Skip to content
BY 4.0 license Open Access Published by De Gruyter December 8, 2020

Regularity for commutators of the local multilinear fractional maximal operators

  • Xiao Zhang and Feng Liu EMAIL logo

Abstract

In this paper we introduce and study the commutators of the local multilinear fractional maximal operators and a vector-valued function b⃗ = (b1, …, bm). Under the condition that each bi belongs to the first order Sobolev spaces, the bounds for the above commutators are established on the first order Sobolev spaces.

MSC 2010: 42B25; 46E35

1 Introduction

1.1 Background

Over the last several years and recently regularity properties of maximal operators have been studied extensively. The first work was due to Kinnunen [13] who showed that the classical Hardy-Littlewood maximal operator

Mf(x)=supr>01|B(x,r)|B(x,r)|f(y)|dy,

is bounded on the first order Sobolev spaces W1,p(ℝn) for 1 < p ≤ ∞, where B(x, r) is the open ball in ℝn centered at x with radius r, and |B(x, r)| denotes the volume of B(x, r). Later on, Kinnunen’s result was extended to the local version in [14], to the fractional version in [16] and to the multisublinear version in [6, 19]. Since we do not have sublinearity for the weak derivatives of the Hardy-Littlewood maximal function, the continuity of M : W1,p(ℝn) → W1,p(ℝn) for 1 < p < ∞ is affirmatively a nontrivial issue, which was addressed by Luiro [23] and later extensions were given in [24]. We can consult [2, 4, 5, 7, 25, 26] for the endpoint Sobolev regularity of maximal operators, as well as [17, 20] for the regularity properties of maximal operators on other smooth function spaces, such as Triebel-Lizorkin spaces, fractional Sobolev spaces and Besov spaces.

It should be pointed out that the commutativity with translations for maximal operators plays a key role in deducing the boundedness of maximal operators on W1,p(ℝn). In fact, it was shown in [9, Theorme 1] that any sub-linear operator which commutes with translations and is bounded on Lp(ℝn) for some p ∈ (1, ∞) forms a bounded operator on W1,p(ℝn). Unfortunately, the local maximal operator does not commute with translations, it makes the investigation on the regularity of the local maximal operators very interesting. In 1998, Kinnunen and Lindqvist [14] firstly studied the regularity of the local maximal operator

MΩf(x)=sup0<r<dist(x,Ωc)1|B(x,r)|B(x,r)|f(y)|dy,

for f Lloc1 (Ω), where Ω is a subdomain in ℝn and Ωc = ℝnΩ. They showed that

|MΩf(x)|2MΩ|f|(x), (1.1)

for almost every xΩ and fW1,p(Ω) with some p ∈ (1, ∞) (see also [9]). As an application of (1.1), we can get the boundedness for MΩ : W1,p(Ω) → W1,p(Ω) for 1 < p < ∞. Here W1,p(Ω), 1 ≤ p ≤ ∞, consists of those functions fLp(Ω) whose weak first order partial derivatives Dif, i = 1, …, n, belong to Lp(Ω). The W1,p(Ω) norm of f is given by

fW1,p(Ω):=fLp(Ω)+fLp(Ω),

where ∇f = (D1f, …, Dnf) is the weak gradient of f and ∥fLp(Ω) = (∫Ω|f(x)|pdx)1/p. Later on, the above result was extended by Heikkinen et al. [11] to the fractional case. Let 0 ≤ α < n and f Lloc1 (Ω), the local fractional maximal operator Mα,Ω is defined as

Mα,Ωf(x)=sup0<r<dist(x,Ωc)rα|B(x,r)|B(x,r)|f(y)|dy.

The authors in [11] showed that if |Ω| < ∞, 1 < p < n and 1 ≤ α < n/p, then the map Mα,Ω : W1,p(Ω) → W1,q(Ω) is bounded, where q = np/(n – (α – 1)p). Moreover,

|Mα,Ωf(x)|2Mα,Ω|f|(x)+αMα1,Ωf(x),

for almost every xΩ if fW1,p(Ω). They also showed that if fLp(Ω) for p ∈ (n/(n – 1), ∞) and 1 ≤ α < min{(n – 1)/p, n – 2n/((n – 1)p)} + 1, then

|Mα,Ωf(x)|n(Mα1,Ωf(x)+Sα1,Ωf(x)), (1.2)

for almost every xΩ. Here 𝓢α,Ω is the local spherical maximal operator, i.e.gf

Sα,Ωf(x)=sup0<r<dist(x,Ωc)rα|B(x,r)|B(x,r)|f(y)|dHn1(y),

where d𝓗n–1 is the normalized (n – 1)-dimensional Hausdorff measure. An immediate consequence of (1.2) shows that the map Mα,Ω : Lp(Ω) → W1,q(Ω) is bounded for n/(n – 1) < p < ∞, 1 ≤ α < n/p and q = np/(n – (α – 1)p) if |Ω| < ∞. Recently, Ramos et al. [29] established the bounds for Mα,Ω : Lp(Ω) → W1,p(Ω) for 1 < pn/(n – 1) and 1 ≤ α < n under certain conditions on Ω.

Recently, Hart et al. [10] investigated the local multilinear maximal operator and its fractional variant. Let m be a positive integer and 0 ≤ α < mn. For f⃗ = (f1, …, fm) with each fj Lloc1 (Ω), the local multilinear fractional type maximal operator 𝔐α,Ω is defined by

Mα,Ω(f)(x)=sup0<r<dist(x,Ωc)rα|B(x,r)|mj=1mB(x,r)|fj(y)|dy,xΩ.

Particularly, when α = 0, the operator 𝔐α,Ω reduces to the local multilinear maximal operator 𝔐Ω. When m = 1, the operator 𝔐Ω (resp., 𝔐α,Ω) is just the local (resp., fractional) maximal operator MΩ (resp., Mα,Ω). It was pointed out in [10] that

Mα,Ω(f)Lq(Ω)α,m,n,p1,,pmj=1mfjLpj(Ω) (1.3)

for all f⃗ = (f1, …, fm) with each fjLpj(Ω), and

Mα,Ω:Lp1(Ω)××Lpm(Ω)Lq(Ω)is continuous (1.4)

if each pj > 1, 0 ≤ α < mn, 1/q = 1/p1 + ⋯ + 1/pmα/n and one of the following conditions holds:

  1. α = 0, 1 ≤ q ≤ ∞ and 1 < p1, …, pm ≤ ∞;

  2. 0 < α < n, 1 ≤ q < ∞ and 1 < p1, …, pm ≤ ∞;

  3. nα < mn, 1 ≤ q < ∞ and 1 < p1, …, pm < ∞.

Based on the above boundedness, the authors of [10] established the following results.

Theorem A

([10]). Let 1 < p1, …, pm < ∞ and f⃗ = (f1, …, fm) with each fjW1,pj(Ω). Assume that one of the following conditions holds:

  1. α = 0, 1/q = 1/p1 + ⋯ + 1/pm < 1;

  2. 1 ≤ α < mn, 1/q = 1/p1 + ⋯ + 1/pm – (α – 1)/n, 1 < q < ∞, |Ω| < ∞;

  3. 1 < p1, ..., pm < n, 1/q = 1/p1 + ⋯ + 1/pm – (α + m – 1)/n, 1/m < q < ∞, Ω admits a pj-Sobolev embedding for each j = 1, ..., m. Here we say that Ω admits a r-Sobolev embedding for 1 < r < n if W1,r(Ω) continuously embeds into L(Ω) with 1/ = 1/r – 1/n, i.e. ∥gL(Ω)r,ngW1,r(Ω).

    Then

    |Mα,Ω(f)(x)|αMα1,Ω(f)+2l=1mMα,Ω(fl)(x),

    for almost every xΩ, where f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm). Moreover,

    Mα,Ω(f)W1,q(Ω)α,m,n,p1,,pmj=1mfjW1,pj(Ω).

Theorem B

([10]). Let n/(n – 1) < p1, …, pm < ∞ and f⃗ = (f1, …, fm) with each fjLpj(Ω). Let 1 ≤ α < + 1 with β = min1≤jm{(n – 1)/pj, n – 2n/((n – 1)pj)}. Let 1/q = 1/p1 + ⋯ + 1/pm – (α – 1)/n and 1 < q < ∞. Then 𝔐α,Ω(f⃗) ∈ W1,q(Ω). Moreover,

|Mα,Ω(f)(x)|α,m,n(Mα1,Ω(f)+l=1mSα¯,Ωfl(x)1jmjlMα¯,Ωfj(x)),

for almost every xΩ, where = (α – 1)/m. If |Ω| < ∞, then

Mα,Ω(f)W1,q(Ω)α,m,n,p1,,pmj=1mfjLpj(Ω).

1.2 Commutators of local multilinear fractional maximal operators

An active extension of current research is to investigate the regularity properties for the commutators of the Hardy-Littlewood maximal function, which is defined in the form

[b,M](f)(x)=b(x)Mf(x)M(bf)(x),xRn,

where b is a locally integral function defined on ℝn. The maximal commutator of M with b is defined by

Mbf(x)=supr>01|B(x,r)|B(x,r)|b(x)b(y)||f(y)|dy,xRn.

The Lp (1 < p < ∞) boundedness for [b, M] was first established by Milman and Schonbek [27] by assuming that b ≥ and b ∈ BMO(ℝn). The same conclusion was also obtained by Bastero et al. [3] without the assumption b ≥ 0. The boundedness of Mb has been studied intensively by many authors (see [1, 12, 30]). Liu et al. [22] proved that

max{[b,M](f)W1,p(Rn),MbfW1,p(Rn)}n,p1,p2bW1,p2(Rn)fW1,p1(Rn),

provided that 1 < p1, p2, p < ∞, 1/p = 1/p1 + 1/p2, fW1,p1(ℝn) and bW1,p2(ℝn). Very recently, Liu and Xi [21] extended the above result to the fractional version. The authors in [21] also studied the commutators of the local fractional maximal operators. To be more precise, for 0 ≤ α < n and a locally integrable function b defined on Ω, we define the commutator of the local fractional maximal function by

[b,Mα,Ω](f)(x)=b(x)Mα,Ωf(x)Mα,Ω(bf)(x),xΩ.

The fractional maximal commutators of Mα,Ω with b is defined by

Mα,Ω,bf(x)=sup0<r<dist(x,Ωc)rα|B(x,r)|B(x,r)|b(x)b(y)||f(y)|dy,xΩ.

Liu and Xi established the following results.

Theorem C

([21]). (i) Let 1 < p1, p2, p1p2/(p1+p2) < ∞, 1 < p1 < n, 1 ≤ α < n/p1 and 1/p = 1/p1 + 1/p2 – (α – 1)/n. Assume that bW1,p2(Ω) and |Ω| < ∞, then [b, Mα,Ω] is bounded from W1,p1(Ω) to W1,p(Ω). Moreover, if fW1,p1(Ω), then

[b,Mα,Ω](f)W1,p(Ω)α,n,p1,p2,|Ω|bW1,p2(Ω)fW1,p1(Ω).

(ii) Let 1 < p1, p2, p1p2/(p1+p2) < ∞, 1 < p1 < n, 1 ≤ α < n/p1 and 1/p = 1/p1 + 1/p2α/n. Assume that bW1,p2(Ω) and Ω admits a p1-Sobolev embedding, then [b, Mα,Ω] is bounded from W1,p1(Ω) to W1,p(Ω). Moreover, if fW1,p1(Ω), then

[b,Mα,Ω](f)W1,p(Ω)α,n,p1,p2bW1,p2(Ω)fW1,p1(Ω).

(iii) Let 1 < p1, p2 < ∞, n/(n – 1) < p1p2/(p1 + p2) < ∞, 1 ≤ α < n/p1 and 1/p = 1/p1 + 1/p2 – (α – 1)/n. Assume that |Ω| < ∞ and bW1,p2(Ω), then [b, Mα,Ω] is bounded from Lp1(Ω) to W1,p(Ω). Moreover, if fLp1(Ω), then

[b,Mα,Ω](f)W1,p(Ω)α,n,p1,p2,|Ω|bW1,p2(Ω)fLp1(Ω).

The same conclusions also hold for the operator Mα,Ω,b.

The man purpose of this work is to investigate the Sobolev regularity of the commutators of local multilinear fractional maximal operators.

Definition 1.1

(Commutators of the local multilinear fractional maximal operators). Let m ≥ 1 and f⃗ = (f1, …, fm) and b⃗ = (b1, …, bm) with each fj Lloc1 (Ω) and bj Lloc1 (Ω). The Commutator of 𝔐α,Ω and b⃗ is defined by

[b,Mα,Ω](f)(x)=i=1m[b,Mα,Ω]i(f)(x),xΩ,

where

[b,Mα,Ω]i(f)(x)=bi(x)Mα,Ω(f)(x)Mα,Ω(f1,,fi1,bifi,fi+1,,fm)(x).

The local multilinear fractional maximal commutator with b⃗ is defined by

Mα,Ω,b(f)(x)=i=1mMα,Ω,bi(f)(x),

where

Mα,Ω,bi(f)(x)=sup0<r<dist(x,Ωc)rα|B(x,r)|m(B(x,r))m|bi(x)bi(yi)|j=1m|fj(yj)|dy1dym.

When α = 0, we denote [b⃗, 𝔐α,Ω] = [b⃗, 𝔐Ω] and 𝔐α,Ω,b⃗ = 𝔐Ω,b⃗. When m = 1, then [b⃗, 𝔐α,Ω] = [b1, Mα,Ω] and 𝔐α,Ω,b⃗ = Mα,Ω,b1. Particularly, when Ω = ℝn, the commutator [b⃗, 𝔐α,Ω] (resp, 𝔐α,Ω,b⃗) is just the commutator of the multilinear fractional maximal operators (resp., the multilinear maximal commutator), which were introduced by Zhang [31] who investigated the multiple weighted estimates for these commutators. It should be pointed out that the multilinear commutator was first studied by Pérez and Torres in [28] and was later developed by many authors (see [18] et al.).

We now point out the following remarks, which are very useful for our proofs.

Remark 1.2

Let 1 < p1, …, pm+1 < ∞ and 1/pi + 1/pm+1 < 1 for i = 1, …, m. Let 0 ≤ α < mn, 1/q = 1/p1 + ⋯ + 1/pm+1α/n, 1/p = 1/q – 1/pm+1, 1 < p, q < ∞. Let f⃗ = (f1, …, fm) with each fjLpj(Ω) and b⃗ = (b1, …, bm) with each bjLpm+1(Ω). Then for all i = 1, …, m, it holds that

max{[b,Mα,Ω]i(f)Lq(Ω),Mα,Ω,bi(f)Lq(Ω)}α,m,n,p1,,pm+1biLpm+1(Ω)j=1mfjLpj(Ω). (1.5)

To see (1.5), let us fix i ∈ {1, …, m} set 1/ti = 1/pi + 1/pm+1. It is clear that 1 < ti < ∞ and 1/q = 1/p1 + ⋯ + 1/pi–1 + 1/ti + 1/pi+1 + ⋯ +1/pmα/n. Observe that

max{|[b,Mα,Ω]i(f)(x)|,|Mα,Ω,bi(f)(x)|}|bi(x)|Mα,Ω(f)(x)+Mα,Ω(f1,,fi1,bifi,fi+1,,fm)(x), (1.6)

for all xΩ. By (1.3), (1.6) and Hölder’s inequality, we have

max{[b,Mα,Ω]i(f)Lq(Ω),Mα,Ω,bi(f)Lq(Ω)}biMα,Ω(f)Lq(Ω)+Mα,Ω(f1,,fi1,bifi,fi+1,,fm)Lq(Ω)biLpm+1(Ω)Mα,Ω(f)Lp(Ω)+CbifiLti(Ω)1jmjifjLpj(Ω)α,m,n,p1,,pm+1biLpm+1(Ω)j=1mfjLpj(Ω),

which proves (1.5). Then we get from (1.5) that

max{[b,Mα,Ω](f)Lq(Ω),Mα,Ω,b(f)Lq(Ω)}α,m,n,p1,,pm+1(i=1mbiLpm+1(Ω))j=1mfjLpj(Ω). (1.7)

Based on the above, we shall establish the following results.

Theorem 1.1

Let 1 < p1, …, pm+1 < ∞ such that 1/pi + 1/pm+1 < 1 for i = 1, …, m. If b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) and one of the following conditions holds:

  1. α = 0, 1/q = 1/p1 + ⋯ + 1/pm+1, q > 1;

  2. 1 ≤ α < mn, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n, q > 1, |Ω| < ∞;

  3. 1 ≤ α < mn, 1 < p1, ..., pm < n, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α + m – 1)/n, 1/m < q < ∞, Ω admits a pj-Sobolev embedding for each j = 1, ..., m.

    Then

    [b,Mα,Ω]:W1,p1(Ω)××W1,pm(Ω)W1,q(Ω)

    is bounded. Precisely, for f⃗ = (f1, …, fm) with each fjW1,pj(Ω), the following estimate holds:

    [b,Mα,Ω](f)W1,q(Ω)α,m,n,p1,,pm+1(i=1mbiW1,pm+1(Ω))j=1mfjW1,pj(Ω). (1.8)

Theorem 1.2

Let 1 < p1, …, pm+1 < ∞ and 1/ti = 1/pi + 1/pm+1 for i = 1, …, m with ti > n/(n – 1). Let 1 ≤ α < + 1 with β = min1≤jm {(n – 1)/pj, n – 2n/((n – 1)tj)} and 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n with q > 1. If b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) and |Ω| < ∞, then

[b,Mα,Ω]:Lp1(Ω)××Lpm(Ω)W1,q(Ω)

is bounded. Precisely, if f⃗ = (f1, …, fm) with each fjLpj(Ω), then

[b,Mα,Ω](f)W1,q(Ω)α,m,n,p1,,pm+1,|Ω|(i=1mbiW1,pm+1(Ω))j=1mfjLpj(Ω). (1.9)

Theorem 1.3

Let 1 < p1, …, pm+1 < ∞ such that 1/pi + 1/pm+1 < 1 for i = 1, …, m. If b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) and one of the following conditions holds:

  1. α = 0, 1/q = 1/p1 + ⋯ +1/pm+1, q > 1;

  2. 1 ≤ α < mn, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n, q > 1, |Ω| < ∞;

  3. 1 ≤ α < mn, 1 < p1, ..., pm < n, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α + m – 1)/n, 1/m < q < ∞, Ω admits a pj-Sobolev embedding for each j = 1, ..., m.

    Then

    Mα,Ω,b:W1,p1(Ω)××W1,pm(Ω)W1,q(Ω)

    is bounded. Precisley, for f⃗ = (f1, …, fm) with each fjW1,pj(Ω), the following estimate holds:

    Mα,Ω,b(f)W1,q(Ω)α,m,n,p1,,pm+1(i=1mbiW1,pm+1(Ω))j=1mfjW1,pj(Ω). (1.10)

Theorem 1.4

Let 1 < p1, …, pm+1 < ∞ and 1/ti = 1/pi + 1/pm+1 for i = 1, …, m with ti > n/(n – 1). Let 1 ≤ α < + 1 with β = min1≤jm{(n – 1)/pj, n – 2n/((n – 1)tj)} and 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n with q > 1. If b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) and |Ω| < ∞, then

Mα,Ω,b:Lp1(Ω)××Lpm(Ω)W1,q(Ω)

is bounded. Precisely, if f⃗ = (f1, …, fm) with each fjLpj(Ω), then

Mα,Ω,b(f)W1,q(Ω)α,m,n,p1,,pm+1,|Ω|(i=1mbiW1,pm+1(Ω))j=1mfjLpj(Ω). (1.11)

Remark 1.3

When m = 1, Theorems 1.1-1.4 implies Theorem C. Thus, the corresponding results in Theorems 1.1-1.4 can be regarded as the multilinear version of Theorem C.

Finally, we shall consider the behaviors of commutators of the local multilinear fractional maximal operators on the Sobolev space with zero boundary values. We denote by W01,p (Ω) with 1 ≤ p < ∞ the Sobolev space with zero boundary values, which is defined as the completion of C0 (Ω) with respect to the Sobolev norm. In 1998, Kinnunen and Lindqvist [14] observed that MΩ : W01,p (Ω) → W01,p (Ω) is bounded for all 1 < p < ∞. This result was later extended to the fractional version by Heikkinen et al. [11]. Recently, Hart et al. [10] established the following results.

Theorem D

([10])

  1. Let f⃗ = (f1, …, fm) with each fj W01,pj (Ω) for some pj ∈ (1, ∞) and 1/q = 1/p1 + ⋯ + 1/pm with 1 < q < ∞. Then 𝔐Ω(f⃗) ∈ W01,q (Ω);

  2. Let f⃗ = (f1, …, fm) with each fjW1,pj(Ω) for some pj ∈ (1, ∞) and 1/q = 1/p1 + ⋯ + 1/pm – (α – 1)/n with 1 < q < ∞. Assume |Ω| < ∞, then 𝔐α,Ω(f⃗) ∈ W01,q (Ω).

  3. Let f⃗ = (f1, …, fm) with each fjLpj(Ω) for some pj ∈ (n/(n – 1), ∞). Let 1 ≤ α < + 1 with β = min1≤jm{(n – 1)/pj, n – 2n/((n – 1)pj)}. If |Ω| < ∞, then 𝔐α,Ω(f⃗) ∈ W01,q (Ω).

Motivated by Theorem D, we shall prove the following results, which are some applications of Theorems 1.1-1.4.

Corollary 1.5

Let 1/q = 1/p1 + … + 1/pm+1 < ∞ where 1 < p1, …, pm+1 < ∞ and 1/pi1/pm+1 < 1 for i = 1, …, m. Let b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω). Then

[b,MΩ]:W01,p1(Ω)××W01,pm(Ω)W01,q(Ω),

is bounded. The same conclusion holds for 𝔐α,Ω,b⃗.

Corollary 1.6

Let 1 < p1, …, pm+1 < ∞ such that 1/pi + 1/pm+1 < 1 for i = 1, …, m. Let b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω). Then

[b,Mα,Ω]:W1,p1(Ω)××W1,pm(Ω)W01,q(Ω),

is bounded, provided that one of the following conditions holds:

  1. 1 ≤ α < mn, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n, q > 1, |Ω| < ∞;

  2. 1 ≤ α < mn, 1 < p1, ..., pm < n, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α + m – 1)/n, 1/m < q < ∞, Ω admits a pj-Sobolev embedding for each j = 1, ..., m.

    The same conclusion holds for 𝔐α,Ω,b⃗.

Corollary 1.7

Let 1 < p1, …, pm+1 < ∞ and 1/ti = 1/pi + 1/pm+1 with ti > n/(n – 1) for i = 1, …, m. Let 1 ≤ α < + 1 with β = min1≤jm{(n – 1)/pj, n – 2n/((n – 1)tj)} and 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n and q > 1. Assume that b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) and |Ω| < ∞, then

[b,Mα,Ω]:Lp1(Ω)××Lpm(Ω)W01,q(Ω)

is bounded. The same conclusion holds for 𝔐α,Ω,b⃗.

The paper is organized as follows. In Section 2, we shall present the proofs of Theorems 1.1 and 1.2. The proofs of Theorems 1.3 and 1.4 will be given in Section 3. Finally, we shall prove Corollaries 1.5-1.7 in Section 4. We would like to remark that the main ideas in our proofs are motivated by [10, 11, 14, 21], but our methods and techniques are more subtle and complex than those of [10, 11, 14, 21]. Particularly, some new techniques will be explored in our main proofs. The main ingredients in the proofs of Theorems 1.1 and 1.2 are a characterization of the product Sobolev spaces in Lemma 2.1 and Theorems A and B. The main ingredients in the proofs of Theorems 1.3 and 1.4 are some pointwise estimates for the weak gradients of the multilinear maximal commutators in Lemmas 3.8-3.11, which are of interest in their own rights.

Throughout this paper, if there exists a constant c > 0 depending only on ϑ such that AcB, we then write Aϑ B or Bϑ A; and if Aϑ Bϑ A, we then write Aϑ B. We use the following conventions ∏j∈∅ aj = 1 and ∑j∈∅ aj = 0.

2 Proofs of Theorems 1.1 and 1.2

This section is devoted to presenting the proofs of Theorems 1.1 and 1.2. Before proving Theorems 1.1 and 1.2, let us introduce the following result, which focuses on a characterization of the product Sobolev spaces in local setting.

Lemma 2.1

([21]) Let 1 < p1, p2, p < ∞ and 1/p = 1/p1 + 1/p2. If fW1,p1(Ω) and gW1,p2(Ω), then fgW1,p(Ω). Moreover,

(fg)=gf+fg,

almost everywhere in Ω. In particular, it holds that

fgW1,p(Ω)fW1,p1(Ω)gW1,p2(Ω).

Now we turn to prove Theorems 1.1 and 1.2.

Proof of Theorem 1.1

To prove (1.8), it suffices to show that

[b,Mα,Ω]i(f)W1,q(Ω)α,m,n,p1,,pm+1biW1,pm+1(Ω)j=1mfjW1,pj(Ω), (2.1)

for all i = 1, 2, …, m. We only work with (2.1) for the case i = 1, since other cases can be proved similarly. We discuss the following cases:

Assume that the condition (a) holds. Let 1/t1 = 1/p1 + 1/pm+1 and 1/p = 1/p1 + ⋯ + 1/pm. It is clear that 1 < t1 < ∞, q < p < p1, 1/q = 1/t1 + 1/p2 + ⋯ + 1/pm and 1/q = 1/p + 1/pm+1. Invoking Lemma 2.1 we have that b1f1W1,t1(Ω). By Theorem A we have that 𝔐Ω(f⃗) ∈ W1,p(Ω) and 𝔐Ω(b1f1, f2, …, fm) ∈ W1,q(Ω). By Theorem A and Lemma 2.1 again,

[b,MΩ]1(f)W1,q(Ω)b1MΩ(f)W1,q(Ω)+MΩ(b1f1,f2,,fm)W1,q(Ω)b1W1,pm+1(Ω)MΩ(f)W1,p(Ω)+Cb1f1W1,t1(Ω)j=2mfjW1,pj(Ω)α,m,n,p1,,pm+1b1W1,pm+1(Ω)j=1mfjW1,pj(Ω), (2.2)

which yields (2.1) for the case i = 1.

Assume that the condition (b) holds. Let 1/t1 = 1/p1 + 1/pm+1 and 1/p = 1/p1 + ⋯ + 1/pm – (α – 1)/n. Clearly, 1 < t1 < ∞, 1/q = 1/t1 + 1/p2 + ⋯ + 1/pm – (α – 1)/n and 1/q = 1/p + 1/pm+1. By Lemma 2.1 and Theorem A we have that b1f1W1,t1(Ω), 𝔐α,Ω(f⃗) ∈ W1,p(Ω) and 𝔐α,Ω(b1f1, f2, …, fm) ∈ W1,q(Ω). The arguments similar to those used to derive (2.2) will give that

[b,Mα,Ω]1(f)W1,q(Ω)α,m,n,p1,,pm+1b1W1,pm+1(Ω)j=1mfjW1,pj(Ω).

This proves (2.1) for the case i = 1 under the condition (b).

Assume that the condition (c) holds. Let p be such that 1/p = 1/p1 + ⋯ + 1/pm – (α + m – 1)/n and pi~ , ti, ti~ be such that 1/pi = 1/pi – 1/n, 1/ti = 1/pi + 1/pm+1 and 1/ ti~ = 1/ pi~ + 1/pm+1 for i = 1, …, m. Then we have 1/p = 1/ p1~ + ⋯ + 1/ pm~ – (α – 1)/n and 1/q = 1/ p1~ + ⋯ + 1/ pm~ + 1/pm+1 – (α – 1)/ = 1/t1 + 1/ p2~ + ⋯ + 1/ pm~ α/n = 1/ t1~ + 1/ p2~ + ⋯ + 1/ pm~ – (α – 1)/n = 1/p1 + 1/ p2~ + ⋯ + 1/ pm~ + 1/pm+1α/n. Let be such that 1/ = 1/p1 + 1/ p2~ + ⋯ + 1/ pm~ α/n. Clearly, 1/q = 1/pm+1 + 1/ and 1/ = 1/ p1~ + ⋯ + 1/ pm~ – (α – 1)/n = 1/ p1~ + ⋯ + 1/ pi1~ + 1/pi + 1/ pi+1~ + ⋯ + 1/ pm~ α/n for all i = 1, …, m. Since Ω admits a pj-Sobolev embedding for all j = 1, …, m, then for each j = 1, …, m,

fjLpj~(Ω)pj,nfjW1,pj(Ω). (2.3)

By Remark 1.2 and (2.3), we have

[b,Mα,Ω]1(f)Lq(Ω)α,m,n,p1,,pm+1b1Lpm+1(Ω)f1Lp1(Ω)j=2mfjLpj~(Ω)α,m,n,p1,,pm+1b1W1,pm+1(Ω)j=1mfjW1,pj(Ω). (2.4)

On the other hand, by Lemma 2.1 and Theorem A, we have

|(b1Mα,Ω(f))(x)||b1(x)|Mα,Ω(f)(x)+|b1(x)||Mα,Ω(f)(x)||b1(x)|Mα,Ω(f)(x)+|b1(x)|(αMα1,Ω(f)(x)+2l=1mMα,Ω(fl)(x)),

for almost every xΩ, where f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm). This together with (2.3), (1.3), Hölder’s inequality and Minkowski’s inequality implies

(b1Mα,Ω(f))Lq(Ω)|b1|Mα,Ω(f)Lq(Ω)+b1(αMα1,Ω(f)+2l=1mMα,Ω(fl))Lq(Ω)b1Lpm+1(Ω)Mα,Ω(f)Lp~(Ω)+b1Lpm+1(Ω)(αMα1,Ω(f)Lp~(Ω)+2l=1mMα,Ω(fl)Lp~(Ω))α,m,n,p1,,pmb1Lpm+1(Ω)f1Lp1(Ω)i=2mfiLpi~(Ω)+b1Lpm+1(Ω)(j=1mfjLpj~(Ω)+l=1mflLpl(Ω)1jmjlfjLpj~(Ω))α,m,n,p1,,pmb1W1,pm+1(Ω)j=1mfjW1,pj(Ω). (2.5)

By Theorem A, we get

|Mα,Ω(b1f1,f2,,fm)(x)|αMα1,Ω(b1f1,f2,,fm)(x)+2Mα,Ω(|(b1f1)|,f2,,fm)(x)+2l=2mMα,Ω(b1f1,f2,,fl1,|fl|,fl+1,,fm)(x),

for almost every xΩ, which together with (2.3), (1.3), Hölder’s inequality and Minkowski’s inequality implies

Mα,Ω(b1f1,f2,,fm)Lq(Ω)αMα1,Ω(b1f1,f2,,fm)Lq(Ω)+2Mα,Ω(|(b1f1)|,f2,,fm)Lq(Ω)+2l=2mMα,Ω(b1f1,f2,,fl1,|fl|,fl+1,,fm)Lq(Ω)α,m,n,p1,,pm+1b1f1Lt1~(Ω)j=2mfjLpj~(Ω)+(b1f1)Lt1(Ω)j=2mfjLpj~(Ω)+l=2mb1f1Lt1~(Ω)flLpl(Ω)2jmjlfjLpj~(Ω)α,m,n,p1,,pm+1b1W1,pm+1(Ω)j=1mfjW1,pj(Ω). (2.6)

It follows from (2.5) and (2.6) that

[b,Mα,Ω]1(f)Lq(Ω)α,m,n,p1,,pm+1b1W1,pm+1(Ω)j=1mfjW1,pj(Ω). (2.7)

Combining (2.7) with (2.4) yields (2.1) for i = 1.□

Proof of Theorem 1.2

To prove (1.9), it suffices to show that

[b,Mα,Ω]i(f)W1,q(Ω)α,m,n,p1,,pm+1biW1,pm+1(Ω)j=1mfjLpj(Ω), (2.8)

for all i = 1, …, m. We only work with (2.8) for the case i = 1, since other cases can be proved similarly. Let

β1=min1jm{(n1)/pj,n2n/((n1)pj)},
β2=max{min2jm{(n1)/pj,n2n/((n1)pj)},(n1)/t1,n2n/((n1)t1)}.

By our assumption, we have that 1 ≤ α < m min{β1, β2} + 1. Let p be such that 1/p = 1/p1 + ⋯ + 1/pm – (α – 1)/n. Then q < p and 1/q = 1/pm+1 + 1/p = 1/t1 + 1/p2 + ⋯ + 1/pm – (α – 1)/n. Hence, 𝔐α,Ω(f⃗) ∈ W1,p(Ω) and 𝔐α,Ω(b1f1, f2, …, fm) ∈ W1,q(Ω) by Theorem B. By Theorem B, Lemma 2.1 and Hölder’s inequality, we have

[b,Mα,Ω]1(f)W1,q(Ω)b1Mα,Ω(f)W1,q(Ω)+Mα,Ω(b1f1,f2,,fm)W1,q(Ω)b1W1,pm+1(Ω)Mα,Ω(f)W1,p(Ω)+Cb1f1Lt1(Ω)j=2mfjLpj(Ω)α,m,n,p1,,pm+1b1W1,pm+1(Ω)j=1mfjLpj(Ω).

This finishes the proof of Theorem 1.2.□

3 Proofs of Theorems 1.3 and 1.4

3.1 Preliminary notation and lemmas

Let δ(x) = dist(x, Ωc). It is well known that δ is a Lipschitz function and |∇δ(x)| = 1 for almost every xΩ. Let t ∈ (0, 1), α ∈ [0, mn) and i = 1, …, m, we define the average function Λα,t,bi(f⃗) : Ω → [–∞, ∞] by

Λα,t,bi(f)(x)=(tδ(x))α|B(x,tδ(x))|m(B(x,tδ(x)))m|bi(x)bi(yi)|j=1mfj(yj)dy1dy2dym.

Particularly, we denote Λα,t,bi(f⃗) = Λt,bi(f⃗) when α = 0. For two arbitrary functions f, g defined on Ω, we set

Af(x)=B(x,tδ(x))f(y)dy,Agf(x)=B(x,tδ(x))|g(x)g(y)|f(y)dy.

Then Λα,t,bi(f⃗) can be rewritten as

Λα,t,bi(f)(x)=(tδ(x))α|B(x,tδ(x))|mAbifi(x)1jmjiAfj(x). (3.1)

By the definition of Mα,Ω,bi (f⃗), the function Mα,Ω,bi (f⃗) can be rewritten as

Mα,Ω,bi(f)(x)=sup0<t<1Λα,t,bi(f)(x), (3.2)

The following lemmas will provide a foundation for our analysis.

Proposition 3.1

([11], [14]) Let 1 ≤ p ≤ ∞. If fkf, gkg weakly in Lp(Ω) and fkgk (k = 1, 2, …) almost everywhere in Ω, then fg almost everywhere in Ω.

Lemma 3.2

([11]) Let n ≥ 2, p > n/(n – 1), q = np/(nαp), 0 ≤ α < min{(n – 1)/p, n – 2n/((n – 1)p)}. Then

Sα,Ωfq,Ωα,n,pfp,Ω.

The following is an easy observation, which is useful for our proofs.

Lemma 3.3

Let {Ti}i=1m be a finite sequence of operators, where all Ti are bounded on Lpi(Ω) for some pi ∈ (1, ∞). Let 1/q = 1/p1 + ⋯ + 1/pm and define the operator T by

T(f)=j=1mTjfj,forf=(f1,,fm).

Then T is bounded from Lp1(Ω) × ⋯ × Lpm(Ω) to Lq(Ω). If each Ti is also sublinear, then T is continuous from Lp1(Ω) × ⋯ × Lpm(Ω) to Lq(Ω).

Proof

Let f⃗ = (f1, …, fm) with each fjLpj(Ω). By Hölder’s inequality, one finds

T(f)Lq(Ω)j=1mTjfjLpj(Ω)Cj=1mfjLpj(Ω),

which gives the boundedness for T. Let f⃗k = (f1,k, …, fm,k) with each fj,kfj in Lpj(Ω) as k → ∞ for all j = 1, 2, …, m. One can easily check that

|T(fk)T(f)|l=1m|Tlfl,kTlfl|μ=1l1|Tμfμ|ν=l+1m|Tνfν,k|.

This together with Hölder’s inequality and Minkowski’s inequality implies that

T(fk)T(f)Lq(Ω)l=1m|Tlfl,kTlfl|μ=1l1|Tμfμ|ν=l+1m|Tνfν,k|Lq(Ω)l=1mTl(fl,kfl)Lpl(Ω)μ=1l1TμfμLpμ(Ω)ν=l+1mTνfν,kLpν(Ω)Cl=1mfl,kflLpl(Ω)μ=1l1fμLpμ(Ω)ν=l+1m(fν,kfνLpν(Ω)+fνLpν(Ω)),

which leads to T(f⃗k) → T(f⃗) in Lq(Ω) as k → ∞.□

3.2 Gradient estimates for average functions

In this subsection we shall present some pointwise gradient estimates for average functions. We start with the following lemma.

Lemma 3.4

Let 1 ≤ α < mn and |Ω| < ∞. Let 1 < p1, ⋯, pm+1 < ∞, 1/pi + 1/pm+1 < 1 for i = 1, …, m, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n with q > 1. If f⃗ = (f1, …, fm) and b⃗ = (b1, …, bm) with each fjW1,pj(Ω) and bjW1,pm+1(Ω), then for any i = 1, …, m, we have Λα,t,bi(f⃗) ∈ W1,q(Ω). Moreover,

|Λα,t,bi(f)(x)|αMα1,Ω,bi(f)(x)+2l=1mMα,Ω,bi(fl)(x)+|bi|(x)Mα,Ω(f)(x)+2Mα,Ω(f1,,fi1,|bi|fi,fi+1,,fm)(x), (3.3)

for almost every xΩ. Here f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm).

Proof

Without loss of generality we only prove (3.3) for the case i = 1 since other cases are analogous.

First of all, we shall prove the following claim:

  1. Let b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) ∩ 𝓒(Ω) and f⃗ = (f1, …, fm) with each fjW1,pj(Ω) ∩ 𝓒(Ω). Then we have

    |Λα,t,b1(f)(x)|αMα1,Ω,b1(f)(x)+2l=1mMα,Ω,b1(fl)(x)+|b1|(x)Mα,Ω(f)(x)+2Mα,Ω(|b1|f1,f2,,fm)(x), (3.4)

    for almost every xΩ, where f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm).

    For convenience, for any arbitrary function F defined on Ω × Ω, we denote by Di,xF the i-th weak parital derivative of F in x. Let ∇x = (D1,x, …, Dn,x) and ϖn = |B(0, 1)|. By (3.1) and the Leibnitz rule, we have

    DiΛα,t,b1(f)(x)=Di((tδ(x))α|B(x,tδ(x))|m)Ab1f1(x)j=2mAfj(x)+(tδ(x))α|B(x,tδ(x))|mDi,xAb1f1(x)j=2mAfj(x)+(tδ(x))α|B(x,tδ(x))|mAb1f1(x)l=2mDi,xAfl(x)2jmjlAfj(x), (3.5)

    for almost every xΩ. Fix 1 ≤ l ≤ m. Using the chain rule we have

    DiAfl(x)=B(x,tδ(x))Difl(y)dy+tDiδ(x)B(x,tδ(x))fl(y)dHn1(y), (3.6)
    Di,xAb1f1(x)=B(x,tδ(x))Di,x|b1(x)b1(y)|f1(y)dy+B(x,tδ(x))Di,y(|b1(x)b1(y)|f1(y))dy+tDiδ(x)B(x,tδ(x))|b1(x)b1(y)|f1(y)dHn1(y), (3.7)

    for almost every xΩ, where we have used the fact rB(x,r)fl(y)dy=B(x,r)fl(y)dHn1(y). Then we get from (3.5)-(3.7) that

    DiΛα,t,b1(f)(x)=(αmn)Diδ(x)δ(x)((tδ(x))α|B(x,tδ(x))|m)Ab1f1(x)j=2mAfj(x)+(tδ(x))α|B(x,tδ(x))|mj=2mAfj(x)(B(x,tδ(x))Di,x|b1(x)b1(y)|f1(y)dy+B(x,tδ(x))Di,y(|b1(x)b1(y)|f1(y))dy+tDiδ(x)B(x,tδ(x))|b1(x)b1(y)|f1(y)dHn1(y))+(tδ(x))α|B(x,tδ(x))|mAb1f1(x)l=2m2jmjlAfj(x)(B(x,tδ(x))Difl(y)dy+tDiδ(x)B(x,tδ(x))fl(y)dHn1(y)), (3.8)

    for almost every xΩ. It follows from (3.8) that

    Λα,t,b1(f)(x)=αδ(x)δ(x)(tδ(x))α|B(x,tδ(x))|mAb1f1(x)j=2mAfj(x)+(tδ(x))α|B(x,tδ(x))|mj=2mAfj(x)(B(x,tδ(x))x|b1(x)b1(y)|f1(y)dy+B(x,tδ(x))y(|b1(x)b1(y)|f1(y))dy)+nδ(x)δ(x)(tδ(x))α|B(x,tδ(x))|m1j=2mAfj(x)×(1|B(x,tδ(x))|B(x,tδ(x))|b1(x)b1(y)|f1(y)dHn1(y)Ab1f1(x)|B(x,tδ(x))|)+nδ(x)δ(x)(tδ(x))α|B(x,tδ(x))|m1l=2mAb1f1(x)2jmjlAfj(x)×(1|B(x,tδ(x))|B(x,tδ(x))fl(y)dHn1(y)Afl(x)|B(x,tδ(x))|)+(tδ(x))α|B(x,tδ(x))|mAb1f1(x)l=2m2jmjlAfj(x)B(x,tδ(x))fl(y)dy, (3.9)

    for almost every xΩ.

    Fix xΩ. Let R > 0 such that B(x, R) ⊂ Ω and F(x, y) be a function defined on Ω × Ω. It was shown in [21, (5.7)] that

    1|B(x,R)|B(x,R)F(x,y)dHn1(y)1|B(x,R)|B(x,R)F(x,y)dy=1n1|B(x,R)|B(x,R)yF(x,y)(yx)dy. (3.10)

    By (3.10) with R = (x) and F(x, y) = |b1(x) – b1(y)|f1(y) or F(x, y) = fl(y), we obtain

    1|B(x,tδ(x))|B(x,tδ(x))|b1(x)b1(y)|f1(y)dHn1(y)1|B(x,tδ(x))|B(x,tδ(x))|b1(x)b1(y)|f1(y)dy=1n1|B(x,tδ(x))|B(x,tδ(x))y(|b1(x)b1(y)|f1(y))(yx)dy, (3.11)
    1|B(x,tδ(x))|B(x,tδ(x))fl(y)dHn1(y)1|B(x,tδ(x))|B(x,tδ(x))fl(y)dy=1n1|B(x,tδ(x))|B(x,tδ(x))fl(y)(yx)dy. (3.12)

    It follows from (3.8), (3.11) and (3.12) that

    Λα,t,b1(f)(x)=αδ(x)δ(x)(tδ(x))α|B(x,tδ(x))|mAb1f1(x)j=2mAfj(x)+(tδ(x))α|B(x,tδ(x))|mj=2mAfj(x)(B(x,tδ(x))x|b1(x)b1(y)|f1(y)dy+B(x,tδ(x))y(|b1(x)b1(y)|f1(y))dy)+δ(x)δ(x)(tδ(x))α|B(x,tδ(x))|mj=2mAfj(x)B(x,tδ(x))y(|b1(x)b1(y)|f1(y))(yx)dy+δ(x)δ(x)(tδ(x))α|B(x,tδ(x))|ml=2mAb1f1(x)2jmjlAfj(x)B(x,tδ(x))fl(y)(yx)dy+(tδ(x))α|B(x,tδ(x))|mAb1f1(x)l=2m2jmjlAfj(x)B(x,tδ(x))fl(y)dy, (3.13)

    for almost every xΩ.

    Since |Ω| < ∞, thus we have b1(x) – b1(⋅) ∈ W1,p2(Ω) and |b1(x) – b1(⋅)| ∈ W1,pm+1(Ω). For i = 1, …, m, let ti be such that 1/ti = 1/pi + 1/pm+1. Clearly, 1 < ti < ∞. By Lemma 2.1, we have that |b1(x) – b1(⋅)|f1(⋅) ∈ W1,t1(Ω) and

    y(|b1(x)b1(y)|f1(y))=f1(y)y|b1(x)b1(y)|+|b1(x)b1(y)|f1(y),

    for almost every yΩ. Moreover, |∇y|b1(x) – b1(y)|| = |∇b1(y)| for almost every yΩ and |∇x|b1(x) – b1(y)|| = |∇b1(x)| for almost every xΩ. These facts together with (3.13) lead to (3.4) and completes the proof of Claim 1.

    Next we proceed with the proof via an approximation argument. Assume that b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) and f⃗ = (f1, …, fm) with each fjW1,pj(Ω). We can find some sequences of functions {φi,j}j=1 in W1,pm+1(Ω) ∩ 𝓒(Ω) and {ψi,j}j=1 in W1,pi(Ω) ∩ 𝓒(Ω) such that φi,jbi in W1,pm+1(Ω) and ψi,jfi in W1,pi(Ω) as j → ∞. Let φ⃗j = (φ1,j, …, φm,j) and ψ⃗j = (ψ1,j, …, ψm,j). By Riesz theorem there exists a subsequence {φ1,jk}k=1{φ1,j}j=1 such that φ1,jk(x) → b1(x) as k → ∞ for almost every xΩ. We shall prove the following claim:

  2. It holds that Λα,t,φ1,jk (ψ⃗jk)(x) → Λα,t,b1(f⃗)(x) as k → ∞ for almost every xΩ.

    Note that

    |Λα,t,φ1,j(ψj)(x)Λα,t,b1(f)(x)|=(tδ(x))α|B(x,tδ(x))|m|Aφ1,jψ1,j(x)j=2mAψi,j(x)Ab1f1(x)j=2mAfj(x)|(tδ(x))α|B(x,tδ(x))|m|Aφ1,jψ1,j(x)Ab1f1(x)|j=2mA|ψi,j|(x)+(tδ(x))α|B(x,tδ(x))|mAb1|f1|(x)|j=2mAψi,j(x)j=2mAfj(x)|(tδ(x))α|B(x,tδ(x))|m|Aφ1,jψ1,j(x)Ab1f1(x)|j=2A|ψi,j|(x)+(tδ(x))α|B(x,tδ(x))|mAb1|f1|(x)l=2mA|ψl,jfl|(x)μ=2l1A|fμ|(x)ν=l+1mA|ψν,j|(x). (3.14)

    Since ψi,jfi in Lpi(Ω) as j → ∞ for all i = 1, 2, …, m, there exists a constant C > 0 such that

    max1imψi,jLpi(Ω)<C. (3.15)

    By Hölder’s inequality, we have

    A|g|(x)gLr(Ω)|B(x,tδ(x))|11/r (3.16)

    for all gLr(Ω) with some r ∈ (1, ∞). In light of (3.15) and (3.16) we would have

    A|ψi,j|(x)ψi,jLpi(Ω)|B(x,tδ(x))|11/pi<C|B(x,tδ(x))|11/pi,
    A|fi|(x)fiLpi(Ω)|B(x,tδ(x))|11/pi,A|ψl,jfl|(x)ψl,jflLpl(Ω)|B(x,tδ(x))|11/pl0asj,
    Ab1|f1|(x)|b1(x)|A|f1|(x)+A(|b1f1|)(x)|b1(x)|f1Lp1(Ω)|B(x,tδ(x))|11/p1+b1f1Lt1(Ω)|B(x,tδ(x))|11/t1f1Lp1(Ω)(|b1(x)||B(x,tδ(x))|11/p1+b1Lpm+1(Ω)|B(x,tδ(x))|11/t1)<,
    |Aφ1,jkψ1,jk(x)Ab1f1(x)|B(x,tδ(x))||φ1,jk(x)φ1,jk(y)|ψ1,jk(y)|b1(x)b1(y)|f1(y)|dyB(x,tδ(x))||φ1,jk(x)φ1,jk(y)||b1(x)b1(y)|||ψ1,jk(y)|dy+Ab1|ψ1,jkf1|(x)|φ1,jk(x)b1(x)|A|ψ1,jk|(x)+A(|(φ1,jkb1)ψ1,jk|)(x)+Ab1|ψ1,jkf1|(x)|φ1,jk(x)b1(x)||B(x,tδ(x))|11/p1ψ1,jkLp1(Ω)+|B(x,tδ(x))|11/t1(φ1,jkb1)ψ1,jkLt1(Ω)+|b1(x)||B(x,tδ(x))|11/p1ψ1,jkf1Lp1(Ω)+|B(x,tδ(x))|11/t1b1(ψ1,jkf1)Lt1(Ω)|B(x,tδ(x))|11/p1(|φ1,jk(x)b1(x)|ψ1,jkLp1(Ω)+|b1(x)|ψ1,jkf1Lp1(Ω))+|B(x,tδ(x))|11/t1(φ1,jkb1Lpm+1(Ω)ψ1,jkLp1(Ω)+b1Lpm+1(Ω)(ψ1,jkf1)Lp1(Ω))0ask

    for almost every xΩ. Above facts together with (3.14) will imply Claim 2.

    Now we shall prove the following claim:

  3. There exists a subsequence {Λα,t,φ1,jk(ψjk)}=1of{Λα,t,φ1,jk(ψjk)}k=1 such that ∇Λα,t,φ1,jk(ψ⃗jk) → ∇Λα,t,b1(f⃗) weakly in Lq(Ω) as k → ∞.

    By Claim 1, one sees that

    |Λα,t,φ1,jk(ψjk)(x)|αMα1,Ω,φjk1(ψjk)(x)+2l=1mMα,Ω,φjk1(ψjk,l)(x)+|φ1,jk|(x)Mα,Ω(ψjk)(x)+2Mα,Ω(|φ1,jk|ψ1,jk,ψ2,jk,,ψm,jk)(x), (3.17)

    for almost every xΩ, where ψ⃗jk,l = (ψ1,jk, …, ψl–1,jk, |∇ψl,jk|, ψl+1,jk, …, ψm,jk). Let 1/q* = 1/p1 + ⋯ + 1/pm+1α/n and 1/p* = 1/p1 + ⋯ + 1/pmα/n. Then q < q* < p* and 1/q* = 1/p* + 1/pm+1 = 1/t1 + 1/p2 + ⋯ + 1/pmα/n. By Remark 1.2 and Hölder’s inequality, we have

    Mα1,Ω,φjk1(ψjk)Lq(Ω)α,m,n,p1,,pm+1φ1,jkLpm+1(Ω)i=1mψi,jkLpi(Ω), (3.18)
    Mα,Ω,φjk1(ψjk,l)Lq(Ω)|Ω|1/q1/qMα,Ω,φjk1(ψjk,l)Lq(Ω)α,m,n,p1,,pm+1,|Ω|φ1,jkLpm+1(Ω)ψl,jkLpl(Ω)1imilψi,jkLpi(Ω)α,m,n,p1,,pm+1,|Ω|φ1,jkLpm+1(Ω)i=1mψi,jkW1,pi(Ω). (3.19)

    We also get by (1.3) and Hölder’s inequality that

    |φ1,jk|Mα,Ω(ψjk)Lq(Ω)|Ω|1/q1/q|φ1,jk|Mα,Ω(ψjk)Lq(Ω)|Ω|1/q1/qφ1,jkLpm+1(Ω)Mα,Ω(ψjk)Lp(Ω)α,m,n,p1,,pm+1,|Ω|φ1,jkW1,pm+1(Ω)i=1mψi,jkLpi(Ω), (3.20)
    Mα,Ω(|φ1,jk|ψ1,jk,ψ2,jk,,ψm,jk)Lq(Ω)|Ω|1/q1/qMα,Ω(|φ1,jk|ψ1,jk,ψ2,jk,,ψm,jk)Lq(Ω)α,m,n,p1,,pm+1,|Ω||φ1,jk|ψ1,jkLt1(Ω)i=2mψi,jkLpi(Ω)α,m,n,p1,,pm+1,|Ω|φ1,jkW1,pm+1(Ω)i=1mψi,jkLpi(Ω). (3.21)

    By (3.17)-(3.21) and Minkowski’s inequality we get

    Λα,t,φ1,jk(ψjk)Lq(Ω)α,m,n,p1,,pm+1,|Ω|φ1,jkW1,pm+1(Ω)i=1mψi,jkW1,pi(Ω),

    which gives that {|Λα,t,φ1,jk(ψjk)|}k=1 is a bounded sequence in Lq(Ω). This together with the Claim 2 implies that the Sobolev derivative ∇Λα,t,φ1,jk(ψ⃗jk) exists almost everywhere in Ω and Claim 3 was proved.

    For convenience, we set

    Fjk(x):=αMα1,Ω,φjk1(ψjk)(x)+2l=1mMα,Ω,φjk1(ψjk,l)(x)+|φ1,jk|(x)Mα,Ω(ψjk)(x)+2Mα,Ω(|φ1,jk|ψ1,jk,ψ2,jk,,ψm,jk)(x),
    G(x):=αMα1,Ω,b1(f)(x)+2l=1mMα,Ω,b1(fl)(x)+|b1|(x)Mα,Ω(f)(x)+2Mα,Ω(|b1|f1,f2,,fm)(x).

    Then we want to prove the following claim:

  4. FjkG in Lq(Ω) as k → ∞.

    It was shown in the proof of Claim 2 that

    |Aφ1,jkψ1,jk(x)Ab1f1(x)||φ1,jk(x)b1(x)|A|ψ1,jk|(x)+A(|(φ1,jkb1)ψ1,jk|)(x)+|b1(x)|A|ψ1,jkf1|(x)+A|b1(ψ1,jkf1)|(x),

which together with (3.2) and (3.14) leads to

|Mα1,Ω,φjk1(ψjk)Mα1,Ω,b1(f)|sup0<t<1|Λα1,t,φ1,jk(ψjk)(x)Λα1,t,b1(f)(x)||φ1,jk(x)b1(x)|Mα1,Ω(ψjk)(x)+Mα1,Ω(|φ1,jk(x)b1(x)|ψ1,jk,ψ2,jk,,ψm,jk)(x)+|b1(x)|Mα1,Ω(|ψ1,jkf1|,ψ2,jk,,ψm,jk)(x)+Mα1,Ω(b1|ψ1,jkf1|,ψ2,jk,,ψm,jk)(x)+l=2mMα1,Ω,b1(Jl)(x), (3.22)

where J⃗l = (f1, …, fl–1, |ψl,jkfl|, ψl+1,jk, …, ψm,jk). By (3.22) and the arguments similar to those used to derive (3.18)-(3.21), we have

Mα1,Ω,φjk1(ψjk,l)Mα1,Ω,b1(fl)Lq(Ω)α,m,n,p1,,pm+1φ1,jkb1W1,pm+1(Ω)i=1mψi,jkLpi(Ω)+b1W1,pm+1(Ω)l=1mψl,jkflLpl(Ω)μ=1l1fμLpμ(Ω)ν=l+1mψν,jkLpν(Ω)0ask. (3.23)

Similarly, a refine analysis will give

Mα,Ω,φjk1(ψjk,l)Mα,Ω,b1(fl)Lq(Ω)0ask,forl=1,,m. (3.24)

It is clear that |∇φ1,jk|ψ1,jk → |∇b1|f1 in Lt1(Ω) as k → ∞ by Hölder’s inequality. This together with (1.4) implies that

Mα,Ω(|φ1,jk|ψ1,jk,ψ2,jk,,ψm,jk)Mα,Ω(|b1|f1,f2,,fm)inLq(Ω)ask. (3.25)

We also get by (1.4) and Hölder’s inequality that

|φ1,jk|Mα,Ω(ψjk)|b1|Mα,Ω(f)inLq(Ω)ask. (3.26)

Combining (3.26) with (3.23)-(3.25) and Minkowski’s inequality implies Claim 4.

By (3.17) we have that

|Λα,t,φ1,jk(ψjk)(x)|Fjk(x), (3.27)

for almost every xΩ. Applying Proposition 3.1 to (3.27), we get from Claim 3 and Claim 4 that (3.3) holds for i = 1. This completes the proof of Lemma 3.4.□

By the arguments similar to those used in deriving Lemma 3.4, we can get the following lemma. The details for the proof are omitted.

Lemma 3.5

Let 1 < p1, ⋯, pm+1 < ∞, 1/q = 1/p1 + ⋯ + 1/pm+1 with q > 1. If f⃗ = (f1, …, fm) and b⃗ = (b1, …, bm) with each fjW1,pj(Ω) and bjW1,pm+1(Ω), then for any i = 1, …, m, we have Λt,bi(f⃗) ∈ W1,q(Ω) and

|Λt,bi(f)(x)||bi|(x)MΩ(f)+2l=1mMΩ,bi(fl)(x)+2MΩ(f1,,fi1,|bi|fi,fi+1,,fm)(x),

for almost every xΩ, where f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm).

It should be pointed out that the condition |Ω| < ∞ is necessary for the validity of Lemma 3.4. Actually, the condition |Ω| < ∞ can be replaced by the condition that Ω admits a pj-Sobolev embedding for each j = 1, ..., m.

Lemma 3.6

Let 1 ≤ α < mn, 1 < p1, ..., pm+1 < n, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α + m – 1)/n, 1/m < q < ∞. Assume that Ω admits a pj-Sobolev embedding for each j = 1, ..., m. If f⃗ = (f1, …, fm) and b⃗ = (b1, …, bm) with each fjW1,pj(Ω) and bjW1,pm+1(Ω), then for any i = 1, …, m, we have Λα,t,bi(f⃗) ∈ W1,q(Ω) and

|Λα,t,bi(f)(x)|αMα1,Ω,bi(f)(x)+2l=1mMα,Ω,bi(fl)(x)+|bi|(x)Mα,Ω(f)+2Mα,Ω(f1,,fi1,|bi|fi,fi+1,,fm)(x), (3.28)

for almost every xΩ. Here f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm).

Proof

We first prove (3.28) for the case i = 1 since other cases are analogous. Assume that b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) and f⃗ = (f1, …, fm) with each fjW1,pj(Ω). We can find some sequences of functions {φi,j}j=1 in W1,pm+1(Ω) ∩ 𝓒(Ω) and {ψi,j}j=1 in W1,pi(Ω) ∩ 𝓒(Ω) such that φi,jbi in W1,pm+1(Ω) and ψi,jfi in W1,pi(Ω) as j → ∞. Let φ⃗j = (φ1,j, …, φm,j) and ψ⃗j = (ψ1,j, …, ψm,j). By Riesz theorem, there exists a subsequence {φ1,jk}k=1{φ1,j}j=1 such that φ1,jk(x) → b1(x) as k → ∞ for almost every xΩ. By the arguments similar to those used to the proof of Claim 2 of Lemma 3.4, we have

limkΛα,t,φ1,jk(ψjk)(x)=Λα,t,b1(f)(x), (3.29)

for almost every xΩ. Same arguments to those in deriving (3.17) will give

|Λα,t,φ1,jk(ψjk)(x)|αMα1,Ω,φjk1(ψjk)(x)+2l=1mMα,Ω,φjk1(ψjk,l)(x)+|φ1,jk|(x)Mα,Ω(ψjk)(x)+2Mα,Ω(|φ1,jk|ψ1,jk,ψ2,jk,,ψm,jk)(x), (3.30)

for almost every xΩ, where ψ⃗jk,l = (ψ1,jk, …, ψl–1,jk, |∇ψl,jk|, ψl+1,jk, …, ψm,jk).

Let p be such that 1/p = 1/p1 + ⋯ + 1/pm – (α + m – 1)/n and pi~ be such that 1/ pi~ = 1/pi – 1/n for i = 1, 2, …, m. Then we have 1/p = 1/ p1~ + ⋯ + 1/ pm~ – (α – 1)/n and 1/q = 1/ p1~ + ⋯ + 1/ pm~ + 1/pm+1 – (α – 1)/n. Let ti be such that 1/ti = 1/pi + 1/pm+1. It is clear that 1/q = 1/t1 + 1/ p2~ + ⋯ + 1/ pm~ α/n = 1/p1 + 1/ p2~ + ⋯ + 1/ pm~ + 1/pm+1α/n. Let be such that 1/ = 1/p1 + 1/ p2~ + ⋯ + 1/ pm~ α/n. It is clear that 1/q = 1/pm+1 + 1/. By (1.3), Remark 1.2, Hölder’s inequality, Minkowski’s inequality, (3.30) and (2.3), we have

Λα,t,φ1,jk(ψjk)Lq(Ω)αMα1,Ω,φjk1(ψjk)Lq(Ω)+|φ1,jk|Mα,Ω(ψjk)Lq(Ω)+2Mα,Ω(|φ1,jk|ψ1,jk,ψ2,jk,,ψm,jk)Lq(Ω)+2l=1mMα,Ω,φjk1(ψjk,l)Lq(Ω)α,m,n,p1,,pm+1φ1,jkLpm+1(Ω)i=1mψi,jkLpi~(Ω)+φ1,jkLpm+1Mα,Ω(ψjk)Lp~(Ω)+ψ1,jk|φ1,jk|Lt1(Ω)i=2mψi,jkLpi~(Ω)+φ1,jkLpm+1(Ω)l=1mψl,jkLpl(Ω)1imilψi,jkLpi~(Ω)α,m,n,p1,,pm+1φ1,jkW1,pm+1(Ω)i=1mψi,jkW1,pi(Ω).

This tells us that {|Λα,t,φ1,jk(ψjk)|}k=1 is a bounded sequence in Lq(Ω), which together with (3.29) implies that the Sobolev derivative ∇Λα,t,φ1,jk(ψ⃗jk) exists almost everywhere in Ω and there exists a subsequence {Λα,t,φ1,jk(ψjk)}=1of{Λα,t,φ1,jk(ψjk)}k=1 such that

|Λα,t,φ1,jk(ψjk)||Λα,t,b1(f)|weakly inLq(Ω)as. (3.31)

Let Fjk and G be given as in the proof of Lemma 3.4. By (2.3), we have that ψi,jkfi in Lpi~ (Ω) as k → ∞ for all i = 1, …, m. Clearly, |∇φ1,jk|ψ1,jk → |∇b1|f1 in Lt1(Ω) as k → ∞ by Hölder’s inequality. These together with (1.4) imply that

Mα,Ω(ψjk)Mα,Ω(f)Lp~(Ω)0ask, (3.32)
Mα,Ω(|φ1,jk|ψ1,jk,ψ2,jk,,ψm,jk)Mα,Ω(|b1|f1,f2,,fm)Lq(Ω)0ask. (3.33)

By (3.32) and Hölder’s inequality, one can get

|φ1,jk|Mα,Ω(ψjk)|b1|Mα,Ω(f)Lq(Ω)0ask. (3.34)

Similar arguments to those in deriving (3.23) and (3.24) will imply

Mα1,Ω,φjk1(ψjk)Mα1,Ω,b1(f)Lq(Ω)0ask, (3.35)
Mα,Ω,φjk1(ψjk,l)Mα,Ω,b1(fl)Lq(Ω)0ask,fori=1,,m. (3.36)

Then we get from (3.33)-(3.36) that

FjkGLq(Ω)0ask. (3.37)

By (3.30) we have

|Λα,t,φ1,jk(ψjk)(x)|Fjk(x), (3.38)

for almost every xΩ. By (3.31), (3.37) and applying Proposition 3.1 to (3.38), we have that (3.28) holds for i = 1. This completes the proof of Lemma 3.6.□

Lemma 3.7

Let 1 < p1, …, pm+1 < ∞ and 1/ti = 1/pi + 1/pm+1 for i = 1, …, m with ti > 1. Let f⃗ = (f1, …, fm) with each fjLpj(Ω) for ti > n/(n – 1) and b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω). Let 1 ≤ α < + 1 with β = min1≤jm{(n – 1)/pj, n – 2n/((n – 1)tj)}. Let 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n and q > 1. Then for any i = 1, 2, …, m, we have that |∇Λα,t,bi(f⃗)| ∈ Lq(Ω) and

|Λα,t,bi(f)(x)|(mnα)Mα1,Ω,bi(f)(x)+|bi(x)|Mα,Ω(f)(x)+2n(|bi(x)|Sα¯,Ω(fi)(x)+Sα¯,Ω(bifi)(x))1jmjiMα¯,Ω(fj)(x)+2n(|bi(x)|Mα¯,Ωfi(x)+Mα¯,Ω(bifi)(x))1lmliSα¯,Ωfl(x)1jmjl,jiMα¯,Ωfj(x), (3.39)

for almost every xΩ, where α = (α – 1)/m.

Proof

We first prove (3.39) for the case b⃗ = (b1, …, bm) and f⃗ = (f1, …, fm) with each biW1,pm+1(Ω) ∩ 𝓒(Ω) and fiLpi(Ω) ∩ 𝓒(Ω). We only consider the case i = 1 and other cases are analogous. By Gauss’s theorem, we get

B(x,tδ(x))fl(y)dy=B(x,tδ(x))fl(y)ν(y)dHn1(y),l=1,,m, (3.40)
B(x,tδ(x))y(|b1(x)b1(y)|f1(y))dy=B(x,tδ(x))|b1(x)b1(y)|f1(y)ν(y)dHn1(y), (3.41)

where ν(y) = (yx)/((x)) is the unit outer normal of B(x, (x)). From (3.9), (3.40) and (3.41), we see that

Λα,t,b1(f)(x)=(αmn)δ(x)δ(x)((tδ(x))α|B(x,tδ(x))|m)Ab1f1(x)j=2mAfj(x)+(tδ(x))α|B(x,tδ(x))|mj=2mAfj(x)(B(x,tδ(x))x|b1(x)b1(y)|f1(y)dy+B(x,tδ(x))|b1(x)b1(y)|f1(y)ν(y)dHn1(y)+tδ(x)B(x,tδ(x))|b1(x)b1(y)|f1(y)dHn1(y))+(tδ(x))α|B(x,tδ(x))|mAb1f1(x)l=2m2jmjlAfj(x)(B(x,tδ(x))fl(y)ν(y)dHn1(y)+tδ(x)B(x,tδ(x))fl(y)dHn1(y)), (3.42)

for almost every xΩ. Further we get from (3.42) that

|Λα,t,b1(f)(x)|(mnα)((tδ(x))α1|B(x,tδ(x))|m)Ab1|f1|(x)j=2mA|fj|(x)+(tδ(x))α|B(x,tδ(x))|mj=2mA|fj|(x)(|b1|(x)A|f1|(x)+2B(x,tδ(x))|b1(x)b1(y)||f1(y)|dHn1(y))+2(tδ(x))α|B(x,tδ(x))|mAb1|f1|(x)l=2m2jmjlA|fj|(x)B(x,tδ(x))|fl|(y)dHn1(y)(mnα)Mα1,Ω,b1(f)(x)+|b1(x)|Mα,Ω(f)(x)+2n(|b1(x)|Sα¯,Ωf1(x)+Sα¯,Ω(b1f1)(x))j=2Mα¯,Ωfj(x)+2nl=2m(|b1|(x)Mα¯,Ωf1(x)+Mα¯,Ω(b1f1)(x))Sα¯,Ωfl(x)2jmjlMα¯,Ωfj(x), (3.43)

for almost every xΩ, where α = (α – 1)/m.

The rest of proof follows an approximation argument. Assume that b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω) and f⃗ = (f1, …, fm) with each fjW1,pj(Ω). There exist some sequences of functions {φi,j}j=1 in W1,pm+1(Ω) ∩ 𝓒(Ω) and {ψi,j}j=1 in W1,pi(Ω) ∩ 𝓒(Ω) such that φi,jbi in W1,pm+1(Ω) and ψi,jfi in W1,pi(Ω) as j → ∞. Let φ⃗j = (φ1,j, …, φm,j) and ψ⃗j = (ψ1,j, …, ψm,j). By Riesz theorem, there exists a subsequence {φ1,jk}k=1{φ1,j}j=1 such that φ1,jk(x) → b1(x) as k → ∞ for almost every xΩ. Applying the arguments as in the proof of Lemma 3.4, we have that Λα,t,φ1,jk (ψ⃗jk)(x) → Λα,t,b1(f⃗)(x) as k → ∞ for almost every xΩ. We have proved that

|Λα,t,φ1,jk(ψjk)(x)|(mnα)Mα1,Ω,φjk1(ψjk)(x)+|φ1,jk(x)|Mα,Ω(ψjk)(x)+2n(|φ1,jk|(x)Sα¯,Ωψ1,jk(x)+Sα¯,Ω(φ1,jkψ1,jk)(x))i=2Mα¯,Ωψi,jk(x)+2nl=2m(|φ1,jk|(x)Mα¯,Ωψ1,jk(x)+Mα¯,Ω(φ1,jkψ1,jk)(x))×Sα¯,Ωψl,jk(x)2imilMα¯,Ωψi,jk(x)=:Ijk(x), (3.44)

for almost every xΩ, where ψ⃗l,jk = (ψ1,jk, ⋯, ψl–1,jk, |∇ψl,jk|, ψl+1,jk, …, ψm,jk).

Let pi~ be such that 1/ pi~ = 1/piα/n for i = 1, 2, …, m. Then we have 1/q = 1/ p1~ + ⋯ + 1/ pm~ + 1/pm+1. Let ti~ be such that 1/ ti~ = 1/tiα/n. Then we have 1/q = 1/ t1~ + 1/ p2~ + ⋯ + 1/ pm~ and 1/ ti~ = 1/ p1~ + 1/pm+1. From (3.18) and (3.20) we get

Mα1,Ω,φjk1(ψjk)Lq(Ω)α,m,n,p1,,pm+1φ1,jkLp1(Ω)i=1mψi,jkLpi(Ω), (3.45)
|φ1,jk|Mα,Ω(ψjk)Lq(Ω)α,m,n,p1,,pm+1φ1,jkW1,pm+1(Ω)i=1mψi,jkLpi(Ω). (3.46)

For convenience, for any arbitrary functions f, g, define the operator

T1(f,g)(x)=(|g(x)|Sα¯,Ωf(x)+Sα¯,Ω(fg)(x)),xΩ.

By Lemma 3.2, Hölder’s inequality and the sublinearity for 𝓢α,Ω we have that T1 is bounded and continuous from Lp1(Ω) × Lpm+1(Ω) to Lt1~ (Ω). Moreover, both 𝓢α,Ω and Mα,Ω are bounded and continuous from Lpi(Ω) to Lpi~ (Ω). These facts together with Lemma 3.3 imply that

(|φ1,jk|Sα¯,Ωψ1,jk+Sα¯,Ω(φ1,jkψ1,jk))i=2Mα¯,Ωφi,jkLq(Ω)α,m,n,p1,,pm+1φ1,jkLpm+1(Ω)i=1mψi,jkLpi(Ω), (3.47)
(|φ1,jk|Sα¯,Ωψ1,jk+Sα¯,Ω(φ1,jkψ1,jk))i=2Mα¯,Ωψi,jk(|b1|Sα¯,Ωf1+Sα¯,Ω(b1f1))i=2Mα¯,ΩfiinLq(Ω)ask. (3.48)

Similarly, we can get

(|φ1,jk|Mα¯,Ωψ1,jk+Mα¯,Ω(φ1,jkψ1,jk))Sα¯,Ωψl,jk2imilMα¯,Ωψi,jkLq(Ω)α,m,n,p1,,pm+1φ1,jkLpm+1(Ω)i=1mψi,jkLpi(Ω), (3.49)
(|φ1,jk|Mα¯,Ωψ1,jk+Mα¯,Ω(φ1,jkψ1,jk))Sα¯,Ωψl,jk2imilMα¯,Ωψi,jk(|b1|Mα¯,Ωf1+Mα¯,Ω(b1f1))Sα¯,Ωfl2imilMα¯,ΩfiinLq(Ω)ask. (3.50)

By (3.44)-(3.47), (3.49) and Minkowski’s inequality, we have

Λα,t,φ1,jk(ψjk)Lq(Ω)IjkLq(Ω)α,m,n,p1,,pm+1φ1,jkW1,pm+1(Ω)i=1mψi,jkLpi(Ω). (3.51)

This yields that {|Λα,t,φ1,jk|}k=1 is a bounded sequence in Lq(Ω), which together with the fact that Λα,t,φ1,jk(ψ⃗jk)(x) → Λα,t,b1(f⃗)(x) as k → ∞ for almost every xΩ implies that the Sobolev derivative ∇Λα,t,φ1,jk(ψ⃗jk) exists almost everywhere in Ω and there exists a subsequence {Λα,t,φ1,jk(ψjk)}=1 of {Λα,t,φ1,jk(ψjk)}k=1 such that

|Λα,t,φ1,jk(ψjk)||Λα,t,b1(f)|weakly inLq(Ω)as. (3.52)

Let

I(x):=(mnα)Mα1,Ω,b1(f)(x)+|b1|(x)Mα,Ω(f)(x)+2n(|b1|(x)Sα¯,Ωf1(x)+Sα¯,Ω(b1f1)(x))j=2Mα¯,Ωfj(x)+2nl=2m(|b1|(x)Mα¯,Ωf1(x)+Mα¯,Ω(b1f1)(x))Sα¯,Ωfl(x)2jmjlMα¯,Ωfj(x).

We get from (3.23) and (3.26) that

Mα1,Ω,φjk1(ψjk)Mα1,Ω,b1(f)Lq(Ω)0ask, (3.53)
|φ1,jk|Mα,Ω(ψjk)|b1|Mα,Ω(f)Lq(Ω)0ask. (3.54)

It follows from (3.48), (3.50), (3.53), (3.54) and Minkowski’s inequality that

IjkIinLq(Ω)ask. (3.55)

By (3.44) we have that

|Λα,t,φ1,jk(ψjk)(x)|Ijk(x), (3.56)

for almost every xΩ. Applying Proposition 3.1 to (3.56), we get from (3.52) and (3.55) that (3.39) holds. This completes the proof of Lemma 3.7.□

3.3 Gradient estimates for the local multilinear fractional maximal commutator

In this subsection we shall present the pointwise gradient estimates for the local multilinear maximal commutator, which are of interest in their own rights.

Lemma 3.8

Let 1 ≤ α < mn and |Ω| < ∞. Let 1 < p1, ⋯, pm+1 < ∞, 1/pi + 1/pm+1 < 1 for i = 1, …, m, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n with q > 1. If f⃗ = (f1, …, fm) and b⃗ = (b1, …, bm) with each fjW1,pj(Ω) and bjW1,pm+1(Ω), then for any i = 1, …, m, we have

|Mα,Ω,bi(f)(x)|αMα1,Ω,bi(f)(x)+2l=1mMα,Ω,bi(fl)(x)+|bi|(x)Mα,Ω(f)(x)+2Mα,Ω(f1,,fi1,|bi|fi,fi+1,,fm)(x),

for almost every xΩ. Here f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm).

Proof

We shall adopt the method of [13] to prove Lemma 3.8. To prove Lemma 3.8, it suffices to show that

|Mα,Ω,bi(f)(x)|αMα1,Ω,bi(f)(x)+2l=1mMα,Ω,bi(fl)(x)+|bi|(x)Mα,Ω(f)(x)+2Mα,Ω(f1,,fi1,|bi|fi,fi+1,,fm)(x), (3.57)

for almost every xΩ. Here f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm).

We only prove (3.57) for the case i = 1 since other cases are analogous. Let tj, j = 1, 2, …, be an enumeration of the rationals between 0 and 1. One can easily check that

Mα,Ω,b1(f)(x)=supj1Λα,tj,b1(f)(x),

for all xΩ. Invoking Lemma 3.4 we have that Λα,tj,b1(f⃗) ∈ W1,q(Ω) and

|Λα,t,b1(f)(x)|αMα1,Ω,b1(f)(x)+2l=1mMα,Ω,b1(fl)(x)+|b1|(x)Mα,Ω(f)(x)+2Mα,Ω(|b1|f1,f2,,fm)(x), (3.58)

for almost every xΩ. Here f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm). For any k ≥ 1, we define the function vk : Ω → [–∞, ∞] by

vk(x)=max1jkΛα,tj,b1(f)(x).

It is easy to see that the sequence {vk}k=1 is an increasing sequence of functions converging pointwise to Mα,Ω,b1(f). Combining (3.58) with the fact that the maximum of two Sobolev functions belongs to the Sobolev space (see [8, Lemma 7.6]) implies that

|vk(x)|αMα1,Ω,b1(f)(x)+2l=1mMα,Ω,b1(fl)(x)+|b1|(x)Mα,Ω(f)(x)+2Mα,Ω(|b1|f1,f2,,fm)(x), (3.59)

for almost every xΩ and all k = 1, 2, …. By the arguments similar to those in the proof of Lemma 3.4, we have

vkLq(Ω)α,m,n,p1,,pm+1,|Ω|b1W1,pm+1(Ω)j=1mfjW1,pj(Ω),

which leads to that {|vk|}k=1 is a bounded sequence in Lq(Ω). Since vk converges pointwise to Mα,Ω,b1 (f⃗) as k → ∞, then the weak gradient Mα,Ω,b1(f) exists and there exists a subsequence {vk}=1of{vk}k=1 such that |vk||Mα,Ω,b1(f)| weakly in Lq(Ω) as → ∞. We may proceed to the weak limit in (3.59). The rest of proof follows from the same argument as in the end of the proof of Lemma 3.4. We omit the details.□

Lemma 3.9

Let 1 < p1, ⋯, pm+1 < ∞, 1/q = 1/p1 + ⋯ + 1/pm+1 with q > 1. If f⃗ = (f1, …, fm) and b⃗ = (b1, …, bm) with each fjW1,pj(Ω) and bjW1,pm+1(Ω), then for any i = 1, …, m, we have

|MΩ,bi(f)(x)||bi|(x)MΩ(f)+2l=1mMΩ,bi(fl)(x)+2MΩ(f1,,fi1,|bi|fi,fi+1,,fm)(x),

for almost every xΩ. Here f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm).

Proof

By Lemma 3.5 and the arguments similar to those used in proving Lemma 3.8, we can get the desire conclusion of Lemma 3.9. The details are omitted.□

Lemma 3.10

Let 1 ≤ α < mn, 1 < p1, ..., pm+1 < n, 1/q = 1/p1 + ⋯ + 1/pm+1 – (α + m – 1)/n, 1/m < q < ∞. Assume that Ω admits a pj-Sobolev embedding for each j = 1, ..., m. If f⃗ = (f1, …, fm) and b⃗ = (b1, …, bm) with each fjW1,pj(Ω) and bjW1,pm+1(Ω), then for any i = 1, …, m, we have

|Mα,Ω,bi(f)(x)|αMα1,Ω,bi(f)(x)+2l=1mMα,Ω,bi(fl)(x)+|bi|(x)Mα,Ω(f)+2Mα,Ω(f1,,fi1,|bi|fi,fi+1,,fm)(x),

for almost every xΩ. Here f⃗l = (f1, …, fl–1, |∇fl|, fl+1, …, fm).

Proof

By Lemma 3.6 and the arguments similar to those used in proving Lemma 3.8, we can get the desire conclusion of Lemma 3.10. The details are omitted.□

Lemma 3.11

Let 1 < p1, …, pm+1 < ∞ and 1/ti = 1/pi + 1/pm+1 for i = 1, …, m with ti > 1. Let f⃗ = (f1, …, fm) with each fjLpj(Ω) for ti > n/(n – 1) and b⃗ = (b1, …, bm) with each bjW1,pm+1(Ω). Let 1 ≤ α < + 1 with β = min1≤jm{(n – 1)/pj, n – 2n/((n – 1)tj)} and |Ω| < ∞. Let 1/q = 1/p1 + ⋯ + 1/pm+1 – (α – 1)/n and q > 1. If b⃗ = (b1, …, bm) and f⃗ = (f1, …, fm) with each biW1,pm+1(Ω) and fiLpi(Ω), then for any i = 1, …, m, we have

|Mα,Ω,bi(f)(x)|(mnα)Mα1,Ω,bi(f)(x)+|bi(x)|Mα,Ω(f)(x)+2n(|bi(x)|Sα¯,Ω(fi)(x)+Sα¯,Ω(bifi)(x))1jmjiMα¯,Ω(fj)(x)+2n(|bi(x)|Mα¯,Ωfi(x)+Mα¯,Ω(bifi)(x))1lmliSα¯,Ωfl(x)1jmjl,jiMα¯,Ωfj(x),

for almost every xΩ, where α = (α – 1)/m.

Proof

Let tj, j = 1, 2, …, be an enumeration of the rationals between 0 and 1. Let q* be such that 1/q* = 1/p1 + ⋯ + 1/pm+1α/n. It is clear that q < q*. By the fact that |Λα,tj,b1(f⃗)| ≤ 𝔐α,Ω,b⃗(f⃗), (1.3) and Hölder’s inequality, we get

Λα,tj,b1(f)Lq(Ω)Mα,Ω,b(f)Lq(Ω)|Ω|1/q1/qMα,Ω,b(f)Lq(Ω)α,m,n,p1,,pm+1,|Ω|(i=1mbiLpm+1(Ω))j=1mfjLpj(Ω).

On the other hand, we get by Lemma 3.7 that |∇Λα,tj,b1(f⃗)| ∈ Lq(Ω). Therefore, Λα,tj,b1(f⃗) ∈ W1,q(Ω). The rest proof follows from Lemma 3.7 and the arguments similar to those used in proving Lemma 3.8, we omit the details.□

3.4 Proofs of Theorems 1.3 and 1.4

The proofs of Theorems 1.3 and 1.4 can be listed as follows:

Proof of Theorem 1.3

To prove (1.10), it suffices to show that

Mα,Ω,bi(f)W1,q(Ω)α,m,n,p1,,pm+1(i=1mbiW1,pm+1(Ω))j=1mfjW1,pj(Ω),i=1,,m. (3.60)

We only work with i = 1 and other cases are analogous.

Assume that the condition (a) holds. Let t1, p be such that 1/t1 = 1/p1 + 1/pm+1 and 1/p = 1/p1 + ⋯ + 1/pm. It is clear that 1/q = 1/t1 + 1/p = 1/t1 + 1/p2 + ⋯ + 1/pm. Then by Lemma 3.9, (1.3) and Remark 1.2, we get

MΩ,b1(f)Lq(Ω)|b1|MΩ(f)Lq(Ω)+2l=1mMΩ,b1(fl)Lq(Ω)+2MΩ(|b1|f1,f2,,fm)Lq(Ω)b1Lpm+1(Ω)MΩ(f)Lp(Ω)+Cl=1mb1Lpm+1(Ω)flLpl(Ω)1jmjlfjLpj(Ω)+C|b1|f1Lt1(Ω)j=2mfjLpj(Ω)α,m,n,p1,,pm+1b1W1,pm+1(Ω)j=1mfjW1,pj(Ω),

which together with (1.7) leads to (3.60) for the case i = 1.

Assume that the condition (b) holds. Let 1/q* = 1/p + 1/pm+1α/n. Clearly, q < q*. By Hölder’s inequality and Remark 1.2, we get

Mα,Ω,b1(f)Lq(Ω)|Ω|1/q1/qMα,Ω,b1(f)Lq(Ω)α,m,n,p1,,pm+1,|Ω|b1Lpm+1(Ω)j=1mfjLpj(Ω). (3.61)

By Lemma 3.8 and the arguments similar to those used to estimate ∥∇Λα,t,φ1,jk(ψ⃗jk)∥Lq(Ω) in the proof of Lemma 3.4, one finds

Mα,Ω,b1(f)Lq(Ω)α,m,n,p1,,pm+1,|Ω|b1W1,pm+1(Ω)j=1mfjW1,pj(Ω),

which together with (3.61) leads to (3.60) for the case i = 1.

Assume that the condition (c) holds. Let pi~ be such that 1/ pi~ = 1/pi – 1/n for i = 1, 2, …, m. Then we have 1/q = 1/p1 + 1/ p2~ + ⋯ + 1/ pm~ + 1/pm+1α/n. By Remark 1.1 and (2.3), we get

Mα,Ω,b1(f)Lq(Ω)α,m,n,p1,,pm+1b1Lpm+1(Ω)f1Lp1(Ω)j=2mfjLpj~(Ω)α,m,n,p1,,pm+1b1Lpm+1(Ω)j=1mfjW1,pj(Ω). (3.62)

By Lemma 3.10 and the arguments similar to those used to estimate ∥∇Λα,t,φ1,jk(ψ⃗jk)∥Lq(Ω) in the proof of Lemma 3.6, we have

Mα,Ω,b1(f)Lq(Ω)α,m,n,p1,,pm+1,|Ω|b1W1,pm+1(Ω)j=1mfjW1,pj(Ω),

which together with (3.62) leads to (3.60) for the case i = 1.□

Proof of Theorem 1.4

To prove (1.11), it suffices to prove that

Mα,Ω,bi(f)W1,q(Ω)α,m,n,p1,,pm+1(i=1mbiW1,pm+1(Ω))j=1mfjLpj(Ω),i=1,,m. (3.63)

We only work with i = 1 and other cases are analogous.

Let q* be such that 1/q* = 1/p1 + ⋯ + 1/pm+1α/n. It is clear that q < q*. By Hölder’s inequality and Remark 1.2, we get

Mα,Ω,b1(f)Lq(Ω)|Ω|1/q1/qMα,Ω,b1(f)Lq(Ω)α,m,n,p1,,pm+1,|Ω|b1Lpm+1(Ω)j=1mfjLpj(Ω). (3.64)

Using Lemma 3.11 and the arguments similar to those used in getting (3.51), we have

Mα,Ω,b1(f)Lq(Ω)α,m,n,p1,,pm+1b1W1,pm+1(Ω)j=1mfjLpj(Ω),

which together with (3.64) leads to (3.63) with i = 1.□

4 Proofs of Corollaries 1.5-1.7

To prove Corollaries 1.5-1.7, the following result is needed.

Lemma 4.1

([15]). Let Ω ⊂ ℝn, Ω ≠ ℝn, be an open set. Let fW1,p(Ω) for 1 < p < ∞ and Ω(f(x)dist(x,Ωc))pdx < ∞. Then f W01,p (Ω).

Proof of Corollary 1.5

Let f⃗ = (f1, …, fm) with each fj W01,pj (Ω) for 1 < pj < ∞. Fix 1 ≤ jm, there is a sequence {gj,k}k∈ℤ of functions in W1,pj(Ω) ∩ C0 (Ω) such that gj,kfj in W1,pj(Ω) as k → ∞. Let g⃗k = (g1,k, …, gm,k).

To prove [b⃗, 𝔐Ω](f⃗) ∈ W01,q (Ω), it suffices to show that [b⃗, 𝔐Ω]i(f⃗) ∈ W01,q (Ω) for all i = 1, …, m. We only consider the case i = 1 since other cases are analogous. Let t1, p be such that 1/t1 = 1/p1 + 1/pm+1 and 1/p = 1/p1 + ⋯ + 1/pm. It is clear that q < p < p1, q < t1 < p1 and 1/q = 1/pm+1 + 1/p = 1/t1 + 1/p2 + ⋯ + 1/pm. We get by the proof of Theorem 1.1 that [b⃗, 𝔐Ω]1(g⃗k) ∈ W1,q(Ω). It is easy to check that [b⃗, 𝔐Ω]1(g⃗k)(x) = 0 whenever dist(x, Ωc) < 12 min1≤jm dist(supp(gj,k), Ωc). It follows that {[b⃗, 𝔐Ω]1(g⃗k)}k∈ℤ is a bounded sequence in W01,q (Ω). By Hölder’s inequality we get b1g1,kb1f1 in Lt1(Ω) as k → ∞. One can easily check that

[b,MΩ]1(gk)[b,MΩ]1(f)||b1||MΩ(gk)MΩ(f)|+|MΩ(b1g1,k,g2,k,,gm,k)MΩ(b1f1,f2,,fm)|,

which together with (1.5) implies that [b⃗, 𝔐Ω]1 (g⃗k) → [b⃗, 𝔐Ω]1(f⃗) in Lq(Ω) as k → ∞. Then [b⃗, 𝔐Ω]1(f⃗) ∈ W01,q (Ω) follows from weak compactness argument.

It remains to consider the operator 𝔐Ω,b⃗. Without loss of generality we only prove that MΩ,b1 (f⃗) ∈ W01,q (Ω). Let t1, p be given as above. By the proof of Theorem 1.3 we have that MΩ,b1 (g⃗k) ∈ W1,q(Ω). It is easy to check that MΩ,b1 (g⃗k)(x) = 0 whenever dist(x, Ωc) < 12 min1≤jm dist(supp(gj,k), Ωc). It follows that {MΩ,b1(gk)}kZ is a bounded sequence in W01,q (Ω). By Hölder’s inequality we get b1g1,kb1f1 in Lt1(Ω) as k → ∞. By the arguments similar to those used in getting (3.22),

|MΩ,b1(gk)MΩ,b1(f)||b1|MΩ(|g1,kf1|,g2,k,,gm,k)+MΩ(b1|g1,kf1|,g2,k,,gm,k)+l=2mMΩ,b1(hl),

where h⃗l = (f1, …, fl–1, |gl,kfl|, gl+1,k, …, gm,k). This together with (1.4), Remark 1.2 and Hölder’s inequality implies that MΩ,b1 (g⃗k) → MΩ,b1 (f⃗) in Lq(Ω) as k → ∞. Then a weak compactness argument yields MΩ,b1 (f⃗) ∈ W01,q (Ω).□

Proof of Corollary 1.6

Let f⃗ = (f1, …, fm) with each fjW1,pj(Ω). It follows from Theorems 1.1 and 1.3 that [b⃗, 𝔐α,Ω](f⃗) ∈ W1,q(Ω) and 𝔐α,Ω,b⃗(f⃗) ∈ W1,q(Ω). Then the conclusion [b⃗, 𝔐α,Ω](f⃗) ∈ W01,q (Ω) and 𝔐α,Ω,b⃗(f⃗) ∈ W01,q (Ω) follow from Lemma 4.1 and the following

Ω(|[b,Mα,Ω](f)(x)|dist(x,Ωc))qdx+Ω(Mα,Ω,b(f)(x)dist(x,Ωc))qdx<. (4.1)

To prove (4.1), it suffices to show that

Ω(|[b,Mα,Ω]i(f)(x)|dist(x,Ωc))qdx+Ω(Mα,Ω,bi(f)(x)dist(x,Ωc))qdx<,i=1,,m. (4.2)

We only verify (4.2) for i = 1 since other cases are analogous. One can easily check that

Mα,Ω(f)(x)dist(x,Ωc)Mα1,Ω(f)(x),xΩ,

which gives that

max{|[b,Mα,Ω]1(f)(x)|,Mα,Ω,b1(f)(x)}dist(x,Ωc)(|b1|(x)Mα1,Ω(f)(x)+Mα1,Ω(b1f1,f2,,fm)(x)), (4.3)

for all xΩ.

Assume that the condition (a) holds. Let t1, p be such that 1/t1 = 1/p1 + 1/pm+1 and 1/p = 1/p1 + ⋯ + 1/pm – (α – 1)/n. Then we have 1 < t1 < p1, q < p < p1 and 1/q = 1/t1 + 1/p2 + ⋯ + 1/pm – (α – 1)/n = 1/pm+1 + 1/p. By (1.3) and Hölder’s inequality, we get

|b1|Mα1,Ω(f)Lq(Ω)q+Mα1,Ω(b1f1,f2,,fm)Lq(Ω)qCqb1Lpm+1(Ω)qMα1,Ω(f)Lp(Ω)q+Cb1f1Lt1(Ω)qj=2mfjLpj(Ω)qCqb1Lpm+1(Ω)qj=1mfjLpj(Ω)q<, (4.4)

which together with (4.3) proves (4.2) for the case i = 1.

Assume that the condition (b) holds. Let pi~ , ti~ be such that 1/ pi~ = 1/pi – 1/n and 1/ ti~ = 1/ pi~ + 1/pm+1 for i = 1, 2, …, m. Let 1/p = 1/ p1~ + ⋯ + 1/ pm~ – (α – 1)/n. Then 1/q = 1/ p1~ + ⋯ + 1/ pm~ + 1/pm+1 – (α – 1)/n = 1/pm+1 + 1/ = 1/ t1~ + 1/ p2~ + ⋯ + 1/ pm~ – (α – 1)/n. By (2.3) and the Hölder’s inequality we have that b1f1 Lt1~ (Ω). By (2.3) and the arguments similar to those used to derive (4.4), one can get

|b1|Mα1,Ω(f)Lq(Ω)q+Mα1,Ω(b1f1,f2,,fm)Lq(Ω)qCqb1Lpm+1(Ω)qj=1mfjLpj~(Ω)qCqb1Lpm+1(Ω)qj=1mfjLpj(Ω)q<,

which together with (4.3) proves (4.2) for the case i = 1.□

Proof of Corollary 1.7

Let f⃗ = (f1, …, fm) with each fjLpj(Ω). It follows from Theorems 1.2 and 1.4 that [b⃗, 𝔐α,Ω](f⃗) ∈ W1,q(Ω) and 𝔐α,Ω,b⃗(f⃗) ∈ W1,q(Ω). We now prove

Ω(|[b,Mα,Ω](f)(x)|dist(x,Ωc))qdx+Ω(Mα,Ω,b(f)(x)dist(x,Ωc))qdx<. (4.5)

To prove (4.5), it suffices to show that

Ω(|[b,Mα,Ω]i(f)(x)|dist(x,Ωc))qdx+Ω(Mα,Ω,bi(f)(x)dist(x,Ωc))qdx<,i=1,,m. (4.6)

We only verify (4.6) for i = 1 since other cases are analogous. By the arguments similar to those used to derive (4.4), we get

|b1|Mα1,Ω(f)Lq(Ω)q+Mα1,Ω(b1f1,f2,,fm)Lq(Ω)q<,

which together with (4.3) leads to (4.6) for i = 1. Combining (4.6) with Lemma 4.1 and the facts [b⃗, 𝔐α,Ω](f⃗) ∈ W1,q(Ω) and 𝔐α,Ω,b⃗(f⃗) ∈ W1,q(Ω) yields the desire conclusions of Corollary 1.7.□

Acknowledgement

This work was supported partly by National Natural Science Foundation of China (Grant No. 11701333).

References

[1] M. Agcayazi, A. Gogatishvili, K. Koca, R. Mustafayev, A note on maximal commutators and commutators of maximal functions, J. Math. Soc. Jpn. 67 (2015), no. 2, 581–593.10.2969/jmsj/06720581Search in Google Scholar

[2] J. M. Aldaz and J. Pérez Lázaro, Functions of bounded variation, the derivative of the one dimensional maximal function, and applications to inequalities, Trans. Amer. Math. Soc. 359 (2007), no. 5, 2443–2461.10.1090/S0002-9947-06-04347-9Search in Google Scholar

[3] J. Bastero, M. Milman, F. J. Ruiz, Commutators of for the maximal and sharp functions, Proc. Amer. Math. Soc. 128 (2000), no. 11, 3329–3334.10.1090/S0002-9939-00-05763-4Search in Google Scholar

[4] E. Carneiro and J. Mardid, Derivative bounds for fractional maximal functions, Trans. Amer. Math. Soc. 369 (2017), no. 6, 4063–4092.10.1090/tran/6844Search in Google Scholar

[5] E. Carneiro, J. Mardid, L. B. Pierce, Endpoint Sobolev and BV continuity for maximal operators, J. Funct. Anal. 273 (2017), no. 10, 3262–3294.10.1016/j.jfa.2017.08.012Search in Google Scholar

[6] E. Carneiro and D. Moreira, On the regularity of maximal operators, Proc. Amer. Math. Soc. 136 (2008), no. 12, 4395–4404.10.1090/S0002-9939-08-09515-4Search in Google Scholar

[7] E. Carneiro and B. F. Svaiter, On the variation of maximal operators of convolution type, J. Funct. Anal. 265 (2013), 837–865.10.1016/j.jfa.2013.05.012Search in Google Scholar

[8] D. Gilbarg and N. S. Trudinger, Elliptic partial differential equations of second order, 2nd edn., Springer-Verlag, Berlin, 1983.Search in Google Scholar

[9] P. Hajłasz and J. Onninen, On boundedness of maximal functions in Sobolev spaces, Ann. Acad. Sci. Fenn. Math. 29 (2004), no. 1, 167–176.Search in Google Scholar

[10] J. Hart, F. Liu, Q. Xue, Regularity and continuity of local multilinear maximal type operators, J. Geom. Anal., to appear. https://doi.org/10.1007/s12220-020-00400-7.10.1007/s12220-020-00400-7Search in Google Scholar

[11] T. Heikkinen, J. Kinnunen, J. Korvenpää, H. Tuominen, Regularity of the local fractional maximal function, Bull. London Math. Soc. 53 (2015), no. 1, 127–154.10.1007/s11512-014-0199-2Search in Google Scholar

[12] G. Hu and D. Yang, Maximal commutators of BMO functions and singular integral operators with nonsmooth kernels on spaces of homogeneous type, J. Math. Anal. Appl. 354 (2009), 249–262.10.1016/j.jmaa.2008.12.066Search in Google Scholar

[13] J. Kinnunen, The Hardy-Littlewood maximal function of a Sobolev function, Israel J. Math. 100 (1997), 117–124.10.1007/BF02773636Search in Google Scholar

[14] J. Kinnunen and P. Lindqvist, The derivative of the maximal function, J. Reine. Angew. Math. 503 (1998), 161–167.10.1515/crll.1998.095Search in Google Scholar

[15] J. Kinnunen and O. Martio, Hardy’s inequalities for Sobolev functions, Math. Res. Lett. 4 (1997), no. 4, 489–500.10.4310/MRL.1997.v4.n4.a6Search in Google Scholar

[16] J. Kinnunen and E. Saksman, Regularity of the fractional maximal function, Bull. London Math. Soc. 35 (2003), no. 4, 529–535.10.1112/S0024609303002017Search in Google Scholar

[17] S. Korry, Boundedness of Hardy-Littlewood maximal operator in the framework of Lizorkin-Triebel spaces, Rev. Mat. Complut. 15 (2002), no. 2, 401–416.10.5209/rev_REMA.2002.v15.n2.16899Search in Google Scholar

[18] A. K. Lerner, S. Ombrosi, C. Pérez, R. H. Torres, R. Trujillo-González, New maximal functions and multiple weights for the multilinear Calderón-Zygmund theory, Adv. Math. 220 (2009), 1222–1264.10.1016/j.aim.2008.10.014Search in Google Scholar

[19] F. Liu and H. Wu, On the regularity of the multisublinear maximal functions, Canad. Math. Bull. 58 (2015), no. 4, 808–817.10.4153/CMB-2014-070-7Search in Google Scholar

[20] F. Liu and H. Wu, On the regularity of maximal operators supported by submanifolds, J. Math. Anal. Appl. 453 (2017), 144–158.10.1016/j.jmaa.2017.03.058Search in Google Scholar

[21] F. Liu and S. Xi, Sobolev regularity for commutators of the fractional maximal functions, Banach J. Math. Anal., to appear. https://doi.org/10.1007/s43037-020-00095-6.10.1007/s43037-020-00095-6Search in Google Scholar

[22] F. Liu, Q. Xue and P. Zhang, Regularity and continuity of commutators of the Hardy-Littlewood maximal function, Math. Nachr. 293 (2020), no. 3, 491–509.10.1002/mana.201900013Search in Google Scholar

[23] H. Luiro, Continuity of the maixmal operator in Sobolev spaces, Proc. Amer. Math. Soc. 135 (2007), no. 1, 243–251.10.1090/S0002-9939-06-08455-3Search in Google Scholar

[24] H. Luiro, On the regularity of the Hardy-Littlewood maximal operator on subdomains of ℝn, Proc. Edinburgh Math. Soc. 53 (2010), no. 1, 211–237.10.1017/S0013091507000867Search in Google Scholar

[25] H. Luiro, The variation of the maximal function of a radial function, Ark. Math. 56 (2018), no. 1, 147–161.10.4310/ARKIV.2018.v56.n1.a9Search in Google Scholar

[26] H. Luiro and J. Madrid, The variation of the fractional maximal function of a radial function, Intern. Math. Research Notices 2019 (2019), no. 17, 5284–5298.10.1093/imrn/rnx277Search in Google Scholar

[27] M. Milman and T. Schonbek, Second order estimates in interpolation theory and applications, Proc. Amer. Math. Soc. 110 (1990), no. 4, 961–969.10.1090/S0002-9939-1990-1075187-4Search in Google Scholar

[28] C. Pérez and R. H. Torres, Sharp maximal function estimates for multilinear singular integrals, Contemp. Math. 320 (2003), 323–331.10.1090/conm/320/05615Search in Google Scholar

[29] J.P. G. Ramos, O. Saari, J. Weigt, Weak differentiability for fractional maximal functions of general Lp functions on domains, Adv. Math. 368 (2020), no. 15, 107–144.10.1016/j.aim.2020.107144Search in Google Scholar

[30] P. Zhang, Characterization of Lipschitz spaces via commutators of the Hardy-Littlewood maximal function, Comptes Rendus Math. 355 (2017), no. 3, 336–344.10.1016/j.crma.2017.01.022Search in Google Scholar

[31] P. Zhang, Multiple weighted estimates for commutators of multilinear maximal function, Acta Math. Sin. Engl. Ser. 31 (2015), no. 6, 973–994.10.1007/s10114-015-4293-6Search in Google Scholar

Received: 2020-10-09
Accepted: 2020-11-12
Published Online: 2020-12-08

© 2021 X. Zhang and F. Liu, published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 11.5.2024 from https://www.degruyter.com/document/doi/10.1515/anona-2020-0162/html
Scroll to top button