϕ ( N 0 x y ) ] 1 u ( x ) u ( y ) B ( x , 4 N 0 x y ) 1 p g L q ( B ( x , 4 N 0 x y ) Ω ) + [ ϕ ( N 0 x y ) ] 1 u ( x ) u ( y ) ( N 0 x y ) n p < .

Thus, v M ˙ ball ϕ , p ( Ω ) . By using p > n ( log 2 α ϕ ) and Remark 2.13(i), we obtain ϕ ( p ) is a well-defined almost increasing function on [ 0 , ) with ϕ ( p ) ( 0 ) = 0 . From (4.4), Lemma 4.3 with u replaced by v , and the definition of g ˜ , it follows that, for almost every x , y Ω with x y λ ,

(4.13) u ( x ) u ( y ) = v ( x ) v ( y ) ϕ ( p ) ( x y ) v M ˙ ball ϕ , p ( Ω ) ϕ ( p ) ( x y ) g ˜ L p ( Ω ) ϕ ( p ) ( x y ) B ( x , 4 N 0 x y ) Ω [ g ( z ) ] p d z 1 p + ϕ ( p ) ( x y ) ϕ ( p ) ( N 0 x y ) u ( x ) u ( y ) ,

where, when p = , ϕ ( p ) and ( B ( x , 4 N 0 x y ) Ω [ g ( z ) ] p d z ) 1 p are replaced, respectively, by ϕ and the essential supremum of g on B ( x , 4 N 0 x y ) Ω . In the remainder of this proof, we denote by C 0 the implicit positive constant in the last inequality of (4.13), which is independent of both x and y . By both α ϕ ( p ) ( 0 , 1 ) and (2.10), we find that, for any i Z + ,

ϕ ( p ) ( t ) ϕ ( p ) ( 2 i t ) α ϕ ( p ) + 1 2 i

with the implicit positive constant independent of both t and i , and hence,

lim i ϕ ( p ) ( t ) ϕ ( p ) ( 2 i t ) = 0 .

Thus, when N 0 2 i and i Z + is large enough, one has ϕ ( p ) ( x y ) ϕ ( p ) ( N 0 x y ) 1 ( 2 C 0 ) , which, combined with (4.13), further implies that, for almost every x , y Ω with x y λ ,

(4.14) u ( x ) u ( y ) ϕ ( p ) ( x y ) B ( x , 4 N 0 x y ) Ω [ g ( z ) ] p d z 1 p ϕ ( x y ) B ( x , 4 N 0 x y ) 1 B ( x , 4 N 0 x y ) Ω [ g ( z ) ] p d z 1 p ϕ ( x y ) { B ( x , 4 N 0 x y ) Ω [ g ( z ) ] p d z } 1 p .

Thus, by (4.11), we obtain

u ( x ) u ( y ) ϕ ( x y ) [ 5 N 0 x y Ω ( g p ) ] 1 p ( x ) + [ 5 N 0 x y Ω ( g p ) ] 1 p ( y )

with the same modification made when p = as earlier. This proves (4.12) with N 5 N 0 , which completes the proof of Lemma 4.8.□

Remark 4.9

(i) In the proof of Lemma 4.8, for the given two points, x and y , under the same assumption that, for any z Ω , 0 u ( y ) u ( z ) u ( x ) , if we let φ be an N 0 1 x y 1 -Lipschitz function such that supp φ B ( y , 2 N 0 x y ) , 0 φ 1 , and φ 1 B ( y , N 0 x y ) 1 , and define v [ u ( x ) u ] φ and

g ˜ { g + [ ϕ ( λ 1 ) ] 1 u ( x ) u ( y ) } 1 B ( y , 4 N 0 x y ) Ω ,

then, by an argument similar to that used in the proof of Lemma 4.3 and v ( x ) v ( y ) = u ( y ) u ( x ) , we conclude that g ˜ is a positive harmless constant multiple of an element in D ball ϕ ( v ) . Thus, similar to (4.14), we also have

(4.15) u ( x ) u ( y ) ϕ ( x y ) { B ( y , 4 N 0 x y ) Ω [ g ( z ) ] p d z } 1 p

with N 0 in (4.14), where the implicit positive constant is independent of x and y .

(ii) If Ω is a bounded M ˙ ball ϕ , p -embedding domain, then Remark 4.5 and an argument similar to that used in the proof of Lemma 4.8 with λ = diam Ω imply that (4.12) holds for almost every x , y Ω .

We now establish the following relation between ϕ -Hajłasz-Sobolev extensions and embedding domains, which in the case of ϕ ( t ) t s for any t [ 0 , ) with s ( 0 , 1 ] is precisely [66, Theorem 4.2].

Theorem 4.10

Let ϕ be a modulus of continuity with α ϕ ( 0 , 1 ) , p ( n ( log 2 α ϕ ) , ] , and Ω be an M ball ϕ , p -embedding (resp. a bounded M ˙ ball ϕ , p -embedding) domain. Then, for any q ( p , ) , Ω is also an M ball ϕ , q -extension (resp. M ˙ ball ϕ , q -extension) domain.

To show Theorem 4.10, we need an auxiliary extension lemma as below. We refer the reader to [58, Theorem 1.3], [29, Theorem 6], and [66, Lemma 4.1] for some related results for the classical (Hajłasz-)Sobolev spaces.

Lemma 4.11

Let ϕ be a modulus of continuity and p [ 1 , ] . If E is a (resp. bounded) regular subset of R n , then there exists a bounded linear extension operator from M ϕ , p ( E ) [resp. M ˙ ϕ , p ( E ) ] to M ϕ , p ( R n ) [resp. M ˙ ϕ , p ( R n ) ], where M ˙ ϕ , p ( E ) is defined in Definition 2.15(i) with the domain Ω therein replaced by E.

Since the proof of [66, Lemma 4.1] was omitted, here, for the convenience of the reader, we give a complete proof of Lemma 4.11. The main strategy for proving Lemma 4.11 is to use the Whitney ball-covering, the reflected quasi-balls associated with this covering, and the fact that the measure of the boundary of any regular set is zero. To this end, we first recall some concepts.

Definition 4.12

Let U be an open subset of R n . A countable family of balls, W U { B ( x i , r i ) : i N } , is called a Whitney ball-covering with respect to U if

  1. U = i B ( x i , r i ) ;

  2. there exist two positive constants C 1 and C 2 such that, for any i N ,

    C 1 r i dist ( B ( x i , r i ) , R n \ U ) C 2 r i ;

  3. there exists a positive constant C , depending only on n , such that

    B W U 1 B C .

Remark 4.13

For any closed set S R n , by [58, Theorem 2.4] (see also [29, Lemma 7]), we find that there exists such a Whitney ball-covering with respect to U R n \ S . Moreover, with the same symbols as in Definition 4.12, by [58, both (b) and (d) of Lemma 2.5], we conclude that, for any i and x 9 8 B ( x i , r i ) , r i dist ( x , Ω ) 28 r i and, for any B W U , { B ˜ W U : 9 8 B 9 8 B ˜ } has finite overlap.

Definition 4.14

Let E be a regular closed subset of R n , R E ( 0 , ) the same as in Definition 2.7, U R n \ E , and W U a Whitney ball-covering with respect to U . A family of measurable subsets,

U { H B i E : B i B ( x i , r i ) W U } i ,

is called a family of reflected quasi-balls with respect to W U if there exist positive constants c 1 , c 2 , and c 3 such that

  1. for any B i W U , H B i ( c 1 B i ) E ;

  2. for any B i W U with r i R E , B i c 2 H B i ;

  3. B i W U 1 H B i c 3 .

By [58, Theorem 2.6], we find that there exists such a family in Definition 4.14. Now, we are ready to prove Lemma 4.11.

Proof of Lemma 4.11

Applying both the regularity of E and [58, Lemma 2.1], we find that E = 0 . By this, without loss of generality, we may assume that E is closed, and hence, U R n \ E is open. Indeed, when E is not closed, we can first consider the extension from E to E ¯ and then from E ¯ to the whole R n because, by Definition 2.14(i) and E = 0 , the existence of the continuous extension from M ˙ ϕ , p ( E ) to M ˙ ϕ , p ( E ¯ ) is trivial.

Let W U { B i B ( x i , r i ) : i } be a Whitney ball-covering with respect to U in Definition 4.12 and U { H B ˜ : B W U } be a family of reflected quasi-balls with respect to W U in Definition 4.14, where N is an index set. Let R E ( 0 , ) be the same as in Definition 2.7.

We first consider the homogeneous case. In this case, E is further assumed to be bounded. By Remark 2.8(iv), we may assume that R E = diam E . Let u M ˙ ϕ , p ( E ) and g D ϕ ( u ) L p ( E ) be such that g L p ( E ) u M ˙ ϕ , p ( E ) .

Let { φ i } i be a sequence of c r i 1 -Lipschitz functions such that supp φ i 9 8 B i , 0 φ i 1 , and i φ i 1 , where c is a positive constant independent of i . Define the extension function u from E to R n by setting

(4.16) u ( x ) u ( x ) if x E , i u H B i φ i ( x ) if x U ,

and define g ˜ by setting

g ˜ g ( x ) if x E , i g H B i 1 9 8 B i ( x ) if x U ,

where, for any i with r i R E , we let H B i H B i ˜ and, for any i with r i > R E , we let H B i E .

To show that g ˜ is a positive harmless constant multiple of a ϕ -Hajłasz gradient of u , we first make the following observations. Let x , y be in the exceptional set of (2.13) and B , B ˜ W U , respectively, with the radii r , r ˜ ( 0 , ) , satisfying that x B and y B ˜ . On the one hand, if both r > R E and r ˜ > R E , then

(4.17) u H B u H B ˜ = u E u E = 0 ;

if r R E or r ˜ R E , then, by Definition 4.14(i) and the almost increasing property of ϕ with the observation that, for any z 1 H B and z 2 H B ˜ ,

z 1 z 2 diam ( H B H B ˜ ) diam H B + x y + diam H B ˜ ,

one has

(4.18) u H B u H B ˜ = H B u ( z 1 ) d z 1 H B ˜ u ( z 2 ) d z 2 H B H B ˜ u ( z 1 ) u ( z 2 ) d z 1 d z 2 H B H B ˜ ϕ ( z 1 z 2 ) [ g ( z 1 ) + g ( z 2 ) ] d z 1 d z 2 ϕ ( diam H B + x y + diam H B ˜ ) [ g H B + g H B ˜ ] .

On the other hand, similar to the aforementioned arguments, from the almost increasing property of ϕ with y z x y + diam H B , we also deduce that, for any r ( 0 , ) ,

(4.19) u H B u ( y ) = H B u ( z ) d z u ( y ) H B ϕ ( y z ) [ g ( z ) + g ( y ) ] d z ϕ ( x y + diam H B ) [ g H B + g ( y ) ] .

Now, we prove that there exists a positive harmless constant C such that C g ˜ D ϕ ( u ) . To this end, for almost every x , y R n , we consider the following four cases.

Case 1. When x , y E = R n \ U , by both the choice of g and the definition of g ˜ , we find that

u ( x ) u ( y ) = u ( x ) u ( y ) ϕ ( x y ) [ g ˜ ( x ) + g ˜ ( y ) ] ,

which is a desired estimate in this case.

Case 2. When x U and y E , by Definition 4.12(i), x belongs to several balls in W U . In this case, by using Definition 4.14(i) when r i R E and using the assumptions that R E = diam E and H B i E when r i > R E , we find that, for any B i W U containing x , diam H B i r i , which, combined with Definition 4.12(ii) with R n \ U replaced by E , x B i , and y E , further implies that diam H B i r i dist ( B i , E ) x y . From this, i φ i 1 , 0 φ i 1 , (4.19), the almost increasing property of ϕ with diam H B i x y , Proposition 3.4 with both c 2 and F replaced by ϕ , and i 1 B i 1 , we infer that

(4.20) u ( x ) u ( y ) = u ( x ) u ( y ) = i u H B i φ i ( x ) i u ( y ) φ i ( x ) i , 9 8 B i x u H B i u ( y ) ϕ ( x y ) i , 9 8 B i x [ g H B i + g ( y ) ] ϕ ( x y ) [ g ˜ ( x ) + g ˜ ( y ) ] ,

which is a desired estimate in this case.

Case 3. When x E and y U , by symmetry, (4.20) remains true in this case.

Case 4. Finally, we consider the case where both x and y are in U . Let i 0 be such that x B i 0 . We then consider two cases of y .

Case 4.1. If y U 9 8 B i 0 , then, on the one hand, by Definition 4.12(ii) and

dist ( x , E ) dist ( y , E ) x y r i 0 ,

we find that dist ( x , E ) r i 0 dist ( y , E ) and, moreover, for any B i containing x or y , by y U 9 8 B i 0 , we obtain

(4.21) x y r i 0 dist ( B i , E ) + r i r i ;

on the other hand, by Definition 4.14(i) in the case of r i R E or by both R E = diam E and H B i E in the case of r i > R E , we have, for any i such that B i { x , y } ,

(4.22) diam H B i r i .

Thus, using i φ i 1 , the Lipschitz continuity of { φ i } i , (4.17), (4.18), Proposition 3.4 with F replaced by ϕ , the almost increasing property of ϕ with both (4.21) and (4.22), the almost decreasing property of ϕ * with (4.21), the bounded overlap of W U , and the definition of g ˜ , we conclude that

(4.23) u ( x ) u ( y ) = i u H B i [ φ i ( x ) φ i ( y ) ] i u H B i 0 [ φ i ( x ) φ i ( y ) ] i u H B i u H B i 0 φ i ( x ) φ i ( y ) i 9 8 B i { x , y } u H B i 0 u H B i x y r i x y i 9 8 B i { x , y } ϕ ( diam H B i 0 + x y + diam H B i ) r i ( g H B i 0 + g H B i ) x y i 9 8 B i { x , y } ϕ * ( r i ) ( g H B i 0 + g H B i ) x y ϕ * ( x y ) i 9 8 B i { x , y } g H B i 0 + g ˜ ( x ) + g ˜ ( y ) ϕ ( x y ) [ g ˜ ( x ) + g ˜ ( y ) ] ,

which is a desired estimate in this case.

Case 4.2. If y U \ ( 9 8 B i 0 ) , then, on the one hand, by Definition 4.14(i) and the definition of H B i 0 , one has

(4.24) x y r i 0 diam H B i 0 ;

on the other hand, using Definitions 4.14(i) and 4.12(ii) and using (4.24), we conclude that, for any B i W U containing x ,

diam H B i r i dist ( B i , E ) dist ( x , E ) dist ( B i 0 , E ) + r i 0 r i 0 x y

and, furthermore, for any B i W U containing y ,

(4.25) diam H B i dist ( B i , E ) dist ( y , E ) x y + dist ( x , E ) x y .

Thus, by i φ i 1 , an argument similar to that used in the estimation of (4.23), 0 φ i 1 , (4.17), (4.18), Proposition 3.4 with F replaced by ϕ , the almost increasing property of ϕ with (4.24)–(4.25), and the bounded overlap of W U , we obtain

u ( x ) u ( y ) i u H B i u H B i 0 φ i ( x ) φ i ( y ) i u H B i u H B i 0 [ φ i ( x ) + φ i ( y ) ] i 9 8 B i { x , y } u H B i u H B i 0 i 9 8 B i { x , y } ϕ ( diam H B i 0 + x y + diam H B i ) [ g H B i + g H B i 0 ] ϕ ( x y ) g ˜ ( x ) + g ˜ ( y ) + i 9 8 B i { x , y } g H B i 0 ϕ ( x y ) [ g ˜ ( x ) + g ˜ ( y ) ] .

Altogether, we conclude that g ˜ is a positive harmless constant multiple of an element in D ϕ ( u ) , which shows the aforementioned claim.

Finally, from [58, Lemma 3.5], we deduce that

i g H B i 1 9 8 B i L p ( R n ) g L p ( E ) .

By this, 0 φ i 1 , and the choice of g , we find that

u M ˙ ϕ , p ( R n ) g ˜ L p ( R n ) g L p ( E ) u M ˙ ϕ , p ( E ) ,

which, combined with (1.2), further implies u M ˙ ϕ , p ( E ) u M ˙ ϕ , p ( R n ) E . Notice that u M ˙ ϕ , p ( R n ) E u M ˙ ϕ , p ( E ) is trivial. We obtain M ˙ ϕ , p ( E ) = M ˙ ϕ , p ( R n ) E . Thus, E supports the M ˙ ϕ , p -extension.

For the inhomogeneous case of this lemma, we introduce the same extension function u as in (4.16) with the modification that H B i for any i with r i > R E . Then, by an argument similar to that used in the proof of the homogeneous case, we prove that the function h ˜ , defined by setting, for any x R n ,

h ˜ g ( x ) if x E , i ( g H B i + u H B i ) 1 9 8 B i ( x ) if x U ,

is a positive harmless constant multiple of an element in D ϕ ( u ) . Thus, by applying [58, Lemma 3.5] again, we obtain

u M ϕ , p ( R n ) u L p ( R n ) + h ˜ L p ( R n ) u L p ( Ω ) + g L p ( E ) u M ϕ , p ( E ) ,

which further implies that E supports the M ϕ , p -extension. This finishes the proof of the inhomogeneous case and hence Lemma 4.11.□

Now, we show Theorem 4.10.

Proof of Theorem 4.10

If Ω is an M ball ϕ , p -embedding domain or a bounded M ˙ ball ϕ , p -embedding domain, from Proposition 4.6, we infer that, in both cases, Ω supports the embedding from M ball ϕ , p ( Ω ) to M ˙ ϕ ( p ) , ( Ω ) . By using Remark 2.13(ii), we obtain p > n ( log 2 α ϕ ) n 2 . Let q ( p , ) . Recall that M ball ϕ , q ( R n ) Ω = M ϕ , q ( R n ) Ω , which follows from both the obvious fact M ball ϕ , q ( R n ) = M ϕ , q ( R n ) and (1.2). Thus, to prove that Ω is an M ball ϕ , q -extension domain, by both Lemma 4.11 with E replaced by Ω and the obvious embedding M ϕ , q ( Ω ) M ball ϕ , q ( Ω ) , it suffices to show M ball ϕ , q ( Ω ) M ϕ , q ( Ω ) . From an argument similar to this, it follows that, to prove that Ω is an M ˙ ball ϕ , q -extension domain, it suffices to show M ˙ ball ϕ , q ( Ω ) M ˙ ϕ , q ( Ω ) .

We first prove M ball ϕ , q ( Ω ) M ϕ , q ( Ω ) . Let u M ball ϕ , q ( Ω ) and g D ball ϕ ( u ) with g L q ( Ω ) u M ˙ ball ϕ , q ( Ω ) . On the one hand, by Lemmas 4.4 and 4.8, we find that there exist positive constants λ ( 0 , ) and N 1 ( 1 , ) such that Ω is a regular domain with the same positive constant as in Definition 2.7 with E replaced by Ω , R Ω 2 λ , and, furthermore, for this λ and almost every x , y Ω satisfying x y λ ,

(4.26) u ( x ) u ( y ) ϕ ( x y ) [ N 1 x y Ω ( g p ) ( x ) ] 1 p + [ N 1 x y Ω ( g p ) ( y ) ] 1 p ϕ ( x y ) [ N 1 λ Ω ( g p ) ( x ) ] 1 p + [ N 1 λ Ω ( g p ) ( y ) ] 1 p ,

where N 1 λ Ω is the same as in (4.11), both [ N 1 λ Ω ( g p ) ] 1 p ( x ) and [ N 1 λ Ω ( g p ) ] 1 p ( y ) in the case of p = denote the essential suprema of g , respectively, on B ( x , N 1 λ ) Ω and B ( y , N 1 λ ) Ω , and the implicit positive constants depend on ϕ , n , p , and Ω .

On the other hand, for almost every x , y Ω satisfying x y > λ , since x z < λ whenever z B ( x , λ ) , from both (4.14) and (4.15) with y replaced by z , the almost increasing property of ϕ with λ < x y , and the estimate that N 2 x z Ω ( g p ) N 2 λ Ω ( g p ) pointwise, we deduce that

u ( x ) u ( x ) u B ( x , λ ) Ω + u B ( x , λ ) Ω B ( x , λ ) Ω u ( x ) u ( z ) d z + B ( x , λ ) Ω u ( z ) d z B ( x , λ ) Ω ϕ ( x z ) B ( x , 4 N 0 x z ) Ω [ g ( w ) ] p d w 1 p d z + λ Ω ( u ) ( x ) ϕ ( λ ) B ( x , λ ) Ω [ N 2 x z Ω ( g p ) ( x ) ] 1 p d z + λ Ω ( u ) ( x ) ϕ ( x y ) [ N 2 λ Ω ( g p ) ( x ) ] 1 p + λ Ω ( u ) ( x )

with the implicit positive constants depending only on ϕ , n , p , and Ω , where N 0 is precisely the positive constant in both (4.14) and (4.15) and where N 2 > 4 N 0 . Letting N max { N 1 , N 2 } , this further implies that

u ( x ) u ( y ) u ( x ) + u ( y ) ϕ ( x y ) [ N λ Ω ( g p ) ( x ) ] 1 p + λ Ω ( u ) ( x ) + [ N λ Ω ( g p ) ( y ) ] 1 p + λ Ω ( u ) ( y ) ,

where the implicit positive constants depend on ϕ , n , p , and Ω .

Altogether, we find that λ Ω ( u ) + [ N λ Ω ( g p ) ] 1 p is a positive harmless constant multiple of an element in D ϕ ( u ) .

Notice that u λ Ω ( u ) almost everywhere. Then, by this, the regularity of Ω , Remark 2.8(iv) with R 1 and R 2 therein replaced, respectively, by λ and N λ , the L q ( Ω ) -boundedness with q ( 1 , ) and L q p ( Ω ) -boundedness with q > p of Ω (Remark 4.7), and the choice of g , we obtain

u M ϕ , q ( Ω ) λ Ω ( u ) + [ N λ Ω ( g p ) ] 1 p L q ( Ω ) + u L q ( Ω ) λ Ω ( u ) L q ( Ω ) + [ N λ Ω ( g p ) ] 1 p L q ( Ω ) Ω ( u ) L q ( Ω ) + [ Ω ( g p ) ] 1 p L q ( Ω ) u M ball ϕ , q ( Ω ) ,

which further implies u M ϕ , q ( Ω ) , and hence, M ball ϕ , q ( Ω ) M ϕ , q ( Ω ) . Thus, Ω is an M ball ϕ , q -extension domain, which completes the proof of the inhomogeneous case.

Next, we show the homogeneous case, that is, M ˙ ball ϕ , q ( Ω ) M ˙ ϕ , q ( Ω ) . Let Ω be a bounded M ˙ ball ϕ , p -embedding domain. By Remark 2.8(iv), without loss of generality, we may assume that the same positive constant R Ω as in Definition 2.7 with E replaced by Ω satisfies R Ω diam Ω . From Remarks 4.5 and 4.9 and from an argument similar to that used in the estimation of (4.26) with λ replaced by R Ω , we infer that [ N R Ω Ω ( g p ) ] 1 p is a positive harmless constant multiple of an element of D ϕ ( u ) , which, combined with the regularity of Ω , Remark 2.8(iv) with R 1 and R 2 therein replaced, respectively, by R Ω and N R Ω , and the L q p ( Ω ) -boundedness of Ω (Remark 4.7) further implies that

u M ˙ ϕ , q ( Ω ) [ N R Ω Ω ( g p ) ] 1 p L q ( Ω ) [ Ω ( g p ) ] 1 p L q ( Ω ) g L q ( Ω ) u M ˙ ball ϕ , q ( Ω ) .

Thus, M ˙ ball ϕ , q ( Ω ) M ˙ ϕ , q ( Ω ) , and hence, Ω is an M ˙ ball ϕ , q -extension domain. This finishes the proof of the homogeneous case and hence Theorem 4.10.□

5 Geometric characterization of M ˙ ball ϕ , p -embedding domains

This section is devoted to proving our main theorem, Theorem 1.2. In Theorems 5.2 and 5.4, using the weak self-improving property of generalized weak cigar domains obtained in Theorem 3.13, we can show that a weak F -cigar domain is an M ˙ ball ϕ , p -embedding domain and hence finish the proof of the sufficiency of Theorem 1.2. This, together with Theorem 4.10, further implies that a weak F -cigar domain is an M ˙ ball ϕ , p -extension domain. The necessity is proved in Theorem 5.5. Actually, these theorems give stronger conclusions than Theorem 1.2 itself.

In what follows, for any ϕ A and r ( 1 , ) , the function G ( ϕ , r ) : ( 0 , ) ( 0 , ) is defined by setting, for any t ( 0 , ) ,

(5.1) G ( ϕ , r ) ( t ) t n r ϕ ( t ) r r 1 .

Furthermore, if lim t 0 + G ( ϕ , r ) ( t ) [ 0 , ] exists, we extend G ( ϕ , r ) to [ 0 , ) by setting

G ( ϕ , r ) ( 0 ) lim t 0 + G ( ϕ , r ) ( t ) .

Let G ( ϕ , r ) * be the same as in (3.1) with F replaced by G ( ϕ , r ) . Observe that

(5.2) G ( ϕ , r ) = [ ϕ ( r ) ] r r 1

with ϕ ( r ) the same as in (1.5) with p replaced by r .

Lemma 5.1

Let ϕ be a modulus of continuity. Then, for any r 1 , r 2 ( 1 , ) with r 1 r 2 , one has

G ( ϕ , r 1 ) * ( ) G ( ϕ , r 2 ) * ( ) = G ( ϕ , r 1 ) ( ) G ( ϕ , r 2 ) ( )

and the function G ( ϕ , r 1 ) * ( ) G ( ϕ , r 2 ) * ( ) is almost decreasing on ( 0 , ) and the function G ( ϕ , r 2 ) * ( ) G ( ϕ , r 1 ) * ( ) is almost increasing on ( 0 , ) , where G ( ϕ , r 1 ) and G ( ϕ , r 2 ) are the same as in (5.1) with r replaced, respectively, by r 1 and r 2 and where G ( ϕ , r 1 ) * and G ( ϕ , r 2 ) * are the same as in (3.1) with F replaced, respectively, by G ( ϕ , r 1 ) and G ( ϕ , r 2 ) .

Proof

By both the almost decreasing property of ϕ * and n 2 , we find that t n ϕ ( t ) is also almost decreasing, which, combined with ( r 2 r 1 ) [ ( r 2 1 ) ( r 1 1 ) ] 0 , further implies that

G ( ϕ , r 1 ) * ( t ) G ( ϕ , r 2 ) * ( t ) = G ( ϕ , r 1 ) ( t ) G ( ϕ , r 2 ) ( t ) = [ t 1 n ϕ * ( t ) ] r 2 r 1 ( r 2 1 ) ( r 1 1 )

is almost decreasing. Furthermore, G ( ϕ , r 2 ) * ( ) G ( ϕ , r 1 ) * ( ) = [ G ( ϕ , r 1 ) * ( ) G ( ϕ , r 2 ) * ( ) ] 1 is obvious almost increasing. This finishes the proof of Lemma 5.1.□

The following theorem gives a better result than the sufficiency of Theorem 1.2.

Theorem 5.2

Let Δ be the same as in Definition 3.5, Λ 1 as in Definition 3.7 with p = 1 , G ( ϕ , p * ) as in (5.1) with r replaced by p * , and G ( ϕ , p * ) * as in (3.1) with F replaced by G ( ϕ , p * ) . Let ϕ be a modulus of continuity with α ϕ ( 0 , 1 ) , p ( n ( log 2 α ϕ ) , ) , F G ( ϕ , p ) Λ 1 be such that F * in (3.1) belongs to Δ , Ω a bounded weak F-cigar domain in R n , and q * the same as in Theorem 3.13. If, for some p * [ n ( log 2 α ϕ ) , p ) , G ( ϕ , p * ) * ( F * ) q * is almost increasing, then, for any p ˜ ( p * , ) , Ω is an M ˙ ball ϕ , p ˜ -extension domain and, furthermore, an M ˙ ball ϕ , p ˜ -embedding domain.

Proof

From p * n ( log 2 α ϕ ) and Proposition 2.13(i), it follows that, for any p ˜ ( p * , ) , ϕ ( p ˜ ) is well defined. To show that, for any p ˜ ( p * , ) , Ω is an M ˙ ball ϕ , p ˜ -extension domain, by Theorem 4.10, it suffices to prove that, for any q ( p * , p ] , Ω is an M ˙ ball ϕ , q -embedding domain [here, ϕ ( q ) is also well defined because q ( p * , p ] ( p * , ) ]. To this end, by both the boundedness of Ω and the Hölder inequality, it suffices to show that, for any u M ˙ ball ϕ , p ( Ω ) , any q ( p * , p ] , and any Lebesgue point x , y Ω of u ,

(5.3) u ( x ) u ( y ) ϕ ( q ) ( x y ) u M ˙ ball ϕ , q ( Ω ) ,

where ϕ ( q ) is the same as in (1.5) with p therein replaced by q . This is because (5.3) implies that u M ˙ ϕ ( q ) , ( Ω ) u M ˙ ball ϕ , q ( Ω ) , and hence, M ˙ ball ϕ , q ( Ω ) M ˙ ϕ ( q ) , ( Ω ) .

We first claim that, for any q ( p * , p ] , G ( ϕ , q ) is a modulus of continuity. Indeed, by Remark 2.13, we find that α ϕ [ 1 2 , 1 ) , and hence, n [ log 2 α ϕ ] n 2 . Notice that, from p * [ n [ log 2 α ϕ ] , p ) , we deduce that, for any q ( p * , p ] ,

log 2 α ϕ ( q ) = n q log 2 α ϕ > n p * log 2 α ϕ 0 ,

and, moreover, from q > 1 , we infer that

log 2 α G ( ϕ , q ) = n q log 2 α ϕ q q 1 > 0 .

Thus, by Remark 2.13(i) with ϕ therein replaced by ϕ ( q ) or by G ( ϕ , q ) , we find that ϕ ( q ) ( 0 ) = 0 or G ( ϕ , q ) ( 0 ) = 0 and that both ϕ ( q ) and G ( ϕ , q ) are almost increasing on [ 0 , ) . Observe that, for any t ( 0 , ) , F * ( t ) = [ t ( 1 n ) p ϕ * ( t ) ] p ( p 1 ) . Thus, by using n > 1 , we conclude that F * is almost decreasing on ( 0 , ) , which, combined with both Lemma 5.1 with q p and Proposition 2.10, further implies that G ( ϕ , q ) * = F * G ( ϕ , q ) * G ( ϕ , p ) * is almost decreasing on ( 0 , ) . Thus, G ( ϕ , q ) is a modulus of continuity.

Now, we turn to prove (5.3). Let u M ˙ ball ϕ , p ( Ω ) . Then, by both the boundedness of Ω and the Hölder inequality, we find that, for any q ( p * , p ] , u M ˙ ball ϕ , q ( Ω ) . Let g 1 D ball ϕ ( u ) L q ( Ω ) be such that g 1 L q ( Ω ) u M ˙ ball ϕ , q ( Ω ) and let g 2 D ball ϕ ( u ) L p ( Ω ) .

Let x and y be any two given Lebesgue points of u . We consider the following two cases.

Case 1. x y 1 4 dist ( x , Ω ) . In this case, B ( y , 2 x y ) B ( x , 4 x y ) Ω . By this, the almost increasing property of ϕ , the Hölder inequality with q > 1 , (2.12) with α ϕ ( q ) ( 0 , 1 ) , and the choice of g , we have

(5.4) u ( y ) u B ( y , 2 x y ) j 1 u B ( y , 2 j 1 x y ) u B ( y , 2 j x y ) j 1 B ( y , 2 j 1 x y ) B ( y , 2 j x y ) u ( z 1 ) u ( z 2 ) d z 1 d z 2 j 1 ϕ ( 2 j + 1 x y ) B ( y , 2 j x y ) g ( z ) d z j 1 ϕ ( q ) ( 2 j + 1 x y ) B ( y , 2 j x y ) [ g ( z ) ] q d z 1 q ϕ ( q ) ( x y ) g L q ( Ω ) ϕ ( q ) ( x y ) u M ˙ ball ϕ , q ( Ω ) ,

and, furthermore, similar to (5.4), by B ( y , 2 x y ) B ( x , 4 x y ) , we obtain

(5.5) u ( x ) u B ( y , 2 x y ) u ( x ) u B ( x , 4 x y ) + u B ( x , 4 x y ) u B ( y , 2 x y ) ϕ ( q ) ( x y ) u M ˙ ball ϕ , q ( Ω ) .

By combining (5.4) and (5.5), we conclude that

u ( x ) u ( y ) u ( y ) u B ( y , 2 x y ) + u B ( y , 2 x y u ( x ) ) ϕ ( q ) ( x y ) u M ˙ ball ϕ , q ( Ω ) ,

which shows (5.3) in the case of x y 1 4 dist ( x , Ω ) .

Case 2. x y > 1 4 dist ( x , Ω ) . In this case, for any given ε ( 0 , ) , let Γ be the curve joining x and y and let Γ ˜ be the subset of Γ in Theorem 3.13. By the Besicovitch covering lemma [61], we find N N and a family { B i B ( z i , r i ) B ( z i , dist ( z i , Ω ) 10 ) } i = 0 N , consisting of the balls centered on Γ , such that z 0 x , z N y , B i B i + 1 for any i { 0 , , N 1 } , and i = 0 N 1 2 B i M for some harmless M N . For any i { 0 , , N 1 } , let ω i B i B i + 1 . Notice that

(5.6) z i ω i < r i = dist ( z i , Ω ) 10 < dist ( z i , Ω )

and

(5.7) z i + 1 ω i < dist ( z i + 1 , Ω ) 10 < dist ( z i + 1 , Ω ) .

Thus, from an argument similar to that used in the estimations of both (5.4) and (5.5) with x and y replaced, respectively, by ω i and z i or by ω i and z i + 1 , the Hölder inequality, and the almost increasing property of ϕ ( q ) with both (5.6) and (5.7), we deduce that, for any q ( p * , p ] ,

(5.8) u ( x ) u ( y ) i = 0 N 1 u ( z i ) u B ( z i , 2 ω i z i ) + i = 0 N 1 u ( ω i ) u B ( z i , 2 ω i z i ) + i = 0 N 1 u ( ω i ) u B ( z i + 1 , 2 ω i z i + 1 ) + i = 0 N 1 u ( z i + 1 ) u B ( z i + 1 , 2 ω i z i + 1 ) i = 0 N 1 ϕ ( q ) ( z i ω i ) min l = 1,2 B ( z i , 2 ω i z i ) [ g l ( z ) ] q d z 1 q + i = 0 N 1 ϕ ( q ) ( z i + 1 ω i ) min l = 1,2 B ( z i + 1 , 2 ω i z i ) [ g l ( z ) ] q d z 1 q i = 0 N ϕ ( q ) ( dist ( z i , Ω ) ) min l = 1,2 2 B i [ g l ( z ) ] q d z 1 q .

Let ˜ { B i : B i Γ ˜ } . For any q ( p * , p ] , applying (5.8), (5.2) with r q , and the Hölder inequality, we conclude that

u ( x ) u ( y ) B i ˜ ϕ ( q ) ( dist ( z i , Ω ) ) 2 B i [ g 1 ( z ) ] q d z 1 q + B i \ ˜ ϕ ( dist ( z i , Ω ) ) 2 B i [ g 2 ( z ) ] q d z 1 q B i ˜ ϕ ( q ) ( dist ( z i , Ω ) ) 2 B i [ g 1 ( z ) ] q d z 1 q + B i \ ˜ ϕ ( p ) ( dist ( z i , Ω ) ) 2 B i [ g 2 ( z ) ] p d z 1 p B i ˜ G ( ϕ , q ) ( dist ( z i , Ω ) ) q 1 q B i ˜ 2 B i [ g 1 ( z ) ] q d z 1 q + B i \ ˜ F ( dist ( z i , Ω ) ) p 1 p B i \ ˜ 2 B i [ g 2 ( z ) ] p d z 1 p I 1 + I 2 .

In the following, we estimate I 1 and I 2 , respectively.

We first estimate I 1 . Since, for any z i ˜ B i ¯ Γ ˜ , B ( z i ˜ , r i ) 2 B i , then, from this and Theorem 3.13(ii) with B replaced by B ( z i ˜ , r i ) , we infer that

(5.9) dist ( z i , Ω ) leng ( B ( z i ˜ , r i ) Γ ˜ ) leng ( 2 B i Γ ˜ ) .

Notice that, for any i { 0 , , N } and z 2 B i ¯ , z z i dist ( z i , Ω ) 5 . Consequently,

(5.10) 4 5 dist ( z i , Ω ) dist ( z , Ω ) 6 5 dist ( z i , Ω ) ,

which, combined with (5.9), the almost decreasing property of G ( ϕ , q ) * , ϕ A , and Proposition 3.4 with F replaced by G ( ϕ , q ) , further implies that, for any q ( p * , p ] ,

(5.11) G ( ϕ , q ) ( dist ( z i , Ω ) ) leng ( 2 B i Γ ˜ ) G ( ϕ , q ) * ( dist ( z i , Ω ) ) leng ( 2 B i Γ ˜ ) min z 2 B i Γ ˜ G ( ϕ , q ) * 5 6 dist ( z , Ω ) 2 B i Γ ˜ G ( ϕ , q ) * ( dist ( z , Ω ) ) d z .

From q * > 1 , we deduce that G ( ϕ , p ) F q * = G ( ϕ , p ) 1 q * is almost decreasing, which, together with Lemma 5.1 with r 1 q and r 2 p , and Proposition 2.10 with f 1 f 2 and f 2 replaced, respectively, by G ( ϕ , q ) G ( ϕ , p ) and G ( ϕ , p ) F q * , further implies that, for any q ( p * , p ] ,

(5.12) G ( ϕ , q ) F q * = G ( ϕ , q ) G ( ϕ , p ) G ( ϕ , p ) F q *

is almost decreasing. Similarly, by the assumption that G ( ϕ , p * ) * ( F * ) q * is almost increasing, Lemma 5.1 with r 1 p * and r 2 q , and Proposition 2.10, we conclude that, for any q ( p * , p ] ,

(5.13) G ( ϕ , q ) * ( F * ) q * = G ( ϕ , q ) * G ( ϕ , p * ) * G ( ϕ , p * ) * ( F * ) q *

is almost increasing. From Remark 2.13(i), it follows that

(5.14) lim t F ( t ) = .

Thus, by using i = 0 N 1 2 B i 1 and applying Theorem 3.13(iii) [which needs (5.14)] to (5.11) with G ( ϕ , q ) being a modulus of continuity, (5.12), and (5.13), we further obtain

I 1 B i ˜ 2 B i Γ ˜ G ( ϕ , q ) * ( dist ( z , Ω ) ) d z q 1 q Ω [ g 1 ( z ) ] q d z 1 q Γ ˜ G ( ϕ , q ) * ( dist ( z , Ω ) ) d z q 1 q g 1 L q ( Ω ) [ G ( ϕ , q ) ( x y ) ] q 1 q g 1 L q ( Ω ) ϕ ( q ) ( x y ) u M ˙ ball ϕ , q ( Ω ) .

To estimate I 2 , by diam 2 B i leng ( B i Γ ) for any i { 0 , , N } , the almost decreasing property of F * with (5.10), Proposition 3.4 with c = 5 6 , i = 0 N 1 2 B i 1 , and (3.9), we also conclude that

(5.15) I 2 B i ( diam B i ) F * ( dist ( z i , Ω ) ) p 1 p B i \ ˜ 2 B i [ g 2 ( z ) ] p d z 1 p B i 2 B i Γ F * ( dist ( z , Ω ) ) d z p 1 p B i \ ˜ 2 B i [ g 2 ( z ) ] p d z 1 p Γ F * ( dist ( z , Ω ) ) d z p 1 p B i \ ˜ 2 B i [ g 2 ( z ) ] p d z 1 p [ F ( x y ) ] p 1 p B i \ ˜ 2 B i [ g 2 ( z ) ] p d z 1 p ϕ ( p ) ( x y ) B i \ ˜ 2 B i [ g 2 ( z ) ] p d z 1 p

with the implicit positive constants independent of x , y , and the choice of the ball family . Since, for any B i \ ˜ , we have B i Γ Γ \ Γ ˜ , then, from (2.1) with the globally assumption n > 1 , i = 0 N 1 2 B i 1 , and (3.10), we infer that

B i \ ˜ 2 B i B i \ ˜ B i 1 n n B i \ ˜ diam B i n B i \ ˜ leng ( B i Γ ) n Γ \ Γ ˜ B i 1 B i ( z ) d z n [ leng ( Γ \ Γ ˜ ) ] n ε n .

By using this, (5.15), and g 2 L p ( Ω ) and applying the absolute continuity of integrals with g L p ( Ω ) , we are able to choose ε ( 0 , ) small enough such that B i \ ˜ 2 B i [ g ( z ) ] p d z is sufficiently small and, furthermore,

I 2 < ϕ ( q ) ( x y ) u M ˙ ball ϕ , q ( Ω ) .

Altogether, we conclude that (5.3) holds for almost every x , y Ω , and hence, for any p ˜ ( p * , ) , Ω is an M ˙ ball ϕ , p ˜ -extension domain, which, combined with Remark 4.2, further implies that Ω is also an M ˙ ball ϕ , p ˜ -embedding domain. This finishes the proof of Theorem 5.2.□

The following gives an example, which confirms the existence of p * in Theorem 5.2.

Remark 5.3

Let all the assumptions be the same as in Theorem 5.2, and let both ϕ and p be such that

(5.16) 1 + n + p log 2 α ϕ p 1 1 + n p log 2 β ϕ p 1 < q * .

Then there exists p * ( n ( log 2 α ϕ ) , p ) such that G ( p * ) * ( F * ) q * is almost increasing; in particular, if log 2 α ϕ = log 2 β ϕ , such a p * always exists.

Indeed, observe that the function θ , defined by setting, for any r ( 1 , ) ,

θ ( r ) 1 + n + r log 2 α ϕ r 1 ,

is continuous and decreases to 1 + n + p log 2 α ϕ p 1 as r > p and r p . Thus, there exists p * ( n ( log 2 α ϕ ) , p ) such that

1 + n + p log 2 α ϕ p 1 1 + n p log 2 β ϕ p 1 < 1 + n + p * log 2 α ϕ p * 1 1 + n p log 2 β ϕ p 1 < q * .

Then, by p > 1 and p 1 + n p log 2 β ϕ > 0 , we obtain

log 2 α G ( ϕ , p * ) * = 1 n + p * log 2 α ϕ p * 1 > q * 1 n p log 2 β ϕ p 1 = log 2 β ( F * ) q * ,

which, together with Remark 2.13(i), further implies that G ( ϕ , p * ) * ( F * ) q * is almost increasing.

In particular, if log 2 α ϕ = log 2 β ϕ , then (5.16) always holds. Thus, p * always exists.

We now have the following conclusion.

Theorem 5.4

Let ϕ be a modulus of continuity with α ϕ ( 0 , 1 ) , p ( n [ log 2 α ϕ ] , ) , F G ( ϕ , p ) be such that F Λ 1 and F * Δ , and Ω a bounded weak F-cigar domain in R n . Then, for any p ˜ [ p , ) , Ω is an M ˙ ball ϕ , p ˜ -embedding domain.

To prove Theorem 5.4, it suffices to show that Ω is an M ˙ ball ϕ , p -embedding domain, which is not hard to prove by applying an argument similar to that used in the proof of Theorem 5.2 with q therein replaced by p . Indeed, since there exists no need to use Theorem 4.10 and an intermediate parameter p * , the proof of Theorem 5.4 is much easier than that of Theorem 5.2. We omit the details.

Next, we turn the attention to the necessity of Theorem 1.2, where an additional assumption that Ω has a slice property is needed.

Theorem 5.5

Let ϕ be a modulus of continuity with α ϕ ( 0 , 1 ) , p [ n [ log 2 α ϕ ] , ) be such that ϕ ( p ) in (1.5) is a well-defined almost increasing function on [ 0 , ) with ϕ ( p ) ( 0 ) = 0 , and Ω a bounded M ˙ ball ϕ , p -embedding domain with a slice property. For any t [ 0 , ) , define

F ( t ) [ ϕ ( p ) ( t ) ] p p 1 = t n p 1 [ ϕ ( t ) ] p p 1 .

Then, for any modulus of continuity, G : [ 0 , ) [ 0 , ) , with G F being almost increasing, Ω is a weak G-cigar domain.

Proof

By Remark 2.13(ii), we have p n ( log 2 α ϕ ) n 2 , and hence, F is almost increasing and F * ( ) [ ( n 1 ) p ϕ * ( ) ] p ( p 1 ) is almost decreasing, which means that F is a modulus of continuity. To show the present theorem, by both the assumption of G and Proposition 3.11, it suffices to prove that Ω is a weak F -cigar domain.

Let Ω have a slice property with the slice constant C S ( 1 , ) in Definition 2.5 and let x , y Ω . Without loss of generality, we may assume that x y . We denote by Γ the rectifiable curve joining x and y and by { S i } i = 0 N with N N the collection of all open subsets of Ω in Definition 2.5 with x S 0 and y S N [By Definition 2.5(i), obviously, N 1 ]. By Remark 2.6, we may also assume that Γ is chosen such that (2.4) holds.

When N = 1 , by using Definition 2.5(iv), we obtain, for any z Γ S 0 ,

(5.17) dist ( z , Ω ) x z + dist ( x , Ω ) diam S 0 + dist ( x , Ω ) dist ( x , Ω ) x y

and, for any z Γ S 1 ,

(5.18) dist ( z , Ω ) y z + dist ( y , Ω ) diam S 1 + dist ( y , Ω ) dist ( y , Ω ) x y .

Thus, from the almost increasing property of F with both (5.17) and (5.18) and from the choice of Γ with (2.4) , it follows that

(5.19) Γ F * ( dist ( z , Ω ) ) d z F ( dist ( x , Ω ) ) Γ S 0 [ dist ( z , Ω ) ] 1 d z + F ( dist ( y , Ω ) ) Γ S 1 [ dist ( z , Ω ) ] 1 d z F ( dist ( x , Ω ) ) + F ( dist ( y , Ω ) ) F ( x y ) ,

where the implicit positive constants depend on both F and C S .

In the following, we let N 2 . For any given i { 1 , , N 1 } , define u i and g i , respectively, by setting, for any z Ω ,

u i ( z ) inf { leng ( Γ x z S i ) : Γ x z is a rectifiable curve joining x and z }

and

g i ( z ) [ ϕ * ( diam S i ) ] 1 1 B ( x i , 2 C S dist ( x i , Ω ) ) ( z ) ,

where ϕ * is the same as in (3.1) and x i the same as in Definition 2.5(iv). Then, by Definition 2.5(iv), one has

(5.20) C S 1 dist ( x i , Ω ) diam S i C S dist ( x i , Ω ) .

We first claim that there exists a positive constant C such that, for any given i { 1 , , N 1 } , C g i D ball ϕ , 1 ( 2 C S 4 ) ( u i ) . Indeed, let i { 1 , , N 1 } and z 1 , z 2 Ω be such that

(5.21) z 1 z 2 1 2 C S 4 min { dist ( z 1 , Ω ) , dist ( z 2 , Ω ) } .

Without loss of generality, we may assume that u i ( z 1 ) > u i ( z 2 ) . Next, we show this claim by considering the following three cases of both z 1 and z 2 .

Case 1. When z 1 B ( x i , 2 C S dist ( x i , Ω ) ) , by (5.21) and C S > 1 , we have

z 2 x i z 2 z 1 + z 1 x i 1 2 C S 4 dist ( z 1 , Ω ) + z 1 x i 1 2 C S 4 dist ( x i , Ω ) + 1 + 1 2 C S 4 z 1 x i < 4 C S dist ( x i , Ω ) ,

which means that z 2 B ( x i , 4 C S dist ( x i , Ω ) ) . From this and (5.20), we deduce that

(5.22) z 1 z 2 z 1 x i + x i z 2 < 6 C S dist ( x i , Ω ) diam S i .

Thus, by the definition of u i , the almost decreasing property of ϕ * with (5.22), and Proposition 3.4, we have

(5.23) u i ( z 1 ) u i ( z 2 ) inf { leng ( Γ z 1 z 2 S i ) : Γ z 1 z 2 is a rectifiable curve joining  z 1 and z 2 } z 1 z 2 ϕ ( z 1 z 2 ) [ ϕ * ( z 1 z 2 ) ] 1 ϕ ( z 1 z 2 ) [ ϕ * ( diam S i ) ] 1 ϕ ( z 1 z 2 ) g i ( z 1 ) ϕ ( z 1 z 2 ) [ g i ( z 1 ) + g i ( z 2 ) ] .

Case 2. When z 2 B ( x i , 2 C S dist ( x i , Ω ) ) , due to a similar argument to Case 1, (5.23) also holds in this case.

Case 3. z 1 , z 2 Ω \ B ( x i , 2 C S dist ( x i , Ω ) ) . In this case, from Definition 2.5(iv), it follows that

S i B ( x i , diam S i ) B ( x i , 2 C S dist ( x i , Ω ) ) ,

and hence, z 1 , z 2 Ω \ S i . In the following, we consider two subcases.

Case 3.1. If z 1 and z 2 belong to the same component of Ω \ S i , then, by the definition of u i , one has either u i ( z 1 ) = 0 = u i ( z 2 ) or u i ( z 1 ) and u i ( z 2 ) are equal to the same constant, both of which imply that u i ( z 1 ) u i ( z 2 ) = 0 .

Case 3.2. If z 1 and z 2 belong to different components of Ω \ S i , then the intersection of the line segment z 1 z 2 ¯ and S i cannot be empty. We choose ω i z 1 z 2 ¯ S i . Then, by Definition 2.5(iii) and (5.20), one has

(5.24) C S 1 dist ( ω i , Ω ) dist ( ω i , S i ) diam S i C S dist ( x i , Ω ) .

From (5.21), we infer that

z 1 z 2 1 2 C S 4 dist ( z 1 , Ω ) 1 2 C S 4 [ z 1 ω i + dist ( ω i , Ω ) ] 1 2 C S 4 [ z 1 z 2 + dist ( ω i , Ω ) ] ,

which, together with (5.24) and C S > 1 , further implies that

z 1 z 2 1 2 C S 4 1 dist ( ω i , Ω ) C S 2 2 C S 4 1 dist ( x i , Ω ) < 1 C S 2 dist ( x i , Ω ) .

However, notice that, by both Definition 2.5(ii) and (5.20), we have

z 1 z 2 C S 1 diam S i C S 2 dist ( x i , Ω ) ,

which leads to a contradiction. Thus, this case does not exist, and hence, both z 1 and z 2 must be in the same component of Ω \ S i . In view of this, we always have u i ( z 1 ) u i ( z 2 ) = 0 when z 1 , z 2 Ω \ B ( x i , 2 C S dist ( x i , Ω ) ) .

Altogether, we conclude that g i is a positive harmless constant multiple of an element in D ball ϕ , 1 ( 2 C S 4 ) ( u i ) , which proves the above claim. Thus, by Proposition 2.18 with θ 1 1 ( 2 C S 4 ) and by (5.20), we obtain u i M ˙ ball ϕ , p ( Ω ) and

u i M ˙ ball ϕ , p ( Ω ) g i L p ( Ω ) [ diam S i ] n p [ ϕ * ( diam S i ) ] 1 .

Next, define

u i = 1 N 1 [ F ( diam S i ) ] 1 p [ diam S i ] n p ϕ * ( diam S i ) u i

and

g i = 1 N 1 [ F ( diam S i ) ] 1 p [ diam S i ] n p ϕ * ( diam S i ) g i = i = 1 N 1 [ F ( diam S i ) ] 1 p [ diam S i ] n p 1 B ( x i , 2 C S dist ( x i , Ω ) ) .

Then g is a positive harmless constant multiple of an element in D ball ϕ , 1 ( 2 C S 4 ) ( u ) .

Notice that, by C S > 1 , one has, for any given i { 1 , , N 1 } and z B ( x i , 2 C S dist ( x i , Ω ) ) ,

B ( x i , C S 1 dist ( x i , Ω ) ) B ( z , 3 C S dist ( x i , Ω ) ) ,

which further implies that

[ F ( diam S i ) ] 1 p [ diam S i ] n p 1 B ( x i , 2 C S dist ( x i , Ω ) ) ( z ) = B ( z , 3 C S dist ( x i , Ω ) ) B ( x i , C S 1 dist ( x i , Ω ) ) 2 B ( z , 3 C S dist ( x i , Ω ) ) [ F ( diam S i ) ] 1 p [ diam S i ] n p 1 ˜ i ( w ) 1 2 d w 2 [ F ( diam S i ) ] 1 p [ diam S i ] n p 1 ˜ i 1 2 ( z ) 2

with the implicit positive constants depending only on both C S and n , here and thereafter, 1 ˜ i 1 B ( x i , C S 1 dist ( x i , Ω ) ) . Thus, from C g D ball ϕ , 1 ( 2 C S 4 ) ( u ) for some positive constant C , Proposition 2.18 with θ 1 1 ( 2 C S 4 ) , the Fefferman-Stein vector-valued maximal inequality, respectively, with respect to the indices 2 and p [20, Theorem 1], Definition 2.5(iv), and the fact that { S i } i = 1 N 1 is pairwise disjoint, we deduce that

(5.25) u M ˙ ball ϕ , p ( Ω ) p g L p ( Ω ) p Ω i = 1 N 1 [ F ( diam S i ) ] 1 p [ diam S i ] n p 1 ˜ i 1 2 ( z ) 2 p d z R n i = 1 N 1 [ F ( diam S i ) ] 1 p [ diam S i ] n p 1 ˜ i ( z ) p d z R n i = 1 N 1 [ F ( diam S i ) ] 1 p [ diam S i ] n p 1 S i ( z ) p d z i = 1 N 1 S i F ( diam S i ) [ diam S i ] n d z i = 1 N 1 F ( diam S i ) ,

and, moreover, u M ˙ ball ϕ , p ( Ω ) . Since Ω is a bounded M ˙ ball ϕ , p -embedding domain, it then follows that u M ˙ ϕ ( p ) , ( Ω ) , which, combined with both Remark 2.16(ii) and (5.25), further implies that, when x , y Ω \ E ,

(5.26) u ( x ) u ( y ) ϕ ( p ) ( x y ) u M ˙ ball ϕ , p ( Ω ) ϕ ( p ) ( x y ) i = 1 N 1 F ( diam S i ) 1 p ,

where E is the exceptional set of (2.13) with ϕ therein replaced by ϕ ( p ) . Observe that, for any given i { 1 , , N 1 } , by both the definition of u i and Definition 2.5(ii), one has

u i ( y ) u i ( x ) = u i ( y ) C S 1 diam S i ,

which, together with the definitions of both F and u , further implies that

i = 1 N 1 F ( diam S i ) i = 1 N 1 F ( diam S i ) [ diam S i ] 1 [ u i ( y ) u i ( x ) ] u ( x ) u ( y ) .

From this, (5.26), and the definition of F , it follows that

(5.27) i = 1 N 1 F ( diam S i ) [ ϕ ( p ) ( x y ) ] p p 1 F ( x y ) .

Finally, notice that, for any z Γ S i , by Definition 2.5(iii), we have

(5.28) dist ( z , Ω ) C S dist ( z , S i ) C S diam S i .

Thus, by applying the almost increasing property of F with (5.28), Proposition 3.4 with c C S , and the assumption of Γ with (2.4) and applying (5.19) and (5.27), we conclude that

Γ F * ( dist ( z , Ω ) ) d z = i = 1 N 1 Γ S i F * ( dist ( z , Ω ) ) d z + Γ S 0 F * ( dist ( z , Ω ) ) d z + Γ S N F * ( dist ( z , Ω ) ) d z i = 1 N 1 F ( diam S i ) Γ S i [ dist ( z , Ω ) ] 1 d z + F ( x y ) i = 1 N 1 F ( diam S i ) + F ( x y ) F ( x y ) .

Now, we have already showed the above inequality for almost every x , y Ω . Moreover, via a limit argument by the continuity of F , it is easy to extend it for any x , y Ω . This means that Ω is a weak F -cigar domain and hence, due to Proposition 3.11, a weak G -cigar domain. This finishes the proof of Theorem 5.5.□

Finally, as an immediate consequence, we obtain Theorem 1.2 by combining Theorem 5.4 with p ˜ p and Theorem 5.5 with G F ; we omit the details.

Remark 5.6

The assumption of the slice property is necessary for Theorem 5.5 and also for the necessity part of Theorem 1.2. To see this, let x 0 R n . Inspired by the counterexample in [3, p. 578], we consider the remaining domain of the ball B ( x 0 , 1 ) after removing some disjoint small balls. That is, define

E k = 1 j = 1 2 k B ( x k , j , 2 k 2 ) and Ω B ( x 0 , 1 ) \ E ,

where, for any k N , { x k , j } j { 1 , , 2 k } are points on B ( x 0 , 2 k ) such that { B ( x k , j , 2 k 2 ) } k , j are pairwise disjoint. Then we claim that Ω does not have the slice property. Indeed, let x Ω be close to B ( x 0 , 1 ) and y Ω be close to x 0 and also close to B ( x k 0 , j k 0 , 2 k 0 2 ) for some k 0 N big enough. We take an arbitrary N N and an arbitrary partition { S i } i = 0 N of Ω satisfying (i), (ii), and (iii) of Definition 2.5. This partition of Ω can also be regarded as a partition of B ( x 0 , 1 ) removing E , that is, for any i { 0 , , N } , S i F i \ E , where { F i } i = 0 N is a partition of B ( x 0 , 1 ) . Then there must exist i { 0 , , N } such that F i contains k = m B ( x k , j k , 2 k 2 ) for some m N , where j k { 1 , , 2 k } . This implies that Definition 2.5(iv) fails because dist ( y , Ω ) is small. Thus, the aforementioned claim holds. In addition, notice that Ω is regular. Then, by Lemma 4.11, Ω is an extension domain and therefore an embedding domain. Also note that Ω is not a weak F -cigar domain [this can be inferred from choosing x 1 and x 2 very close to the boundary of some ball B ( x k , j , 2 k 2 ) but satisfying x 1 x 2 2 k 1 ]. This example shows that the assumption of the slice property in Theorem 5.5, and hence, Theorem 1.2 cannot be absent.

Acknowledgements

The authors would like to thank both referees for their carefully reading and useful comments which improve the quality of this article.

  1. Funding information: This project is partially supported by the National Key Research and Development Program of China (Grant No. 2020YFA0712900), the National Natural Science Foundation of China (Grant Nos. 12431006 and 12371093), and the Fundamental Research Funds for the Central Universities (Grant No. 2233300008).

  2. Author contributions: All authors developed and discussed the results and contributed to the final manuscript.

  3. Conflict of interest: The authors state no conflict of interest.

  4. Data availability statement: Data sharing not applicable to this article as no data sets were generated or analyzed during the current study.

References

[1] R. A. Adams, On the Orlicz-Sobolev imbedding theorem, J. Funct. Anal. 24 (1977), 241–257, https://doi.org/10.1016/0022-1236(77)90055-6. Suche in Google Scholar

[2] R. Alvarado, P. Górka, and P. Hajłasz, Sobolev embedding for M1,p spaces is equivalent to a lower bound of the measure, J. Funct. Anal. 279 (2020), Paper No. 108628, 39 pp. https://doi.org/10.1016/j.jfa.2020.108628. Suche in Google Scholar

[3] S. M. Buckley and P. Koskela, Sobolev-Poincaré implies John, Math. Res. Lett. 2 (1995), 577–593, DOI: https://dx.doi.org/10.4310/MRL.1995.v2.n5.a5. 10.4310/MRL.1995.v2.n5.a5Suche in Google Scholar

[4] S. M. Buckley and P. Koskela, Criteria for imbeddings of Sobolev-Poincaré type, Internat. Math. Res. Not. 18 (1996), 881–901, 10.1155/S1073792896000542. Suche in Google Scholar

[5] A. M. Caetano, A. Gogatishvili, and B. Opic, Sharp embeddings of Besov spaces involving only logarithmic smoothness, J. Approx. Theory 152 (2008), 188–214, https://doi.org/10.1016/j.jat.2007.12.003. Suche in Google Scholar

[6] F. Cobos and Ó. Domínguez, Embeddings of Besov spaces of logarithmic smoothness, Studia Math. 223 (2014), 193–204, https://doi.org/10.1016/j.jat.2007.12.003. Suche in Google Scholar

[7] F. Cobos and Ó. Domínguez, On Besov spaces of logarithmic smoothness and Lipschitz spaces, J. Math. Anal. Appl. 425 (2015), 71–84, https://doi.org/10.1016/j.jmaa.2014.12.034. Suche in Google Scholar

[8] F. Cobos and Ó. Domínguez, On the relationship between two kinds of Besov spaces with smoothness near zero and some other applications of limiting interpolation, J. Fourier Anal. Appl. 22 (2016), 1174–1191, 10.1007/s00041-015-9454-6. Suche in Google Scholar

[9] F. Cobos, Ó Domínguez, and H. Triebel, Characterizations of logarithmic Besov spaces in terms of differences, Fourier-analytical decompositions, wavelets and semi-groups, J. Funct. Anal. 270 (2016), 4386–4425, https://doi.org/10.1016/j.jfa.2016.03.007. Suche in Google Scholar

[10] F. Cobos and D. L. Fernandez, Hardy-Sobolev spaces and Besov spaces with a function parameter, in: Function Spaces and Applications (Lund, 1986), Lecture Notes in Mathematics, vol. 1302, Springer, Berlin, 1988, pp. 158–170, https://doi.org/10.1007/BFb0078872. Suche in Google Scholar

[11] R. R. Coifman and G. Weiss, Analyse Harmonique Non-commutative sur Certains Espaces Homogènes. Étude de Certaines Intégrales Singulières (French), Lecture Notes in Mathematics, vol. 242, Springer-Verlag, Berlin-New York, 1971, 10.1007/BFb0058953. Suche in Google Scholar

[12] T. Deheuvels, Sobolev extension property for tree-shaped domains with self-contacting fractal boundary, Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 15 (2016), 209–247, 10.2422/2036-2145.201307_008. Suche in Google Scholar

[13] Ó Domínguez, Sharp embeddings of Besov spaces with logarithmic smoothness in sub-critical cases, Anal. Math. 43 (2017), 219–240, https://doi.org/10.1007/s10476-017-0305-3. Suche in Google Scholar

[14] Ó Domínguez, Sharp estimates of the norms of embeddings between Besov spaces, Z. Anal. Anwend. 37 (2018), 127–149, 10.4171/ZAA/1606. Suche in Google Scholar

[15] Ó Domínguez, D. D. Haroske, and S. Tikhonov, Embeddings and characterizations of Lipschitz spaces, J. Math. Pures Appl. (9) 144 (2020), 69–105, https://doi.org/10.1016/j.matpur.2020.11.004. Suche in Google Scholar

[16] Ó Domínguez and S. Tikhonov, Function spaces of logarithmic smoothness: embeddings and characterizations, Mem. Amer. Math. Soc. 282 (2023), no. 1393, 166, https://doi.org/10.1090/memo/1393. Suche in Google Scholar

[17] T. K. Donaldson and N. S. Trudinger, Orlicz-Sobolev spaces and imbedding theorems, J. Funct. Anal. 8 (1971), 52–75, https://doi.org/10.1016/0022-1236(71)90018-8. Suche in Google Scholar

[18] W. Farkas, N. Jacob, and R. L. Schilling, Function spaces related to continuous negative definite functions: ψ-Bessel potential spaces, Dissertationes Math. (Rozprawy Mat.) 393 (2001), 1–62, 10.4064/dm393-0-1. Suche in Google Scholar

[19] W. Farkas and H.-G. Leopold, Characterisationsof function spaces of generalized smoothness, Ann. Mat. Pura Appl. (4) 185 (2006), 1–62, https://doi.org/10.1007/s10231-004-0110-z. Suche in Google Scholar

[20] C. Fefferman and E. M. Stein, Some maximal inequalities, Amer. J. Math. 93 (1971), 107–115, https://doi.org/10.2307/2373450. Suche in Google Scholar

[21] F. W. Gehring, Uniform domains and the ubiquitous quasidisk, Jahresber. Deutsch. Math. Verein. 89 (1987), 88–103. Suche in Google Scholar

[22] F. W. Gehring and O. Martio, Lipschitz classes and quasiconformal mappings, Ann. Acad. Sci. Fenn. Ser. A I Math. 10 (1985), 203–219, https://doi.org/10.5186/aasfm.1985.1022. Suche in Google Scholar

[23] F. W. Gehring and B. G. Osgood, Uniform domains and the quasihyperbolic metric, J. Analyse Math. 36 (1979), 50–74, https://doi.org/10.1007/BF02798768. Suche in Google Scholar

[24] M. L. Goldman, A description of the trace space for functions of a generalized Hölder class, Dokl. Akad. Nauk. SSSR 231 (1976), 525–528. Suche in Google Scholar

[25] P. Górka, In metric-measure spaces Sobolev embedding is equivalent to a lower bound for the measure, Potential Anal. 47 (2017), 13–19, https://doi.org/10.1007/s11118-016-9605-7. Suche in Google Scholar

[26] P. Gurka and B. Opic, Sharp embeddings of Besov spaces with logarithmic smoothness, Rev. Mat. Complut. 18 (2005), 81–110, http://eudml.org/doc/44545. 10.5209/rev_REMA.2005.v18.n1.16715Suche in Google Scholar

[27] P. Hajłasz, Sobolev spaces on an arbitrary metric space, Potential Anal. 5 (1996), 403–415, https://doi.org/10.1007/BF00275475. Suche in Google Scholar

[28] P. Hajłasz, P. Koskela, and H. Tuominen, Sobolev embeddings, extensions and measure density condition, J. Funct. Anal. 254 (2008), 1217–1234, https://doi.org/10.1016/j.jfa.2007.11.020. Suche in Google Scholar

[29] P. Hajłasz, P. Koskela, and H. Tuominen, Measure density and extendability of Sobolev functions, Rev. Mat. Iberoam. 24 (2008), 645–669, 10.4171/RMI/551. Suche in Google Scholar

[30] D. D. Haroske, Z. Liu, S. D. Moura, and L. Skrzypczak, Embeddings of generalised Morrey smoothness spaces, Acta Math. Sin. (Engl. Ser.) 41 (2025), 413–456, https://doi.org/10.1007/s10114-025-3553-3. Suche in Google Scholar

[31] T. Heikkinen and N. Karak, Orlicz-Sobolev embeddings, extensions and Orlicz-Poincaré inequalities, J. Funct. Anal. 282 (2022), Paper No. 109292, 53 pp. 10.1016/j.jfa.2021.109292. Suche in Google Scholar

[32] J. Hu, A note on Hajłasz-Sobolev spaces on fractals, J. Math. Anal. Appl. 280 (2003), 91–101, 10.1016/S0022-247X(03)00039-8. Suche in Google Scholar

[33] X. Hu and L. Liu, Restricting Riesz-logarithmic-Gagliardo-Lipschitz potentials, Acta Math. Sin. (Engl. Ser.) 41 (2025), 131–148, https://doi.org/10.1007/s10114-025-3458-1. Suche in Google Scholar

[34] A. Jonsson and H. Wallin, Function Spaces on Subsets of RN, Math. Rep. 2 (1984), 221, xiv+221 . Suche in Google Scholar

[35] G. A. Kaljabin, Descriptions of functions from classes of Besov-Lizorkin-Triebel type, Trudy Mat. Inst. Steklov. 156 (1980), 82–109. Suche in Google Scholar

[36] N. Karak, Generalized Lebesgue points for Sobolev functions, Czechoslovak Math. J. 67 (2017), 143–150, https://doi.org/10.21136/CMJ.2017.0405-15. Suche in Google Scholar

[37] N. Karak, Measure density and embeddings of Hajłasz-Besov and Hajłasz-Triebel-Lizorkin spaces, J. Math. Anal. Appl. 475 (2019), 966–984, DOI: https://doi.org/10.1016/j.jmaa.2018.11.086. 10.1016/j.jmaa.2018.11.086Suche in Google Scholar

[38] N. Karak, Lower bound of measure and embeddings of Sobolev, Besov and Triebel-Lizorkin spaces, Math. Nachr. 293 (2020), 120–128, https://doi.org/10.1002/mana.201800121. Suche in Google Scholar

[39] N. Karak, Triebel-Lizorkin capacity and Hausdorff measure in metric spaces, Math. Slovaca 70 (2020), 617–624, https://doi.org/10.1515/ms-2017-0376. Suche in Google Scholar

[40] N. Karak, Sobolev embedding implies regularity of measure in metric measure spaces, arXiv: 1903.02342. Suche in Google Scholar

[41] V. Kokilashvili, A. Meskhi, H. Rafeiro, and S. Samko, Integral Operators in Non-standard Function Spaces, Vol. 1, Variable Exponent Lebesgue and Amalgam Spaces, Operator Theory: Advances and Applications 248, Birkhäuser/Springer, Cham, 2016, DOI: https://doi.org/10.1007/978-3-319-21015-5. 10.1007/978-3-319-21015-5_1Suche in Google Scholar

[42] P. Koskela, Extensions and imbeddings, J. Funct. Anal. 159 (1998), 369–383, https://doi.org/10.1006/jfan.1998.3331. Suche in Google Scholar

[43] P. Koskela, A. Koski, and J. Onninen, Sobolev homeomorphic extensions onto John domains, J. Funct. Anal. 279 (2020), Paper No. 108719, 17 pp. https://doi.org/10.1016/j.jfa.2020.108719. Suche in Google Scholar

[44] P. Koskela and E. Saksman, Pointwise characterizations of Hardy-Sobolev functions, Math. Res. Lett. 15 (2008), 727–744, 10.4310/MRL.2008.v15.n4.a11. Suche in Google Scholar

[45] P. Koskela, D. Yang, and Y. Zhou, A characterizationof Hajłasz-Sobolev and Triebel-Lizorkin spaces via grand Littlewood-Paley functions, J. Funct. Anal. 258 (2010), 2637–2661, https://doi.org/10.1016/j.jfa.2009.11.004. Suche in Google Scholar

[46] P. Koskela, D. Yang, and Y. Zhou, A Jordan Sobolev extension domain, Ann. Acad. Sci. Fenn. Math. 35 (2010), 309–320, 10.5186/aasfm.2010.3519. Suche in Google Scholar

[47] V. Lappalainen, Liph-extension domains, Ann. Acad. Sci. Fenn. Ser. A I Math. Dissertationes 56 (1985), 52, https://doi.org/10.5186/aasfmd.00. Suche in Google Scholar

[48] H.-G. Leopold, Embeddings and entropy numbers in Besov spaces of generalized smoothness, in: Function Spaces, Lecture Notes in Pure and Applied Mathematics, vol. 213, Dekker, New York, 2000, pp. 323–336, DOI: https://doi.org/10.1201/9781482270501. 10.1201/9781482270501Suche in Google Scholar

[49] Z. Li, D. Yang, and W. Yuan, Lebesgue points for Besov and Triebel-Lizorkin spaces with generalized smoothness, Mathematics 9 (2021), Art. No. 2724, https://doi.org/10.3390/math9212724. Suche in Google Scholar

[50] Z. Li, D. Yang, and W. Yuan, Pointwise characterizations of Besov and Triebel-Lizorkin spaces with generalized smoothness and their applications, Acta Math. Sin. (Engl. Ser.) 38 (2022), 623–661, https://doi.org/10.1007/s10114-022-1086-6. Suche in Google Scholar

[51] O. Martio, Definitions for uniform domains, Ann. Acad. Sci. Fenn. Ser. A I Math. 5 (1980), 197–205, https://doi.org/10.5186/aasfm.1980.0517. Suche in Google Scholar

[52] O. Martio and J. Sarvas, Injectivity theorems in plane and space, Ann. Acad. Sci. Fenn. Ser. A I Math. 4 (1979), 383–401, https://doi.org/10.5186/aasfm.1978-79.0413. Suche in Google Scholar

[53] A. D. Melas, A sharp Lp inequality for dyadic A1 weights in Rn, Bull. London Math. Soc. 37 (2005), 919–926, https://doi.org/10.1112/S0024609305004765. Suche in Google Scholar

[54] S. D. Moura, Function spaces of generalised smoothness, Dissertationes Math. (Rozprawy Mat.) 398 (2001), 1–88, DOI: https://doi.org/10.4064/dm398-0-1. 10.4064/dm398-0-1Suche in Google Scholar

[55] Y. Sawano and T. Shimomura, Sobolev embeddings for Riesz potentials of functions in non-doubling Morrey spaces of variable exponents, Collect. Math. 64 (2013), 313–350, https://doi.org/10.1007/s13348-013-0082-7. Suche in Google Scholar

[56] Y. Sawano and T. Shimomura, Sobolev embeddings for generalized Riesz potentials of functions in Morrey spaces L(1,ϕ)(G) over nondoubling measure spaces, J. Funct. Spaces Appl. 2013 (2013), Art. ID 984259, 12 pp. 10.1155/2013/984259. Suche in Google Scholar

[57] Y. Sawano and T. Shimomura, Sobolev embeddings for Riesz potentials of functions in Musielak-Orlicz-Morrey spaces over non-doubling measure spaces, Integral Transf. Spec. Funct. 25 (2014), 976–991, https://doi.org/10.1080/10652469.2014.955099. Suche in Google Scholar

[58] P. Shvartsman, On extensions of Sobolev functions defined on regular subsets of metric measure spaces, J. Approx. Theory 144 (2007), 139–161, https://doi.org/10.1016/j.jat.2006.05.005. Suche in Google Scholar

[59] P. Shvartsman, On Sobolev extension domains in Rn, J. Funct. Anal. 258 (2010), 2205–2245, https://doi.org/10.1016/j.jfa.2010.01.002. Suche in Google Scholar

[60] P. Shvartsman and N. Zobin, On planar Sobolev Lpm-extension domains, Adv. Math. 287 (2016), 237–346, 10.1016/J.AIM.2015.08.031. Suche in Google Scholar

[61] E. M. Stein, Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscillatory Integrals, Princeton Mathematical Series 43, Monographs in Harmonic Analysis, vol. III, Princeton University Press, Princeton, NJ, 1993, https://www.jstor.org/stable/j.ctt1bpmb3s. 10.1515/9781400883929Suche in Google Scholar

[62] H. Triebel, Theory of Function Spaces, Monographs in Mathematics 78, Birkhäuser Verlag, Basel, 1983, DOI: https://doi.org/10.1007/978-3-0346-0416-1. 10.1007/978-3-0346-0416-1Suche in Google Scholar

[63] H. Triebel, The Structure of Functions, Monographs in Mathematics 97, Birkhäuser Verlag, Basel, 2001, DOI: https://doi.org/10.1007/978-3-0348-8257-6. 10.1007/978-3-0348-8257-6Suche in Google Scholar

[64] J. Väisälä, Uniform domains, Tohoku Math. J. (2) 40 (1988), 101–118, 10.2748/tmj/1178228081. Suche in Google Scholar

[65] D. Yang, New characterizations of Hajłasz-Sobolev spaces on metric spaces, Sci. China Ser. A 46 (2003), 675–689, https://doi.org/10.1360/02ys0343. Suche in Google Scholar

[66] Y. Zhou, Hajłasz-Sobolev imbedding and extension, J. Math. Anal. Appl. 382 (2011), 577–593, https://doi.org/10.1016/j.jmaa.2011.04.067. Suche in Google Scholar

[67] Y. Zhou, Fractional Sobolev extension and imbedding, Trans. Amer. Math. Soc. 367 (2015), 959–979, http://www.jstor.org/stable/24512870. Suche in Google Scholar

Received: 2024-06-17
Revised: 2025-02-09
Accepted: 2025-02-25
Published Online: 2025-03-27

© 2025 the author(s), published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Artikel in diesem Heft

  1. Research Articles
  2. Incompressible limit for the compressible viscoelastic fluids in critical space
  3. Concentrating solutions for double critical fractional Schrödinger-Poisson system with p-Laplacian in ℝ3
  4. Intervals of bifurcation points for semilinear elliptic problems
  5. On multiplicity of solutions to nonlinear Dirac equation with local super-quadratic growth
  6. Entire radial bounded solutions for Leray-Lions equations of (p, q)-type
  7. Combinatorial pth Calabi flows for total geodesic curvatures in spherical background geometry
  8. Sharp upper bounds for the capacity in the hyperbolic and Euclidean spaces
  9. Positive solutions for asymptotically linear Schrödinger equation on hyperbolic space
  10. Existence and non-degeneracy of the normalized spike solutions to the fractional Schrödinger equations
  11. Existence results for non-coercive problems
  12. Ground states for Schrödinger-Poisson system with zero mass and the Coulomb critical exponent
  13. Geometric characterization of generalized Hajłasz-Sobolev embedding domains
  14. Subharmonic solutions of first-order Hamiltonian systems
  15. Sharp asymptotic expansions of entire large solutions to a class of k-Hessian equations with weights
  16. Stability of pyramidal traveling fronts in time-periodic reaction–diffusion equations with degenerate monostable and ignition nonlinearities
  17. Ground-state solutions for fractional Kirchhoff-Choquard equations with critical growth
  18. Existence results of variable exponent double-phase multivalued elliptic inequalities with logarithmic perturbation and convections
  19. Homoclinic solutions in periodic partial difference equations
  20. Classical solution for compressible Navier-Stokes-Korteweg equations with zero sound speed
  21. Properties of minimizers for L2-subcritical Kirchhoff energy functionals
  22. Asymptotic behavior of global mild solutions to the Keller-Segel-Navier-Stokes system in Lorentz spaces
  23. Blow-up solutions to the Hartree-Fock type Schrödinger equation with the critical rotational speed
  24. Qualitative properties of solutions for dual fractional parabolic equations involving nonlocal Monge-Ampère operator
  25. Regularity for double-phase functionals with nearly linear growth and two modulating coefficients
  26. Uniform boundedness and compactness for the commutator of an extension of Riesz transform on stratified Lie groups
  27. Normalized solutions to nonlinear Schrödinger equations with mixed fractional Laplacians
  28. Existence of positive radial solutions of general quasilinear elliptic systems
  29. Low Mach number and non-resistive limit of magnetohydrodynamic equations with large temperature variations in general bounded domains
  30. Sharp viscous shock waves for relaxation model with degeneracy
  31. Bifurcation and multiplicity results for critical problems involving the p-Grushin operator
  32. Asymptotic behavior of solutions of a free boundary model with seasonal succession and impulsive harvesting
  33. Blowing-up solutions concentrated along minimal submanifolds for some supercritical Hamiltonian systems on Riemannian manifolds
  34. Stability of rarefaction wave for relaxed compressible Navier-Stokes equations with density-dependent viscosity
  35. Singularity for the macroscopic production model with Chaplygin gas
  36. Global strong solution of compressible flow with spherically symmetric data and density-dependent viscosities
  37. Global dynamics of population-toxicant models with nonlocal dispersals
  38. α-Mean curvature flow of non-compact complete convex hypersurfaces and the evolution of level sets
  39. High-energy solutions for coupled Schrödinger systems with critical growth and lack of compactness
  40. On the structure and lifespan of smooth solutions for the two-dimensional hyperbolic geometric flow equation
  41. Well-posedness for physical vacuum free boundary problem of compressible Euler equations with time-dependent damping
  42. On the existence of solutions of infinite systems of Volterra-Hammerstein-Stieltjes integral equations
  43. Remark on the analyticity of the fractional Fokker-Planck equation
  44. Continuous dependence on initial data for damped fourth-order wave equation with strain term
  45. Unilateral problems for quasilinear operators with fractional Riesz gradients
  46. Boundedness of solutions to quasilinear elliptic systems
  47. Existence of positive solutions for critical p-Laplacian equation with critical Neumann boundary condition
  48. Non-local diffusion and pulse intervention in a faecal-oral model with moving infected fronts
Heruntergeladen am 30.9.2025 von https://www.degruyterbrill.com/document/doi/10.1515/anona-2025-0077/html
Button zum nach oben scrollen