Sobolev Versus Homogeneous Sobolev Extension

By Proposition 3.1, for 1\leq q\leq p\leq \infty, a bounded (L^{1,p},L^{1,q})-extension domain is a (W^{1,p},W^{1,q})-extension domain.

Towards a partial converse, let B\subset{{{\mathbb R}}^n} be a large enough ball such that {\Omega}\subset \subset B. Since {\Omega} is a (W^{1, p}, W^{1, q})-extension domain, for every u\in W^{1, p}({\Omega}), there exists a function E(u)\in W^{1, q}(B) such that E(u)\big|_{{\Omega}}\equiv u and

where C>0. Let \{u_j\} be a bounded sequence in W^{1,p}(\Omega). By (?), \{E(u_j)\} is also a bounded sequence in W^{1,q}(B). Since 1\leq q\leq p<q^\star, we can use the Rellich-Kondrachov theorem to conclude that there is a subsequence which converges in L^p(B) to some v\in L^p(B). Especially, this subsequence converges to v|_{\Omega} in L^p(\Omega). Hence W^{1,p}(\Omega) is compactly embedded in L^p(\Omega). Then, by using Theorem (2.1), we conclude that, for every u\in W^{1, p}({\Omega}), we have

where C>0 is independent of u.

Fix u\in L^{1, p}({\Omega}). For every n\in\mathbb N, define a function u_n on {\Omega} by setting

Since |\Omega|< \infty and u\in L^{1}_{loc}(\Omega), by using an exhaustion argument we get that for some m_0>0 it holds that

Set

For m,n \geq m_0, u_m -u_n =0 on A and u_m -u_n \in W^{1,p}(\Omega). If v\in W^{1,p}(\Omega), then by using Lemma 2.1, we get that

By using inequality (?) for v=u_m-u_n, we conclude that

This yields

Hence, \{u_m\} is a Cauchy sequence in W^{1,p}(\Omega) and by definition \{u_m\} converges point-wise to u in \Omega, which implies that u \in W^{1,p}(\Omega). So, for u\in L^{1,p}(\Omega), we also have (u-u_{\Omega})\in W^{1,p}(\Omega). We define a function on B by setting

Then T(u)\in W^{1, q}(B) with T(u)\big|_\Omega\equiv u. By using inequality (?) and the fact {\Omega} is a (W^{1, p}, W^{1, q})-extension domain, we have

Inequality (?) and the fact that B is an (L^{1,q}, L^{1,q})-extension domain show that {\Omega} is an (L^{1, p}, L^{1, q})-extension domain.

We have

Let {\Omega}\subset{\mathbb R}^n be a bounded (L^{1, p}, L^{1, q})-extension domain for 1\leq q\leq p\leq \infty and let B\subset{\mathbb R}^n be a ball with {\Omega}\subset\subset B. Since {\Omega} is an (L^{1, p}, L^{1, q})-extension domain, for every u\in W^{1,p}({\Omega})\subset L^{1,p}({\Omega}), we find E(u)\in L^{1,q}({\mathbb R}^n) with E(u)\big|_{\Omega}\equiv u and

where C>0. By [M, Section 1.1.2], we have E(u)\big|_B\in W^{1, q}(B). Suppose first that q<\infty. Using Lemma 2.1 and (Proof), we obtain

The triangle inequality and Hölder’s inequality together with (Proof) give

Define an extension operator \widetilde E on W^{1, p}({\Omega}) by setting

Inequality (Proof) tells us that \tilde E is a bounded extension operator from W^{1, p}({\Omega}) to W^{1, q}(B). Finally, the result that {\Omega} is a (W^{1, p}, W^{1, q})-extension domain comes from the fact that each ball B is a (W^{1, q}, W^{1, q})-extension domain.

The Case q=\infty follows similarly from the estimate

Also when q<\infty and p=\infty, we repeat the same arguments by replacing (Proof) with

Fix a continuous u\in W^{1,\infty}(\Omega) and x_{0},y_{0}\in \Omega. We may assume that u(x_{0})=1 and u(y_{0})=0. By truncation, we may further assume that 0\leq u(x)\leq 1 on \Omega. Then \|u\|_{L^{\infty}(\Omega)}\leq 1. Let B\subset\mathbb{R}^n be a ball such that \overline{\Omega}\subset B. Since \Omega is a bounded (W^{1,\infty},W^{1,q})-extension domain with q>n, there exists a function E(u)\in W^{1,q}(B) such that E(u)|_{\Omega}\equiv u and

where C>0.
By Morrey’s inequality and (Proof), we have

By using (Proof) and our normalization, we get that

Define \delta >0 by the equation

By using (Proof), we get that

whenever |x_0-y_0|<\delta.
Hence (Proof) gives

under the assumption that |x_0-y_0|<\delta.

Fix j such that \sqrt{n}2^{-j}\leq \delta and consider a dyadic decomposition \mathcal{D} of \mathbb{R}^n into closed cubes of edge length 2^{-j}. Since {\Omega}\subset\mathbb R^n is bounded, there are only finitely many cubes, say M\in \mathbb{N} in \mathcal D which intersect {\Omega}. Let x, y\in{\Omega} be arbitrary and \gamma\subset{\Omega} be a locally rectifiable curve joining x and y. There exists a finite class of dyadic cubes \{Q_j\}_{j=1}^N\subset\mathcal D which intersect \gamma such that x\in Q_1 and y\in Q_N with N\leq M. Set x_0:=x and x_{N}:=y. Choose x_1 to be the last point on along the curve \gamma which is contained in the union of the cubes Q_j containing x_0. We continue with x_1 taking the role of x_0. By repeating this argument (at most \hat{N} times), we obtain \{x_0=x,x_1,x_2,\cdots,x_{\hat{N}}\}, \hat{N}\leq N with x_j\in \Omega and so that

Hence (Proof) together with triangle inequality gives

for all x,y \in \Omega.

Let \Omega\subset \mathbb{R}^n be a bounded (W^{1,\infty},W^{1,q})-extension domain with q>n. Then by Lemma 4.1, there exists a constant C such that

for all x,y \in \Omega and every continuous u\in W^{1,\infty}(\Omega).

Hence for u\in W^{1,\infty}(\Omega), by considering a continuous representative of u, we conclude that

where C>0 is independent of u.

Since every u\in W^{1,\infty}(\Omega) satisfies (?), we conclude that \Omega is a (L^{1,\infty}, L^{1,q})-extension domain by again defining Tu:= E(u-u_\Omega)+u_\Omega as at the end of the proof of Theorem 1.2.

It follows from the construction that \Omega is quasiconvex and hence (W^{1,\infty}, W^{1,\infty})-extension domain [HKT:JFA]. It is clear that {\Omega}_1 is an (\epsilon, \delta)-domain for some \epsilon, \delta>0. By [Jones:acta], we have a bounded linear extension operator from W^{1, q}({\Omega}_1) to W^{1, q}({{{\mathbb R}}^n}) for every 1\leq q\leq\infty. Hence, it suffices to construct a bounded extension operator from W^{1, p}({\Omega}) to W^{1, q}({\Omega}_1) when q^\star<p<\infty. It is not difficult to check that {\Omega} satisfies the segment condition defined in Definition 2.3. By Lemma 2.2, C^\infty(F)\cap W^{1, p}({\Omega}) is dense in W^{1, p}({\Omega}). To begin, we define a bounded linear extension operator \widetilde{E_W} from C^\infty(F)\cap W^{1, p}({\Omega}) to W^{1, q}({\Omega}_1) and later extend it to W^{1, p}({\Omega}). Given u\in C^\infty(F)\cap W^{1, p}({\Omega}), we define a function \widetilde{E_W}(u) on {\Omega}_1 by setting

where

is the integral average of the function u over the reflected cube Q^\star and \{\psi_Q\} is the partition of unity associated to the Whitney decomposition of \mathcal W(F^c).

Since \psi_Q are smooth and, for every compact subset of {\Omega}_1\setminus F, only finitely many \psi_Q are non-zero on it, \widetilde{E_W}(u) is smooth on \Omega_1\setminus F. Hence, it suffices to show that the Sobolev norm of \widetilde{E_W}(u) is uniformly bounded from above by the Sobolev norm of u. To begin, we have

Since the volume of {\Omega} is finite, the Hölder inequality implies

Now

and for every x\in C_k\setminus\mathscr C_k, there are only finitely many Q\in\mathcal W^2_k such that

Since

for every k\in\mathbb N, we have

Combining (?), (?) and Hölder’s inequality, we obtain

Using (?) and (?), we deduce that

By using inequalities (?) and (?), we get that

Similarly, we have

Let k\in\mathbb N. For x\in C_k\setminus\mathscr C_k , there are only finitely many Q\in\mathcal W^2_k such that

Hence, for x\in\bigcup_{k} C_k\setminus\mathscr C_k, we have

Since

we have

For Q\in\bigcup_k\mathcal W_k^1, we have

with a positive constant C independent of u and Q. By using (?) and the properties of our partition of unity, there exists a constant C>1 such that for every x\in Q and each \tilde Q\in\mathcal W_Q, we have

Given \tilde Q\in\mathcal W_Q^1, there exists a chain \{Q_i^{\star}\}_{i=1}^m of length at most M which connects Q^\star and \tilde Q^\star and satisfies the properties in Lemma 5.1. The triangle inequality implies

Since there exists a constant C>1 such that for every 1\leq i\leq m-1, we have

the triangle inequality and [Jones:acta, Lemma 2.2] give that

By Lemma 5.1, there exists a constant C>1 such that for all Q_i^\star, Q_j^\star\in\big\{Q_k ^{\star}\big \}_{k=1}^m, we have

Since n\leq M, by combining inequalities (?) and (?), we obtain

Hence, for \tilde Q\in\mathcal W_Q^1, we have

If \tilde Q\in\mathcal W_Q^2, then Q^\star and \tilde Q^\star are at different sides of the hyperplane P_k. Then the triangle inequality, (?) and the Hölder inequality imply that

By combining (?), (?), (?) and (?), we conclude that

For each Q\in\bigcup_k\mathcal W_k^1, the cardinality of \mathcal W_Q^1 is uniformly bounded by a constant C. By combining (5.1) and the Hölder inequality, we obtain

Given j\in\mathbb N, we define

and

Let j\in\mathbb N and k\leq j. Let Q\in\mathcal W_k^1 be such that Q^\star\in\mathcal W_j({\Omega}) and \mathcal W_Q^2 is non-empty. Then, for such a Q we have

where 0<c<1 is independent of j,k. By the geometry of {\Omega}, there exists C>1 such that for each Q\in\mathcal W_k^1 with non-empty \mathcal W_Q^2, we have

Hence, for every j\in\mathbb N and each k\leq j, the cardinalities of the sets \widetilde{\mathcal W}_j(F^c)\cap\mathcal W_k^1 and \mathcal W_j({\Omega}) are uniformly bounded from above by a constant C. Hence, the cardinality of the set \widetilde{\mathcal W}_j(F^c) is bounded from above by Cj. For each Q\in\bigcup_k\mathcal W_k^1, the cardinality of the set \mathcal W_Q^2 is uniformly bounded from above. Since 1\leq q<n and nq/(n-q)<p<\infty, by (?), we have

By combining (?), (?) and (?), we obtain

Using Hölder’s inequality and inequalities (?) and (?), we conclude that

By combining (?) and (?), we obtain that the extension operator constructed in (Proof) is a bounded linear extension operator from C^\infty(F)\cap W^{1, p}({\Omega}) to W^{1, q}({\Omega}_1), which means that for every u\in C^\infty(F)\cap W^{1, p}({\Omega}), we have \widetilde{E_W}(u)\in W^{1, q}({\Omega}_1) with

where C>0 is independent of u. Since C^{\infty}(F)\cap W^{1, p}({\Omega}) is dense in W^{1, p}({\Omega}), for every u\in W^{1, p}({\Omega}), there exists a Cauchy sequence \{u_m\}\subset C^\infty(F)\cap W^{1, p}({\Omega}) which converges to u in W^{1, p}({\Omega}). Furthermore, there exists a subsequence of \{u_m\} which converges to u almost everywhere in {\Omega}. To simplify the notation, we still refer to this subsequence by \{u_m\}. By (?), \{\widetilde{E_W}(u_m)\} is a Cauchy sequence in the Sobolev space W^{1, q}({\Omega}_1) and hence converges to some function v\in W^{1, q}({\Omega}_1). Furthermore, there exists a subsequence of \{\widetilde{E_W}(u_m)\} which converges to v almost everywhere in {\Omega}_1. On the other hand, by the definition of E(u_m) and E(u), we have

for almost every x\in{\Omega}_1. Hence v(x)=E_W(u)(x) almost everywhere. This implies that E_W(u)\in W^{1, q}({\Omega}_1) with

where C>0 is independent of u. Hence, E_W defined in (5.1) is a bounded linear extension operator from W^{1,p}({\Omega}) to W^{1, q}({\Omega}_1). The fact that {\Omega}_1 is a (W^{1, q}, W^{1, q})-extension domain implies {\Omega} is a (W^{1, p}, W^{1, q})-extension domain for all 1\leq q< q^\star<p<\infty.

To prove that {\Omega} is not an (L^{1, p}, L^{1, q})-extension domain, it suffices to show that, given M<\infty, there is u \in L^{1, p}({\Omega}) whose every extension satisfies

Let k\in\mathbb N. We define

where 0<l_k<2^{-3k-1} is sufficiently small and will be fixed later.

We define a function \tilde{u}_k on \mathscr C_{k} by setting

And we define u_k on \Omega by setting

According to p equals to n or not, we divide the following argument into two cases.

Case 1(p=n): In this case, we have

Hence u_k \in L^{1,n}(\Omega).

Let B_k be ball with radius \sqrt{2}\cdot2^{-k} and which is centered at x_k= (z_k, -2^{-k-1}). Suppose that E(u_k)\in L^{1,q}(\mathbb{R}^n) is an extension of u_k with q<n. The Sobolev-Poincaré inequality gives

Let A_1=((B^{n}_{k} \cap \widetilde{\mathscr C}_{k}^{+})\cup \mathscr C_{k}^{-})\cap \Omega and A_2= \mathscr C_{k}^{+}\cap \Omega\setminus (\tilde B^{n}_{{k}}\cap \widetilde{\mathscr C}_{k}^{+}). Then

Furthermore

Hence, we have

By combining (?) and (?), we conclude that

By using (?) and (?) and choosing

we obtain

In conclusion,

when k is sufficiently large.

Case 2(p<n): We use Hölder’s inequality to estimate

Hence, by using (?), we have

Similarly to the case above, Let E(u_k) be an extension of u_k. Then by using (?) and (?) and choosing

we obtain that for 1\leq q< p<n, we have

Hence, again \|Eu_k\|_{L^{1,q}(\mathbb{R}^n)}\geq M \|u_k\|_{L^{1,p}(\Omega)} when k is sufficiently large.