Volume 2010, Article ID 584521,38pages doi:10.1155/2010/584521
Research Article
Sharp Constants of Br ´ezis-Gallou ¨et-Wainger
Type Inequalities with a Double Logarithmic Term on Bounded Domains in Besov and
Triebel-Lizorkin Spaces
Kei Morii,
1Tokushi Sato,
2Yoshihiro Sawano,
3and Hidemitsu Wadade
41Heian Jogakuin St. Agnes’ School, 172-2, Gochomecho, Kamigyo-ku, Kyoto 602-8013, Japan
2Mathematical Institute, Tohoku University, Sendai 980-8578, Japan
3Department of Mathematics, Kyoto University, Kyoto 606-8502, Japan
4Department of Mathematics, Osaka City University, 3-3-138 Sugimoto, Sumiyoshi-ku, Osaka 558-8585, Japan
Correspondence should be addressed to Yoshihiro Sawano,yoshihiro-sawano@celery.ocn.ne.jp Received 4 October 2009; Revised 15 September 2010; Accepted 12 October 2010
Academic Editor: Veli B. Shakhmurov
Copyrightq2010 Kei Morii et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The present paper concerns the Sobolev embedding in the endpoint case. It is known that the embeddingW1,nRn→L∞Rnfails forn≥2. Br´ezis-Gallou¨et-Wainger and some other authors quantified why this embedding fails by means of the H ¨older-Zygmund norm. In the present paper we will give a complete quantification of their results and clarify the sharp constants for the coefficients of the logarithmic terms in Besov and Triebel-Lizorkin spaces.
1. Introduction and Known Results
We establish sharp Br´ezis-Gallou¨et-Wainger type inequalities in Besov and Triebel-Lizorkin spaces as well as fractional Sobolev spaces on a bounded domainΩ ⊂ Rn. Throughout the present paper, we place ourselves in the setting ofRnwithn≥ 2. We treat only real-valued functions.
First we recall the Sobolev embedding theorem in the critical case. For 1< q <∞, it is well known that the embeddingWn/q,qRn→LrRnholds for anyq≤r <∞, and does not hold forr ∞, that is, one cannot estimate theL∞-norm by theWn/q,q-norm. However, the
Br´ezis-Gallou¨et-Wainger inequality states that theL∞-norm can be estimated by theWn/q,q- norm with the partial aid of theWs,p-norm withs > n/pand 1≤p≤ ∞as follows:
uq/q−1L∞Rn ≤λ
1log
1uWs,pRn
1.1
holds whenever u ∈ Wn/q,qRn∩Ws,pRnsatisfies uWn/q,qRn 1, where 1 ≤ p ≤ ∞, 1 < q < ∞, ands > n/p. Inequality 1.1 for the casen p q s 2 dates back to Br´ezis-Gallou¨et1 . Later on, Br´ezis and Wainger2 obtained1.1for the general case, and remarked that the power q/q−1in 1.1is maximal; equation 1.1 fails for any larger power. Ozawa 3 proved 1.1 with the Sobolev normuWs,pRn in 1.1 replaced by the homogeneous Sobolev normuW˙s,pRn. An attempt of replacinguWs,pRn with the other norms has been made in several papers. For instance, Kozono et al.4 generalized1.1with both ofWn/q,qRnandWs,pRnreplaced by the Besov spaces and applied it to the regularity problem for the Navier-Stokes equation and the Euler equation. Moreover, Ogawa5 proved 1.1in terms of Triebel-Lizorkin spaces for the purpose to investigate the regularity to the gradient flow of the harmonic map into a sphere. We also mention that1.1was obtained in the Besov-Morrey spaces in6 .
In what follows, we concentrate on the case q n and replace the function space Wn/q,qRnbyW01,nΩwith a bounded domainΩinRn. Note that the norm ofW01,nΩis equivalent to∇uLnΩbecause of the Poincar´e inequality. When the differential ordersm is an integer with 1≤m≤n, andn/m < p≤n/m−1, the first, second and fourth authors 7 generalized the inequality corresponding to1.1and discussed how optimal the constant λis. To describe the sharpness of the constantλ, they made a formulation more precise as follows:
For given constantsλ1>0 andλ2 ∈R, does there exist a constantCsuch that un/n−1L∞Ω ≤λ1log
1uWs,pΩ λ2log
1log
1uWs,pΩ C
1.2
holds for allu∈W01,nΩ∩Ws,pΩwith∇uLnΩ1?
Here for the sake of definiteness, define
∇uLnΩ|∇u|LnΩ, |∇u|
n
i1
∂u
∂xi 2
1/2
. 1.3
We call the first term and the second term of the right-hand side of 1.2 the single logarithmic term and the double logarithmic term, respectively. We remark that the double logarithmic term grows weaker than the single one asuWs,pΩ → ∞.
Then they proved the following theorem, which gives the sharp constants forλ1and λ2in1.2. Here and below,Λ1andΛ2are constants defined by
Λ1 1
ωn−11/n−1, Λ2 Λ1
n 1
nω1/n−1n−1 , 1.4
whereωn−12πn/2/Γn/2is the surface area ofSn−1 {x∈Rn; |x|1}. SeeDefinition 2.5 below for the definition of the strong local Lipschitz condition for a domainΩ.
Theorem 1.17, Theorem 1.2 . Letn ≥ 2, 0 < α <1,m ∈ {1,2, . . . , n}, and,Ωbe a bounded domain inRnsatisfying the strong local Lipschitz condition.
iAssume that either
Iλ1 > Λ1
α , λ2∈R or IIλ1 Λ1
α , λ2 ≥ Λ2
α 1.5
holds. Then there exists a constant Csuch that inequality 1.2 with s m and p n/m−αholds for allu∈W01,nΩ∩Wm,n/m−αΩwith∇uLnΩ1.
iiAssume that either
IIIλ1 < Λ1
α , λ2∈R or IVλ1 Λ1
α , λ2< Λ2
α 1.6
holds. Then for any constantC, inequality1.2withs mandp n/m−αfails for someu∈W01,nΩ∩W0m,n/m−αΩwith∇uLnΩ1.
We note that the differential ordermof the higher order Sobolev space inTheorem 1.1 had to be an integer. The primary aim of the present paper is to pass Theorem 1.1 to those which include Sobolev spaces of fractional differential order. Meanwhile, higher-order Sobolev spaces are continuously embedded into corresponding H ¨older spaces. Standing on such a viewpoint, the first, second, and fourth authors 8 proved a result similar to Theorem 1.1 for the homogeneous H ¨older space ˙C0,αΩ instead of the Sobolev space Wm,n/m−αΩ. Furthermore, it is known that the H ¨older spaceC0,αΩ is expressed as the marginal case of the Besov spaceBα,∞,∞Ω provided that 0 < α < 1, which allows us to extendTheorem 1.1with the same sharp constants in Besov spaces.
In general, we set up the following problem in a fixed function spaceXΩ, which is contained inL∞Ω.
Fix a function space XΩ. For given constants λ1 > 0 andλ2 ∈ R, does there exist a constantCsuch that
un/n−1L∞Ω ≤λ1log
1uXΩ λ2log
1log
1uXΩ
C
1.7
holds for allu∈W01,nΩ∩XΩunder the normalization∇uLnΩ1?
We callWs,pRnan auxiliary space of1.7. First we state the following proposition, which is an immediate consequence of an elementary inequality,
log1st≤logsst log1t logs fort≥0, s≥1. 1.8
Proposition 1.2. LetΩbe a domain inRn, and letX1Ω,X2Ωbe function spaces satisfying uX1Ω≤MuX2Ω foru∈X2Ω 1.9
with some constantM≥1.
iIf inequality1.7holds inXΩ X1Ωwith a constantC, then so does1.7inXΩ X2Ωwith another constantC,
or equivalently,
iif inequality 1.7 fails in XΩ X2Ω with any constant C, then so does 1.7 in XΩ X1Ωwith any constantC.
From the proposition above, the sharp constants forλ1andλ2in1.7are independent of the choice of the equivalent norms of the auxiliary spaceXΩ. On the other hand, note that these sharp constants may depend on the definition of∇uLnΩ; there are several manners to define∇uLnΩ. In what follows, we choose1.3as the definition of∇uLnΩ.
In the present paper we will include Besov and Triebel-Lizorkin spaces as an auxiliary spaceXΩ. To describe the definition of Besov and Triebel-Lizorkin spaces, we denote byBR
the open ball inRncentered at the origin with radiusR >0, that is,BR {x∈Rn; |x|< R}.
Define the Fourier transformFand its inverseF−1by Fuξ 1
2πn/2
Rne−√−1x·ξuxdx, F−1ux 1 2πn/2
Rne√−1x·ξuξdξ 1.10 for u ∈ SRn, respectively, and they are also extended on SRn by the usual way. For ϕ∈ SRn, define an operatorϕDby
ϕDuF−1 ϕFu 1
2πn/2
F−1ϕ
∗u. 1.11
Next, we fix functionsψ0, ϕ0 ∈ C∞c Rnwhich are supported in the ballB4, in the annulus B4\B1, respectively, and satisfying
∞ k−∞
ϕ0kx χRn\{0}x, ψ0x 1−∞
k0
ϕ0kx forx∈Rn, 1.12
where we set ϕ0k ϕ0·/2k. Here,χE is the characteristic function of a setE and C∞c Ω denotes the class of compactly supportedC∞-functions onΩ. We also denote byCcΩthe class of compactly supported continuous functions onΩ.
Definition 1.3. Takeψ0, ϕ0satisfying1.12, and letu∈ SRn.
iLet 0< s <∞, 0< p≤ ∞, and 0< q≤ ∞. The Besov spaceBs,p,qRnis normed by
uBs,p,qRnψ0Du
LpRn ∞
k0
2skϕ0kDu
LpRn
q1/q
1.13
with the obvious modification whenq∞.
iiLet 0 < s <∞, 0< p < ∞, and 0 < q ≤ ∞. The Triebel-Lizorkin spaceFs,p,qRnis normed by
uFs,p,qRnψ0Du
LpRn
∞
k0
2skϕ0kDuq1/q
LpRn
1.14
with the obvious modification whenq∞; one excludes the casep∞.
Different choices of ψ0 andϕ0 satisfying1.12yield equivalent norms in1.13and 1.14. We refer to9 for exhaustive details of this fact. Here and below, we denote byAs,p,q the spacesBs,p,qwith 0< s <∞, 0< p≤ ∞, 0< q≤ ∞, orFs,p,qwith 0< s <∞, 0< p <∞, 0<
q≤ ∞. Unless otherwise stated, the letterAmeans the same scale throughout the statement.
As in9,10 , we adopt a traditional method of defining function spaces on a domain Ω⊂Rn.
Definition 1.4. Let 0< s <∞and 0< p, q≤ ∞.
iThe function spaceAs,p,qΩis defined as the subset ofDΩobtained by restricting elements inAs,p,qRntoΩ, and the norm is given by
uAs,p,qΩinf
vAs,p,qRn;v∈As,p,qRn, v|Ω uinDΩ
. 1.15
iiThe function space As,p,q0 Ω is defined as the closure of Cc∞Ω in the norm of As,p,qΩ.
iiiThe potential spaceHs,pΩstands forFs,p,2Ω.
Now we state our main result, which claims that the sharp constants in1.7are given by the same ones as inTheorem 1.1whenXΩ As,pα,s,qΩorAs,p0 α,s,qΩ, where in what follows we denote
pα,s
⎧⎨
⎩ n
s−α fors > α,
∞ forsα.
1.16
Here, conditionsI–IVare the same as inTheorem 1.1. We should remark thatAs,p0 α,s,qΩ⊂ As,pα,s,qΩ ⊂ L∞Ω and the formulation of Theorem 1.5 remains unchanged no matter what equivalent norms we choose for the norm of the function spaceAs,pα,s,qΩ. Indeed, Proposition 1.2i resp.,iishows that the condition onλ1andλ2for which inequality1.7 holdsresp., failsremains unchanged if we replace the definition of the norm · As,pα,s,qΩ
with any equivalent norm.
In the case 0< α <1, we can determine the condition completely.
Theorem 1.5. Letn≥ 2, 0< α < 1,s ≥α, 0< q≤ ∞, and letΩbe a bounded domain inRnand XΩ As,pα,s,qΩ.
iAssume that either (I) or (II) holds. Then there exists a constantCsuch that inequality1.7 holds for allu∈W01,nΩ∩As,pα,s,qΩwith∇uLnΩ1.
iiAssume that either (III) or (IV) holds. Then for any constantC, the inequality1.7fails for someu∈C∞c Ωwith∇uLnΩ1.
Remark 1.6. If Ω has a Lipschitz boundary, then the Stein total extension theorem 11, Theorem 5.24 implies thatWm,pΩ Hm,pΩ Fm,p,2Ωfor m ∈ Nand 1 < p < ∞.
HenceTheorem 1.5impliesTheorem 1.1.
In order to state our results in the caseα ≥ 1 for a general bounded domain Ω, we replace assumptionIIby the slightly stronger one
IIλ1 Λ1
α , λ2 >Λ2. 1.17
Unfortunately, we do not know whether the result in this case corresponding to the caseα≥1 inTheorem 1.5holds.
Theorem 1.7. Letn≥2,α≥1,s ≥α, 0 < q≤ ∞, letΩbe a bounded domain inRnsatisfying the strong local Lipschitz condition andXΩ As,pα,s,qΩ.
iAssume that either (I) orIIholds. Then there exists a constantCsuch that inequality 1.7holds for allu∈W01,nΩ∩As,pα,s,qΩwith∇uLnΩ1.
iiAssume that either (III) or (IV) holds. Then for any constantC, the inequality1.7fails for someu∈C∞c Ωwith∇uLnΩ1.
Remark 1.8. We have to impose the strong local Lipschitz condition inTheorem 1.7, because we use the universal extension theorem obtained by Rychkov12, Theorem 2.2 .
However, in the case 1< α <2, we can also determine the condition completely as in the case 0< α <1 provided that we restrict the functions toCcΩ.
Theorem 1.9. Letn ≥ 2, 1 < α < 2,s ≥ α, 0 < q ≤ ∞, letΩbe a bounded domain inRn, and XΩ As,pα,s,qΩ.
iAssume that either (I) or (II) holds. Then there exists a constantCsuch that inequality1.7 holds for allu∈W01,nΩ∩As,pα,s,qΩ∩CcΩwith∇uLnΩ1.
iiAssume that either (III) or (IV) holds. Then for any constantC, inequality1.7fails for someu∈Cc∞Ωwith∇uLnΩ1.
We also obtain the following corollary becauseC∞c Ω⊂As,p,q0 Ω⊂As,p,qΩ.
Corollary 1.10. Theorems1.5,1.7, and1.9still hold true if one replacesAs,pα,s,qΩbyAs,p0 α,s,qΩ.
Remark 1.11. iThe assertion inCorollary 1.10corresponding toTheorem 1.7still holds even if we do not impose the strong local Lipschitz condition, because there is a trivial extension operator fromAs,p,q0 ΩintoAs,p,qRn.
iiIf∂Ωis smooth, then we can see that u∈C
Ω
, u0 on∂Ω foru∈W01,nΩ∩As,pα,s,qΩ. 1.18
However,W01,nΩ∩As,pα,s,qΩis not contained inAs,p0 α,s,qΩ, in general.
Remark 1.12. The powern/n−1on the left-hand side of1.7is optimal in the sense that rn/n−1is the largest power for which there existλ1andCsuch that
urL∞Ω≤λ1log
1uXΩ
C 1.19
can hold for allu∈ W01,nΩ∩XΩwith∇uLnΩ 1. Here,XΩis as in Theorems1.5, 1.7, and1.9andCorollary 1.10. Indeed, ifr > n/n−1, then for anyλ1>0 and any constant C,1.19does not hold for someu∈W01,nΩ∩XΩwith∇uLnΩ1, which is shown by carrying out a similar calculation to the proof of Theorems1.5,1.7, and1.9ii; seeRemark 3.9 below for the details. To the contrary, if 1 ≤ r < n/n−1, then for anyλ1 > 0, there exists a constantCsuch that1.19holds for allu∈W01,nΩ∩XΩwith∇uLnΩ 1. This fact follows from the embedding described below and the same assertion concerning the Br´ezis- Gallou¨et-Wainger type inequality in the H ¨older space, which is shown in8, Remark 3.5 for 0< α <1andRemark 4.3forα≥1.
Finally let us describe the organization of the present paper. In Section 2, we introduce some notation of function spaces and state embedding theorems. Section 3 is devoted to proving the negative assertions of Theorems 1.5–1.9. Section 4 describes the affirmative assertions of Theorems1.5and 1.7.Section 5concerns the affirmative assertion ofTheorem 1.9. In the appendix, we prove elementary calculus which we stated inSection 5.
2. Preliminaries
First we provide a brief view of H ¨older and H ¨older-Zygmund spaces. Throughout the present paper,Cdenotes a constant which may vary from line to line.
For 0 < α ≤ 1, ˙C0,αRndenotes the homogeneous H ¨older space of orderαendowed with the seminorm
uC˙0,αRn sup
x,y∈Rn x /y
ux−u y
x−yα , 2.1
andC0,αRndenotes the nonhomogeneous H ¨older space of orderαendowed with the norm uC0,αRnuL∞RnuC˙0,αRn. 2.2
Define also
uC˙0,αRn;Rn sup
x,y∈Rn x /y
ux−u y
x−yα 2.3
for an Rn-valued function u. For 1 ≤ α ≤ 2, ˙C1,α−1Rn denotes the homogeneous H ¨older- Zygmund space of orderα, the set of all continuous functionsuendowed with the seminorm
uC˙1,α−1Rn sup
x,y∈Rn x /y
ux−2u xy
/2
u y
x−yα , 2.4
andC1,α−1Rndenotes the nonhomogeneous H ¨older-Zygmund space of orderα, the set of all continuous functionsuendowed with the norm
uC1,α−1RnuL∞RnuC˙1,α−1Rn. 2.5
Note that ˙C0,1Rnis a proper subset of ˙C1,0Rn. We remark that, in defining ˙C1,α−1Rn, it is necessary that we assume the functions continuous. Here we will exhibit an example of a discontinuous functionusatisfyinguC˙1,α−1Rn 0 in the appendix. We will not need to define the H ¨older-Zygmund space of the higher order. We need an auxiliary function space;
for 1< α≤2, let ˙C∇1,α−1Rndenote the analogue of ˙C1,α−1Rnendowed with the seminorm
uC˙1,α−1∇ Rn∇uC˙0,α−1Rn;Rn sup
x,y∈Rn x /y
∇ux− ∇u y
x−yα−1 . 2.6
The other function spaces on a domain Ω ⊂ Rn are made analogously to As,p,qΩ. For example, define
uC˙0,αΩinf
vC˙0,αRn;v∈C˙0,αRn, v|ΩuinDΩ , uC˙1,α−1Ωinf
vC˙1,α−1Rn;v∈C˙1,α−1Rn, v|Ωuin DΩ , uC˙1,α−1∇ Ωinf
∇vC˙0,α−1Rn;Rn;v∈C˙1,α−1∇ Rn, ∇v|Ω∇uinDΩ .
2.7
A moment’s reflection shows that for 0< α≤1,uC˙0,αΩcan be written as
uC˙0,αΩ sup
x,y∈Ω x /y
ux−u y
x−yα foru∈C˙0,αΩ 2.8
since the function
vx inf
y∈Ω
u y
uC˙0,αΩx−yα
forx∈Rn 2.9
attains the infimum defininguC˙0,αΩsee13, Theorem 3.1.1 . Moreover, we also observe that
uC˙1,α−1∇ Ω∇uC˙0,α−1Ω;Rn sup
x,y∈Ω x /y
∇ux− ∇u y
x−yα−1 foru∈C˙1,α−1∇ Ω∩CcΩ 2.10 since the zero-extended functionvofuonRn\Ωattains the infimum defining∇uC˙0,α−1Ω;Rn.
An elementary relation between these spaces andBα,∞,∞Rnis as follows.
Lemma 2.1Taibleson,14, Theorem 4 . Let 0< α <2. Then one has the norm equivalence
Bα,∞,∞RnCα,α−αRn, 2.11
whereαdenotes the integer part ofα;αmax{k∈N∪ {0}; k≤α}.
We remark that Lemma 2.1is still valid for α ≥ 2 after defining the function space Cα,α−αRnappropriately. However, we do not go into detail, since we will use the space Cα,α−αRnonly with 0< α <2.
We will invoke the following fact on the Sobolev type embedding for Besov and Triebel-Lizorkin spaces:
Lemma 2.2. Let 0< s <∞, 0< p <p≤ ∞, 0< q <q≤ ∞, and letΩbe a domain inRn. Then
Bs,p,qΩ→Bs,p,qΩ, Bs,p,qΩ→Bs−n1/p−1/p,p,qΩ, Bs,p,min{p,q}Ω→Fs,p,qΩ→Bs,p,max{p,q}Ω
2.12
in the sense of continuous embedding.
Proof. We accept all the embeddings whenΩ Rn; see9 for instance. The case whenΩhas smooth boundary is covered in9 . However, as the proof below shows, the results are still valid even when the boundary ofΩis not smooth. For the sake of convenience, let us prove the second one. To this end we takeu∈Bs,p,qΩ. Then by the definition ofBs,p,qΩand its norm, we can findv∈Bs,p,qRnso that
v|Ω u inDΩ, uBs,p,qΩ≤ vBs,p,qRn≤2uBs,p,qΩ. 2.13
Now that we acceptvBs−n1/p−1/p,p,qRn≤Cs,p,p,qvBs,p,qRn, we have
uBs−n1/p−1/p,p,qΩ≤ vBs−n1/p−1/p,p,qRn≤Cs,p,p,qvBs,p,qRn. 2.14
Combining these observations, we see that the second embedding holds.
We need the following proposition later, which claims that ˙C1,α−1Rn→C˙∇1,α−1Rnfor 1< α <2 in the sense of continuous embedding.
Proposition 2.3. Let 1< α <2. Then there existsCα>0 such that
uC˙1,α−1∇ Rn≤CαuC˙1,α−1Rn foru∈C˙1,α−1Rn. 2.15
The proof is somehow well knownsee15, Chapter 0 whenn1. Here for the sake of convenience we include it in the appendix. We will show that this fact is also valid on a domainΩ⊂Rn.
Proposition 2.4. Let 1< α <2 andΩbe a domain inRn. Then there existsCα>0 such that
uC˙1,α−1∇ Ω≤CαuC˙1,α−1Ω foru∈C˙1,α−1Ω. 2.16
Proof. For anyu∈C˙1,α−1Ω, there exists an extensionvu∈C˙1,α−1RnofuonRnsuch that
vu|Ω u inDΩ, uC˙1,α−1Ω≤ vuC˙1,α−1Rn≤2uC˙1,α−1Ω. 2.17
In particular,∇vu|Ω∇uinDΩ. By applyingProposition 2.3, we have uC˙1,α−1∇ Ω;Rn
inf
∇vC˙0,α−1Rn;Rn;v∈C˙∇1,α−1Rn, ∇v|Ω∇uinDΩ
≤ ∇vuC˙0,α−1Rn;RnvuC˙∇1,α−1Rn;Rn
≤CαvuC˙1,α−1Rn ≤2CαuC˙1,α−1Ω
2.18
and obtain the desired result.
Let us establish the following proposition. Here, unlike a bounded domainΩ, for the whole spaceRnwe adopt the following definition of the norm ofW1,nRn:
uW1,nRnuLnRn∇uLnRn. 2.19
Definition 2.5. One says that a bounded domainΩsatisfies the strong local Lipschitz condition if Ω has a locally Lipschitz boundary, that is, each point x on the boundary of Ω has a neighborhood Ux whose intersection with the boundary of Ω is the graph of a Lipschitz continuous function.
The definition for a general domain is more complicated; see11 for details.
Proposition 2.6. Let 0< γ < α. Then one has
uBγ,∞,∞Rn≤Cγuγ/αBα,∞,∞Rnu1−γ/αW1,nRn foru∈W1,nRn∩Bα,∞,∞Rn. 2.20
Furthermore, letΩbe a bounded domain inRn satisfying the strong local Lipschitz condition. Then one has
uBγ,∞,∞Ω≤Cγuγ/αBα,∞,∞Ω∇u1−γ/αLnΩ foru∈W01,nΩ∩Bα,∞,∞Ω. 2.21
Proposition 2.6can be obtained directly from a theory of interpolation. However, the proof being simple, we include it for the sake of reader’s convenience.
Proof ofProposition 2.6. Let us takeζ∈C∞c Rnso thatζ1 onB4\B1and suppζ⊂B8\B1/2. Set
ζiξ ξi
|ξ|2ζξ, ζi,kξ ζi
ξ 2k
2kξi
|ξ|2ζ ξ
2k
fork∈N∪ {0}, i∈ {1, . . . , n}. 2.22
Recall thatϕ0kis supported onB2k2\B2k, and observe that
ϕ0kξ 1 2k
n i1
ξiζi,kξϕ0kξ. 2.23
Hence we have
ϕ0kDu
L∞Rn
1 2k
n i1
ζi,kDϕ0kD∂u
∂xi
L∞Rn
1 2πn2k
n i1
F−1ζi,k∗F−1ϕ0k∗ ∂u
∂xi
L∞Rn
≤ 1 2πn2k
F−1ϕ0k
L1Rn
n i1
F−1ζi,k
Ln/n−1Rn
∂u
∂xi
LnRn
1
2πnF−1ϕ0
L1Rn
n i1
F−1ζi
Ln/n−1Rn
∂u
∂xi
LnRn
≤C∇uLnRn≤CuW1,nRn.
2.24
A similar estimate forψ0is also available:
ψ0Du
L∞Rn 1 2πn/2
F−1ψ0
∗u
L∞Rn
≤CuLnRn≤CuW1,nRn.
2.25
Hence we have
uBγ,∞,∞Rnψ0Du
L∞Rn sup
k∈N∪{0}
2γlϕ0kDu
L∞Rn
≤Cγ sup
k∈N∪{0}min 1
2α−γkuBα,∞,∞Rn,2γkuW1,nRn
2.26
since 2αkϕ0kDuL∞Rn≤ uBα,∞,∞Rn. Hence we have
uBγ,∞,∞Rn≤Cγsup
t>0
min 1
tα−γuBα,∞,∞Rn, tγuW1,nRn
Cγuγ/αBα,∞,∞Rnu1−γ/αW1,nRn.
2.27
It remains to prove2.21. The universal extension theorem obtained by Rychkov12, Theorem 2.2 yields that there exists a common extension operatorE:W01,nΩBγ,∞,∞Ω → W1,nRn Bγ,∞,∞Rnsuch that
uBβ,∞,∞Ω≤ EuBβ,∞,∞Rn≤CβuBβ,∞,∞Ω foru∈Bβ,∞,∞Ω,
∇uLnΩ≤ EuW1,nRn≤C∇uLnΩ foru∈W01,nΩ 2.28
for allγ≤β <∞. Then2.21is an immediate consequence of2.20.
3. Counterexample for the Inequality
In this section, we will give the proof of assertioniiof Theorems1.5–1.9.Lemma 2.2shows that
Bs,p,min{p,q}Ω→Fs,p,qΩ, 3.1
and hence it suffices to consider the caseAs,pα,s,qΩ Bs,pα,s,qΩin view ofProposition 1.2 i. Furthermore,Lemma 2.2also shows that
Bs,p α,s,min{pα,s,q}Ω→Bs,pα,s,qΩ for s > s, 3.2 and hence we have only to consider the case 0< q≤pα,sn/s−α≤1. Therefore, it suffices to show the following theorem for the proof ofiiof Theorems1.5–1.9.
Theorem 3.1. Letn≥2,α >0,s≥nα, 0< q≤pα,s, and letΩbe a bounded domain inRnand XΩ Bs,pα,s,qΩ. Assume that either (III) or (IV) holds. Then for any constantC, inequality1.7 fails for someu∈C∞c Ωwith∇uLnΩ1.
Here and below, we use the notation
s log1s fors≥0 3.3
for short, and then ◦s log1 log1s for s ≥ 0. We note that inequality 1.7 with XΩ Bs,pα,s,qΩ holds for allu ∈ W01,nΩ∩Bs,pα,s,qΩwith ∇uLnΩ 1 if and only if there exists a constantCindependent ofusuch thatFα,s,qu;λ1, λ2 ≤ Cholds for all u∈W01,nΩ∩Bs,pα,s,qΩ\ {0}, where
Fα,s,qu;λ1, λ2
uL∞Ω
∇uLnΩ
n/n−1
−λ1
uBs,pα,s,qΩ
∇uLnΩ
−λ2◦
uBs,pα,s,qΩ
∇uLnΩ
foru∈W01,nΩ∩Bs,pα,s,qΩ\ {0}.
3.4
For the proof ofTheorem 3.1, we have to find a sequence{uj}∞j1 ⊂ C∞c Ω\ {0}such thatFα,s,quj;λ1, λ2 → ∞asj → ∞under assumptionIIIorIV. In the case thatΩ Rn and that all the functions are supported inB1, we can choose such a sequence.
Lemma 3.2. Letn≥ 2,α > 0,s≥ nα, 0< q ≤ pα,s, andΩ Rn. Then there exists a family of functions{uj}∞j1⊂Cc∞Rn\ {0}with suppuj⊂B1for allj∈Nsuch that
Fα,s,q
uj;λ1, λ2
−→ ∞ asj−→ ∞ 3.5
under assumption (III) or (IV) ofTheorem 3.1.
We can now proveTheorem 3.1once we acceptLemma 3.2.
Proof ofTheorem 3.1. Examining1.7fails, so we may assume thatλ1, λ2 ≥0. Fixz0 ∈Ωand R0 ≥1 such that
B
x∈Rn;|x−z0|< 1 R0
⊂Ω. 3.6
Let{uj}∞j1be a family of functions as inLemma 3.2. If we set
vjx
⎧⎨
⎩
ujR0x−z0 forx∈B,
0 forx∈Ω\B, 3.7
thenvj∈C∞cΩ, and there exists a constantCα,s,R0≥1 such that vj
L∞Ωuj
L∞Rn, ∇vj
LnΩ∇uj
LnRn, vj
Bs,pα,s,qΩ≤Cα,s,R0uj
Bs,pα,s,qRn.
3.8
The first and the second equalities are immediate, while the third inequality is a direct consequence of the fact that the dilationu → uR0· is an isomorphism over Bs,pα,s,qRn. Using1.8and the fact thatλ1, λ2≥0, we have
Fα,s,q
uj;λ1, λ2
≤ uj
L∞Rn
∇uj
LnRn
n/n−1
−λ1 1
Cα,s,R0 uj
Bs,pα,s,qRn
∇uj
LnRn
−λ2◦ 1
Cα,s,R0
uj
Bs,pα,s,qRn
∇uj
LnRn
≤ vj
L∞Ω
∇vj
LnΩ
n/n−1
−λ1 vj
Bs,pα,s,qΩ
∇vj
LnΩ
−λ2◦ vj
Bs,pα,s,qΩ
∇vj
LnΩ
Cα,s,R0,λ1,λ2
Fα,s,q
vj;λ1, λ2
Cα,s,R0,λ1,λ2,
3.9
from which we conclude thatFα,s,qvj;λ1, λ2 → ∞asj → ∞.
We now concentrate on the proof ofLemma 3.2, and we first prepare several lemmas.
Let ϕ0 ∈ C∞c 0,∞ be a smooth function that is nonnegative, supported on the interval1,4 and satisfies
∞ l−∞
ϕ0
2l2t
1 fort >0. 3.10
Observe that3.10forcesϕ02 1.
Proposition 3.3. iIt holds
χ1/2j1,1/4 t≤j
l1
ϕ0
2l2t
≤χ1/2j2,1/2 t forj∈N. 3.11
iiIt holds
∞
0
ϕ0t
t dtlog 2. 3.12
Proof. iIn view of the size of the support ofϕ0, we easily obtain3.11.
iiIf we integrate both the sides of inequality3.11, then we have 1
j 1/4
1/2j1
1 tdt≤ 1
j j
l1
∞
0
ϕ0 2l2t
t dt
∞
0
ϕ0t
t dt≤ 1 j
1/2
1/2j2
1
tdt. 3.13
As a consequence, it follows that
log 2 1−1
j
≤ ∞
0
ϕ0t
t dt≤ log 2 11
j
. 3.14
A passage to the limit asj → ∞therefore yields3.12.
Define
wlx ∞
|x|
ϕ0 2l2t
t dt forx∈Rn, l∈N. 3.15
Note thatwlw12l−1·. Set
ujx 1 log 2
j j
l1
wlx forx∈Rn, j∈N. 3.16
We also note that suppuj ⊂B1/2since suppwl⊂B1/2l.
When we are going to specify the best constant,3.19is the heart of the matter.
Lemma 3.4. Letn≥2 and 0< p <∞. Then one has uj
L∞Rn1 forj∈N, 3.17
1−1/jp
nΛn−11 2nj1 ≤ujp
LpRn ≤ 11/jp
nΛn−11 forj∈N, 3.18
1−1/j log 2
Λ1
n−1
jn−1 ≤∇ujn
LnRn ≤ 11/j log 2
Λ1
n−1
jn−1 forj ∈N. 3.19
Proof. It is not so hard to prove3.17. Indeed, a change of variables yields
uj
L∞Rnuj0 1 log 2
j ∞
0
j l1
ϕ0 2l2t
t dt 1
log 2 ∞
0
ϕ0t
t dt. 3.20
Thus, we obtain3.17by applying3.12.
We next verify3.18. Recall that Λ1 is defined byΛ1 1/ω1/n−1n−1 . If we insert the definitions3.15and3.16, then we have
ujp
LpRn 1
Λn−11 log 2
jp 1
0
∞
r
1 t
j l1
ϕ0
2l2t dt
p
rn−1dr. 3.21