Introduction
The fractional Orlicz-Sobolev spaces W s , A ( R n ) W^{s,A}(\mathbb{R}^n) W s , A ( R n ) introduced in [4 , 17 ] are the natural generalizations of the classical fractional order Sobolev spaces W s , p ( R n ) W^{s,p}(\mathbb{R}^n) W s , p ( R n ) , p > 1 p>1 p > 1 . Given an N-function A A A and a fractional parameter s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) , these spaces are defined as
W s , A ( R n ) = { u : R n → R : ∫ R n A ( ∣ u ∣ ) d x + ∬ R n × R n A ( ∣ u ( x ) − u ( y ) ∣ ∣ x − y ∣ s ) d x d y ∣ x − y ∣ n < ∞ } . W^{s,A}(\mathbb{R}^n)=\left\{u:\mathbb{R}^n\to\mathbb{R}:\int_{\mathbb{R}^n}A(|u|)\,dx+\iint_{\mathbb{R}^n\times\mathbb{R}^n}A\left(\frac{|u(x)-u(y)|}{|x-y|^s}\right)\frac{dx\,dy}{|x-y|^n}<\infty\right\}. W s , A ( R n ) = { u : R n → R : ∫ R n A ( ∣ u ∣ ) d x + ∬ R n × R n A ( ∣ x − y ∣ s ∣ u ( x ) − u ( y ) ∣ ) ∣ x − y ∣ n d x d y < ∞ } .
For properties and further research on these spaces we refer to [4 , 5 , 7 , 8 , 17 ] .
In [28 ] , a characterization of the fractional Orlicz-Sobolev spaces W s , A ( R n ) W^{s,A}(\mathbb{R}^n) W s , A ( R n ) via Bessel potentials is introduced. More precisely, the Bessel-Orlicz potential space H s , A ( R n ) H^{s,A}(\mathbb{R}^n) H s , A ( R n ) is defined as the set
H s , A ( R n ) : = { u ∈ L A ( R n ) : u = G s ∗ f , for some f ∈ L A ( R n ) } , H^{s,A}(\mathbb{R}^n):=\left\{u\in L^A(\mathbb{R}^n):u=G_s*f,\text{ for some }f\in L^A(\mathbb{R}^n)\right\}, H s , A ( R n ) := { u ∈ L A ( R n ) : u = G s ∗ f , for some f ∈ L A ( R n ) } ,
endowed with the norm ∥ u ∥ H s , A ( R n ) : = ∥ f ∥ L A ( R n ) \|u\|_{H^{s,A}(\mathbb{R}^n)}:=\|f\|_{L^A(\mathbb{R}^n)} ∥ u ∥ H s , A ( R n ) := ∥ f ∥ L A ( R n ) , where ∥ ⋅ ∥ L A ( R n ) \|\cdot\|_{L^A(\mathbb{R}^n)} ∥ ⋅ ∥ L A ( R n ) denotes the Luxemburg norm associated with the Orlicz space L A ( R n ) L^A(\mathbb{R}^n) L A ( R n ) . These spaces are a direct generalization of the spaces studied in [2 , 3 , 36 ] . In particular, as proved in [28 Theorems 3.7 and 3.9] , for 0 < s < s ′ < 1 0<s<s'<1 0 < s < s ′ < 1 the following continuous embeddings hold:
H s ′ , A ( R n ) ↪ W s , A ( R n ) and W s ′ , A ( R n ) ↪ H s , A ( R n ) . (1) H^{s',A}(\mathbb{R}^n)\hookrightarrow W^{s,A}(\mathbb{R}^n)\quad\text{and}\quad W^{s',A}(\mathbb{R}^n)\hookrightarrow H^{s,A}(\mathbb{R}^n). \tag*{(1)} H s ′ , A ( R n ) ↪ W s , A ( R n ) and W s ′ , A ( R n ) ↪ H s , A ( R n ) . (1)
Furthermore, both spaces coincide in the limit case s = 1 s=1 s = 1 , i.e., H 1 , A ( R n ) = W 1 , A ( R n ) H^{1,A}(\mathbb{R}^n)=W^{1,A}(\mathbb{R}^n) H 1 , A ( R n ) = W 1 , A ( R n ) with equivalent norms.
Regarding embeddings in Orlicz spaces, in [4 Theorem 6.1] the authors established the optimal Orlicz target space for the Orlicz-Sobolev embedding. They prove that the continuous embedding W s , A ( R n ) ↪ L A n / s ( R n ) W^{s,A}(\mathbb{R}^n)\hookrightarrow L^{A_{n/s}}(\mathbb{R}^n) W s , A ( R n ) ↪ L A n / s ( R n ) holds, where A n / s A_{n/s} A n / s is defined as (see § 2.1 for further details)
A n / s ( t ) = A ( H − 1 ( t ) ) , H ( t ) = ( ∫ 0 t ( τ A ( τ ) ) s n − s d τ ) n − s n , t ≥ 0. A_{n/s}(t)=A(H^{-1}(t)),\qquad H(t)=\left(\int_0^t\left(\frac{\tau}{A(\tau)}\right)^{\frac{s}{n-s}}d\tau\right)^{\frac{n-s}{n}},\qquad t\geq0. A n / s ( t ) = A ( H − 1 ( t )) , H ( t ) = ( ∫ 0 t ( A ( τ ) τ ) n − s s d τ ) n n − s , t ≥ 0.
Using the aforementioned embedding, and combining the potential characterization of Orlicz spaces from [28 ] with the boundedness of the Riesz potential established in [12 ] , we obtain in Theorem 3.1 the optimal target space for the embedding of the potential space H s , A ( R n ) H^{s,A}(\mathbb{R}^{n}) H s , A ( R n ) into L A n / s ( R n ) L^{A_{n/s}}(\mathbb{R}^{n}) L A n / s ( R n ) . This result is applied in Theorem 3.4 to get embeddings of H 0 s , A ( Ω ) H_{0}^{s,A}(\Omega) H 0 s , A ( Ω ) into an unweighted Orlicz space L B ( Ω ) L^{B}(\Omega) L B ( Ω ) for a suitable N-function growing slower than A A A , being Ω \Omega Ω a bounded domain.
In the study of nonlinear elliptic equations on R n \mathbb{R}^{n} R n , a fundamental mathematical obstacle is the lack of compactness in standard Sobolev embeddings. However, restricting the functional to radial functions introduces a natural spatial decay that recovers compactness. In this direction, Strauss in [35 ] establishes a pointwise decay of radial functions in H 1 ( R n ) H^{1}(\mathbb{R}^{n}) H 1 ( R n ) bounded away from the origin. Building upon this principle, in [27 ] Ni obtained sharp pointwise bounds which became critical for establishing the asymptotic behavior and uniqueness of positive solutions. These type of estimates were generalized to the potential spaces settings, see for instance [14 , 36 ] . More precisely, if u ∈ H r a d s , p ( R n ) u \in H_{\mathrm{rad}}^{s,p}(\mathbb{R}^{n}) u ∈ H rad s , p ( R n ) with n > 1 n>1 n > 1 and 1 < p < ∞ 1<p<\infty 1 < p < ∞ then
∣ u ( x ) ∣ ≤ { c n ∣ x ∣ 1 − n p ∥ u ∥ H s , p ( R n ) 1 p < s < n (Strauss’ estimate) C n ∣ x ∣ s − n p ∥ u ∥ H s , p ( R n ) 1 p < s < n p (Ni’s estimate) . |u(x)| \le\begin{cases}
c_n |x|^{1-\frac{n}{p}} \lVert u\rVert_{H^{s,p}(\mathbb{R}^{n})} & \frac{1}{p}<s<n \quad\text{(Strauss' estimate)}\\
C_n |x|^{s-\frac{n}{p}} \lVert u\rVert_{H^{s,p}(\mathbb{R}^{n})} & \frac{1}{p}<s<\frac{n}{p} \quad\text{(Ni's estimate)}.
\end{cases} ∣ u ( x ) ∣ ≤ { c n ∣ x ∣ 1 − p n ∥ u ∥ H s , p ( R n ) C n ∣ x ∣ s − p n ∥ u ∥ H s , p ( R n ) p 1 < s < n (Strauss’ estimate) p 1 < s < p n (Ni’s estimate) .
In the context of potential spaces in the Orlicz settings, assuming that A A A satisfies the condition
1 < p − ≤ t A ′ ( t ) A ( t ) ≤ p + < ∞ 1<p^{-}\le\frac{tA'(t)}{A(t)}\le p^{+}<\infty 1 < p − ≤ A ( t ) t A ′ ( t ) ≤ p + < ∞
for some constants p ± p^{\pm} p ± , in [28 ] , Theorem 4.3 we obtained the following Strauss type inequality for u ∈ H r a d s , A ( R n ) u\in H_{\mathrm{rad}}^{s,A}(\mathbb{R}^{n}) u ∈ H rad s , A ( R n ) :
(Strauss’ estimate) If 1 p − < s ≤ 1 \frac{1}{p^{-}}<s\le1 p − 1 < s ≤ 1 , there exists C > 0 C>0 C > 0 such that
∣ u ( x ) ∣ ≤ C ∣ x ∣ 1 − n A ^ − 1 ( ∣ x ∣ 1 − n ) ∥ u ∥ H s , A ( R n ) . |u(x)|\le C\frac{|x|^{1-n}}{\widehat{A}^{-1}(|x|^{1-n})}\lVert u\rVert_{H^{s,A}(\mathbb{R}^{n})}. ∣ u ( x ) ∣ ≤ C A − 1 ( ∣ x ∣ 1 − n ) ∣ x ∣ 1 − n ∥ u ∥ H s , A ( R n ) .
To complement this analysis, in Theorem 4.3 we establish a Ni-type decay inequality for radial functions in the potential space H s , A ( R n ) H^{s,A}(\mathbb{R}^{n}) H s , A ( R n ) :
(Ni’s estimate) If 1 p − < s < n p + \frac{1}{p^{-}}<s<\frac{n}{p^{+}} p − 1 < s < p + n , there exists C > 0 C>0 C > 0 such that
∣ u ( x ) ∣ ≤ C ∣ x ∣ s A − 1 ( ∣ x ∣ − n ) ∥ u ∥ H s , A ( R n ) . |u(x)|\le C|x|^{s}A^{-1}(|x|^{-n})\lVert u\rVert_{H^{s,A}(\mathbb{R}^{n})}. ∣ u ( x ) ∣ ≤ C ∣ x ∣ s A − 1 ( ∣ x ∣ − n ) ∥ u ∥ H s , A ( R n ) .
By using these inequalities, we establish in Theorem 4.5 that the embedding
H r a d s , A ( R n ) ⊂ L B ( R n , ∣ x ∣ α d x ) H_{\mathrm{rad}}^{s,A}(\mathbb{R}^{n})\subset L^{B}(\mathbb{R}^{n},|x|^{\alpha}\,dx) H rad s , A ( R n ) ⊂ L B ( R n , ∣ x ∣ α d x )
is compact for appropriate relations between the Δ 2 \Delta_{2} Δ 2 -exponents of the Orlicz functions:
( p B − − p A + ) ( 1 − n p A + ) + α < 0 , and p B + < ( α + n ) p A − n − s p A − . (2) (p_{B}^{-}-p_{A}^{+})\left(\frac{1-n}{p_{A}^{+}}\right)+\alpha<0,\qquad\text{and}\qquad p_{B}^{+}<\frac{(\alpha+n)p_{A}^{-}}{n-sp_{A}^{-}}. \tag*{(2)} ( p B − − p A + ) ( p A + 1 − n ) + α < 0 , and p B + < n − s p A − ( α + n ) p A − . (2)
These conditions are extensions of the standard-growth case (see for instance [34 ] and [29 Theorem 3.7] ). Consequently, this allow us to derive in Theorem 4.7 that the embedding
W s , A ( R n ) ⊂ L B ( R n , ∣ x ∣ α d x ) W^{s,A}(\mathbb{R}^{n})\subset L^{B}(\mathbb{R}^{n},|x|^{\alpha}\,dx) W s , A ( R n ) ⊂ L B ( R n , ∣ x ∣ α d x )
is also compact.
As an application of the above results, we consider a general class of mixed local and nonlocal Hénon type problems of the form:
{ L s , A , γ u + a ( ∣ u ∣ ) ∣ u ∣ u = ∣ x ∣ α b ( ∣ u ∣ ) ∣ u ∣ u in R n , u > 0 , (3) \begin{cases}
\mathcal{L}_{s,A,\gamma}u+\dfrac{a(|u|)}{|u|}u=|x|^\alpha\dfrac{b(|u|)}{|u|}u & \text{in }\mathbb{R}^n,\\
u>0, \tag*{(3)}
\end{cases} ⎩ ⎨ ⎧ L s , A , γ u + ∣ u ∣ a ( ∣ u ∣ ) u = ∣ x ∣ α ∣ u ∣ b ( ∣ u ∣ ) u u > 0 , in R n , (3)
where u u u belongs to a suitable functional space of radial functions W s , A ( R n ) W^{s,A}(\mathbb{R}^n) W s , A ( R n ) (see (37) ). Given γ ∈ [ 0 , 1 ] \gamma\in[0,1] γ ∈ [ 0 , 1 ] , s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ] (depending on γ \gamma γ ), the operator L s , A , γ \mathcal{L}_{s,A,\gamma} L s , A , γ is
L s , A , γ u : = { ( − Δ a ) u , if γ = 1 , γ ( − Δ a ) u + ( 1 − γ ) ( − Δ a ) s u , if γ ∈ ( 0 , 1 ) , s ∈ ( 0 , 1 ) , ( − Δ a ) s u , if γ = 0 , s ∈ ( 0 , 1 ) , \mathcal{L}_{s,A,\gamma}u:=\begin{cases}
(-\Delta_a)u, & \text{if }\gamma=1,\\
\gamma(-\Delta_a)u+(1-\gamma)(-\Delta_a)^su, & \text{if }\gamma\in(0,1),\ s\in(0,1),\\
(-\Delta_a)^su, & \text{if }\gamma=0,\ s\in(0,1),
\end{cases} L s , A , γ u := ⎩ ⎨ ⎧ ( − Δ a ) u , γ ( − Δ a ) u + ( 1 − γ ) ( − Δ a ) s u , ( − Δ a ) s u , if γ = 1 , if γ ∈ ( 0 , 1 ) , s ∈ ( 0 , 1 ) , if γ = 0 , s ∈ ( 0 , 1 ) ,
where ( − Δ a ) s (-\Delta_a)^s ( − Δ a ) s is the fractional a a a -Laplacian of order s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ] (see Section 5 for details). The classical Hénon’s equation [20 ] with Dirichlet boundary conditions is
− Δ u = ∣ x ∣ α u q − 1 in B , u > 0 in B , u = 0 on ∂ B , (4) -\Delta u=|x|^\alpha u^{q-1}\ \text{in }B,\qquad u>0\ \text{in }B,\qquad u=0\ \text{on }\partial B, \tag*{(4)} − Δ u = ∣ x ∣ α u q − 1 in B , u > 0 in B , u = 0 on ∂ B , (4)
where B B B is the unit ball in R n \mathbb{R}^n R n , q > 1 q>1 q > 1 and α > 0 \alpha>0 α > 0 . This was used to investigate numerically the stability of spherical steady-state stellar systems with respect to spherical disturbances, and it has been the starting point of extensive research. A remarkable progress in the study of (4) was made in the work [27 ] , where it was proved that a radial solution exists for (4) if
2 < q < 2 α ∗ : = 2 ∗ + 2 α n − 2 = 2 ( α + n ) n − 2 . (5) 2<q<2_\alpha^*:=2^*+\frac{2\alpha}{n-2}=\frac{2(\alpha+n)}{n-2}. \tag*{(5)} 2 < q < 2 α ∗ := 2 ∗ + n − 2 2 α = n − 2 2 ( α + n ) . (5)
Other extensions of problem (4) have been studied in recent years, including nonlinear operators [26 , 16 ] , different boundary conditions [18 ] , nonlocal operators [33 , 9 ] , and mixed local–nonlocal operators [31 , 29 ] , to mention just a few.
Further generalizations to the fractional Laplacian have been obtained in the unbounded setting, where the ball is replaced by the whole space R n \mathbb{R}^n R n . In [9 ] it was shown that the equation
( − Δ ) s u = ∣ x ∣ α u q − 1 in R n , u > 0 , (-\Delta)^su=|x|^\alpha u^{q-1}\ \text{in }\mathbb{R}^n,\qquad u>0, ( − Δ ) s u = ∣ x ∣ α u q − 1 in R n , u > 0 ,
admits no positive solutions in the range 2 < q < 2 s , α ∗ : = 2 ( n + α ) n − 2 s 2<q<2_{s,\alpha}^*:=\dfrac{2(n+\alpha)}{n-2s} 2 < q < 2 s , α ∗ := n − 2 s 2 ( n + α ) , where α > − 2 s \alpha>-2s α > − 2 s , due to the loss of mass at infinity. In order to obtain existence of solutions in R n \mathbb{R}^n R n , a confining potential can be added to prevent the loss of mass at infinity (see for instance [34 ] , [25 ] or [29 ] ). Following these references, in Section 5 we apply the continuous and compact embeddings developed in this work to employ critical point theory to extend the above existence results for the Hénon-type model (3) . As a result, in Theorem 5.1 we obtain that under condition (2) , problem (3) has a radial solution. Observe that (2) constitutes a generalization of the Ni’s conditions (5) .
The article is organized as follows. In Section 2 we introduce the basic notation and known results on fractional Orlicz-Sobolev spaces, rearrangements and the Bessel kernel. Section 3 are devoted to prove our continuous embedding results while that Section 4 contains our compact embedding results for potential spaces and radial functions. Finally, in Section 5 we address an application related with the Hénon problem for a mixed local-nonlocal operator in R n \mathbb{R}^n R n .
Preliminaries
In this section, we introduce basic definitions and preliminary results used in the paper.
N-functions
Definition 2.1. We say that A : [ 0 , ∞ ) → [ 0 , ∞ ) A : [0,\infty) \to[0,\infty) A : [ 0 , ∞ ) → [ 0 , ∞ ) is an N N N -function if satisfies the following conditions
(i) A A A is convex, increasing, and A ( 0 ) = 0 A(0)=0 A ( 0 ) = 0 .
(ii) A A A is super-linear at zero and at infinity, that is
lim t → 0 + A ( t ) t = 0 and lim t → ∞ A ( t ) t = ∞ . \lim_{t\to0+} \frac{A(t)}{t}=0 \quad\text{and} \quad\lim_{t\to\infty} \frac{A(t)}{t}=\infty. t → 0 + lim t A ( t ) = 0 and t → ∞ lim t A ( t ) = ∞. Moreover, an N N N -function A A A admits the representation
A ( t ) = ∫ 0 t a ( τ ) d τ , A(t)=\int_0^t a(\tau)\,d\tau, A ( t ) = ∫ 0 t a ( τ ) d τ , where the function a a a is right-continuous for t ≥ 0 t\geq0 t ≥ 0 , positive for t > 0 t>0 t > 0 , non-decreasing and satisfies the conditions
a ( 0 ) = 0 , lim t → ∞ a ( t ) = ∞ . a(0)=0,\qquad\lim_{t\to\infty}a(t)=\infty. a ( 0 ) = 0 , t → ∞ lim a ( t ) = ∞. We extend A A A to R \mathbb{R} R as an even function, and we assume without loss of generality that A ( 1 ) = 1 A(1)=1 A ( 1 ) = 1 . Associated to A A A is the complementary N-function function to it which is defined as follows:
A ^ ( t ) : = sup { t w − A ( w ) : w > 0 } . (6) \widehat{A}(t):=\sup\{tw-A(w):w>0\}. \tag*{(6)} A ( t ) := sup { tw − A ( w ) : w > 0 } . (6) The following Young-type inequality holds
t 1 t 2 ≤ A ( t 1 ) + A ^ ( t 2 ) for every t 1 , t 2 ≥ 0. (7) t_1t_2\leq A(t_1)+\widehat{A}(t_2)\quad\text{for every }t_1,t_2\geq0. \tag*{(7)} t 1 t 2 ≤ A ( t 1 ) + A ( t 2 ) for every t 1 , t 2 ≥ 0. (7) An N N N -function A A A satisfies the Δ 2 \Delta_2 Δ 2 condition if there exists C > 2 C>2 C > 2 such that
A ( 2 t ) ≤ C A ( t ) for all t ≥ 0. A(2t)\leq CA(t)\quad\text{for all }t\geq0. A ( 2 t ) ≤ C A ( t ) for all t ≥ 0. By [23 Theorems 4.1 and 4.3, Chapter 1] , an N N N -function A A A and its complementary N N N -function A ^ \widehat{A} A satisfy the Δ 2 \Delta_2 Δ 2 condition if and only if there is 1 < p − ≤ p + < ∞ 1<p^-\leq p^+<\infty 1 < p − ≤ p + < ∞ such that
p − ≤ t a ( t ) A ( t ) ≤ p + , ∀ t > 0. (8) p^- \leq\frac{ta(t)}{A(t)} \leq p^+,\ \forall\,t>0. \tag*{(8)} p − ≤ A ( t ) t a ( t ) ≤ p + , ∀ t > 0. (8) In particular, (8) implies that
min { t p − , t p + } ≤ A ( t ) ≤ max { t p − , t p + } , t ≥ 0. (9) \min\{t^{p^-},t^{p^+}\}\leq A(t)\leq\max\{t^{p^-},t^{p^+}\},\quad t\geq0. \tag*{(9)} min { t p − , t p + } ≤ A ( t ) ≤ max { t p − , t p + } , t ≥ 0. (9) For further reference, we next define a special N N N -function. Let A A A be an N N N -function such that
∫ ∞ ( t A ( t ) ) s n − s d t = ∞ , ∫ 0 ( t A ( t ) ) s n − s d t < ∞ , (10) \int^\infty\left(\frac{t}{A(t)}\right)^{\frac{s}{n-s}}dt=\infty,\qquad\int_0 \left(\frac{t}{A(t)}\right)^{\frac{s}{n-s}}dt<\infty, \tag*{(10)} ∫ ∞ ( A ( t ) t ) n − s s d t = ∞ , ∫ 0 ( A ( t ) t ) n − s s d t < ∞ , (10) then the N N N -function A n / s A_{n/s} A n / s is defined as (see [4 ] )
A n / s ( t ) = A ( H − 1 ( t ) ) , t ≥ 0. (11) A_{n/s}(t)=A(H^{-1}(t)),\qquad t\geq0. \tag*{(11)} A n / s ( t ) = A ( H − 1 ( t )) , t ≥ 0. (11)
Here, the auxiliary function H H H is given by
H ( t ) = ( ∫ 0 t ( τ A ( τ ) ) s n − s d τ ) n − s n , t ≥ 0. H(t)=\left(\int_0^t\left(\frac{\tau}{A(\tau)}\right)^{\frac{s}{n-s}}d\tau\right)^{\frac{n-s}{n}}, \qquad t\geq0. H ( t ) = ( ∫ 0 t ( A ( τ ) τ ) n − s s d τ ) n n − s , t ≥ 0.
As we will see in the next sections, A n / s A_{n/s} A n / s is the optimal critical N N N -function in the embedding of Orlicz-Sobolev spaces into Orlicz spaces.
Orlicz-Sobolev spaces
Given an N N N -function A A A , we define the Orlicz space
L A ( R n ) = { u : R n → R : ρ A ( u ) < ∞ } , L^A(\mathbb{R}^n)=\{u:\mathbb{R}^n\to\mathbb{R}:\rho_A(u)<\infty\}, L A ( R n ) = { u : R n → R : ρ A ( u ) < ∞ } ,
and the fractional Orlicz-Sobolev space of order s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ]
W s , A ( R n ) = { u : R n → R : ρ A ( u ) + ρ s , A ( u ) < ∞ } , W^{s,A}(\mathbb{R}^n)=\{u:\mathbb{R}^n\to\mathbb{R}:\rho_A(u)+\rho_{s,A}(u)<\infty\}, W s , A ( R n ) = { u : R n → R : ρ A ( u ) + ρ s , A ( u ) < ∞ } ,
where ρ A \rho_A ρ A and ρ s , A \rho_{s,A} ρ s , A are the modulars given by
ρ A ( u ) : = ∫ R n A ( ∣ u ∣ ) d x , ρ s , A ( u ) : = { ∬ R n × R n A ( ∣ D s u ∣ ) d μ , if s ∈ ( 0 , 1 ) , ∫ R n A ( ∣ ∇ u ∣ ) d x , if s = 1 , \rho_A(u):=\int_{\mathbb{R}^n}A(|u|)\,dx,\qquad
\rho_{s,A}(u):=
\begin{cases}
\iint_{\mathbb{R}^n\times\mathbb{R}^n}A(|D^s u|)\,d\mu, & \text{if }s\in(0,1),\\
\int_{\mathbb{R}^n}A(|\nabla u|)\,dx, & \text{if }s=1,
\end{cases} ρ A ( u ) := ∫ R n A ( ∣ u ∣ ) d x , ρ s , A ( u ) := { ∬ R n × R n A ( ∣ D s u ∣ ) d μ , ∫ R n A ( ∣∇ u ∣ ) d x , if s ∈ ( 0 , 1 ) , if s = 1 ,
where d μ = ∣ x − y ∣ − n d x d y d\mu=|x-y|^{-n}dxdy d μ = ∣ x − y ∣ − n d x d y and D s u = u ( x ) − u ( y ) ∣ x − y ∣ s D^s u=\frac{u(x)-u(y)}{|x-y|^s} D s u = ∣ x − y ∣ s u ( x ) − u ( y ) .
The spaces L A ( R n ) L^A(\mathbb{R}^n) L A ( R n ) and W s , A ( R n ) W^{s,A}(\mathbb{R}^n) W s , A ( R n ) are endowed with the Luxemburg norms defined as
∥ u ∥ L A ( R n ) = inf { λ > 0 : ρ A ( u λ ) ≤ 1 } \|u\|_{L^A(\mathbb{R}^n)}=\inf\left\{\lambda>0:\rho_A\left(\frac{u}{\lambda}\right)\leq1\right\} ∥ u ∥ L A ( R n ) = inf { λ > 0 : ρ A ( λ u ) ≤ 1 }
and
∥ u ∥ W s , A ( R n ) = ∥ u ∥ L A ( R n ) + [ u ] W s , A ( R n ) , \|u\|_{W^{s,A}(\mathbb{R}^n)}=\|u\|_{L^A(\mathbb{R}^n)}+[u]_{W^{s,A}(\mathbb{R}^n)}, ∥ u ∥ W s , A ( R n ) = ∥ u ∥ L A ( R n ) + [ u ] W s , A ( R n ) ,
being [ u ] W s , A ( R n ) [u]_{W^{s,A}(\mathbb{R}^n)} [ u ] W s , A ( R n ) the so-called Gagliardo-like seminorm defined as
[ u ] W s , A ( R n ) = inf { λ > 0 : ρ s , A ( u λ ) ≤ 1 } . [u]_{W^{s,A}(\mathbb{R}^n)}=\inf\left\{\lambda>0:\rho_{s,A}\left(\frac{u}{\lambda}\right)\leq1\right\}. [ u ] W s , A ( R n ) = inf { λ > 0 : ρ s , A ( λ u ) ≤ 1 } .
The following Hölder’s type inequality holds for Orlicz spaces (see [22 Theorem 3.7.5] ).
Lemma 2.2. Let A A A be an N N N -function with complementary function A ^ \widehat{A} A . If u ∈ L A ( R n ) u\in L^A(\mathbb{R}^n) u ∈ L A ( R n ) and v ∈ L A ^ ( R n ) v\in L^{\widehat{A}}(\mathbb{R}^n) v ∈ L A ( R n ) then u v ∈ L 1 ( R n ) uv\in L^1(\mathbb{R}^n) uv ∈ L 1 ( R n ) and
∫ R n ∣ u v ∣ d x ≤ 2 ∥ u ∥ L A ( R n ) ∥ v ∥ L A ^ ( R n ) . \int_{\mathbb{R}^n}|uv|\,dx\leq2\|u\|_{L^A(\mathbb{R}^n)}\|v\|_{L^{\widehat{A}}(\mathbb{R}^n)}. ∫ R n ∣ uv ∣ d x ≤ 2∥ u ∥ L A ( R n ) ∥ v ∥ L A ( R n ) .
In [13 Theorem 4.1] it is proved the following inequality for convolutions in Orlicz spaces.
Lemma 2.3. Let A A A be an N N N -function. Then
∥ u ∗ v ∥ L A ( R n ) ≤ ∥ v ∥ L 1 ( R n ) ∥ u ∥ L A ( R n ) \|u*v\|_{L^A(\mathbb{R}^n)}\leq\|v\|_{L^1(\mathbb{R}^n)}\|u\|_{L^A(\mathbb{R}^n)} ∥ u ∗ v ∥ L A ( R n ) ≤ ∥ v ∥ L 1 ( R n ) ∥ u ∥ L A ( R n ) for every u ∈ L A ( R n ) u\in L^A(\mathbb{R}^n) u ∈ L A ( R n ) and v ∈ L 1 ( R n ) v\in L^1(\mathbb{R}^n) v ∈ L 1 ( R n ) . Moreover,
∫ R n A ( ∣ u ∗ v ∣ ) d x ≤ ∫ R n A ( ∥ v ∥ L 1 ( R n ) ∣ u ∣ ) d x . \int_{\mathbb{R}^n}A(|u*v|)\,dx\leq\int_{\mathbb{R}^n}A(\|v\|_{L^1(\mathbb{R}^n)}|u|)\,dx. ∫ R n A ( ∣ u ∗ v ∣ ) d x ≤ ∫ R n A ( ∥ v ∥ L 1 ( R n ) ∣ u ∣ ) d x .
Bessel kernels
Given s ∈ R s \in\mathbb{R} s ∈ R , we consider the Bessel kernel
G s ( x ) = F − 1 ( ( 1 + ∣ x ∣ 2 ) − s / 2 ) , G_s(x)=\mathcal{F}^{-1}\left((1+|x|^2)^{-s/2}\right), G s ( x ) = F − 1 ( ( 1 + ∣ x ∣ 2 ) − s /2 ) ,
where F ( f ) \mathcal{F}(f) F ( f ) denotes the Fourier Transform of f f f . Here, we allow s ∈ R s \in\mathbb{R} s ∈ R since G s ^ ( ξ ) = ( 1 + ∣ ξ ∣ 2 ) − s / 2 \widehat{G_s}(\xi)=(1+|\xi|^2)^{-s/2} G s ( ξ ) = ( 1 + ∣ ξ ∣ 2 ) − s /2 identified with a tempered distribution for all s s s . According to [32 Proposition 2] and [3 pages 13 and 27] , we recall some useful properties of G s G_s G s .
Lemma 2.4. The Bessel kernel satisfies the following properties:
i) G s > 0 G_s>0 G s > 0 , G s ∈ L 1 ( R n ) G_s\in L^1(\mathbb{R}^n) G s ∈ L 1 ( R n ) for all s > 0 s>0 s > 0 and ∥ G s ∥ L 1 ( R n ) = 1 \|G_s\|_{L^1(\mathbb{R}^n)}=1 ∥ G s ∥ L 1 ( R n ) = 1 ,
ii) ∇ G s ∈ L 1 ( R n ) \nabla G_s\in L^1(\mathbb{R}^n) ∇ G s ∈ L 1 ( R n ) for all s > 1 s>1 s > 1 ,
iii) G s ( x ) = G s ( − x ) G_s(x)=G_s(-x) G s ( x ) = G s ( − x ) for all x ∈ R n x\in\mathbb{R}^n x ∈ R n ,
iv) G s ( x ) G_s(x) G s ( x ) is radial and radially decreasing.
Rearrangements
We recall some definitions concerning rearrangement for a measurable function u : R n → R u:\mathbb{R}^n\to\mathbb{R} u : R n → R . The distribution function μ \mu μ of u u u is defined as
μ ( t ) = ∣ { x ∈ R n : ∣ u ( x ) ∣ > t } ∣ , for t > 0. \mu(t)=|\{x\in\mathbb{R}^n:|u(x)|>t\}|,\qquad\text{for }t>0. μ ( t ) = ∣ { x ∈ R n : ∣ u ( x ) ∣ > t } ∣ , for t > 0.
We observe that μ \mu μ is non-increasing and right-continuous.
The decreasing rearrangement of u u u , denoted by u ∗ u^* u ∗ , is the function
u ∗ ( s ) = inf { λ ≥ 0 : μ ( λ ) ≤ s } , for s ≥ 0. u^*(s)=\inf\{\lambda\geq0:\mu(\lambda)\leq s\},\qquad\text{for }s\geq0. u ∗ ( s ) = inf { λ ≥ 0 : μ ( λ ) ≤ s } , for s ≥ 0.
The symmetric rearrangement of u u u , denoted by u ⋆ u^\star u ⋆ , is the function
u ⋆ ( x ) = u ∗ ( ω n ∣ x ∣ n ) , x ∈ R n , u^\star(x)=u^*(\omega_n|x|^n),\qquad x\in\mathbb{R}^n, u ⋆ ( x ) = u ∗ ( ω n ∣ x ∣ n ) , x ∈ R n ,
where ω n \omega_n ω n is the volume of the n n n -dimensional unit ball. The following integral representations are valid for u ∗ u^* u ∗ and u ⋆ u^\star u ⋆ :
u ∗ ( s ) = ∫ 0 ∞ χ [ 0 , μ ( t ) ) ( s ) d t , u^*(s)=\int_0^\infty\chi_{[0,\mu(t))}(s)\,dt, u ∗ ( s ) = ∫ 0 ∞ χ [ 0 , μ ( t )) ( s ) d t ,
u ⋆ ( x ) = ∫ 0 ∞ χ { y : ω n ∣ y ∣ n < μ ( t ) } ( x ) d t = ∫ 0 ∞ χ { y : ω n ∣ y ∣ n < μ ( t ) } ∗ ( x ) d t , u^\star(x)=\int_0^\infty\chi_{\{y:\omega_n|y|^n<\mu(t)\}}(x)\,dt
=\int_0^\infty\chi_{\{y:\omega_n|y|^n<\mu(t)\}}^*(x)\,dt, u ⋆ ( x ) = ∫ 0 ∞ χ { y : ω n ∣ y ∣ n < μ ( t )} ( x ) d t = ∫ 0 ∞ χ { y : ω n ∣ y ∣ n < μ ( t )} ∗ ( x ) d t ,
where { y : ω n ∣ y ∣ n < μ ( t ) } ∗ \{y:\omega_n|y|^n<\mu(t)\}^* { y : ω n ∣ y ∣ n < μ ( t ) } ∗ is the symmetric rearrangement of the set { y : ω n ∣ y ∣ n < μ ( t ) } \{y:\omega_n|y|^n<\mu(t)\} { y : ω n ∣ y ∣ n < μ ( t )} , that is, it is the open ball centered at the origin whose volume is that of { y : ω n ∣ y ∣ n < μ ( t ) } \{y:\omega_n|y|^n<\mu(t)\} { y : ω n ∣ y ∣ n < μ ( t )} .
Continuous embeddings of Hs, A into Orlicz spces
By combining the potential characterization of Orlicz spaces from [28 ] with the boundedness of the Riesz potential established in [12 ] , we obtain a very simple proof of the optimal target space for the embedding of the potential space H s , A H^{s,A} H s , A .
Theorem 3.1. Let A A A be an N N N -function satisfying (10) , and let A n / s A_{n/s} A n / s be defined as in (11) . Then
H s , A ( R n ) ↪ L A n / s ( R n ) H^{s,A}(\mathbb{R}^n)\hookrightarrow L^{A_{n/s}}(\mathbb{R}^n) H s , A ( R n ) ↪ L A n / s ( R n ) and there exists C > 0 C>0 C > 0 such that
∥ u ∥ L A n / s ( R n ) ≤ C ∥ u ∥ H s , A ( R n ) . \|u\|_{L^{A_{n/s}}(\mathbb{R}^n)}\leq C\|u\|_{H^{s,A}(\mathbb{R}^n)}. ∥ u ∥ L A n / s ( R n ) ≤ C ∥ u ∥ H s , A ( R n ) .
Proof. Let u ∈ H s , A ( R n ) u \in H^{s,A}(\mathbb{R}^n) u ∈ H s , A ( R n ) . By definition, there exists a function f ∈ L A ( R n ) f \in L^A(\mathbb{R}^n) f ∈ L A ( R n ) such that u = G s ∗ f u=G_s*f u = G s ∗ f and ∥ u ∥ H s , A ( R n ) = ∥ f ∥ L A ( R n ) \|u\|_{H^{s,A}(\mathbb{R}^n)}=\|f\|_{L^A(\mathbb{R}^n)} ∥ u ∥ H s , A ( R n ) = ∥ f ∥ L A ( R n ) . Since the Bessel kernel G s G_s G s and the Riesz kernel I s I_s I s satisfy the pointwise inequality G s ( x ) ≤ I s ( x ) G_s(x)\le I_s(x) G s ( x ) ≤ I s ( x ) for all x ∈ R n ∖ { 0 } x\in\mathbb{R}^n\setminus\{0\} x ∈ R n ∖ { 0 } , we obtain
∣ ( G s ∗ f ) ( x ) ∣ ≤ ( G s ∗ ∣ f ∣ ) ( x ) ≤ ( I s ∗ ∣ f ∣ ) ( x ) for a.e. x ∈ R n . |(G_s*f)(x)|\le(G_s*|f|)(x)\le(I_s*|f|)(x)\quad\text{for a.e. }x\in\mathbb{R}^n. ∣ ( G s ∗ f ) ( x ) ∣ ≤ ( G s ∗ ∣ f ∣ ) ( x ) ≤ ( I s ∗ ∣ f ∣ ) ( x ) for a.e. x ∈ R n .
By the monotonicity of the Orlicz norm, this pointwise inequality implies
∥ G s ∗ f ( x ) ∥ L A n / s ( R n ) ≤ ∥ I s ∗ ∣ f ∣ ∥ L A n / s ( R n ) . \|G_s*f(x)\|_{L^{A_{n/s}}(\mathbb{R}^n)}\le\|I_s*|f|\|_{L^{A_{n/s}}(\mathbb{R}^n)}. ∥ G s ∗ f ( x ) ∥ L A n / s ( R n ) ≤ ∥ I s ∗ ∣ f ∣ ∥ L A n / s ( R n ) .
According to [12 Theorem 3.6] , the Riesz potential operator is bounded from L A ( R n ) L^A(\mathbb{R}^n) L A ( R n ) to L A n / s ( R n ) L^{A_{n/s}}(\mathbb{R}^n) L A n / s ( R n ) . Thus, there exists a constant c = c ( s , n ) > 0 c=c(s,n)>0 c = c ( s , n ) > 0 such that
∥ I s ∗ ∣ f ∣ ∥ L A n / s ( R n ) ≤ c ∥ ∣ f ∣ ∥ L A ( R n ) = c ∥ f ∥ L A ( R n ) . \|I_s*|f|\|_{L^{A_{n/s}}(\mathbb{R}^n)}\le c\||f|\|_{L^A(\mathbb{R}^n)}=c\|f\|_{L^A(\mathbb{R}^n)}. ∥ I s ∗ ∣ f ∣ ∥ L A n / s ( R n ) ≤ c ∥∣ f ∣ ∥ L A ( R n ) = c ∥ f ∥ L A ( R n ) .
Combining these inequalities, we conclude that
∥ u ∥ L A n / s ( R n ) = ∥ G s ∗ f ( x ) ∥ L A n / s ( R n ) ≤ c ∥ f ∥ L A ( R n ) = c ∥ u ∥ H s , A ( R n ) . \|u\|_{L^{A_{n/s}}(\mathbb{R}^n)}=\|G_s*f(x)\|_{L^{A_{n/s}}(\mathbb{R}^n)}\le c\|f\|_{L^A(\mathbb{R}^n)}=c\|u\|_{H^{s,A}(\mathbb{R}^n)}. ∥ u ∥ L A n / s ( R n ) = ∥ G s ∗ f ( x ) ∥ L A n / s ( R n ) ≤ c ∥ f ∥ L A ( R n ) = c ∥ u ∥ H s , A ( R n ) .
which completes the proof.
Corollary 3.2. Let B B B be an N N N -function. Assume there exist constants c , t 0 > 0 c,t_0>0 c , t 0 > 0 such that:
i) B ( t ) ≤ A ( c t ) B(t)\le A(ct) B ( t ) ≤ A ( c t ) for all 0 ≤ t ≤ t 0 0\le t\le t_0 0 ≤ t ≤ t 0 ,
ii) B ( t ) ≤ A n / s ( c t ) B(t)\le A_{n/s}(ct) B ( t ) ≤ A n / s ( c t ) for all t ≥ t 0 t\ge t_0 t ≥ t 0 .
Then, the embedding H s , A ( R n ) ↪ L B ( R n ) H^{s,A}(\mathbb{R}^n)\hookrightarrow L^B(\mathbb{R}^n) H s , A ( R n ) ↪ L B ( R n ) is continuous. In particular, there exists a constant C > 0 C>0 C > 0 such that
∥ u ∥ L B ( R n ) ≤ C ∥ u ∥ H s , A ( R n ) for all u ∈ H s , A ( R n ) . \|u\|_{L^B(\mathbb{R}^n)}\le C\|u\|_{H^{s,A}(\mathbb{R}^n)}\quad\text{for all }u\in H^{s,A}(\mathbb{R}^n). ∥ u ∥ L B ( R n ) ≤ C ∥ u ∥ H s , A ( R n ) for all u ∈ H s , A ( R n ) .
Proof. Since u ∈ H s , A ( R n ) u\in H^{s,A}(\mathbb{R}^n) u ∈ H s , A ( R n ) , there exists f ∈ L A ( R n ) f\in L^A(\mathbb{R}^n) f ∈ L A ( R n ) such that u = G s ∗ f u=G_s*f u = G s ∗ f and ∥ u ∥ H s , A ( R n ) = ∥ f ∥ L A ( R n ) \|u\|_{H^{s,A}(\mathbb{R}^n)}=\|f\|_{L^A(\mathbb{R}^n)} ∥ u ∥ H s , A ( R n ) = ∥ f ∥ L A ( R n ) . First, since N N N -functions are nonnegative, conditions (i) and (ii) imply that
B ( t ) ≤ A ( c t ) + A n / s ( c t ) , t ≥ 0. B(t)\le A(ct)+A_{n/s}(ct),\qquad t\ge0. B ( t ) ≤ A ( c t ) + A n / s ( c t ) , t ≥ 0.
By Theorem 3.1 , there exists a constant C 1 > 0 C_1>0 C 1 > 0 such that
∥ u ∥ L A n / s ( R n ) ≤ C 1 ∥ u ∥ H s , A ( R n ) . (12) \|u\|_{L^{A_{n/s}}(\mathbb{R}^n)}\le C_1\|u\|_{H^{s,A}(\mathbb{R}^n)}. \tag*{(12)} ∥ u ∥ L A n / s ( R n ) ≤ C 1 ∥ u ∥ H s , A ( R n ) . (12)
Moreover, since G s ∈ L 1 ( R n ) G_s\in L^1(\mathbb{R}^n) G s ∈ L 1 ( R n ) by Lemma 2.4 , Young’s inequality for convolution in Orlicz spaces yields
∥ u ∥ L A ( R n ) = ∥ G s ∗ f ∥ L A ( R n ) ≤ ∥ G s ∥ L 1 ( R n ) ∥ f ∥ L A ( R n ) = ∥ u ∥ H s , A ( R n ) . (13) \|u\|_{L^A(\mathbb{R}^n)}=\|G_s*f\|_{L^A(\mathbb{R}^n)}\le\|G_s\|_{L^1(\mathbb{R}^n)}\|f\|_{L^A(\mathbb{R}^n)}=\|u\|_{H^{s,A}(\mathbb{R}^n)}. \tag*{(13)} ∥ u ∥ L A ( R n ) = ∥ G s ∗ f ∥ L A ( R n ) ≤ ∥ G s ∥ L 1 ( R n ) ∥ f ∥ L A ( R n ) = ∥ u ∥ H s , A ( R n ) . (13)
Next, let λ = max { ∥ u ∥ L A ( R n ) , ∥ u ∥ L A n / s ( R n ) } \lambda=\max\{\|u\|_{L^A(\mathbb{R}^n)},\|u\|_{L^{A_{n/s}}(\mathbb{R}^n)}\} λ = max { ∥ u ∥ L A ( R n ) , ∥ u ∥ L A n / s ( R n ) } . By the definition of the Luxemburg norm, we have
∫ R n A ( ∣ u ( x ) ∣ λ ) d x ≤ 1 and ∫ R n A n / s ( ∣ u ( x ) ∣ λ ) d x ≤ 1. (14) \int_{\mathbb{R}^n}A\left(\frac{|u(x)|}{\lambda}\right)\,dx\le1\quad\text{and}\quad\int_{\mathbb{R}^n}A_{n/s}\left(\frac{|u(x)|}{\lambda}\right)\,dx\le1. \tag*{(14)} ∫ R n A ( λ ∣ u ( x ) ∣ ) d x ≤ 1 and ∫ R n A n / s ( λ ∣ u ( x ) ∣ ) d x ≤ 1. (14)
Using the convexity of the N N N -functions together with (14) , we obtain
∫ R n B ( ∣ u ( x ) ∣ 2 λ c ) d x ≤ ∫ R n A ( c ∣ u ( x ) ∣ 2 λ c ) d x + ∫ R n A n / s ( c ∣ u ( x ) ∣ 2 λ c ) d x ≤ 1 2 ∫ R n A ( ∣ u ( x ) ∣ λ ) d x + 1 2 ∫ R n A n / s ( ∣ u ( x ) ∣ λ ) d x = 1. \begin{aligned}
\int_{\mathbb{R}^n}B\left(\frac{|u(x)|}{2\lambda c}\right)\,dx
&\le\int_{\mathbb{R}^n}A\left(\frac{c|u(x)|}{2\lambda c}\right)\,dx+\int_{\mathbb{R}^n}A_{n/s}\left(\frac{c|u(x)|}{2\lambda c}\right)\,dx\\
&\le\frac{1}{2}\int_{\mathbb{R}^n}A\left(\frac{|u(x)|}{\lambda}\right)\,dx+\frac{1}{2}\int_{\mathbb{R}^n}A_{n/s}\left(\frac{|u(x)|}{\lambda}\right)\,dx=1.
\end{aligned} ∫ R n B ( 2 λ c ∣ u ( x ) ∣ ) d x ≤ ∫ R n A ( 2 λ c c ∣ u ( x ) ∣ ) d x + ∫ R n A n / s ( 2 λ c c ∣ u ( x ) ∣ ) d x ≤ 2 1 ∫ R n A ( λ ∣ u ( x ) ∣ ) d x + 2 1 ∫ R n A n / s ( λ ∣ u ( x ) ∣ ) d x = 1.
Therefore, by the definition of the Luxemburg norm combined with (12) and (13) gives
∥ u ∥ L B ( R n ) ≤ 2 c max { ∥ u ∥ L A ( R n ) , ∥ u ∥ L A n / s ( R n ) } ≤ 2 c C 1 ∥ u ∥ H s , A ( R n ) , \|u\|_{L^B(\mathbb{R}^n)} \le2c\max\left\{\|u\|_{L^A(\mathbb{R}^n)},\|u\|_{L^{A_{n/s}}(\mathbb{R}^n)}\right\}\le2cC_1\|u\|_{H^{s,A}(\mathbb{R}^n)}, ∥ u ∥ L B ( R n ) ≤ 2 c max { ∥ u ∥ L A ( R n ) , ∥ u ∥ L A n / s ( R n ) } ≤ 2 c C 1 ∥ u ∥ H s , A ( R n ) ,
which completes the proof.
Using [28 Theorem 3.9] , we easily obtain the following result.
Corollary 3.3. Let s ′ > s > 0 s' > s > 0 s ′ > s > 0 . Then W s ′ , A ( R n ) ↪ L A n / s ( R n ) W^{s',A}(\mathbb{R}^n)\hookrightarrow L^{A_{n/s}}(\mathbb{R}^n) W s ′ , A ( R n ) ↪ L A n / s ( R n ) , and there exists C > 0 C > 0 C > 0 such that
∥ u ∥ L A n / s ( R n ) ≤ C ∥ u ∥ W s ′ , A ( R n ) . \|u\|_{L^{A_{n/s}}(\mathbb{R}^n)} \le C\|u\|_{W^{s',A}(\mathbb{R}^n)}. ∥ u ∥ L A n / s ( R n ) ≤ C ∥ u ∥ W s ′ , A ( R n ) .
In the end of the section, we consider embedding of potential spaces into Orlicz spaces in bounded domains.
Let Ω ⊂ R n \Omega\subset\mathbb{R}^n Ω ⊂ R n be an open bounded set. Define
H 0 s , A ( Ω ) : = { u ∈ H s , A ( R n ) : u = f ∗ G s with f ∈ L A ( Ω ) and f = 0 in R n ∖ Ω } . H_0^{s,A}(\Omega):=\{u\in H^{s,A}(\mathbb{R}^n):u=f*G_s\text{ with }f\in L^A(\Omega)\text{ and }f=0\text{ in }\mathbb{R}^n\setminus\Omega\}. H 0 s , A ( Ω ) := { u ∈ H s , A ( R n ) : u = f ∗ G s with f ∈ L A ( Ω ) and f = 0 in R n ∖ Ω } .
Theorem 3.4. If there are c > 0 c>0 c > 0 and t 0 > 0 t_0>0 t 0 > 0 such that B ( t ) ≤ A ( c t ) B(t)\le A(ct) B ( t ) ≤ A ( c t ) for all t ≥ t 0 t\ge t_0 t ≥ t 0 , then H 0 s , A ( Ω ) ⊂ L B ( Ω ) H_0^{s,A}(\Omega)\subset L^B(\Omega) H 0 s , A ( Ω ) ⊂ L B ( Ω ) .
To prove Theorem 3.4 , we will state a preliminary inequality for rearrangements. Given a function u u u , we recall that u ∗ u^* u ∗ denotes its decreasing rearrangment, and u ⋆ u^\star u ⋆ is the symmetric rearrangement of u u u .
We quote the following rearrangement inequality (see for instance [24 Theorem 3.7] ).
Proposition 3.5 (Riesz rearrangement inequality). Any three non-negative functions f : R n → R + f:\mathbb{R}^n\to\mathbb{R}^+ f : R n → R + , g : R n → R + g:\mathbb{R}^n\to\mathbb{R}^+ g : R n → R + and h : R n → R + h:\mathbb{R}^n\to\mathbb{R}^+ h : R n → R + satisfy the inequality
∬ R n × R n f ( y ) g ( x − y ) h ( x ) d y d x ≤ ∬ R n × R n f ⋆ ( y ) g ⋆ ( x − y ) h ⋆ ( x ) d y d x , \iint_{\mathbb{R}^n\times\mathbb{R}^n}f(y)g(x-y)h(x)\,dy\,dx\le\iint_{\mathbb{R}^n\times\mathbb{R}^n}f^\star(y)g^\star(x-y)h^\star(x)\,dy\,dx, ∬ R n × R n f ( y ) g ( x − y ) h ( x ) d y d x ≤ ∬ R n × R n f ⋆ ( y ) g ⋆ ( x − y ) h ⋆ ( x ) d y d x , where f ⋆ : R n → R + f^\star:\mathbb{R}^n\to\mathbb{R}^+ f ⋆ : R n → R + , g ⋆ : R n → R + g^\star:\mathbb{R}^n\to\mathbb{R}^+ g ⋆ : R n → R + and h ⋆ : R n → R + h^\star:\mathbb{R}^n\to\mathbb{R}^+ h ⋆ : R n → R + are the symmetric decreasing rearrangements of the functions f f f , g g g and h h h respectively.
Theorem 3.6. Let u ∈ H s , A ( R n ) u\in H^{s,A}(\mathbb{R}^n) u ∈ H s , A ( R n ) be such that u = f ∗ G s u=f*G_s u = f ∗ G s for some f ∈ L A ( R n ) f\in L^A(\mathbb{R}^n) f ∈ L A ( R n ) . Define v = f ⋆ ∗ G s v=f^\star*G_s v = f ⋆ ∗ G s . Then
∫ 0 t u ∗ ( r ) d r ≤ ∫ 0 t v ∗ ( r ) d r . \int_0^t u^*(r)\,dr\le\int_0^t v^*(r)\,dr. ∫ 0 t u ∗ ( r ) d r ≤ ∫ 0 t v ∗ ( r ) d r .
Proof. Let u ∈ H s , A ( R n ) u\in H^{s,A}(\mathbb{R}^n) u ∈ H s , A ( R n ) . Then there is f ∈ L A ( R n ) f\in L^A(\mathbb{R}^n) f ∈ L A ( R n ) such that u = f ∗ G s u=f*G_s u = f ∗ G s . Since G s ≥ 0 G_s\ge0 G s ≥ 0 , ∣ u ∣ ≤ ∣ f ∣ ∗ G s |u|\le|f|*G_s ∣ u ∣ ≤ ∣ f ∣ ∗ G s pointwise, and the decreasing rearrangement is monotone, so u ∗ ≤ ( ∣ f ∣ ∗ G s ) ∗ u^*\le(|f|*G_s)^* u ∗ ≤ ( ∣ f ∣ ∗ G s ) ∗ . Hence it suffices to prove the theorem with u u u replaced by u ~ : = ∣ f ∣ ∗ G s ≥ 0 \widetilde{u}:=|f|*G_s\ge0 u := ∣ f ∣ ∗ G s ≥ 0 , whose associated function is the same v = f ⋆ ∗ G s v=f^\star*G_s v = f ⋆ ∗ G s . From now on we assume f ≥ 0 f\ge0 f ≥ 0 and u = f ∗ G s ≥ 0 u=f*G_s\ge0 u = f ∗ G s ≥ 0 .
We consider the function f ⋆ f^\star f ⋆ , the symmetric rearrangement of f f f . Then, ∥ f ⋆ ∥ L A ( R n ) = ∥ f ∥ L A ( R n ) \|f^\star\|_{L^A(\mathbb{R}^n)}=\|f\|_{L^A(\mathbb{R}^n)} ∥ f ⋆ ∥ L A ( R n ) = ∥ f ∥ L A ( R n ) .
Since f ⋆ ∈ L A ( R n ) f^\star\in L^A(\mathbb{R}^n) f ⋆ ∈ L A ( R n ) , we can define the function v = f ⋆ ∗ G s ∈ H s , A ( R n ) v=f^\star*G_s\in H^{s,A}(\mathbb{R}^n) v = f ⋆ ∗ G s ∈ H s , A ( R n ) , and
∥ v ∥ H s , A ( R n ) = ∥ f ⋆ ∥ L A ( R n ) = ∥ f ∥ L A ( R n ) = ∥ u ∥ H s , A ( R n ) \|v\|_{H^{s,A}(\mathbb{R}^n)}=\|f^\star\|_{L^A(\mathbb{R}^n)}=\|f\|_{L^A(\mathbb{R}^n)}=\|u\|_{H^{s,A}(\mathbb{R}^n)} ∥ v ∥ H s , A ( R n ) = ∥ f ⋆ ∥ L A ( R n ) = ∥ f ∥ L A ( R n ) = ∥ u ∥ H s , A ( R n )
Observe that, since both G s G_s G s is positive, radial and decreasing, v v v is radial and strictly decreasing. From Proposition 3.5, and since G s ⋆ = G s G_s^\star=G_s G s ⋆ = G s , we get
∬ R n × R n f ( y ) G s ( x − y ) h ( x ) d y d x ≤ ∬ R n × R n f ⋆ ( y ) G s ( x − y ) h ⋆ ( x ) d y d x . \iint_{\mathbb{R}^n\times\mathbb{R}^n} f(y)G_s(x-y)h(x)\,\mathrm{d}y\,\mathrm{d}x \le\iint_{\mathbb{R}^n\times\mathbb{R}^n} f^\star(y)G_s(x-y)h^\star(x)\,\mathrm{d}y\,\mathrm{d}x. ∬ R n × R n f ( y ) G s ( x − y ) h ( x ) d y d x ≤ ∬ R n × R n f ⋆ ( y ) G s ( x − y ) h ⋆ ( x ) d y d x .
Let E ⊂ R n E\subset\mathbb{R}^n E ⊂ R n be an open set such that ∣ E ∣ = t |E|=t ∣ E ∣ = t , and take h ( x ) = χ E ( x ) h(x)=\chi_E(x) h ( x ) = χ E ( x ) . In this case we have that h ⋆ ( x ) = χ B R t ( x ) h^\star(x)=\chi_{B_{R_t}}(x) h ⋆ ( x ) = χ B R t ( x ) , where R t R_t R t is such that ∣ B R t ∣ = t |B_{R_t}|=t ∣ B R t ∣ = t . The previous expression gives that
∫ E ∫ R n f ( y ) G s ( x − y ) d y d x ≤ ∫ B R t ∫ R n f ⋆ ( y ) G s ( x − y ) d y d x , \int_E\int_{\mathbb{R}^n}f(y)G_s(x-y)\,\mathrm{d}y\,\mathrm{d}x\le\int_{B_{R_t}}\int_{\mathbb{R}^n}f^\star(y)G_s(x-y)\,\mathrm{d}y\,\mathrm{d}x, ∫ E ∫ R n f ( y ) G s ( x − y ) d y d x ≤ ∫ B R t ∫ R n f ⋆ ( y ) G s ( x − y ) d y d x ,
which can be written as
∫ E f ∗ G s ( x ) d x ≤ ∫ B R t f ⋆ ∗ G s ( x ) d x , \int_E f\ast G_s(x)\,\mathrm{d}x\le\int_{B_{R_t}}f^\star\ast G_s(x)\,\mathrm{d}x, ∫ E f ∗ G s ( x ) d x ≤ ∫ B R t f ⋆ ∗ G s ( x ) d x ,
that is,
∫ E u ( x ) d x ≤ ∫ B R t v ( x ) d x . (15) \int_Eu(x)\,\mathrm{d}x\le\int_{B_{R_t}}v(x)\,\mathrm{d}x. \tag*{(15)} ∫ E u ( x ) d x ≤ ∫ B R t v ( x ) d x . (15)
By [10 Lemma 2.5] ,
∫ 0 t u ⋆ ( r ) d r = sup ∣ E ∣ = t ∫ E u ( x ) d x and ∫ 0 t v ⋆ ( r ) d r = sup ∣ E ∣ = t ∫ E v ( x ) d x . \int_0^t u^\star(r)\,\mathrm{d}r=\sup_{|E|=t}\int_Eu(x)\,\mathrm{d}x\quad\text{and}\quad\int_0^t v^\star(r)\,\mathrm{d}r=\sup_{|E|=t}\int_Ev(x)\,\mathrm{d}x. ∫ 0 t u ⋆ ( r ) d r = ∣ E ∣ = t sup ∫ E u ( x ) d x and ∫ 0 t v ⋆ ( r ) d r = ∣ E ∣ = t sup ∫ E v ( x ) d x .
Taking supremum over all E E E such that ∣ E ∣ = t |E|=t ∣ E ∣ = t in (15) , we obtain that
∫ 0 t u ⋆ ( r ) d r ≤ ∫ B R t v ( x ) d x ≤ sup ∣ E ∣ = t ∫ E v ( x ) d x = ∫ 0 t v ⋆ ( r ) d r . \int_0^t u^\star(r)\,\mathrm{d}r\le\int_{B_{R_t}}v(x)\,\mathrm{d}x\le\sup_{|E|=t}\int_Ev(x)\,\mathrm{d}x=\int_0^t v^\star(r)\,\mathrm{d}r. ∫ 0 t u ⋆ ( r ) d r ≤ ∫ B R t v ( x ) d x ≤ ∣ E ∣ = t sup ∫ E v ( x ) d x = ∫ 0 t v ⋆ ( r ) d r .
This completes the proof. ■
Now, we are ready to prove Theorem 3.4 .
Proof of Theorem 3.4 . Let u ∈ H 0 s , A ( Ω ) u\in H_0^{s,A}(\Omega) u ∈ H 0 s , A ( Ω ) . Then there exists f ∈ L A ( R n ) f\in L^A(\mathbb{R}^n) f ∈ L A ( R n ) such that f = 0 f=0 f = 0 in R n ∖ Ω \mathbb{R}^n\setminus\Omega R n ∖ Ω . Then f ⋆ ∈ L A ( R n ) f^\star\in L^A(\mathbb{R}^n) f ⋆ ∈ L A ( R n ) and f ⋆ = 0 f^\star=0 f ⋆ = 0 in R n ∖ Ω ⋆ \mathbb{R}^n\setminus\Omega^\star R n ∖ Ω ⋆ for some ball Ω ⋆ \Omega^\star Ω ⋆ . By Theorem 3.6 ,
∫ 0 t u ⋆ ( r ) d r ≤ ∫ 0 t v ⋆ ( r ) d r . \int_0^t u^\star(r)\,\mathrm{d}r\le\int_0^t v^\star(r)\,\mathrm{d}r. ∫ 0 t u ⋆ ( r ) d r ≤ ∫ 0 t v ⋆ ( r ) d r .
Therefore, by [10 Corollary 4.7] , since L B ( R n ) L^B(\mathbb{R}^n) L B ( R n ) is a rearrangement-invariant space, if v ∈ L B ( Ω ) v\in L^B(\Omega) v ∈ L B ( Ω ) then u ∈ L B ( Ω ) u\in L^B(\Omega) u ∈ L B ( Ω ) and
∥ u ∥ L B ( Ω ) ≤ ∥ u ∥ L B ( R n ) ≤ ∥ v ∥ L B ( R n ) ≤ C ∥ f ∥ L B ( R n ) ≤ C ( ∣ Ω ∣ ) ∥ f ∥ L A ( R n ) = C ∥ u ∥ H s , A ( R n ) . (16) \|u\|_{L^B(\Omega)}\le\|u\|_{L^B(\mathbb{R}^n)}\le\|v\|_{L^B(\mathbb{R}^n)}\le C\|f\|_{L^B(\mathbb{R}^n)}\le C(|\Omega|)\|f\|_{L^A(\mathbb{R}^n)}=C\|u\|_{H^{s,A}(\mathbb{R}^n)}. \tag*{(16)} ∥ u ∥ L B ( Ω ) ≤ ∥ u ∥ L B ( R n ) ≤ ∥ v ∥ L B ( R n ) ≤ C ∥ f ∥ L B ( R n ) ≤ C ( ∣Ω∣ ) ∥ f ∥ L A ( R n ) = C ∥ u ∥ H s , A ( R n ) . (16)
where we have used Young’s to bound
∥ v ∥ L B ( Ω ) ≤ ∥ v ∥ L B ( R n ) ≤ ∥ f ⋆ ∗ G s ∥ L B ( R n ) ≤ ∥ f ⋆ ∥ L B ( R n ) ∥ G s ∥ L 1 ( R n ) = ∥ f ⋆ ∥ L B ( Ω ⋆ ) ∥ G s ∥ L 1 ( R n ) = ∥ f ∥ L B ( R n ) ∥ G s ∥ L 1 ( R n ) , \begin{aligned}
\|v\|_{L^B(\Omega)}\le\|v\|_{L^B(\mathbb{R}^n)}\le\|f^\star\ast G_s\|_{L^B(\mathbb{R}^n)}\le\|f^\star\|_{L^B(\mathbb{R}^n)}\|G_s\|_{L^1(\mathbb{R}^n)} \\
&=\|f^\star\|_{L^B(\Omega^\star)}\|G_s\|_{L^1(\mathbb{R}^n)}=\|f\|_{L^B(\mathbb{R}^n)}\|G_s\|_{L^1(\mathbb{R}^n)},
\end{aligned} ∥ v ∥ L B ( Ω ) ≤ ∥ v ∥ L B ( R n ) ≤ ∥ f ⋆ ∗ G s ∥ L B ( R n ) ≤ ∥ f ⋆ ∥ L B ( R n ) ∥ G s ∥ L 1 ( R n ) = ∥ f ⋆ ∥ L B ( Ω ⋆ ) ∥ G s ∥ L 1 ( R n ) = ∥ f ∥ L B ( R n ) ∥ G s ∥ L 1 ( R n ) ,
and the embedding L A ( R n ) ⊂ L B ( R n ) L^A(\mathbb{R}^n)\subset L^B(\mathbb{R}^n) L A ( R n ) ⊂ L B ( R n ) . This completes the proof. ■
Compact embeddings for potential spaces of radial functions
In this section, we will provide conditions for the compactness of the inclusion of H s , A H^{s,A} H s , A into Orlicz spaces. First, we will discuss the unweighted case and then we will give conditions for the validity of the compact inclusion into weighted Orlicz spaces.
Consider the following class of radial potential functions
H r a d s , A ( R n ) = { u ∈ H s , A ( R n ) : u is radially symmetric } . H^{s,A}_{rad}(\mathbb{R}^{n})=\{u\in H^{s,A}(\mathbb{R}^{n}):\ u\text{ is radially symmetric}\}. H r a d s , A ( R n ) = { u ∈ H s , A ( R n ) : u is radially symmetric } .
In order to obtain compact embeddings of potential spaces into weighted Orlicz spaces, we will apply results regarding the decay of radial functions. Firstly, in [28 ] , we obtained the following Strauss type inequality
Theorem 4.1. Let A A A be a N N N -function so that A A A and A ^ \widehat{A} A satisfy the Δ 2 \Delta_{2} Δ 2 -condition. Let u ∈ H r a d s , A ( R n ) u\in H^{s,A}_{rad}(\mathbb{R}^{n}) u ∈ H r a d s , A ( R n ) with 1 p − < s ≤ 1 \frac{1}{p^{-}}<s\leq1 p − 1 < s ≤ 1 . Then there exists C > 0 C>0 C > 0 depending of n n n , p + p^{+} p + and p − p^{-} p − such that
∣ u ( x ) ∣ ≤ C ∣ x ∣ 1 − n A ^ − 1 ( ∣ x ∣ 1 − n ) ∥ u ∥ H s , A ( R n ) . |u(x)|\leq C\frac{|x|^{1-n}}{\widehat{A}^{-1}(|x|^{1-n})}\|u\|_{H^{s,A}(\mathbb{R}^{n})}. ∣ u ( x ) ∣ ≤ C A − 1 ( ∣ x ∣ 1 − n ) ∣ x ∣ 1 − n ∥ u ∥ H s , A ( R n ) .
The first compactness result is the following.
Theorem 4.2. Assume that A A A and A ^ \widehat{A} A are N N N -function satisfying the Δ 2 \Delta_{2} Δ 2 -condition. Let 1 p − < s < n p + \frac{1}{p^{-}}<s<\frac{n}{p^{+}} p − 1 < s < p + n . Let B B B be an N N N -function such that
lim t → 0 + B ( t ) A ( t ) = 0 and B ≪ A n / s . \lim_{t\to0^{+}}\frac{B(t)}{A(t)}=0\quad\text{and}\quad B\ll A_{n/s}. t → 0 + lim A ( t ) B ( t ) = 0 and B ≪ A n / s . Then, the embedding H r a d s , A ( R n ) ↪ L B ( R n ) H^{s,A}_{rad}(\mathbb{R}^{n})\hookrightarrow L^{B}(\mathbb{R}^{n}) H r a d s , A ( R n ) ↪ L B ( R n ) is compact.
Proof. Let { u k } k ∈ N ⊂ H r a d s , A ( R n ) \{u_{k}\}_{k\in\mathbb{N}}\subset H^{s,A}_{rad}(\mathbb{R}^{n}) { u k } k ∈ N ⊂ H r a d s , A ( R n ) be a bounded sequence. Then, up to a subsequence u k ⇀ u u_{k}\rightharpoonup u u k ⇀ u . Set v k : = u k − u v_{k}:=u_{k}-u v k := u k − u . Then v k ∈ H r a d s , A ( R n ) v_{k}\in H^{s,A}_{rad}(\mathbb{R}^{n}) v k ∈ H r a d s , A ( R n ) , v k ⇀ 0 v_{k}\rightharpoonup0 v k ⇀ 0 and M : = sup k ∥ v k ∥ H s , A ( R n ) < ∞ M:=\sup_{k}\|v_{k}\|_{H^{s,A}(\mathbb{R}^{n})}<\infty M := sup k ∥ v k ∥ H s , A ( R n ) < ∞ . We must show v k → 0 v_{k}\to0 v k → 0 in L B ( R n ) L^{B}(\mathbb{R}^{n}) L B ( R n ) .
Step 1: uniform bounds. By Young’s inequality for convolutions, ∥ v k ∥ L A ( R n ) ≤ M \|v_{k}\|_{L^{A}(\mathbb{R}^{n})}\leq M ∥ v k ∥ L A ( R n ) ≤ M , so
∫ R n A ( ∣ v k ∣ ) d x ≤ C 0 : = max { M p − , M p + } . \int_{\mathbb{R}^{n}}A(|v_{k}|)\,dx\leq C_{0}:=\max\{M^{p^{-}},M^{p^{+}}\}. ∫ R n A ( ∣ v k ∣ ) d x ≤ C 0 := max { M p − , M p + } .
By Theorem 3.1 , K : = sup k ∥ v k ∥ L A n / s ( R n ) < ∞ K:=\sup_{k}\|v_{k}\|_{L^{A_{n/s}}(\mathbb{R}^{n})}<\infty K := sup k ∥ v k ∥ L A n / s ( R n ) < ∞ , hence
∫ R n A n / s ( ∣ v k ∣ / K ) d x ≤ 1. \int_{\mathbb{R}^{n}}A_{n/s}(|v_{k}|/K)\,dx\leq1. ∫ R n A n / s ( ∣ v k ∣/ K ) d x ≤ 1.
Step 2: uniform decay. By the Strauss-type Theorem 4.1 and the inequality t ≤ A − 1 ( t ) A ^ − 1 ( t ) t\leq A^{-1}(t)\widehat{A}^{-1}(t) t ≤ A − 1 ( t ) A − 1 ( t ) ,
∣ v k ( x ) ∣ ≤ C ∣ x ∣ 1 − n A ^ − 1 ( ∣ x ∣ 1 − n ) M ≤ C M A − 1 ( ∣ x ∣ 1 − n ) , x ≠ 0. |v_{k}(x)|\leq C\frac{|x|^{1-n}}{\widehat{A}^{-1}(|x|^{1-n})}M\leq CM A^{-1}(|x|^{1-n}),\qquad x\neq0. ∣ v k ( x ) ∣ ≤ C A − 1 ( ∣ x ∣ 1 − n ) ∣ x ∣ 1 − n M ≤ C M A − 1 ( ∣ x ∣ 1 − n ) , x = 0.
Since n ≥ 2 n\geq2 n ≥ 2 , A − 1 ( ∣ x ∣ 1 − n ) → 0 A^{-1}(|x|^{1-n})\to0 A − 1 ( ∣ x ∣ 1 − n ) → 0 as ∣ x ∣ → ∞ |x|\to\infty ∣ x ∣ → ∞ . Thus, for every δ > 0 \delta>0 δ > 0 there is R = R ( δ ) > 0 R=R(\delta)>0 R = R ( δ ) > 0 such that
∣ v k ( x ) ∣ ≤ δ for a.e. ∣ x ∣ ≥ R , for all k . |v_{k}(x)|\leq\delta\quad\text{for a.e. }|x|\geq R,\ \text{for all }k. ∣ v k ( x ) ∣ ≤ δ for a.e. ∣ x ∣ ≥ R , for all k .
Step 3: local compactness. By [28 ] , Theorem 3.7, { v k } \{v_{k}\} { v k } is bounded in W s ′ , A ( R n ) W^{s',A}(\mathbb{R}^{n}) W s ′ , A ( R n ) for 0 < s ′ < s 0<s'<s 0 < s ′ < s , hence in W s ′ , A ( B R ) W^{s',A}(B_{R}) W s ′ , A ( B R ) . By [4 ] , Theorem 9.1, it is precompact in L A ( B R ) L^{A}(B_{R}) L A ( B R ) . Since also v k ⇀ 0 v_{k}\rightharpoonup0 v k ⇀ 0 in L A ( B R ) L^A(B_R) L A ( B R ) , every subsequence has a further subsequence converging strongly to 0 0 0 ; hence v k → 0 v_k \to0 v k → 0 in L A ( B R ) L^A(B_R) L A ( B R ) and, as A ∈ Δ 2 A \in\Delta_2 A ∈ Δ 2 ,
∫ B R A ( ∣ v k ∣ ) d x → 0 as k → ∞ . \int_{B_R} A(|v_k|)\,dx \to0 \quad\text{as } k \to\infty. ∫ B R A ( ∣ v k ∣ ) d x → 0 as k → ∞.
Step 4: conclusion. Fix λ > 0 \lambda> 0 λ > 0 and ε > 0 \varepsilon> 0 ε > 0 . Since A ( λ t ) ≤ max { λ p − , λ p + } A ( t ) A(\lambda t) \le\max\{\lambda^{p^-},\lambda^{p^+}\}A(t) A ( λ t ) ≤ max { λ p − , λ p + } A ( t ) ,
B ( λ t ) A ( t ) = B ( λ t ) A ( λ t ) A ( λ t ) A ( t ) → 0 ( t → 0 + ) . \frac{B(\lambda t)}{A(t)}=\frac{B(\lambda t)}{A(\lambda t)}\frac{A(\lambda t)}{A(t)}\to0 \quad(t\to0^+). A ( t ) B ( λ t ) = A ( λ t ) B ( λ t ) A ( t ) A ( λ t ) → 0 ( t → 0 + ) .
We have the following situations:
(a) Choose δ ∈ ( 0 , 1 ) \delta\in(0,1) δ ∈ ( 0 , 1 ) with B ( λ t ) ≤ ε A ( t ) B(\lambda t) \le\varepsilon A(t) B ( λ t ) ≤ ε A ( t ) for 0 ≤ t ≤ δ 0 \le t \le\delta 0 ≤ t ≤ δ .
(b) Since B ≪ A n / s B \ll A_{n/s} B ≪ A n / s , choose T > δ T > \delta T > δ with B ( λ t ) ≤ ε A n / s ( t / K ) B(\lambda t) \le\varepsilon A_{n/s}(t/K) B ( λ t ) ≤ ε A n / s ( t / K ) for t ≥ T t \ge T t ≥ T .
(c) For δ ≤ t ≤ T \delta\le t \le T δ ≤ t ≤ T , B ( λ t ) ≤ B ( λ T ) ≤ C ∗ A ( t ) B(\lambda t) \le B(\lambda T) \le C_*A(t) B ( λ t ) ≤ B ( λ T ) ≤ C ∗ A ( t ) , with C ∗ : = B ( λ T ) / A ( δ ) C_* := B(\lambda T)/A(\delta) C ∗ := B ( λ T ) / A ( δ ) .
Take R = R ( δ ) R=R(\delta) R = R ( δ ) from Step 2.
On { ∣ x ∣ ≥ R } \{|x|\ge R\} { ∣ x ∣ ≥ R } we have ∣ v k ∣ ≤ δ |v_k|\le\delta ∣ v k ∣ ≤ δ , so
∫ ∣ x ∣ ≥ R B ( λ ∣ v k ∣ ) d x ≤ ε ∫ R n A ( ∣ v k ∣ ) d x ≤ ε C 0 . \int_{|x|\ge R} B(\lambda|v_k|)\,dx \le\varepsilon\int_{\mathbb{R}^n} A(|v_k|)\,dx \le\varepsilon C_0. ∫ ∣ x ∣ ≥ R B ( λ ∣ v k ∣ ) d x ≤ ε ∫ R n A ( ∣ v k ∣ ) d x ≤ ε C 0 .
On B R B_R B R , by the three cases above,
∫ B R B ( λ ∣ v k ∣ ) d x ≤ ( ε + C ∗ ) ∫ B R A ( ∣ v k ∣ ) d x + ε ∫ R n A n / s ( ∣ v k ∣ K ) d x ≤ ( ε + C ∗ ) ∫ B R A ( ∣ v k ∣ ) d x + ε . \int_{B_R} B(\lambda|v_k|)\,dx \le(\varepsilon+C_*)\int_{B_R} A(|v_k|)\,dx+\varepsilon\int_{\mathbb{R}^n} A_{n/s}\left(\frac{|v_k|}{K}\right)\,dx \le(\varepsilon+C_*)\int_{B_R} A(|v_k|)\,dx+\varepsilon. ∫ B R B ( λ ∣ v k ∣ ) d x ≤ ( ε + C ∗ ) ∫ B R A ( ∣ v k ∣ ) d x + ε ∫ R n A n / s ( K ∣ v k ∣ ) d x ≤ ( ε + C ∗ ) ∫ B R A ( ∣ v k ∣ ) d x + ε .
By Step 3 the first term tends to 0 0 0 , so lim sup k ∫ R n B ( λ ∣ v k ∣ ) d x ≤ ε ( C 0 + 1 ) \limsup_{k}\int_{\mathbb{R}^n} B(\lambda|v_k|)\,dx\le\varepsilon(C_0+1) lim sup k ∫ R n B ( λ ∣ v k ∣ ) d x ≤ ε ( C 0 + 1 ) . As ε \varepsilon ε is arbitrary,
∫ R n B ( λ ∣ v k ∣ ) d x → 0 \int_{\mathbb{R}^n} B(\lambda|v_k|)\,dx\to0 ∫ R n B ( λ ∣ v k ∣ ) d x → 0
for every λ > 0 \lambda>0 λ > 0 . Taking λ = 1 / η \lambda=1/\eta λ = 1/ η gives ∥ v k ∥ L B ( R n ) ≤ η \|v_k\|_{L^B(\mathbb{R}^n)}\le\eta ∥ v k ∥ L B ( R n ) ≤ η for k k k large, for every η > 0 \eta>0 η > 0 . Hence u k → u u_k\to u u k → u in L B ( R n ) L^B(\mathbb{R}^n) L B ( R n ) .
To complement our analysis, we now establish a Ni-type decay inequality for radial functions in the potential space.
Theorem 4.3. Let s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ] and let A A A be an N N N -function such that 1 p − < s < n p + \frac{1}{p^-}<s<\frac{n}{p^+} p − 1 < s < p + n . Then, for u ∈ H r a d s , A ( R n ) u\in H^{s,A}_{rad}(\mathbb{R}^n) u ∈ H r a d s , A ( R n )
∣ u ( x ) ∣ ≤ C ∣ x ∣ s A − 1 ( ∣ x ∣ − n ) ∥ u ∥ H s , A ( R n ) . |u(x)|\le C|x|^s A^{-1}(|x|^{-n})\|u\|_{H^{s,A}(\mathbb{R}^n)}. ∣ u ( x ) ∣ ≤ C ∣ x ∣ s A − 1 ( ∣ x ∣ − n ) ∥ u ∥ H s , A ( R n ) . where C > 0 C>0 C > 0 depends on the bounds of A A A , n n n and s s s .
Before proving Theorem 4.3, we prove the following useful estimate for norms.
Lemma 4.4. Let r > 0 r>0 r > 0 , S r = { r / 2 < ∣ x ∣ < 2 r } S_r=\{r/2<|x|<2r\} S r = { r /2 < ∣ x ∣ < 2 r } . There is C 0 ≥ 1 C_0\ge1 C 0 ≥ 1 , independent of r r r , such that
∥ ∣ x ∣ s − n ∥ L A ^ ( R n ∖ B r ) ≤ C 0 r s A − 1 ( r − n ) if s < n / p + , s ∈ ( 0 , 1 ] , (17) \left\||x|^{s-n}\right\|_{L^{\widehat A}(\mathbb{R}^n\setminus B_r)}\le C_0r^sA^{-1}(r^{-n}) \qquad\text{if } s<n/p^+,\ s\in(0,1], \tag*{(17)} ∣ x ∣ s − n L A ( R n ∖ B r ) ≤ C 0 r s A − 1 ( r − n ) if s < n / p + , s ∈ ( 0 , 1 ] , (17) ∥ ∣ x ∣ − r ∥ L A ^ ( S r ) s − 1 ≤ C 0 r n − 1 + s A − 1 ( r − n ) if 1 / p − < s < 1. (18) \left\||x|-r\right\|_{L^{\widehat A}(S_r)}^{s-1}\le C_0r^{n-1+s}A^{-1}(r^{-n}) \qquad\text{if } 1/p^-<s<1. \tag*{(18)} ∥ ∣ x ∣ − r ∥ L A ( S r ) s − 1 ≤ C 0 r n − 1 + s A − 1 ( r − n ) if 1/ p − < s < 1. (18)
Proof. Since t ≤ A − 1 ( t ) A ^ − 1 ( t ) t\le A^{-1}(t)\widehat A^{-1}(t) t ≤ A − 1 ( t ) A − 1 ( t ) , it gives
1 r n A − 1 ( r − n ) ≤ A ^ − 1 ( r − n ) , A ^ ( A ^ − 1 ( r − n ) ) = r − n . (19) \frac{1}{r^nA^{-1}(r^{-n})}\le\widehat A^{-1}(r^{-n}),\qquad\widehat A\left(\widehat A^{-1}(r^{-n})\right)=r^{-n}. \tag*{(19)} r n A − 1 ( r − n ) 1 ≤ A − 1 ( r − n ) , A ( A − 1 ( r − n ) ) = r − n . (19)
Let us prove (17) . Let λ = C 0 r s A ^ − 1 ( r − n ) \lambda=C_0r^s\widehat{A}^{-1}(r^{-n}) λ = C 0 r s A − 1 ( r − n ) , C 0 ≥ 1 C_0\geq1 C 0 ≥ 1 . With x = r y x=ry x = r y ,
∫ R n ∖ B r A ^ ( ∣ x ∣ s − n λ ) d x = r n ∫ ∣ y ∣ > 1 A ^ ( ∣ y ∣ s − n C 0 r n A ^ − 1 ( r − n ) ) d y ≤ r n ∫ ∣ y ∣ > 1 A ^ ( C 0 − 1 ∣ y ∣ s − n A ^ − 1 ( r − n ) ) d y ≤ C 0 p + p + − 1 ∫ ∣ y ∣ > 1 ∣ y ∣ ( s − n ) p + p + − 1 d y . \begin{aligned}
\int_{\mathbb{R}^n\setminus B_r}\widehat{A}\left(\frac{|x|^{s-n}}{\lambda}\right)\,dx=r^n\int_{|y|>1}\widehat{A}\left(\frac{|y|^{s-n}}{C_0r^n\widehat{A}^{-1}(r^{-n})}\right)\,dy\leq r^n\int_{|y|>1}\widehat{A}\left(C_0^{-1}|y|^{s-n}\widehat{A}^{-1}(r^{-n})\right)\,dy \\
&\leq C_0^{\frac{p^+}{p^+-1}}\int_{|y|>1}|y|^{(s-n)\frac{p^+}{p^+-1}}\,dy.
\end{aligned} ∫ R n ∖ B r A ( λ ∣ x ∣ s − n ) d x = r n ∫ ∣ y ∣ > 1 A ( C 0 r n A − 1 ( r − n ) ∣ y ∣ s − n ) d y ≤ r n ∫ ∣ y ∣ > 1 A ( C 0 − 1 ∣ y ∣ s − n A − 1 ( r − n ) ) d y ≤ C 0 p + − 1 p + ∫ ∣ y ∣ > 1 ∣ y ∣ ( s − n ) p + − 1 p + d y .
The last integral is equal to n ω n ( n − s ) p + p + − 1 − n \frac{n\omega_n}{(n-s)\frac{p^+}{p^+-1}-n} ( n − s ) p + − 1 p + − n n ω n , which is finite provided that ( n − s ) p + p + − 1 > n (n-s)\frac{p^+}{p^+-1}>n ( n − s ) p + − 1 p + > n , i.e. s < n / p + s<n/p^+ s < n / p + . Taking C 0 ≥ 1 C_0\geq1 C 0 ≥ 1 large enough, we get the result.
Let us prove (18) . Let λ = C 0 r n − 1 + s A ^ − 1 ( r − n ) \lambda=C_0r^{n-1+s}\widehat{A}^{-1}(r^{-n}) λ = C 0 r n − 1 + s A − 1 ( r − n ) . Using polar coordinates, ρ = r ( 1 + τ ) \rho=r(1+\tau) ρ = r ( 1 + τ ) with τ ∈ ( − 1 2 , 1 ) \tau\in\left(-\frac{1}{2},1\right) τ ∈ ( − 2 1 , 1 ) and d ρ = r d τ d\rho=r\,d\tau d ρ = r d τ ,
∫ S r A ^ ( ∣ ∣ x ∣ − r ∣ s − 1 λ ) d x ≤ n ω n ( 2 r ) n − 1 ∫ r / 2 2 r A ^ ( ∣ r − ρ ∣ s − 1 λ ) d ρ = C r n ∫ − 1 / 2 1 A ^ ( ∣ τ ∣ s − 1 C 0 r n A ^ − 1 ( r − n ) ) d τ ≤ C r n ∫ − 1 / 2 1 A ^ ( C 0 − 1 ∣ τ ∣ s − 1 A ^ − 1 ( r − n ) ) d τ ≤ C ( C 0 p + p + − 1 + C 0 − p − p − − 1 ) ∫ − 1 / 2 1 ( ∣ τ ∣ − ( 1 − s ) p + p + − 1 + ∣ τ ∣ − ( 1 − s ) p − p − − 1 ) d τ . \begin{aligned}
\int_{S_r}\widehat{A}\left(\frac{\left||x|-r\right|^{s-1}}{\lambda}\right)\,dx\leq n\omega_n(2r)^{n-1}\int_{r/2}^{2r}\widehat{A}\left(\frac{|r-\rho|^{s-1}}{\lambda}\right)\,d\rho=C r^n\int_{-1/2}^{1}\widehat{A}\left(\frac{|\tau|^{s-1}}{C_0r^n\widehat{A}^{-1}(r^{-n})}\right)\,d\tau\\
&\leq Cr^n\int_{-1/2}^{1}\widehat{A}\left(C_0^{-1}|\tau|^{s-1}\widehat{A}^{-1}(r^{-n})\right)\,d\tau\\
&\leq C\left(C_0^{\frac{p^+}{p^+-1}}+C_0^{-\frac{p^-}{p^--1}}\right)\int_{-1/2}^{1}\left(|\tau|^{-(1-s)\frac{p^+}{p^+-1}}+|\tau|^{-(1-s)\frac{p^-}{p^--1}}\right)\,d\tau.
\end{aligned} ∫ S r A ( λ ∣ ∣ x ∣ − r ∣ s − 1 ) d x ≤ n ω n ( 2 r ) n − 1 ∫ r /2 2 r A ( λ ∣ r − ρ ∣ s − 1 ) d ρ = C r n ∫ − 1/2 1 A ( C 0 r n A − 1 ( r − n ) ∣ τ ∣ s − 1 ) d τ ≤ C r n ∫ − 1/2 1 A ( C 0 − 1 ∣ τ ∣ s − 1 A − 1 ( r − n ) ) d τ ≤ C ( C 0 p + − 1 p + + C 0 − p − − 1 p − ) ∫ − 1/2 1 ( ∣ τ ∣ − ( 1 − s ) p + − 1 p + + ∣ τ ∣ − ( 1 − s ) p − − 1 p − ) d τ .
The integral is finite if and only if ( 1 − s ) p − p − − 1 < 1 (1-s)\frac{p^-}{p^--1}<1 ( 1 − s ) p − − 1 p − < 1 , i.e., s > 1 / p − s>1/p^- s > 1/ p − . Taking C 0 C_0 C 0 large the whole expression is lower than 1, which proves (18) .
Proof of Theorem 4.3 . Assume first that s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) . Let 0 < s ′ < s 0<s'<s 0 < s ′ < s such that s ′ p − > 1 s'p^->1 s ′ p − > 1 . Let u ∈ H r a d s , A ( R n ) u\in H_{\mathrm{rad}}^{s,A}(\mathbb{R}^n) u ∈ H rad s , A ( R n ) . Then by [28 Theorem 3.7] the embedding H s , A ( R n ) ⊂ W s ′ , A ( R n ) H^{s,A}(\mathbb{R}^n)\subset W^{s',A}(\mathbb{R}^n) H s , A ( R n ) ⊂ W s ′ , A ( R n ) holds and so u ∈ W r a d s ′ , A ( R n ) u\in W_{\mathrm{rad}}^{s',A}(\mathbb{R}^n) u ∈ W rad s ′ , A ( R n ) . Since s ′ p − > 1 s'p^->1 s ′ p − > 1 , by [6 Theorem 4.2] u u u is almost everywhere equal to a continuous function in R n ∖ { 0 } \mathbb{R}^n\setminus\{0\} R n ∖ { 0 } .
Since u ∈ H r a d s , A ( R n ) u\in H_{\mathrm{rad}}^{s,A}(\mathbb{R}^n) u ∈ H rad s , A ( R n ) , by [28 Lemma 4.1] there is f ∈ L A ( R n ) f\in L^A(\mathbb{R}^n) f ∈ L A ( R n ) radial such that
u ( x ) = ∫ R n f ( y ) G s ( x − y ) d y = ∫ 0 ∞ f ( ρ ) ρ n − 1 ( ∫ S n − 1 G s ( r e 1 − ρ ω ) d σ ω ) d ρ , u(x)=\int_{\mathbb{R}^n}f(y)G_s(x-y)\,dy=\int_0^\infty f(\rho)\rho^{n-1}\left(\int_{S^{n-1}}G_s(re_1-\rho\omega)\,d\sigma_\omega\right)\,d\rho, u ( x ) = ∫ R n f ( y ) G s ( x − y ) d y = ∫ 0 ∞ f ( ρ ) ρ n − 1 ( ∫ S n − 1 G s ( r e 1 − ρ ω ) d σ ω ) d ρ ,
where r = ∣ x ∣ r=|x| r = ∣ x ∣ . In polar coordinates we can write
∣ u ( x ) ∣ ≤ ∫ 0 ∞ ∣ f ( ρ ) ∣ ρ n − 1 K ( r , ρ ) d ρ where K ( r , ρ ) : = ∫ S n − 1 G s ( r e 1 − ρ ω ) d σ ω . |u(x)|\leq\int_0^\infty|f(\rho)|\rho^{n-1}K(r,\rho)\,d\rho\qquad\text{where }K(r,\rho):=\int_{S^{n-1}}G_s(re_1-\rho\omega)\,d\sigma_\omega. ∣ u ( x ) ∣ ≤ ∫ 0 ∞ ∣ f ( ρ ) ∣ ρ n − 1 K ( r , ρ ) d ρ where K ( r , ρ ) := ∫ S n − 1 G s ( r e 1 − ρ ω ) d σ ω .
We claim that
K ( r , ρ ) ≤ C max { r , ρ } 1 − n ∣ r − ρ ∣ s − 1 . K(r,\rho)\leq C\max\{r,\rho\}^{1-n}|r-\rho|^{s-1}. K ( r , ρ ) ≤ C max { r , ρ } 1 − n ∣ r − ρ ∣ s − 1 .
Indeed, since G s ( z ) ≤ c 1 ∣ z ∣ s − n G_s(z)\leq c_1|z|^{s-n} G s ( z ) ≤ c 1 ∣ z ∣ s − n , it suffices to bound J ( r , ρ ) = ∫ S n − 1 ∣ r e 1 − ρ ω ∣ s − n d σ ω J(r,\rho)=\int_{S^{n-1}}|re_1-\rho\omega|^{s-n}\,d\sigma_\omega J ( r , ρ ) = ∫ S n − 1 ∣ r e 1 − ρ ω ∣ s − n d σ ω .
Case ρ ∉ ( r / 2 , 2 r ) \rho\notin(r/2,2r) ρ ∈ / ( r /2 , 2 r ) . Then ∣ r e 1 − ρ ω ∣ ≥ ∣ r − ρ ∣ ≥ max { r , ρ } / 2 |re_1-\rho\omega|\geq|r-\rho|\geq\max\{r,\rho\}/2 ∣ r e 1 − ρ ω ∣ ≥ ∣ r − ρ ∣ ≥ max { r , ρ } /2 , so J ≤ C max { r , ρ } s − n J\leq C\max\{r,\rho\}^{s-n} J ≤ C max { r , ρ } s − n . Since ∣ r − ρ ∣ ≤ max { r , ρ } |r-\rho|\leq\max\{r,\rho\} ∣ r − ρ ∣ ≤ max { r , ρ } and s − 1 < 0 s-1<0 s − 1 < 0 ,
max { r , ρ } s − n = max { r , ρ } 1 − n max { r , ρ } s − 1 ≤ max { r , ρ } 1 − n ∣ r − ρ ∣ s − 1 . \max\{r,\rho\}^{s-n}=\max\{r,\rho\}^{1-n}\max\{r,\rho\}^{s-1}\leq\max\{r,\rho\}^{1-n}|r-\rho|^{s-1}. max { r , ρ } s − n = max { r , ρ } 1 − n max { r , ρ } s − 1 ≤ max { r , ρ } 1 − n ∣ r − ρ ∣ s − 1 .
Case ρ ∈ ( r / 2 , 2 r ) \rho\in(r/2,2r) ρ ∈ ( r /2 , 2 r ) . Let θ ∈ [ 0 , π ] \theta\in[0,\pi] θ ∈ [ 0 , π ] be the angle between e 1 e_1 e 1 and ω \omega ω . Then
∣ r e 1 − ρ ω ∣ 2 = ( r − ρ ) 2 + 2 r ρ ( 1 − cos θ ) ≥ ( r − ρ ) 2 + 4 π 2 r ρ θ 2 , |re_1-\rho\omega|^2=(r-\rho)^2+2r\rho(1-\cos\theta)\ge(r-\rho)^2+\frac{4}{\pi^2}r\rho\theta^2, ∣ r e 1 − ρ ω ∣ 2 = ( r − ρ ) 2 + 2 r ρ ( 1 − cos θ ) ≥ ( r − ρ ) 2 + π 2 4 r ρ θ 2 ,
and d σ ≤ C θ n − 2 d θ d\sigma\le C\theta^{n-2}d\theta d σ ≤ C θ n − 2 d θ . Substituting θ = ∣ r − ρ ∣ t / r ρ \theta=|r-\rho|t/\sqrt{r\rho} θ = ∣ r − ρ ∣ t / r ρ ,
J ≤ C ( r ρ ) − n − 1 2 ∣ r − ρ ∣ s − 1 ∫ 0 ∞ ( 1 + t 2 ) s − n 2 t n − 2 d t ≤ C r 1 − n ∣ r − ρ ∣ s − 1 . (20) J\le C(r\rho)^{-\frac{n-1}{2}}|r-\rho|^{s-1}\int_0^\infty(1+t^2)^{\frac{s-n}{2}}t^{n-2}\,dt\le Cr^{1-n}|r-\rho|^{s-1}. \tag*{(20)} J ≤ C ( r ρ ) − 2 n − 1 ∣ r − ρ ∣ s − 1 ∫ 0 ∞ ( 1 + t 2 ) 2 s − n t n − 2 d t ≤ C r 1 − n ∣ r − ρ ∣ s − 1 . (20)
The integral is finite because the integrand behaves like t s − 2 t^{s-2} t s − 2 at infinity and s < 1 s<1 s < 1 .
Therefore, we obtain the pointwise inequality
∣ u ( x ) ∣ ≤ ∫ 0 ∞ ∣ f ( ρ ) ∣ ρ n − 1 K ( r , ρ ) d ρ = : I 1 + I 2 + I 3 , |u(x)|\le\int_0^\infty|f(\rho)|\rho^{n-1}K(r,\rho)\,d\rho=:I_1+I_2+I_3, ∣ u ( x ) ∣ ≤ ∫ 0 ∞ ∣ f ( ρ ) ∣ ρ n − 1 K ( r , ρ ) d ρ =: I 1 + I 2 + I 3 ,
where I 1 , I 2 , I 3 I_1,I_2,I_3 I 1 , I 2 , I 3 correspond to ρ < r / 2 \rho<r/2 ρ < r /2 , r / 2 ≤ ρ ≤ 2 r r/2\le\rho\le2r r /2 ≤ ρ ≤ 2 r and ρ > 2 r \rho>2r ρ > 2 r , respectively.
Bound for I 1 I_1 I 1 , (i.e., ρ < r / 2 \rho<r/2 ρ < r /2 ). Here max { r , ρ } = r \max\{r,\rho\}=r max { r , ρ } = r and ∣ r − ρ ∣ ≥ r / 2 |r-\rho|\ge r/2 ∣ r − ρ ∣ ≥ r /2 , so the kernel is bounded by C ρ n − 1 r 1 − n r s − 1 = C ρ n − 1 r s − n C\rho^{n-1}r^{1-n}r^{s-1}=C\rho^{n-1}r^{s-n} C ρ n − 1 r 1 − n r s − 1 = C ρ n − 1 r s − n , and
I 1 ≤ C r s − n ∫ B r ∣ f ∣ d x ≤ 2 C r s − n ∥ f ∥ L A ( B r ) ∥ χ B r ∥ L A ^ ≤ C r s A − 1 ( r − n ) ∥ f ∥ L A ( B r ) , I_1\le Cr^{s-n}\int_{B_r}|f|\,dx\le2Cr^{s-n}\|f\|_{L^A(B_r)}\|\chi_{B_r}\|_{L^{\widehat A}}\le Cr^sA^{-1}(r^{-n})\|f\|_{L^A(B_r)}, I 1 ≤ C r s − n ∫ B r ∣ f ∣ d x ≤ 2 C r s − n ∥ f ∥ L A ( B r ) ∥ χ B r ∥ L A ≤ C r s A − 1 ( r − n ) ∥ f ∥ L A ( B r ) ,
since ∥ χ B r ∥ L A ^ = 1 / A ^ − 1 ( 1 / ∣ B r ∣ ) ≤ ∣ B r ∣ A − 1 ( 1 / ∣ B r ∣ ) ≤ C r n A − 1 ( r − n ) \|\chi_{B_r}\|_{L^{\widehat A}}=1/\widehat A^{-1}(1/|B_r|)\le|B_r|A^{-1}(1/|B_r|)\le Cr^nA^{-1}(r^{-n}) ∥ χ B r ∥ L A = 1/ A − 1 ( 1/∣ B r ∣ ) ≤ ∣ B r ∣ A − 1 ( 1/∣ B r ∣ ) ≤ C r n A − 1 ( r − n ) (by concavity of A − 1 A^{-1} A − 1 : A − 1 ( λ t ) ≤ max { 1 , λ } A − 1 ( t ) A^{-1}(\lambda t)\le\max\{1,\lambda\}A^{-1}(t) A − 1 ( λ t ) ≤ max { 1 , λ } A − 1 ( t ) ).
Bound for I 3 I_3 I 3 (i.e., ρ > 2 r \rho>2r ρ > 2 r ). Here max { r , ρ } = ρ \max\{r,\rho\}=\rho max { r , ρ } = ρ and ∣ r − ρ ∣ ≥ ρ / 2 |r-\rho|\ge\rho/2 ∣ r − ρ ∣ ≥ ρ /2 , so the kernel is ≤ C ρ n − 1 ρ 1 − n ρ s − 1 = C ρ s − 1 \le C\rho^{n-1}\rho^{1-n}\rho^{s-1}=C\rho^{s-1} ≤ C ρ n − 1 ρ 1 − n ρ s − 1 = C ρ s − 1 . Then, by Hölder's inequality in Orlicz spaces and (17) ,
I 3 ≤ C ∫ 2 r ∞ ∣ f ( ρ ) ∣ ρ s − 1 d ρ = C ∫ ∣ x ∣ > 2 r ∣ f ( x ) ∣ ∣ x ∣ s − n d x ≤ 2 C ′ ∥ f ∥ L A ( R n ∖ B 2 r ) ∥ ∣ x ∣ s − n ∥ L A ^ ( R n ∖ B r ) ≤ C r s A − 1 ( r − n ) ∥ f ∥ L A ( R n ∖ B 2 r ) . \begin{aligned}
I_3&\le C\int_{2r}^\infty|f(\rho)|\rho^{s-1}\,d\rho=C\int_{|x|>2r}|f(x)||x|^{s-n}\,dx\\
&\le2C'\|f\|_{L^A(\mathbb{R}^n\setminus B_{2r})}\||x|^{s-n}\|_{L^{\widehat A}(\mathbb{R}^n\setminus B_r)}\\
&\le Cr^sA^{-1}(r^{-n})\|f\|_{L^A(\mathbb{R}^n\setminus B_{2r})}.
\end{aligned} I 3 ≤ C ∫ 2 r ∞ ∣ f ( ρ ) ∣ ρ s − 1 d ρ = C ∫ ∣ x ∣ > 2 r ∣ f ( x ) ∣∣ x ∣ s − n d x ≤ 2 C ′ ∥ f ∥ L A ( R n ∖ B 2 r ) ∥∣ x ∣ s − n ∥ L A ( R n ∖ B r ) ≤ C r s A − 1 ( r − n ) ∥ f ∥ L A ( R n ∖ B 2 r ) .
Bound for I 2 I_2 I 2 (i.e., r / 2 ≤ ρ ≤ 2 r r/2\le\rho\le2r r /2 ≤ ρ ≤ 2 r ). By (20) and (18) ,
I 2 ≤ C ∫ r / 2 2 r ∣ f ( ρ ) ∣ ρ n − 1 r 1 − n ∣ r − ρ ∣ s − 1 d ρ I_2\le C\int_{r/2}^{2r}|f(\rho)|\rho^{n-1}r^{1-n}|r-\rho|^{s-1}\,d\rho I 2 ≤ C ∫ r /2 2 r ∣ f ( ρ ) ∣ ρ n − 1 r 1 − n ∣ r − ρ ∣ s − 1 d ρ
= C r 1 − n ∫ S r ∣ f ( y ) ∣ ∣ r − ∣ y ∣ ∣ s − 1 d y ≤ 2 C r 1 − n ∥ f ∥ L A ( S r ) ∥ ∣ r − ∣ y ∣ ∣ s − 1 ∥ L A ^ ( S r ) ≤ C r s A − 1 ( r − n ) ∥ f ∥ L A ( S r ) . (21) \begin{aligned}
&=Cr^{1-n}\int_{S_r}|f(y)||r-|y||^{s-1}\,dy\\
&\le2Cr^{1-n}\|f\|_{L^A(S_r)}\||r-|y||^{s-1}\|_{L^{\widehat A}(S_r)}\\
&\le Cr^sA^{-1}(r^{-n})\|f\|_{L^A(S_r)}. \tag*{(21)}
\end{aligned} = C r 1 − n ∫ S r ∣ f ( y ) ∣∣ r − ∣ y ∣ ∣ s − 1 d y ≤ 2 C r 1 − n ∥ f ∥ L A ( S r ) ∥∣ r − ∣ y ∣ ∣ s − 1 ∥ L A ( S r ) ≤ C r s A − 1 ( r − n ) ∥ f ∥ L A ( S r ) . (21)
Therefore, from the bounds for I 1 I_1 I 1 , I 2 I_2 I 2 and I 3 I_3 I 3 the result follows for s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) .
Assume now that s = 1 s=1 s = 1 . By [28 Theorems 3.4 and 3.5] , H 1 , A ( R n ) = W 1 , A ( R n ) H^{1,A}(\mathbb{R}^n)=W^{1,A}(\mathbb{R}^n) H 1 , A ( R n ) = W 1 , A ( R n ) with equivalent norms. By density, assume that u u u is smooth and has compact support. Write u ( x ) = u ( ∣ x ∣ ) = u(x)=u(|x|)= u ( x ) = u ( ∣ x ∣ ) = u ( r ) u(r) u ( r ) . Moreover, u ( R ) = 0 u(R)=0 u ( R ) = 0 for R > 0 R>0 R > 0 large enough. Observe that u ( r ) = − ∫ r R u ′ ( t ) d t u(r)=-\int_r^R u'(t)\,dt u ( r ) = − ∫ r R u ′ ( t ) d t . Consider the measure d μ = t n − 1 d t d\mu=t^{n-1}dt d μ = t n − 1 d t . Then, Hölder’s inequality yields
∣ u ( r ) ∣ ≤ ∫ r R ∣ u ′ ( t ) ∣ d t = ∫ r R ∣ u ′ ( t ) ∣ t n − 1 t 1 − n d t = ∫ r R ∣ u ′ ( t ) ∣ t 1 − n d μ |u(r)|\leq\int_r^R|u'(t)|\,dt=\int_r^R|u'(t)|t^{n-1}t^{1-n}\,dt=\int_r^R|u'(t)|t^{1-n}\,d\mu ∣ u ( r ) ∣ ≤ ∫ r R ∣ u ′ ( t ) ∣ d t = ∫ r R ∣ u ′ ( t ) ∣ t n − 1 t 1 − n d t = ∫ r R ∣ u ′ ( t ) ∣ t 1 − n d μ
≤ 2 ∥ u ′ ∥ L A ( ( 0 , R ) ; d μ ) ∥ t 1 − n ∥ L A ^ ( ( r , R ) , d μ ) . \leq2\|u'\|_{L^A((0,R);d\mu)}\|t^{1-n}\|_{L^{\widehat A}((r,R),d\mu)}. ≤ 2∥ u ′ ∥ L A (( 0 , R ) ; d μ ) ∥ t 1 − n ∥ L A (( r , R ) , d μ ) .
Using polar coordinates, we get that
ω n ∫ 0 R A ( ∣ u ′ ( t ) ∣ λ ) t n − 1 d t = ∫ 0 R ∫ S n − 1 A ( ∣ u ′ ( t ) ∣ λ ) t n − 1 d σ d t = ∫ B R A ( ∣ ∇ u ( x ) ∣ λ ) d x . \omega_n\int_0^R A\left(\frac{|u'(t)|}{\lambda}\right)t^{n-1}\,dt=\int_0^R\int_{S^{n-1}}A\left(\frac{|u'(t)|}{\lambda}\right)t^{n-1}\,d\sigma dt=\int_{B_R}A\left(\frac{|\nabla u(x)|}{\lambda}\right)\,dx. ω n ∫ 0 R A ( λ ∣ u ′ ( t ) ∣ ) t n − 1 d t = ∫ 0 R ∫ S n − 1 A ( λ ∣ u ′ ( t ) ∣ ) t n − 1 d σ d t = ∫ B R A ( λ ∣∇ u ( x ) ∣ ) d x .
from where it is obtained that
∫ 0 R A ( ∣ u ′ ( t ) ∣ λ ) t n − 1 d t = 1 ⟺ ∫ B R 1 ω n A ( ∣ ∇ u ( x ) ∣ λ ) d x = 1. \int_0^R A\left(\frac{|u'(t)|}{\lambda}\right)t^{n-1}\,dt=1\Longleftrightarrow\int_{B_R}\frac{1}{\omega_n}A\left(\frac{|\nabla u(x)|}{\lambda}\right)\,dx=1. ∫ 0 R A ( λ ∣ u ′ ( t ) ∣ ) t n − 1 d t = 1 ⟺ ∫ B R ω n 1 A ( λ ∣∇ u ( x ) ∣ ) d x = 1.
This gives
∥ u ′ ∥ L A ( ( 0 , R ) ; d μ ) = ∥ ∇ u ∥ L 1 ω n A ( B R ) ≤ c n ∥ ∇ u ∥ L A ( B R ) . \|u'\|_{L^A((0,R);d\mu)}=\|\nabla u\|_{L^{\frac{1}{\omega_n}A}(B_R)}\leq c_n\|\nabla u\|_{L^A(B_R)}. ∥ u ′ ∥ L A (( 0 , R ) ; d μ ) = ∥∇ u ∥ L ω n 1 A ( B R ) ≤ c n ∥∇ u ∥ L A ( B R ) .
Similarly,
∥ t 1 − n ∥ L A ^ ( ( r , R ) , d μ ) = ∥ ∣ x ∣ 1 − n ∥ L 1 ω n A ^ ( B R ∖ B r ) ≤ c n ∥ ∣ x ∣ 1 − n ∥ L A ^ ( B R ∖ B r ) . \|t^{1-n}\|_{L^{\widehat A}((r,R),d\mu)}=\||x|^{1-n}\|_{L^{\frac{1}{\omega_n}\widehat A}(B_R\setminus B_r)}\leq c_n\||x|^{1-n}\|_{L^{\widehat A}(B_R\setminus B_r)}. ∥ t 1 − n ∥ L A (( r , R ) , d μ ) = ∥∣ x ∣ 1 − n ∥ L ω n 1 A ( B R ∖ B r ) ≤ c n ∥∣ x ∣ 1 − n ∥ L A ( B R ∖ B r ) .
Hence, by (17) ,
∣ u ( x ) ∣ ≤ C ∥ ∇ u ∥ L A ( B R ) ∥ ∣ x ∣ 1 − n ∥ L A ^ ( B R ∖ B r ) ≤ C r A − 1 ( r − n ) ∥ u ∥ H 1 , A ( R n ) |u(x)|\leq C\|\nabla u\|_{L^A(B_R)}\||x|^{1-n}\|_{L^{\widehat A}(B_R\setminus B_r)}\leq CrA^{-1}(r^{-n})\|u\|_{H^{1,A}(\mathbb{R}^n)} ∣ u ( x ) ∣ ≤ C ∥∇ u ∥ L A ( B R ) ∥∣ x ∣ 1 − n ∥ L A ( B R ∖ B r ) ≤ C r A − 1 ( r − n ) ∥ u ∥ H 1 , A ( R n )
and the result follows for s = 1 s=1 s = 1 as well.
The following theorem is an application of the Ni-type inequality Theorem 4.3 to compactness of the embedding between potential spaces and weighted Orlicz-Lebesgue spaces.
Theorem 4.5. Assume that A A A , A ^ \widehat A A , and B B B are N N N -functions satisfying the Δ 2 \Delta_2 Δ 2 -condition. Assume that s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ] is such that 1 p A − < s < n p A + \frac{1}{p_A^-}<s<\frac{n}{p_A^+} p A − 1 < s < p A + n and that
p A + < p B − . (22) p_A^+<p_B^-. \tag*{(22)} p A + < p B − . (22) Let α > − n \alpha>-n α > − n . Suppose further that
( p B − − p A + ) ( 1 − n p A + ) + α < 0. (23) \left(p_B^--p_A^+\right)\left(\frac{1-n}{p_A^+}\right)+\alpha<0. \tag*{(23)} ( p B − − p A + ) ( p A + 1 − n ) + α < 0. (23) and p B + < ( α + n ) p A − n − s p A + p_B^+<\frac{(\alpha+n)p_A^-}{n-sp_A^+} p B + < n − s p A + ( α + n ) p A − . Then, the embedding H r a d s , A ( R n ) ↪ L B ( R n , ∣ x ∣ α d x ) H_{\mathrm{rad}}^{s,A}(\mathbb{R}^n)\hookrightarrow L^B(\mathbb{R}^n,|x|^\alpha\,dx) H rad s , A ( R n ) ↪ L B ( R n , ∣ x ∣ α d x ) is compact.
Remark 4.6. Observe that, for appropriate α \alpha α , the N N N -function B = A n / s B=A_{n/s} B = A n / s satisfies the above conditions under the additional assumptions that p + < ( p − ) ∗ p^+<(p^-)^* p + < ( p − ) ∗ and
n − s p − n − s p + n p + p − < n − 1 n − s p − n p − p + + 1. (24) \frac{n-sp^-}{n-sp^+}\frac{np^+}{p^-}<\frac{n-1}{n-sp^-}\frac{np^-}{p^+}+1. \tag*{(24)} n − s p + n − s p − p − n p + < n − s p − n − 1 p + n p − + 1. (24)
Indeed, we have
p B + = ( p + ) ∗ : = n p + n − s p + and p A − = p − . p_B^+=(p^+)^*:=\frac{np^+}{n-sp^+}\qquad\text{and}\qquad p_A^-=p^-. p B + = ( p + ) ∗ := n − s p + n p + and p A − = p − .
Assumptions p B + < ( α + n ) p A − n − s p A − p_B^+ < \frac{(\alpha+n)p_A^-}{n-sp_A^-} p B + < n − s p A − ( α + n ) p A − and (23) are possible if we can find α \alpha α verifying
( p + ) ∗ ( n − s p − p − ) − n < α < ( ( p ) ∗ − p + ) ( n − 1 p + ) . (25) (p^+)^*\left(\frac{n-sp^-}{p^-}\right)-n<\alpha<((p)^*-p^+)\left(\frac{n-1}{p^+}\right). \tag*{(25)} ( p + ) ∗ ( p − n − s p − ) − n < α < (( p ) ∗ − p + ) ( p + n − 1 ) . (25)
As can be checked, this is true in view of (24) .
Proof. Let u k u_k u k be a sequence in H r a d s , A ( R n ) H_{\mathrm{rad}}^{s,A}(\mathbb{R}^n) H rad s , A ( R n ) such that u k ⇀ 0 u_k\rightharpoonup0 u k ⇀ 0 . Write for 0 < ε < δ ≤ 1 0<\varepsilon<\delta\leq1 0 < ε < δ ≤ 1 ,
∫ R n B ( ∣ u k ( x ) ∣ ) ∣ x ∣ α d x = ∫ ∣ x ∣ < ε B ( ∣ u k ( x ) ∣ ) ∣ x ∣ α d x + ∫ ε ≤ ∣ x ∣ < 1 / ε B ( ∣ u k ( x ) ∣ ) ∣ x ∣ α d x + ∫ ∣ x ∣ ≥ 1 / ε B ( ∣ u k ( x ) ∣ ) ∣ x ∣ α d x = : I 1 + I 2 + I 3 . (26) \begin{aligned}
\int_{\mathbb{R}^n}B(|u_k(x)|)|x|^\alpha\,dx=\int_{|x|<\varepsilon}B(|u_k(x)|)|x|^\alpha\,dx+\int_{\varepsilon\leq|x|<1/\varepsilon}B(|u_k(x)|)|x|^\alpha\,dx \\
&+\int_{|x|\geq1/\varepsilon}B(|u_k(x)|)|x|^\alpha\,dx=:I_1+I_2+I_3. \tag*{(26)}
\end{aligned} ∫ R n B ( ∣ u k ( x ) ∣ ) ∣ x ∣ α d x = ∫ ∣ x ∣ < ε B ( ∣ u k ( x ) ∣ ) ∣ x ∣ α d x + ∫ ε ≤ ∣ x ∣ < 1/ ε B ( ∣ u k ( x ) ∣ ) ∣ x ∣ α d x + ∫ ∣ x ∣ ≥ 1/ ε B ( ∣ u k ( x ) ∣ ) ∣ x ∣ α d x =: I 1 + I 2 + I 3 . (26)
Bound for I 1 I_1 I 1 . Observe that s p A − < n sp_A^-<n s p A − < n , then by Ni’s Theorem 4.3,
I 1 ≤ ∫ ∣ x ∣ < ε B ( C ∣ x ∣ s − n A − 1 ( ∣ x ∣ − n ) ∥ u k ∥ H s , A ( R n ) ) ∣ x ∣ α d x ≤ C n ∫ 0 ε B ( r s − n A − 1 ( r − n ) ) r n + α − 1 d r ≤ C n ∫ 0 ε B ( r s − n p A − ) r n + α − 1 d r ≤ C n ∫ 0 ε r p B + ( s − n p A − ) + α + n − 1 d r . (27) \begin{aligned}
I_1\leq\int_{|x|<\varepsilon}B\left(C|x|^{s-n}A^{-1}(|x|^{-n})\|u_k\|_{H^{s,A}(\mathbb{R}^n)}\right)|x|^\alpha\,dx\leq C_n\int_0^\varepsilon B\left(r^{s-n}A^{-1}(r^{-n})\right)r^{n+\alpha-1}\,dr \\
&\leq C_n\int_0^\varepsilon B\left(r^{s-\frac{n}{p_A^-}}\right)r^{n+\alpha-1}\,dr\leq C_n\int_0^\varepsilon r^{p_B^+\left(s-\frac{n}{p_A^-}\right)+\alpha+n-1}\,dr. \tag*{(27)}
\end{aligned} I 1 ≤ ∫ ∣ x ∣ < ε B ( C ∣ x ∣ s − n A − 1 ( ∣ x ∣ − n ) ∥ u k ∥ H s , A ( R n ) ) ∣ x ∣ α d x ≤ C n ∫ 0 ε B ( r s − n A − 1 ( r − n ) ) r n + α − 1 d r ≤ C n ∫ 0 ε B ( r s − p A − n ) r n + α − 1 d r ≤ C n ∫ 0 ε r p B + ( s − p A − n ) + α + n − 1 d r . (27)
where we have used the inequality A − 1 ( t ) ≤ t 1 / p A − A^{-1}(t)\leq t^{1/p_A^-} A − 1 ( t ) ≤ t 1/ p A − for t > 1 t>1 t > 1 . Since p B + < ( α + n ) p A − n − s p A − p_B^+<\frac{(\alpha+n)p_A^-}{n-sp_A^-} p B + < n − s p A − ( α + n ) p A − and α > − n \alpha>-n α > − n , we get
p B + ( s − n p A − ) + α + n > ( α + n ) p A − n − s p A − ( s − n p A − ) + α + n = 0. p_B^+\left(s-\frac{n}{p_A^-}\right)+\alpha+n>\frac{(\alpha+n)p_A^-}{n-sp_A^-}\left(s-\frac{n}{p_A^-}\right)+\alpha+n=0. p B + ( s − p A − n ) + α + n > n − s p A − ( α + n ) p A − ( s − p A − n ) + α + n = 0.
Hence, from (27) ,
I 1 ≤ C ε p B + ( s − n p A − ) + α + n . (28) I_1\leq C\varepsilon^{p_B^+\left(s-\frac{n}{p_A^-}\right)+\alpha+n}. \tag*{(28)} I 1 ≤ C ε p B + ( s − p A − n ) + α + n . (28)
Bound for I 2 I_2 I 2 . First, observe that (22) implies that there is t 0 > 0 t_0>0 t 0 > 0 such that for all 0 < t < t 0 0<t<t_0 0 < t < t 0 , it follows B ( t ) ≤ A ( t ) B(t)\leq A(t) B ( t ) ≤ A ( t ) . Moreover, by Strauss Theorem 4.1 and the inequality
t ≤ A − 1 ( t ) A ^ − 1 ( t ) (29) t\leq A^{-1}(t)\widehat{A}^{-1}(t) \tag*{(29)} t ≤ A − 1 ( t ) A − 1 ( t ) (29)
we get for ε ≤ ∣ x ∣ < 1 / ε \varepsilon\leq|x|<1/\varepsilon ε ≤ ∣ x ∣ < 1/ ε that
∣ u k ( x ) ∣ ≤ C ∣ x ∣ 1 − n A ^ − 1 ( ∣ x ∣ 1 − n ) ≤ C A − 1 ( ∣ x ∣ 1 − n ) ≤ C A − 1 ( ε 1 − n ) ≤ C ε 1 − n p A − . |u_k(x)|\leq C\frac{|x|^{1-n}}{\widehat{A}^{-1}(|x|^{1-n})}\leq CA^{-1}(|x|^{1-n})\leq CA^{-1}(\varepsilon^{1-n})\leq C\varepsilon^{\frac{1-n}{p_A^-}}. ∣ u k ( x ) ∣ ≤ C A − 1 ( ∣ x ∣ 1 − n ) ∣ x ∣ 1 − n ≤ C A − 1 ( ∣ x ∣ 1 − n ) ≤ C A − 1 ( ε 1 − n ) ≤ C ε p A − 1 − n .
So, there is a constant C ( ε ) > 1 C(\varepsilon)>1 C ( ε ) > 1 such that for t 0 ≤ t ≤ C ε 1 − n p A − t_0\leq t\leq C\varepsilon^{\frac{1-n}{p_A^-}} t 0 ≤ t ≤ C ε p A − 1 − n ,
B ( t ) ≤ C ( ε ) B ~ ( t ) B(t)\leq C(\varepsilon)\widetilde{B}(t) B ( t ) ≤ C ( ε ) B ( t )
where B ~ \widetilde{B} B is any N N N -functions verifying the assumptions of Theorem 4.2. In this way, we get for t ≤ C ε 1 − n p A − t\leq C\varepsilon^{\frac{1-n}{p_A^-}} t ≤ C ε p A − 1 − n , that
B ( t ) ≤ C ( ε ) ( A ( t ) + B ~ ( t ) ) . B(t)\leq C(\varepsilon)(A(t)+\widetilde{B}(t)). B ( t ) ≤ C ( ε ) ( A ( t ) + B ( t )) .
Hence, by Theorem 4.2, and up to a subsequence that we do not re label, u k → 0 u_k\to0 u k → 0 in L B ~ ( R n ) L^{\widetilde{B}}(\mathbb{R}^n) L B ( R n ) . On the other hand, since H r a d s , A ( R n ) ⊂ W s ′ , A ( R n ) H_{\mathrm{rad}}^{s,A}(\mathbb{R}^n)\subset W^{s',A}(\mathbb{R}^n) H rad s , A ( R n ) ⊂ W s ′ , A ( R n ) for 0 < s ′ < s 0<s'<s 0 < s ′ < s ([28 Theorem 3.7] ), we get that u k u_k u k is bounded in W s ′ , A ( R n ) W^{s',A}(\mathbb{R}^{n}) W s ′ , A ( R n ) , u k ⇀ 0 u_k \rightharpoonup0 u k ⇀ 0 in W s ′ , A ( R n ) W^{s',A}(\mathbb{R}^{n}) W s ′ , A ( R n ) and, by [4 Theorem 9.1] , u k → 0 u_k \to0 u k → 0 in L A ( B 1 / ε ) L^{A}(B_{1/\varepsilon}) L A ( B 1/ ε ) . As a result,
I 2 ≤ C ( ε ) ε − ∣ α ∣ ∫ ε < ∣ x ∣ ≤ 1 / ε ( A ( ∣ u k ( x ) ∣ ) + B ~ ( ∣ u k ( x ) ∣ ) ) d x → 0 as k → ∞ . (30) I_2 \leq C(\varepsilon)\varepsilon^{-|\alpha|}\int_{\varepsilon<|x|\leq1/\varepsilon}\left(A(|u_k(x)|)+\widetilde{B}(|u_k(x)|)\right)\,dx \to0 \quad\text{as } k\to\infty. \tag*{(30)} I 2 ≤ C ( ε ) ε − ∣ α ∣ ∫ ε < ∣ x ∣ ≤ 1/ ε ( A ( ∣ u k ( x ) ∣ ) + B ( ∣ u k ( x ) ∣ ) ) d x → 0 as k → ∞. (30)
Bound for I 3 I_3 I 3 . Observe that
I 3 ≤ C ∫ 1 / ε ∞ B ( ∣ u k ( r ) ∣ ) r n + α − 1 d r = C ∫ 1 / ε ∞ B ( ∣ u k ( r ) ∣ ) A ( ∣ u k ( r ) ∣ ) A ( ∣ u k ( r ) ∣ ) r n + α − 1 d r . (31) I_3 \leq C\int_{1/\varepsilon}^{\infty}B(|u_k(r)|)r^{n+\alpha-1}\,dr=C\int_{1/\varepsilon}^{\infty}\frac{B(|u_k(r)|)}{A(|u_k(r)|)}A(|u_k(r)|)r^{n+\alpha-1}\,dr. \tag*{(31)} I 3 ≤ C ∫ 1/ ε ∞ B ( ∣ u k ( r ) ∣ ) r n + α − 1 d r = C ∫ 1/ ε ∞ A ( ∣ u k ( r ) ∣ ) B ( ∣ u k ( r ) ∣ ) A ( ∣ u k ( r ) ∣ ) r n + α − 1 d r . (31)
Now, condition (22) implies that p B − > p A + p_B^- > p_A^+ p B − > p A + . Hence, B ( t ) / A ( t ) B(t)/A(t) B ( t ) / A ( t ) is increasing since its derivative satisfies
b ( t ) A ( t ) − B ( t ) a ( t ) A ( t ) 2 = t b ( t ) A ( t ) − B ( t ) t a ( t ) t A ( t ) 2 ≥ ( p B − − p A + ) A ( t ) B ( t ) t A ( t ) 2 ≥ 0. \frac{b(t)A(t)-B(t)a(t)}{A(t)^2}=\frac{tb(t)A(t)-B(t)ta(t)}{tA(t)^2}\geq\frac{(p_B^- -p_A^+)A(t)B(t)}{tA(t)^2}\geq0. A ( t ) 2 b ( t ) A ( t ) − B ( t ) a ( t ) = t A ( t ) 2 t b ( t ) A ( t ) − B ( t ) t a ( t ) ≥ t A ( t ) 2 ( p B − − p A + ) A ( t ) B ( t ) ≥ 0.
Observe that using the Strauss’ Lemma 4.1 and (29) we get
∣ u k ( r ) ∣ ≤ A − 1 ( r 1 − n ) ∥ u ∥ H s , A ( R n ) , |u_k(r)|\leq A^{-1}(r^{1-n})\|u\|_{H^{s,A}(\mathbb{R}^{n})}, ∣ u k ( r ) ∣ ≤ A − 1 ( r 1 − n ) ∥ u ∥ H s , A ( R n ) ,
which leads to
B ( ∣ u k ( r ) ∣ ) A ( ∣ u k ( r ) ∣ ) ≤ B ( A − 1 ( r 1 − n ) ∥ u k ∥ H s , A ( R n ) ) A ( A − 1 ( r 1 − n ) ∥ u k ∥ H s , A ( R n ) ) ≤ max { ∥ u k ∥ H s , A p B + , ∥ u k ∥ H s , A ( R n ) p B − } min { ∥ u k ∥ H s , A p A + , ∥ u k ∥ H s , A ( R n ) p A − } B ( A − 1 ( r 1 − n ) ) r 1 − n ≤ C B ( A − 1 ( r 1 − n ) ) r 1 − n , \begin{aligned}
\frac{B(|u_k(r)|)}{A(|u_k(r)|)}
\leq
\frac{B\left(A^{-1}(r^{1-n})\|u_k\|_{H^{s,A}(\mathbb{R}^{n})}\right)}
{A\left(A^{-1}(r^{1-n})\|u_k\|_{H^{s,A}(\mathbb{R}^{n})}\right)} \\
&\leq
\frac{\max\left\{\|u_k\|_{H^{s,A}}^{p_B^+},\|u_k\|_{H^{s,A}(\mathbb{R}^{n})}^{p_B^-}\right\}}
{\min\left\{\|u_k\|_{H^{s,A}}^{p_A^+},\|u_k\|_{H^{s,A}(\mathbb{R}^{n})}^{p_A^-}\right\}}
\frac{B\left(A^{-1}(r^{1-n})\right)}{r^{1-n}}
\leq C\frac{B\left(A^{-1}(r^{1-n})\right)}{r^{1-n}},
\end{aligned} A ( ∣ u k ( r ) ∣ ) B ( ∣ u k ( r ) ∣ ) ≤ A ( A − 1 ( r 1 − n ) ∥ u k ∥ H s , A ( R n ) ) B ( A − 1 ( r 1 − n ) ∥ u k ∥ H s , A ( R n ) ) ≤ min { ∥ u k ∥ H s , A p A + , ∥ u k ∥ H s , A ( R n ) p A − } max { ∥ u k ∥ H s , A p B + , ∥ u k ∥ H s , A ( R n ) p B − } r 1 − n B ( A − 1 ( r 1 − n ) ) ≤ C r 1 − n B ( A − 1 ( r 1 − n ) ) ,
where C = C ( ∥ u k ∥ H s , A ( R n ) , p A ± , p B ± ) C=C(\|u_k\|_{H^{s,A}(\mathbb{R}^{n})},p_A^\pm,p_B^\pm) C = C ( ∥ u k ∥ H s , A ( R n ) , p A ± , p B ± ) since p B − > p A + p_B^- > p_A^+ p B − > p A + . Hence,
∫ 1 / ε ∞ B ( ∣ u k ( r ) ∣ ) A ( ∣ u k ( r ) ∣ ) A ( ∣ u k ( r ) ∣ ) r n + α − 1 d r ≤ C ∫ 1 / ε ∞ B ( A − 1 ( r 1 − n ) ) r 1 − n A ( ∣ u k ( r ) ∣ ) r n + α − 1 d r ≤ C ∫ 1 / ε ∞ r p B − ( 1 − n p A + ) − 1 + n + α A ( ∣ u k ( r ) ∣ ) r n − 1 d r ≤ C ε − [ ( p B − − p A + ) ( 1 − n p A + ) + α ] ∫ R n A ( ∣ u k ( x ) ∣ ) d x , (32) \begin{aligned}
\int_{1/\varepsilon}^{\infty}\frac{B(|u_k(r)|)}{A(|u_k(r)|)}A(|u_k(r)|)r^{n+\alpha-1}\,dr
\leq C\int_{1/\varepsilon}^{\infty}\frac{B\left(A^{-1}(r^{1-n})\right)}{r^{1-n}}A(|u_k(r)|)r^{n+\alpha-1}\,dr \\
&\leq C\int_{1/\varepsilon}^{\infty}r^{p_B^-\left(\frac{1-n}{p_A^+}\right)-1+n+\alpha}A(|u_k(r)|)r^{n-1}\,dr \\
&\leq C\varepsilon^{-\left[(p_B^- -p_A^+)\left(\frac{1-n}{p_A^+}\right)+\alpha\right]}\int_{\mathbb{R}^{n}}A(|u_k(x)|)\,dx, \tag*{(32)}
\end{aligned} ∫ 1/ ε ∞ A ( ∣ u k ( r ) ∣ ) B ( ∣ u k ( r ) ∣ ) A ( ∣ u k ( r ) ∣ ) r n + α − 1 d r ≤ C ∫ 1/ ε ∞ r 1 − n B ( A − 1 ( r 1 − n ) ) A ( ∣ u k ( r ) ∣ ) r n + α − 1 d r ≤ C ∫ 1/ ε ∞ r p B − ( p A + 1 − n ) − 1 + n + α A ( ∣ u k ( r ) ∣ ) r n − 1 d r ≤ C ε − [ ( p B − − p A + ) ( p A + 1 − n ) + α ] ∫ R n A ( ∣ u k ( x ) ∣ ) d x , (32)
where ( p B − − p A + ) ( 1 − n p A + ) + α < 0 (p_B^- -p_A^+)\left(\frac{1-n}{p_A^+}\right)+\alpha<0 ( p B − − p A + ) ( p A + 1 − n ) + α < 0 by assumption (23) . Hence, the strong convergence in L B ( R n , ∣ x ∣ α d x ) L^{B}(\mathbb{R}^{n},|x|^\alpha\,dx) L B ( R n , ∣ x ∣ α d x ) follows from (28) , (30) , (31) and (32) .
Theorem 4.7. Under the assumptions of Theorem 4.5 , the embedding W r a d s , A ( R n ) ↪ L B ( R n , ∣ x ∣ α d x ) W_{\mathrm{rad}}^{s,A}(\mathbb{R}^{n})\hookrightarrow L^{B}(\mathbb{R}^{n},|x|^\alpha\,dx) W rad s , A ( R n ) ↪ L B ( R n , ∣ x ∣ α d x ) is compact.
Proof. Let 0 < s ′ < s 0<s'<s 0 < s ′ < s be close enough to s s s such that 1 p A − < s ′ \frac{1}{p_A^-}<s' p A − 1 < s ′ and p B + < ( α + n ) p A − n − s ′ p A − p_B^+<\frac{(\alpha+n)p_A^-}{n-s'p_A^-} p B + < n − s ′ p A − ( α + n ) p A − . Then, according to Theorem 4.5 , H r a d s ′ , A ( R n ) ↪ L B ( R n , ∣ x ∣ α d x ) H_{\mathrm{rad}}^{s',A}(\mathbb{R}^{n})\hookrightarrow L^{B}(\mathbb{R}^{n},|x|^\alpha\,dx) H rad s ′ , A ( R n ) ↪ L B ( R n , ∣ x ∣ α d x ) compactly. Since by Theorem 3.9 in [28 ] , W s , A ( R n ) ↪ H s ′ , A ( R n ) W^{s,A}(\mathbb{R}^{n})\hookrightarrow H^{s',A}(\mathbb{R}^{n}) W s , A ( R n ) ↪ H s ′ , A ( R n ) continuously, we end the proof. ■
:::
Application to a mixed local-nonlocal problem
The goal of this section is to extend existence results of [29 ] for Hénon-type equations in R n \mathbb{R}^{n} R n to operators exhibiting both local and nonlocal behavior in the Orlicz setting.
We assume that A A A and B B B are N N N -functions satisfying the assumptions of Theorem 4.5 . We consider
{ L s , A , γ u + a ( ∣ u ∣ ) ∣ u ∣ u = ∣ x ∣ α b ( ∣ u ∣ ) ∣ u ∣ u in R n , u > 0 , (33) \left\{
\begin{aligned}
\mathcal{L}_{s,A,\gamma}u+\frac{a(|u|)}{|u|}u&=|x|^{\alpha}\frac{b(|u|)}{|u|}u &&\text{in }\mathbb{R}^{n},\\
u&>0,
\end{aligned}
\right. \tag*{(33)} ⎩ ⎨ ⎧ L s , A , γ u + ∣ u ∣ a ( ∣ u ∣ ) u u = ∣ x ∣ α ∣ u ∣ b ( ∣ u ∣ ) u > 0 , in R n , (33)
where u u u belongs to a suitable functional space of radial functions W s , A ( R n ) W^{s,A}(\mathbb{R}^{n}) W s , A ( R n ) (see (37) ). Given γ ∈ [ 0 , 1 ] \gamma\in[0,1] γ ∈ [ 0 , 1 ] , s ∈ ( 0 , 1 ] s \in(0,1] s ∈ ( 0 , 1 ] (depending on γ \gamma γ ), the operator L s , A , γ \mathcal{L}_{s,A,\gamma} L s , A , γ is given by
L s , A , γ u : = { ( − Δ a ) u , if γ = 1 , γ ( − Δ a ) u + ( 1 − γ ) ( − Δ a ) s u , if γ ∈ ( 0 , 1 ) , s ∈ ( 0 , 1 ) , ( − Δ a ) s u , if γ = 0 , s ∈ ( 0 , 1 ) , \mathcal{L}_{s,A,\gamma}u:=
\begin{cases}
(-\Delta_{a})u, & \text{if }\gamma=1,\\
\gamma(-\Delta_{a})u+(1-\gamma)(-\Delta_{a})^{s}u, & \text{if }\gamma\in(0,1),\ s\in(0,1),\\
(-\Delta_{a})^{s}u, & \text{if }\gamma=0,\ s\in(0,1),
\end{cases} L s , A , γ u := ⎩ ⎨ ⎧ ( − Δ a ) u , γ ( − Δ a ) u + ( 1 − γ ) ( − Δ a ) s u , ( − Δ a ) s u , if γ = 1 , if γ ∈ ( 0 , 1 ) , s ∈ ( 0 , 1 ) , if γ = 0 , s ∈ ( 0 , 1 ) ,
where we denote the fractional a a a -Laplacian of order s ∈ ( 0 , 1 ] s \in(0,1] s ∈ ( 0 , 1 ] by
( − Δ a ) s u ( x ) : = { 2 p.v. ∫ R n a ( ∣ D s u ( x , y ) ∣ ) ∣ D s u ( x , y ) ∣ D s u ( x , y ) d y ∣ x − y ∣ n + s , if s ∈ ( 0 , 1 ) , − div ( a ( ∣ ∇ u ∣ ) ∣ ∇ u ∣ ∇ u ( x ) ) , if s = 1 , (-\Delta_{a})^{s}u(x):=
\begin{cases}
2\operatorname{p.v.}\int_{\mathbb{R}^{n}}\frac{a(|D_{s}u(x,y)|)}{|D_{s}u(x,y)|}D_{s}u(x,y)\frac{\mathrm{d}y}{|x-y|^{n+s}}, & \text{if }s\in(0,1),\\
-\operatorname{div}\left(\frac{a(|\nabla u|)}{|\nabla u|}\nabla u(x)\right), & \text{if }s=1,
\end{cases} ( − Δ a ) s u ( x ) := ⎩ ⎨ ⎧ 2 p.v. ∫ R n ∣ D s u ( x , y ) ∣ a ( ∣ D s u ( x , y ) ∣ ) D s u ( x , y ) ∣ x − y ∣ n + s d y , − div ( ∣∇ u ∣ a ( ∣∇ u ∣ ) ∇ u ( x ) ) , if s ∈ ( 0 , 1 ) , if s = 1 ,
where
D s u ( x , y ) = u ( x ) − u ( y ) ∣ x − y ∣ s . D_{s}u(x,y)=\frac{u(x)-u(y)}{|x-y|^{s}}. D s u ( x , y ) = ∣ x − y ∣ s u ( x ) − u ( y ) .
Hence, the operator L s , A , γ \mathcal{L}_{s,A,\gamma} L s , A , γ is a mixed local–nonlocal operator, reducing when A ( t ) = t p A(t)=t^{p} A ( t ) = t p , t ≥ 0 t\ge0 t ≥ 0 to the classical p p p –Laplacian for γ = 1 \gamma=1 γ = 1 , and to the fractional p p p –Laplacian when γ = 0 \gamma=0 γ = 0 . We will usually take s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ] , identifying the case γ = 1 \gamma=1 γ = 1 with s = 1 s=1 s = 1 .
Given A A A an N N N -function and s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ] we recall that
W s , A ( R n ) = { u ∈ L A ( R n ) : [ u ] s , A < ∞ } W^{s,A}(\mathbb{R}^{n})=\{u\in L^{A}(\mathbb{R}^{n}):[u]_{s,A}<\infty\} W s , A ( R n ) = { u ∈ L A ( R n ) : [ u ] s , A < ∞ }
endowed with the norm
∥ u ∥ W s , A ( R n ) : = ∥ u ∥ L A ( R n ) + [ u ] W s , A ( R n ) , \|u\|_{W^{s,A}(\mathbb{R}^{n})}:=\|u\|_{L^{A}(\mathbb{R}^{n})}+[u]_{W^{s,A}(\mathbb{R}^{n})}, ∥ u ∥ W s , A ( R n ) := ∥ u ∥ L A ( R n ) + [ u ] W s , A ( R n ) ,
where
∥ u ∥ L A ( R n ) : = inf { λ > 0 : ρ A ( u λ ) ≤ 1 } , \|u\|_{L^{A}(\mathbb{R}^{n})}:=\inf\left\{\lambda>0:\rho_{A}\left(\frac{u}{\lambda}\right)\le1\right\}, ∥ u ∥ L A ( R n ) := inf { λ > 0 : ρ A ( λ u ) ≤ 1 } ,
and the seminorm is given by
[ u ] W s , A ( R n ) : = { inf { λ > 0 : ρ s , A ( u λ ) ≤ 1 } , if s ∈ ( 0 , 1 ) , inf { λ > 0 : ρ A ( ∣ ∇ u ∣ λ ) ≤ 1 } , if s = 1. [u]_{W^{s,A}(\mathbb{R}^{n})}:=
\begin{cases}
\inf\left\{\lambda>0:\rho_{s,A}\left(\frac{u}{\lambda}\right)\le1\right\}, & \text{if }s\in(0,1),\\
\inf\left\{\lambda>0:\rho_{A}\left(\frac{|\nabla u|}{\lambda}\right)\le1\right\}, & \text{if }s=1.
\end{cases} [ u ] W s , A ( R n ) := { inf { λ > 0 : ρ s , A ( λ u ) ≤ 1 } , inf { λ > 0 : ρ A ( λ ∣∇ u ∣ ) ≤ 1 } , if s ∈ ( 0 , 1 ) , if s = 1.
where the modular functions are defined as follows
ρ s , A ( u ) : = ∫ R n ∫ R n A ( ∣ D s u ∣ ) d x d y ∣ x − y ∣ n , ρ A ( u ) : = ∫ R n A ( ∣ u ∣ ) d x . \rho_{s,A}(u):=\int_{\mathbb{R}^{n}}\int_{\mathbb{R}^{n}}A(|D_{s}^{\!}u|)\frac{\mathrm{d}x\mathrm{d}y}{|x-y|^{n}},\qquad
\rho_{A}(u):=\int_{\mathbb{R}^{n}}A(|u|)\,\mathrm{d}x. ρ s , A ( u ) := ∫ R n ∫ R n A ( ∣ D s u ∣ ) ∣ x − y ∣ n d x d y , ρ A ( u ) := ∫ R n A ( ∣ u ∣ ) d x .
For Orlicz weighted spaces, we consider the norm
∥ u ∥ L A , α ( R n ) : = inf { λ > 0 : ρ A , α ( u λ ) ≤ 1 } \|u\|_{L^{A,\alpha}(\mathbb{R}^{n})}:=\inf\left\{\lambda>0:\rho_{A,\alpha}\left(\frac{u}{\lambda}\right)\leq1\right\} ∥ u ∥ L A , α ( R n ) := inf { λ > 0 : ρ A , α ( λ u ) ≤ 1 }
where
ρ A , α ( u ) : = ∫ R n A ( ∣ u ∣ ) ∣ x ∣ α d x . \rho_{A,\alpha}(u):=\int_{\mathbb{R}^{n}}A(|u|)|x|^{\alpha}\,dx. ρ A , α ( u ) := ∫ R n A ( ∣ u ∣ ) ∣ x ∣ α d x .
We recall the following relation between norms and modular, provided that A A A satisfies (8)
∥ u ∥ L A ( R n ) p − ≤ ρ A ( u ) ≤ ∥ u ∥ L A ( R n ) p + if ∥ u ∥ L A ( R n ) ≤ 1 , ∥ u ∥ L A ( R n ) p + ≤ ρ A ( u ) ≤ ∥ u ∥ L A ( R n ) p − if ∥ u ∥ L A ( R n ) > 1. (34) \|u\|_{L^{A}(\mathbb{R}^{n})}^{p^{-}}\leq\rho_{A}(u)\leq\|u\|_{L^{A}(\mathbb{R}^{n})}^{p^{+}}\quad\text{if }\|u\|_{L^{A}(\mathbb{R}^{n})}\leq1,\qquad\|u\|_{L^{A}(\mathbb{R}^{n})}^{p^{+}}\leq\rho_{A}(u)\leq\|u\|_{L^{A}(\mathbb{R}^{n})}^{p^{-}}\quad\text{if }\|u\|_{L^{A}(\mathbb{R}^{n})}>1. \tag*{(34)} ∥ u ∥ L A ( R n ) p − ≤ ρ A ( u ) ≤ ∥ u ∥ L A ( R n ) p + if ∥ u ∥ L A ( R n ) ≤ 1 , ∥ u ∥ L A ( R n ) p + ≤ ρ A ( u ) ≤ ∥ u ∥ L A ( R n ) p − if ∥ u ∥ L A ( R n ) > 1. (34)
Moreover, (see for instance [17 Lemma 4.2] ), for u ∈ W 1 , A ( R n ) u\in W^{1,A}(\mathbb{R}^{n}) u ∈ W 1 , A ( R n ) , and s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) , it holds that
ρ s , A ( u ) ≤ n ω n 1 − s ρ A ( ∣ ∇ u ∣ ) + 2 C n ω n s ρ A ( u ) . (35) \rho_{s,A}(u)\leq\frac{n\omega_{n}}{1-s}\rho_{A}(|\nabla u|)+\frac{2Cn\omega_{n}}{s}\rho_{A}(u). \tag*{(35)} ρ s , A ( u ) ≤ 1 − s n ω n ρ A ( ∣∇ u ∣ ) + s 2 C n ω n ρ A ( u ) . (35)
Then,
[ u ] W s , A ( R n ) ≤ C ∥ u ∥ W 1 , A ( R n ) . (36) [u]_{W^{s,A}(\mathbb{R}^{n})}\leq C\|u\|_{W^{1,A}(\mathbb{R}^{n})}. \tag*{(36)} [ u ] W s , A ( R n ) ≤ C ∥ u ∥ W 1 , A ( R n ) . (36)
With the aforementioned definitions, the natural framework for introducing weak solutions of (33) is the space
W s , A ( R n ) = { W r a d s , A ( R n ) if γ = 0 , s ∈ ( 0 , 1 ) , W r a d 1 , A ( R n ) if γ ∈ ( 0 , 1 ] , (37) \mathcal{W}^{s,A}(\mathbb{R}^{n})=
\begin{cases}
W_{\mathrm{rad}}^{s,A}(\mathbb{R}^{n}) & \text{if }\gamma=0,\ s\in(0,1),\\
W_{\mathrm{rad}}^{1,A}(\mathbb{R}^{n}) & \text{if }\gamma\in(0,1],
\end{cases} \tag*{(37)} W s , A ( R n ) = { W rad s , A ( R n ) W rad 1 , A ( R n ) if γ = 0 , s ∈ ( 0 , 1 ) , if γ ∈ ( 0 , 1 ] , (37)
endowed with ∥ ⋅ ∥ W s , A ( R n ) \|\cdot\|_{W^{s,A}(\mathbb{R}^{n})} ∥ ⋅ ∥ W s , A ( R n ) when γ = 0 \gamma=0 γ = 0 and s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) , and ∥ ⋅ ∥ W 1 , A ( R n ) \|\cdot\|_{W^{1,A}(\mathbb{R}^{n})} ∥ ⋅ ∥ W 1 , A ( R n ) when γ ∈ ( 0 , 1 ] \gamma\in(0,1] γ ∈ ( 0 , 1 ] .
We point out that since W 1 , A ( R n ) ⊂ W s , A ( R n ) W^{1,A}(\mathbb{R}^{n})\subset W^{s,A}(\mathbb{R}^{n}) W 1 , A ( R n ) ⊂ W s , A ( R n ) for all s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) (see Lemma 4.2 in [17 ] ), the following representation formula holds: given u ∈ W s , A ( R n ) u\in\mathcal{W}^{s,A}(\mathbb{R}^{n}) u ∈ W s , A ( R n ) ,
⟨ L s , A , γ u , v ⟩ = γ ∫ R n a ( ∣ ∇ u ∣ ) ∣ ∇ u ∣ ∇ u ⋅ ∇ v d x + ( 1 − γ ) ∫ R n ∫ R n a ( ∣ D s u ( x , y ) ∣ ) ∣ D s u ( x , y ) ∣ D s u ( x , y ) D s v ( x , y ) d x d y ∣ x − y ∣ n (38) \langle L_{s,A,\gamma}u,v\rangle=\gamma\int_{\mathbb{R}^{n}}\frac{a(|\nabla u|)}{|\nabla u|}\nabla u\cdot\nabla v\,dx+(1-\gamma)\int_{\mathbb{R}^{n}}\int_{\mathbb{R}^{n}}\frac{a(|D_{s}u(x,y)|)}{|D_{s}u(x,y)|}D_{s}u(x,y)D_{s}v(x,y)\frac{dx\,dy}{|x-y|^{n}} \tag*{(38)} ⟨ L s , A , γ u , v ⟩ = γ ∫ R n ∣∇ u ∣ a ( ∣∇ u ∣ ) ∇ u ⋅ ∇ v d x + ( 1 − γ ) ∫ R n ∫ R n ∣ D s u ( x , y ) ∣ a ( ∣ D s u ( x , y ) ∣ ) D s u ( x , y ) D s v ( x , y ) ∣ x − y ∣ n d x d y (38)
for all v ∈ W s , A ( R n ) v\in\mathcal{W}^{s,A}(\mathbb{R}^{n}) v ∈ W s , A ( R n ) . Therefore, we say that u ∈ W s , A ( R n ) u\in\mathcal{W}^{s,A}(\mathbb{R}^{n}) u ∈ W s , A ( R n ) is a weak solution of (33) if
⟨ L s , A , γ u , v ⟩ + ∫ R n a ( ∣ u ∣ ) ∣ u ∣ u v d x = ∫ R n ∣ x ∣ α b ( ∣ u ∣ ) ∣ u ∣ u v d x ∀ v ∈ W s , A ( R n ) . \langle L_{s,A,\gamma}u,v\rangle+\int_{\mathbb{R}^{n}}\frac{a(|u|)}{|u|}uv\,dx=\int_{\mathbb{R}^{n}}|x|^{\alpha}\frac{b(|u|)}{|u|}uv\,dx\quad\forall v\in\mathcal{W}^{s,A}(\mathbb{R}^{n}). ⟨ L s , A , γ u , v ⟩ + ∫ R n ∣ u ∣ a ( ∣ u ∣ ) uv d x = ∫ R n ∣ x ∣ α ∣ u ∣ b ( ∣ u ∣ ) uv d x ∀ v ∈ W s , A ( R n ) .
The corresponding energy functional associated to (33) is given by
J ( u ) : = γ ρ A ( ∣ ∇ u ∣ ) + ( 1 − γ ) ρ s , A ( u ) + ρ A ( u ) − ρ B , α ( u ) , γ ∈ [ 0 , 1 ] . \mathcal{J}(u):=\gamma\rho_{A}(|\nabla u|)+(1-\gamma)\rho_{s,A}(u)+\rho_{A}(u)-\rho_{B,\alpha}(u),\qquad\gamma\in[0,1]. J ( u ) := γ ρ A ( ∣∇ u ∣ ) + ( 1 − γ ) ρ s , A ( u ) + ρ A ( u ) − ρ B , α ( u ) , γ ∈ [ 0 , 1 ] .
Indeed, J \mathcal{J} J is a C 1 C^{1} C 1 functional whose Fréchet derivatives is given by
⟨ J ′ ( u ) , v ⟩ = ⟨ L s , A , γ u , v ⟩ + ∫ R n a ( ∣ u ∣ ) ∣ u ∣ u v d x − ∫ R n b ( ∣ u ∣ ) ∣ u ∣ u v ∣ x ∣ α d x ∀ u , v ∈ W s , A ( R n ) . \langle\mathcal{J}'(u),v\rangle=\langle L_{s,A,\gamma}u,v\rangle+\int_{\mathbb{R}^{n}}\frac{a(|u|)}{|u|}uv\,dx-\int_{\mathbb{R}^{n}}\frac{b(|u|)}{|u|}uv|x|^{\alpha}\,dx\quad\forall u,v\in\mathcal{W}^{s,A}(\mathbb{R}^{n}). ⟨ J ′ ( u ) , v ⟩ = ⟨ L s , A , γ u , v ⟩ + ∫ R n ∣ u ∣ a ( ∣ u ∣ ) uv d x − ∫ R n ∣ u ∣ b ( ∣ u ∣ ) uv ∣ x ∣ α d x ∀ u , v ∈ W s , A ( R n ) .
Our main application concerns the existence of radial solutions of problem (33) .
Theorem 5.1. Assume that A A A , A ^ \widehat{A} A , and B B B are N N N -functions satisfying the Δ 2 \Delta_2 Δ 2 -condition and suppose that A ( t ) A(\sqrt{t}) A ( t ) is convex. Assume that 1 p A − < s < n p A + \frac{1}{p_A^-}<s<\frac{n}{p_A^+} p A − 1 < s < p A + n and that p A + < p B − p_A^+<p_B^- p A + < p B − . Let α > − n \alpha>-n α > − n such that
( p B − − p A + ) ( 1 − n p A + ) + α < 0 , and p B + < ( α + n ) p A − n − s p A − . \left(p_B^- - p_A^+\right)\left(\frac{1-n}{p_A^+}\right)+\alpha<0,\qquad\text{and}\qquad p_B^+<\frac{(\alpha+n)p_A^-}{n-sp_A^-}. ( p B − − p A + ) ( p A + 1 − n ) + α < 0 , and p B + < n − s p A − ( α + n ) p A − . Then, problem (33) admits a radial weak solution.
Proof. We check the conditions of the Brezis-Nirenberg Lemma 5.3 with E = W s , A ( R n ) E=W^{s,A}(\mathbb{R}^n) E = W s , A ( R n ) .
(i) Case γ = 0 \gamma=0 γ = 0 and s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) . Here the underlying space is W r a d s , A ( R n ) W_{\mathrm{rad}}^{s,A}(\mathbb{R}^n) W rad s , A ( R n ) . Let u u u be such that ∥ u ∥ W s , A ( R n ) ≤ 1 \|u\|_{W^{s,A}(\mathbb{R}^n)}\leq1 ∥ u ∥ W s , A ( R n ) ≤ 1 . By Theorem 4.5 we can assume that ∥ u ∥ W s , A ( R n ) ≤ 1 \|u\|_{W^{s,A}(\mathbb{R}^n)}\leq1 ∥ u ∥ W s , A ( R n ) ≤ 1 is small enough such that ∥ u ∥ B , α ≤ 1 \|u\|_{B,\alpha}\leq1 ∥ u ∥ B , α ≤ 1 . Then, using (34) and Theorem 4.5
J ( u ) = ρ s , A ( u ) + ρ A ( u ) − ρ B , α ( u ) ≥ ∥ u ∥ W s , A ( R n ) p A + − ∥ u ∥ L B , α ( R n ) p B − ≥ ∥ u ∥ W s , A ( R n ) p A + − C ∥ u ∥ W s , A ( R n ) p B − . \mathcal{J}(u)=\rho_{s,A}(u)+\rho_A(u)-\rho_{B,\alpha}(u)\geq\|u\|_{W^{s,A}(\mathbb{R}^n)}^{p_A^+}-\|u\|_{L^{B,\alpha}(\mathbb{R}^n)}^{p_B^-}\geq\|u\|_{W^{s,A}(\mathbb{R}^n)}^{p_A^+}-C\|u\|_{W^{s,A}(\mathbb{R}^n)}^{p_B^-}. J ( u ) = ρ s , A ( u ) + ρ A ( u ) − ρ B , α ( u ) ≥ ∥ u ∥ W s , A ( R n ) p A + − ∥ u ∥ L B , α ( R n ) p B − ≥ ∥ u ∥ W s , A ( R n ) p A + − C ∥ u ∥ W s , A ( R n ) p B − . Since p A + < p B − p_A^+<p_B^- p A + < p B − , we have that J ( u ) > 0 \mathcal{J}(u)>0 J ( u ) > 0 for ∥ u ∥ W s , A ( R n ) = r \|u\|_{W^{s,A}(\mathbb{R}^n)}=r ∥ u ∥ W s , A ( R n ) = r small enough.
Case γ = 1 \gamma=1 γ = 1 (s = 1 s=1 s = 1 ). Let u ∈ W r a d 1 , A ( R n ) u\in W_{\mathrm{rad}}^{1,A}(\mathbb{R}^n) u ∈ W rad 1 , A ( R n ) be such that ∥ u ∥ W 1 , A ( R n ) ≤ 1 \|u\|_{W^{1,A}(\mathbb{R}^n)}\leq1 ∥ u ∥ W 1 , A ( R n ) ≤ 1 . Similarly, we can assume that ∥ u ∥ B , α ≤ 1 \|u\|_{B,\alpha}\leq1 ∥ u ∥ B , α ≤ 1 . From (34) and Theorem 4.5
J ( u ) = ρ A ( ∣ ∇ u ∣ ) + ρ A ( u ) − ρ B , α ( u ) ≥ ∥ u ∥ W 1 , A ( R n ) p A + − ∥ u ∥ L B , α ( R n ) p B − ≥ ∥ u ∥ W 1 , A ( R n ) p A + − C ∥ u ∥ W 1 , A ( R n ) p B − . \mathcal{J}(u)=\rho_A(|\nabla u|)+\rho_A(u)-\rho_{B,\alpha}(u)\geq\|u\|_{W^{1,A}(\mathbb{R}^n)}^{p_A^+}-\|u\|_{L^{B,\alpha}(\mathbb{R}^n)}^{p_B^-}\geq\|u\|_{W^{1,A}(\mathbb{R}^n)}^{p_A^+}-C\|u\|_{W^{1,A}(\mathbb{R}^n)}^{p_B^-}. J ( u ) = ρ A ( ∣∇ u ∣ ) + ρ A ( u ) − ρ B , α ( u ) ≥ ∥ u ∥ W 1 , A ( R n ) p A + − ∥ u ∥ L B , α ( R n ) p B − ≥ ∥ u ∥ W 1 , A ( R n ) p A + − C ∥ u ∥ W 1 , A ( R n ) p B − . Since p A + < p B − p_A^+<p_B^- p A + < p B − , we have that J ( u ) > 0 \mathcal{J}(u)>0 J ( u ) > 0 for ∥ u ∥ W 1 , A ( R n ) = r \|u\|_{W^{1,A}(\mathbb{R}^n)}=r ∥ u ∥ W 1 , A ( R n ) = r small enough.
Case γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) and s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) . Let u ∈ W r a d 1 , A ( R n ) u\in W_{\mathrm{rad}}^{1,A}(\mathbb{R}^n) u ∈ W rad 1 , A ( R n ) . By Theorem 4.5 we can assume that ∥ u ∥ W 1 , A ( R n ) ≤ 1 \|u\|_{W^{1,A}(\mathbb{R}^n)}\leq1 ∥ u ∥ W 1 , A ( R n ) ≤ 1 is small enough such that ∥ u ∥ L B , α ( R n ) ≤ 1 \|u\|_{L^{B,\alpha}(\mathbb{R}^n)}\leq1 ∥ u ∥ L B , α ( R n ) ≤ 1 . Then, using (34) and Theorem 4.5
J ( u ) = γ ρ A ( ∣ ∇ u ∣ ) + ( 1 − γ ) ρ s , A ( u ) + ρ A ( u ) − ρ B , α ( u ) ≥ γ ∥ u ∥ W 1 , A ( R n ) p A + − ∥ u ∥ L B , α ( R n ) p B − ≥ γ ∥ u ∥ W 1 , A ( R n ) p A + − C ∥ u ∥ W 1 , A ( R n ) p B − . \begin{aligned}
\mathcal{J}(u)&=\gamma\rho_A(|\nabla u|)+(1-\gamma)\rho_{s,A}(u)+\rho_A(u)-\rho_{B,\alpha}(u)\\
&\geq\gamma\|u\|_{W^{1,A}(\mathbb{R}^n)}^{p_A^+}-\|u\|_{L^{B,\alpha}(\mathbb{R}^n)}^{p_B^-}\geq\gamma\|u\|_{W^{1,A}(\mathbb{R}^n)}^{p_A^+}-C\|u\|_{W^{1,A}(\mathbb{R}^n)}^{p_B^-}.
\end{aligned} J ( u ) = γ ρ A ( ∣∇ u ∣ ) + ( 1 − γ ) ρ s , A ( u ) + ρ A ( u ) − ρ B , α ( u ) ≥ γ ∥ u ∥ W 1 , A ( R n ) p A + − ∥ u ∥ L B , α ( R n ) p B − ≥ γ ∥ u ∥ W 1 , A ( R n ) p A + − C ∥ u ∥ W 1 , A ( R n ) p B − . Since p A + < p B − p_A^+<p_B^- p A + < p B − , we have that J ( u ) > 0 \mathcal{J}(u)>0 J ( u ) > 0 for ∥ u ∥ W 1 , A ( R n ) = r \|u\|_{W^{1,A}(\mathbb{R}^n)}=r ∥ u ∥ W 1 , A ( R n ) = r small enough.
(ii) Let γ ∈ [ 0 , 1 ] \gamma\in[0,1] γ ∈ [ 0 , 1 ] and s ∈ ( 0 , 1 ] s\in(0,1] s ∈ ( 0 , 1 ] (where we identify γ = 1 \gamma=1 γ = 1 with s = 1 s=1 s = 1 ). Let w > 0 w>0 w > 0 fixed, and define v : = t w v:=tw v := tw , t ≥ 1 t\geq1 t ≥ 1 . Then, using (34)
J ( v ) = γ ρ A ( t ∣ ∇ w ∣ ) + ( 1 − γ ) ρ s , A ( t w ) + ρ A ( t w ) − ρ B , α ( t w ) ≤ t p A + γ ρ A ( ∣ ∇ w ∣ ) + t p A + ( 1 − γ ) ρ s , A ( w ) + t p A + ρ A ( w ) − t p B − ρ B , α ( w ) ≤ C 1 t p A + − C 2 t p B − . \begin{aligned}
\mathcal{J}(v)&=\gamma\rho_A(t|\nabla w|)+(1-\gamma)\rho_{s,A}(tw)+\rho_A(tw)-\rho_{B,\alpha}(tw)\\
&\leq t^{p_A^+}\gamma\rho_A(|\nabla w|)+t^{p_A^+}(1-\gamma)\rho_{s,A}(w)+t^{p_A^+}\rho_A(w)-t^{p_B^-}\rho_{B,\alpha}(w)\\
&\leq C_1t^{p_A^+}-C_2t^{p_B^-}.
\end{aligned} J ( v ) = γ ρ A ( t ∣∇ w ∣ ) + ( 1 − γ ) ρ s , A ( tw ) + ρ A ( tw ) − ρ B , α ( tw ) ≤ t p A + γ ρ A ( ∣∇ w ∣ ) + t p A + ( 1 − γ ) ρ s , A ( w ) + t p A + ρ A ( w ) − t p B − ρ B , α ( w ) ≤ C 1 t p A + − C 2 t p B − . for C 1 C_1 C 1 and C 2 C_2 C 2 positive constants depending on w w w , γ \gamma γ and A A A . Since p A + < p B − p_A^+<p_B^- p A + < p B − , there exists t t t such that J ( v ) < 0 \mathcal{J}(v)<0 J ( v ) < 0 and ∥ v ∥ > r \|v\|>r ∥ v ∥ > r .
(iii) Let γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) and s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) . Let { u k } k ∈ N ⊂ W r a d 1 , A ( R n ) \{u_k\}_{k\in\mathbb{N}}\subset W_{\mathrm{rad}}^{1,A}(\mathbb{R}^n) { u k } k ∈ N ⊂ W rad 1 , A ( R n ) be such that J ′ ( u k ) → 0 \mathcal{J}'(u_k)\to0 J ′ ( u k ) → 0 and J ( u k ) → c \mathcal{J}(u_k)\to c J ( u k ) → c . Since J ′ ( u k ) → 0 \mathcal{J}'(u_k)\to0 J ′ ( u k ) → 0 , using (36) we get that
∣ ⟨ J ′ ( u k ) , u k ⟩ ∣ = ∣ ⟨ L s , A , γ u k , u k ⟩ + ∫ R n a ( ∣ u k ∣ ) ∣ u k ∣ d x − ∫ R n b ( ∣ u k ∣ ) ∣ u k ∣ ∣ x ∣ α d x ∣ \left|\left\langle\mathcal{J}'(u_k),u_k\right\rangle\right|
=\left|\left\langle\mathcal{L}_{s,A,\gamma}u_k,u_k\right\rangle+\int_{\mathbb{R}^n}a(|u_k|)|u_k|\,dx-\int_{\mathbb{R}^n}b(|u_k|)|u_k||x|^\alpha\,dx\right| ∣ ⟨ J ′ ( u k ) , u k ⟩ ∣ = ⟨ L s , A , γ u k , u k ⟩ + ∫ R n a ( ∣ u k ∣ ) ∣ u k ∣ d x − ∫ R n b ( ∣ u k ∣ ) ∣ u k ∣∣ x ∣ α d x ≤ ∥ J ′ ( u k ) ∥ W r a d 1 , A ( R n ) ′ ∥ u k ∥ 1 , A ≤ ∥ J ′ ( u k ) ∥ W r a d 1 , A ( R n ) ′ ∥ u k ∥ W 1 , A ( R n ) \leq\|\mathcal{J}'(u_k)\|_{W_{\mathrm{rad}}^{1,A}(\mathbb{R}^n)'}\|u_k\|_{1,A}
\leq\|\mathcal{J}'(u_k)\|_{W_{\mathrm{rad}}^{1,A}(\mathbb{R}^n)'}\|u_k\|_{W^{1,A}(\mathbb{R}^n)} ≤ ∥ J ′ ( u k ) ∥ W rad 1 , A ( R n ) ′ ∥ u k ∥ 1 , A ≤ ∥ J ′ ( u k ) ∥ W rad 1 , A ( R n ) ′ ∥ u k ∥ W 1 , A ( R n )
for k k k large enough. In particular, if we define
E k : = γ ρ A ( ∣ ∇ u k ∣ ) + ( 1 − γ ) ρ s , A ( u k ) + ρ A ( u k ) , E_k := \gamma\rho_A(|\nabla u_k|) + (1-\gamma)\rho_{s,A}(u_k) + \rho_A(u_k), E k := γ ρ A ( ∣∇ u k ∣ ) + ( 1 − γ ) ρ s , A ( u k ) + ρ A ( u k ) ,
for k k k large enough we have that
p B − ρ B , α ( u k ) ≤ ∥ u k ∥ W 1 , A ( R n ) + p A + E k . (39) p_B^- \rho_{B,\alpha}(u_k) \le\|u_k\|_{W^{1,A}(\mathbb{R}^n)} + p_A^+ E_k. \tag*{(39)} p B − ρ B , α ( u k ) ≤ ∥ u k ∥ W 1 , A ( R n ) + p A + E k . (39)
Since ∣ J ( u k ) ∣ ≤ C |\mathcal{J}(u_k)| \le C ∣ J ( u k ) ∣ ≤ C , we have that
∣ J ( u k ) ∣ = ∣ E k − ρ B , α ( u k ) ∣ ≤ C . (40) |\mathcal{J}(u_k)| = |E_k-\rho_{B,\alpha}(u_k)| \le C. \tag*{(40)} ∣ J ( u k ) ∣ = ∣ E k − ρ B , α ( u k ) ∣ ≤ C . (40)
From (39) and (40) we get that
E k ≤ C + ρ B , α ( u k ) ≤ C + 1 p B − ∥ u k ∥ W 1 , A ( R n ) + p A + p B − E k E_k \le C+\rho_{B,\alpha}(u_k) \le C+\frac{1}{p_B^-}\|u_k\|_{W^{1,A}(\mathbb{R}^n)}+\frac{p_A^+}{p_B^-}E_k E k ≤ C + ρ B , α ( u k ) ≤ C + p B − 1 ∥ u k ∥ W 1 , A ( R n ) + p B − p A + E k
from where, since p A + < p B − p_A^+<p_B^- p A + < p B − ,
( 1 − p A + p B − ) E k ≤ C + 1 p B − ∥ u k ∥ W 1 , A ( R n ) \left(1-\frac{p_A^+}{p_B^-}\right)E_k \le C+\frac{1}{p_B^-}\|u_k\|_{W^{1,A}(\mathbb{R}^n)} ( 1 − p B − p A + ) E k ≤ C + p B − 1 ∥ u k ∥ W 1 , A ( R n )
This gives that for some c > 0 c>0 c > 0 depending on p A + p_A^+ p A + and p B − p_B^- p B −
E k ≤ C 1 ( 1 + ∥ u k ∥ W 1 , A ( R n ) ) E_k \le C_1(1+\|u_k\|_{W^{1,A}(\mathbb{R}^n)}) E k ≤ C 1 ( 1 + ∥ u k ∥ W 1 , A ( R n ) )
if ∥ u k ∥ W 1 , A ( R n ) > 1 \|u_k\|_{W^{1,A}(\mathbb{R}^n)}>1 ∥ u k ∥ W 1 , A ( R n ) > 1 , using (34)
∥ u k ∥ W 1 , A ( R n ) p A − ≤ C 1 ( 1 + ∥ u k ∥ W 1 , A ( R n ) ) ⟹ ∥ u k ∥ W 1 , A ( R n ) ≤ ( 2 C 1 ) 1 p A − − 1 . \|u_k\|_{W^{1,A}(\mathbb{R}^n)}^{p_A^-}
\le C_1(1+\|u_k\|_{W^{1,A}(\mathbb{R}^n)})
\Longrightarrow
\|u_k\|_{W^{1,A}(\mathbb{R}^n)}
\le(2C_1)^{\frac{1}{p_A^- -1}}. ∥ u k ∥ W 1 , A ( R n ) p A − ≤ C 1 ( 1 + ∥ u k ∥ W 1 , A ( R n ) ) ⟹ ∥ u k ∥ W 1 , A ( R n ) ≤ ( 2 C 1 ) p A − − 1 1 .
Therefore, it follows that { u k } k ∈ N \{u_k\}_{k\in\mathbb{N}} { u k } k ∈ N is bounded in W r a d 1 , A ( R n ) W^{1,A}_{\mathrm{rad}}(\mathbb{R}^n) W rad 1 , A ( R n ) .
Then, up to a subsequence, u k ⇀ u u_k\rightharpoonup u u k ⇀ u weakly in W r a d 1 , A ( R n ) W^{1,A}_{\mathrm{rad}}(\mathbb{R}^n) W rad 1 , A ( R n ) . From the compact embedding given in Theorem 4.5 , we get that u k → u u_k\to u u k → u strongly in L B ( R n , ∣ x ∣ α d x ) L^B(\mathbb{R}^n,|x|^\alpha dx) L B ( R n , ∣ x ∣ α d x ) , which implies, due to convexity, that
∫ R n ∣ x ∣ α b ( u k ) ( u k − u ) d x → 0 \int_{\mathbb{R}^n}|x|^\alpha b(u_k)(u_k-u)\,dx\to0 ∫ R n ∣ x ∣ α b ( u k ) ( u k − u ) d x → 0
as k → ∞ k\to\infty k → ∞ . Therefore
∣ ⟨ L s , A , γ u k , u k − u ⟩ + ∫ R n a ( ∣ u k ∣ ) ∣ u k ∣ u k ( u k − u ) d x ∣ = ∣ ⟨ J ′ ( u k ) , u k − u ⟩ + ∫ R n b ( ∣ u k ∣ ) ∣ u k ∣ u k ( u k − u ) ∣ x ∣ α d x ∣ ≤ ∥ J ′ ( u k ) ∥ W 1 , A ( R n ) ′ ∥ u k − u ∥ W 1 , A ( R n ) + ∫ R n b ( ∣ u k ∣ ) ∣ u k ∣ u k ( u k − u ) ∣ x ∣ α d x → 0 \begin{aligned}
\left|\left\langle\mathcal{L}_{s,A,\gamma}u_k,u_k-u\right\rangle+
\int_{\mathbb{R}^n}\frac{a(|u_k|)}{|u_k|}u_k(u_k-u)\,dx\right|
=
\left|\left\langle\mathcal{J}'(u_k),u_k-u\right\rangle+
\int_{\mathbb{R}^n}\frac{b(|u_k|)}{|u_k|}u_k(u_k-u)|x|^\alpha\,dx\right| \\
&\le\|\mathcal{J}'(u_k)\|_{W^{1,A}(\mathbb{R}^n)'}\|u_k-u\|_{W^{1,A}(\mathbb{R}^n)}
+\int_{\mathbb{R}^n}\frac{b(|u_k|)}{|u_k|}u_k(u_k-u)|x|^\alpha\,dx\to0
\end{aligned} ⟨ L s , A , γ u k , u k − u ⟩ + ∫ R n ∣ u k ∣ a ( ∣ u k ∣ ) u k ( u k − u ) d x = ⟨ J ′ ( u k ) , u k − u ⟩ + ∫ R n ∣ u k ∣ b ( ∣ u k ∣ ) u k ( u k − u ) ∣ x ∣ α d x ≤ ∥ J ′ ( u k ) ∥ W 1 , A ( R n ) ′ ∥ u k − u ∥ W 1 , A ( R n ) + ∫ R n ∣ u k ∣ b ( ∣ u k ∣ ) u k ( u k − u ) ∣ x ∣ α d x → 0
as k → ∞ k\to\infty k → ∞ , where W 1 , A ( R n ) ′ W^{1,A}(\mathbb{R}^n)' W 1 , A ( R n ) ′ denotes the dual space of W 1 , A ( R n ) W^{1,A}(\mathbb{R}^n) W 1 , A ( R n ) .
Therefore, the operator L ~ \widetilde{\mathcal{L}} L defined as
⟨ L ~ u , v ⟩ : = ⟨ L s , A , γ u , v ⟩ + ∫ R n a ( ∣ u ∣ ) ∣ u ∣ u v d x \langle\widetilde{\mathcal{L}}u,v\rangle
:=\langle\mathcal{L}_{s,A,\gamma}u,v\rangle+
\int_{\mathbb{R}^n}\frac{a(|u|)}{|u|}uv\,dx ⟨ L u , v ⟩ := ⟨ L s , A , γ u , v ⟩ + ∫ R n ∣ u ∣ a ( ∣ u ∣ ) uv d x
fulfills that
⟨ L ~ u k , u k − u ⟩ → 0 as k → ∞ . \langle\widetilde{\mathcal{L}}u_k,u_k-u\rangle\to0
\quad\text{as }k\to\infty. ⟨ L u k , u k − u ⟩ → 0 as k → ∞.
Hence, by Lemma 5.2 below, we get that u k → u u_k\to u u k → u strongly in W 1 , A ( R n ) W^{1,A}(\mathbb{R}^n) W 1 , A ( R n ) .
When γ = 0 \gamma=0 γ = 0 and s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) , and when γ = 1 \gamma=1 γ = 1 (identifying this case with s = 1 s=1 s = 1 ), the reasoning is analogous.
(iv) Let γ ∈ ( 0 , 1 ) \gamma\in(0,1) γ ∈ ( 0 , 1 ) and s ∈ ( 0 , 1 ) s \in(0,1) s ∈ ( 0 , 1 ) . To obtain a positive solution, we choose the mapping P P P defined as P ( u ) = ∣ u ∣ P(u)=|u| P ( u ) = ∣ u ∣ . Trivially, P ( 0 ) = 0 P(0)=0 P ( 0 ) = 0 and P ( v ) = v P(v)=v P ( v ) = v , where v v v is given in (ii). Moreover, it is direct that ∥ ∇ ∣ u ∣ ∥ L A ( R n ) = ∥ ∇ u ∥ L A ( R n ) \|\nabla|u|\|_{L^A(\mathbb{R}^n)}=\|\nabla u\|_{L^A(\mathbb{R}^n)} ∥∇∣ u ∣ ∥ L A ( R n ) = ∥∇ u ∥ L A ( R n ) , ∥ ∣ u ∣ ∥ L A ( R n ) = ∥ u ∥ L A ( R n ) \||u|\|_{L^A(\mathbb{R}^n)}=\|u\|_{L^A(\mathbb{R}^n)} ∥∣ u ∣ ∥ L A ( R n ) = ∥ u ∥ L A ( R n ) and ∥ ∣ u ∣ ∥ L A , α ( R n ) = ∥ u ∥ L A , α ( R n ) \||u|\|_{L^{A,\alpha}(\mathbb{R}^n)}=\|u\|_{L^{A,\alpha}(\mathbb{R}^n)} ∥∣ u ∣ ∥ L A , α ( R n ) = ∥ u ∥ L A , α ( R n ) . Furthermore, since ∣ ∣ u ( x ) ∣ − ∣ u ( y ) ∣ ∣ ≤ ∣ u ( x ) − u ( y ) ∣ \big||u(x)|-|u(y)|\big|\leq|u(x)-u(y)| ∣ u ( x ) ∣ − ∣ u ( y ) ∣ ≤ ∣ u ( x ) − u ( y ) ∣ , and t ↦ A ( t ) t\mapsto A(t) t ↦ A ( t ) is monotone increasing for t ≥ 0 t\geq0 t ≥ 0 , A ( ∣ D s ∣ u ∣ ∣ ) ≤ A ( ∣ D s u ∣ ) A(|D^s|u||)\leq A(|D^s u|) A ( ∣ D s ∣ u ∣∣ ) ≤ A ( ∣ D s u ∣ ) . This gives that ρ s , A ( ∣ u ∣ ) ≤ ρ s , A ( u ) \rho_{s,A}(|u|)\leq\rho_{s,A}(u) ρ s , A ( ∣ u ∣ ) ≤ ρ s , A ( u ) and hence [ ∣ u ∣ ] W s , A ( R n ) ≤ [ u ] W s , A ( R n ) [|u|]_{W^{s,A}(\mathbb{R}^n)}\leq[u]_{W^{s,A}(\mathbb{R}^n)} [ ∣ u ∣ ] W s , A ( R n ) ≤ [ u ] W s , A ( R n ) . All these relations give that for any u ∈ W 1 , A ( R n ) u\in W^{1,A}(\mathbb{R}^n) u ∈ W 1 , A ( R n ) ,
J ( P ( u ) ) ≤ J ( u ) . \mathcal{J}(P(u))\leq\mathcal{J}(u). J ( P ( u )) ≤ J ( u ) .
An analogous (and simpler) analysis can be made in the cases γ = 0 \gamma=0 γ = 0 and s ∈ ( 0 , 1 ) s\in(0,1) s ∈ ( 0 , 1 ) , and γ = 1 \gamma=1 γ = 1 (identified with s = 1 s=1 s = 1 ).
Therefore, the conclusion follow from Lemma 5.3 . This concludes the proof.
Lemma 5.2. Suppose that A A A is an N N N -function satisfying the Δ 2 \Delta_2 Δ 2 -condition and suppose that A ( t ) A(\sqrt{t}) A ( t ) is convex. Let { u k } k ∈ N ∈ W s , A ( R n ) \{u_k\}_{k\in\mathbb{N}}\in W^{s,A}(\mathbb{R}^n) { u k } k ∈ N ∈ W s , A ( R n ) be a sequence such that u k ⇀ u u_k\rightharpoonup u u k ⇀ u and
⟨ L ~ u k , u k − u ⟩ → 0 as k → ∞ , ⟨ L ~ u , v ⟩ : = ⟨ L s , A , γ u , v ⟩ + ∫ R n a ( ∣ u ∣ ) ∣ u ∣ u v d x . \langle\widetilde{\mathcal{L}}u_k,u_k-u\rangle\to0\quad\text{as }k\to\infty,\qquad\langle\widetilde{\mathcal{L}}u,v\rangle:=\langle\mathcal{L}_{s,A,\gamma}u,v\rangle+\int_{\mathbb{R}^n}\frac{a(|u|)}{|u|}uv\,dx. ⟨ L u k , u k − u ⟩ → 0 as k → ∞ , ⟨ L u , v ⟩ := ⟨ L s , A , γ u , v ⟩ + ∫ R n ∣ u ∣ a ( ∣ u ∣ ) uv d x . where v v v is defined in Theorem 5.1 . Then u k → u u_k\to u u k → u strongly in W s , A ( R n ) W^{s,A}(\mathbb{R}^n) W s , A ( R n ) .
Proof. Observe that the weak convergence implies that
⟨ L ~ u k − L ~ u , u k − u ⟩ → 0 as k → ∞ . (41) \langle\widetilde{\mathcal{L}}u_k-\widetilde{\mathcal{L}}u,u_k-u\rangle\to0\quad\text{as }k\to\infty. \tag*{(41)} ⟨ L u k − L u , u k − u ⟩ → 0 as k → ∞. (41)
We will appeal to the following well-known inequality for ξ , η ∈ R n \xi,\eta\in\mathbb{R}^n ξ , η ∈ R n (see [30 Lemma 8] ):
( a ( ∣ ξ ∣ ) ∣ ξ ∣ ξ − a ( ∣ η ∣ ) ∣ η ∣ η ) ( ξ − η ) ≥ C A ( ∣ ξ − η ∣ ) . (42) \left(\frac{a(|\xi|)}{|\xi|}\xi-\frac{a(|\eta|)}{|\eta|}\eta\right)(\xi-\eta)\geq C A(|\xi-\eta|). \tag*{(42)} ( ∣ ξ ∣ a ( ∣ ξ ∣ ) ξ − ∣ η ∣ a ( ∣ η ∣ ) η ) ( ξ − η ) ≥ C A ( ∣ ξ − η ∣ ) . (42)
First, assume γ = 0 \gamma=0 γ = 0 and hence W s , A ( R n ) = W r a d s , A ( R n ) W^{s,A}(\mathbb{R}^n)=W_{\mathrm{rad}}^{s,A}(\mathbb{R}^n) W s , A ( R n ) = W rad s , A ( R n ) . Then
⟨ L ~ u k − L ~ u , u k − u ⟩ = ∫ R n ∫ R n ( a ( ∣ D s u k ( x , y ) ∣ ) ∣ D s u k ( x , y ) ∣ D s u k ( x , y ) − a ( ∣ D s u ( x , y ) ∣ ) ∣ D s u ( x , y ) ∣ D s u ( x , y ) ) ⋅ ( D s u k ( x , y ) − D s u ( x , y ) ) d x d y ∣ x − y ∣ n + ∫ R n ( a ( ∣ u k ( x ) ∣ ) ∣ u k ( x ) ∣ u k ( x ) − a ( ∣ u ( x ) ∣ ) ∣ u ( x ) ∣ u ( x ) ) ( u k ( x ) − u ( x ) ) d x ≥ C ( ρ s , A ( u k − u ) + ρ A ( u k − u ) ) . \begin{aligned}
\langle\widetilde{\mathcal{L}}u_k-\widetilde{\mathcal{L}}u,u_k-u\rangle
={}&\int_{\mathbb{R}^n}\int_{\mathbb{R}^n}\left(\frac{a(|D_su_k(x,y)|)}{|D_su_k(x,y)|}D_su_k(x,y)-\frac{a(|D_su(x,y)|)}{|D_su(x,y)|}D_su(x,y)\right)\\
&\qquad\cdot\left(D_su_k(x,y)-D_su(x,y)\right)\frac{dx\,dy}{|x-y|^n}\\
&+\int_{\mathbb{R}^n}\left(\frac{a(|u_k(x)|)}{|u_k(x)|}u_k(x)-\frac{a(|u(x)|)}{|u(x)|}u(x)\right)(u_k(x)-u(x))\,dx\\
&\geq C\left(\rho_{s,A}(u_k-u)+\rho_A(u_k-u)\right).
\end{aligned} ⟨ L u k − L u , u k − u ⟩ = ∫ R n ∫ R n ( ∣ D s u k ( x , y ) ∣ a ( ∣ D s u k ( x , y ) ∣ ) D s u k ( x , y ) − ∣ D s u ( x , y ) ∣ a ( ∣ D s u ( x , y ) ∣ ) D s u ( x , y ) ) ⋅ ( D s u k ( x , y ) − D s u ( x , y ) ) ∣ x − y ∣ n d x d y + ∫ R n ( ∣ u k ( x ) ∣ a ( ∣ u k ( x ) ∣ ) u k ( x ) − ∣ u ( x ) ∣ a ( ∣ u ( x ) ∣ ) u ( x ) ) ( u k ( x ) − u ( x )) d x ≥ C ( ρ s , A ( u k − u ) + ρ A ( u k − u ) ) .
Hence, (41) gives the strong convergence in W s , A ( R n ) W^{s,A}(\mathbb{R}^n) W s , A ( R n ) . If now γ ∈ ( 0 , 1 ] \gamma\in(0,1] γ ∈ ( 0 , 1 ] , then W s , A ( R n ) = W r a d 1 , A ( R n ) W^{s,A}(\mathbb{R}^n)=W_{\mathrm{rad}}^{1,A}(\mathbb{R}^n) W s , A ( R n ) = W rad 1 , A ( R n ) . Then, as before
⟨ L ~ u k − L ~ u , u k − u ⟩ ≥ C ( γ ρ A ( ∇ u k − ∇ u ) + ρ A ( u k − u ) ) . (43) \langle\widetilde{\mathcal{L}}u_k-\widetilde{\mathcal{L}}u,u_k-u\rangle\geq C\left(\gamma\rho_A(\nabla u_k-\nabla u)+\rho_A(u_k-u)\right). \tag*{(43)} ⟨ L u k − L u , u k − u ⟩ ≥ C ( γ ρ A ( ∇ u k − ∇ u ) + ρ A ( u k − u ) ) . (43)
and the conclusion holds.
Appendix
The following variant of the Mountain-Pass Theorem will be of use for our arguments. See [11 Theorem 10] .
Lemma 5.3 (Brezis–Nirenberg). Let E E E be a Banach space and Φ ∈ C 1 ( E , R ) \Phi\in C^{1}(E,\mathbb{R}) Φ ∈ C 1 ( E , R ) satisfy Φ ( 0 ) = 0 \Phi(0)=0 Φ ( 0 ) = 0 and the following conditions:
(i) there exists r > 0 r>0 r > 0 such that inf ∥ u ∥ = r Φ ( u ) > 0 \inf_{\lVert u\rVert=r}\Phi(u)>0 inf ∥ u ∥ = r Φ ( u ) > 0 ;
(ii) there exists v v v with ∥ v ∥ > r \lVert v\rVert>r ∥ v ∥ > r such that Φ ( v ) < 0 \Phi(v)<0 Φ ( v ) < 0 ;
(iii) the Palais–Smale condition: if Φ ′ ( u k ) → 0 \Phi'(u_k)\to0 Φ ′ ( u k ) → 0 and Φ ( u k ) → c \Phi(u_k)\to c Φ ( u k ) → c , then { u k } k ∈ N \{u_k\}_{k\in\mathbb{N}} { u k } k ∈ N has a subsequence converging to u ∈ E u\in E u ∈ E ;
(iv) there exists a continuous mapping P : E → E P:E\to E P : E → E such that P ( 0 ) = 0 P(0)=0 P ( 0 ) = 0 , P ( v ) = v P(v)=v P ( v ) = v , and Φ ( P ( u ) ) ≤ Φ ( u ) \Phi(P(u))\leq\Phi(u) Φ ( P ( u )) ≤ Φ ( u ) for every u ∈ E u\in E u ∈ E .
Then there exists a critical point u ∗ ∈ P ( E ) ‾ u^*\in\overline{P(E)} u ∗ ∈ P ( E ) such that
Φ ( u ∗ ) ≥ inf ∥ u ∥ = r Φ ( u ) . \Phi(u^*)\geq\inf_{\lVert u\rVert=r}\Phi(u). Φ ( u ∗ ) ≥ ∥ u ∥ = r inf Φ ( u ) .
References
P. Ochoa.
Universidad Nacional de Cuyo, Fac. de Ingeniería. CONICET. Universidad J. A. Maza, Parque Gral. San Martín 5500, Mendoza, Argentina.
Email address: pablo.ochoa@ingenieria.uncuyo.edu.ar
A. Salort.
Departamento de Matemáticas Aplicadas y Ciencia de Datos, Universidad San Pablo-CEU, CEU Universities, Julián Romea 2, 28003 Madrid, Spain
Email address: ariel.salort@ceu.es