We now turn our attention to W1,p(Ω) in the case p>n. In this case, we obtain considerably more regularity. We first establish the decisive inequality.
Theorem 5.1 (Morrey's inequality). Suppose n>1 and p∈]n,∞], and let γ=1−pn. There exists a constant C depending only on n and p such that for any u∈C1(Rn),
∣∣u∣∣0+γ≤C⋅∣∣u∣∣1,p.
Proof. Fix x∈Rn and r>0. We proceed in steps:
Step 1. There exists a constant C>0 depending only on n such that
It now suffices to show that μ(Ω)μ(B(x,∣x−y∣)) is bounded from above by a constant depending only on n. To see that this is the case, note that z:=x+21(y−x)∈Ω and B(z,21∣y−x∣)⊂Ω so that
Combining all of the above then establishes the claim.
As with the Gagliardo-Nirenberg-Sobolev inequality, we may use this inequality to prove an embedding theorem.
Corollary 5.2. Suppose n>1, V⊂Rn is open, p∈]n,∞[ and γ=1−pn. There exists a continuous embedding
ι:H01,p(V)↪Cγ(V)
such that for all u∈H01,p(V), ι(u)≡u almost everywhere. In particular, if Ω⊂Rn is a bounded domain of class C1, then we also obtain the existence of a continuous embedding ι:W1,p(Ω)↪Cγ(Ω) such that for all u∈W1,p(Ω), ι(u)≡u almost everywhere.
Proof. For clarity, we write the elements of H01,p(V) as equivalence classes [u].
Consider the inclusion ι:C01(V)⊂H01,p(V)→Cγ(Ω) defined by [u]↦u. Morrey's inequality implies that for all u∈C01(V)⊂C1(Rn),
[ι([u])]γ≤c(n,p)⋅∣∣[u]∣∣1,p.
By Lemma 4.5, ι extends uniquely to a continuous linear mapping ι:H01,p(V)→Cγ(Ω). If {un} is an approximating sequence for some [u]∈H01,p(V), then since both {[un]} and {[ι([un])]=[un]} converge in Lloc1(V) by Hölder's inequality, we must have that ι([u])≡u almost everywhere for all u∈H01,p(V), i.e. ι is a continuous embedding with the desired property. The latter claim follows from choosing V such that Ω⋐V and considering the following chain of continuous embeddings, the last one being a restriction:
W1,p(Ω)↪H0k,p(V)↪Cγ(V)↪Cγ(Ω)
Remark 5.3. In the case where n=1, we have the inclusion
W1,p(]a,b[)↪C1−p1(]a,b[),
which depends only on an application of Hölder's inequality and the characterisation of W1,1(]a,b[) in terms of absolutely continuity. In fact, for u∈W1,p(]a,b[), there exists a v∈C1−p1([a,b]) such that u≡v a.e. and
∣v(y)−v(x)∣≤∣∣u′∣∣p⋅∣y−x∣1−p1
for all x,y∈]a,b[.
Remark 5.4. Note that if Ω is a bounded domain of class C1 and u∈W1,∞(Ω), then u∈W1,p(Ω) for all p<∞ so that u∈C1−pn(Ω) for every p>n after possibly redefining u on a set of measure zero. By the proof of Morrey's inequality, we actually have the estimate
for x,y∈Ω with x=y. We may take the limit p→∞ on both sides to obtain
∣x−y∣∣u(x)−u(y)∣≤2n⋅∣∣Du∣∣∞.
This implies that u is actually Lipschitz continuous. The same argument applies for u∈H01,p(V) provided V is bounded.
Higher-order Sobolev inequalities
By appropriately successively applying the Gagliardo-Nirenberg-Sobolev inequality and Morrey's inequality, we obtain the following higher-order Sobolev embeddings:
Theorem 5.5. Suppose Ω⊂Rn is a bounded domain of class C1.
If k<pn and q∈R satisfies q1=p1−nk, then there is a continuous embedding ι:Wk,p(Ω)↪Lq(Ω) with ι(u)≡u almost everywhere.
If k>pn, then there is a continuous embedding ι:Wk,p(Ω)↪CK,γ(Ω) with ι(u)≡u almost everywhere, where K=k−⌊pn⌋−1 and
γ={⌊pn⌋−pn+1,any positive number <1,pn∈Npn∈N.
Remark 5.6. The idea here is that depending on how weakly differentiable u is relative to n and p, it is either in a higher Lp space or a (higher) Hölder space. For example, fixing m∈N and p=2, we may ask the following question: How large should k be so that Wk,2(Ω)↪Cm(Ω)? According to the latter case, we ought to have K≥m, i.e. k≥m+⌊2n⌋+1. Therefore, if u∈Wk,2(Ω) for all k∈N, then u∈C∞(Ω). These higher embedding theorems will be crucial in proving classical differentiability of weak solutions to elliptic PDE later on in the course.
Compact embeddings
Recall that the Gagliardo-Nirenberg-Sobolev inequality implies the continuous embedding
W1,p(Ω)↪Lp∗(Ω)(↪Lq(Ω))
whenever Ω is a bounded domain of class C1 and p∈[1,n[, where p∗=n−pnp and q∈[1,p∗[. It turns out we can say more about the latter embedding.
Theorem 5.7 (Rellich-Kondrachov). Suppose Ω⊂Rn is a bounded domain of class C1. If p∈[1,n[ and q∈[1,p∗[, then the embedding W1,p(Ω)↪Lq(Ω) is compact.
Proof. We shall show that the embedding H01,p(V)↪Lq(V) is compact for V⊂Rn open and bounded. Choosing V so that V⋑Ω then establishes the claim (why?). Let {um}m=1∞⊂H01,p(V) be a bounded sequence. We proceed in steps.
where θ∈]0,1[ is such that q1=θ+p∗1−θ. Since ∣∣Jεum∣∣p∗≤∣∣um∣∣p∗ and by Gagliardo-Nirenberg-Sobolev ∣∣um∣∣p∗≤const⋅∣∣Dum∣∣p≤const, we immediately see that Jεumε↘0um in Lq uniformly in m.
Step 2. For fixed ε>0, the sequence {Jεum}m=1∞ is uniformly bounded and equicontinuous.
Step 3. For fixed δ>0 there exists a subsequence {umj} of {um} such that j,k→∞limsup∣∣umj−umk∣∣q≤δ.
Subproof. By Step 1, there exists an ε>0 such that ∣∣Jεum−um∣∣q≤2δ for allm∈N, and by Step 2, we may appeal to the Arzelà-Ascoli theorem to pass to a subsequence {Jεumj} converging in C(V), i.e.
j,k→∞limVsup∣Jεumj−Jεumk∣=0.
This implies that limj,k→∞∣∣Jεumj−Jεumk∣∣q=0. Altogether, we obtain that
Now, applying Step 3 with δ=1, we extract a subsequence {umj1} of {um} with j,k→∞limsup∣∣umj1−umk1∣∣q≤1. More generally, for each l∈{2,3,…}, we inductively define the subsequence {umjl}j=1∞ of {umjl−1}j=1∞ obtained from Step 3 with δ=l1, i.e. such that
j,k→∞limsup∣∣umjl−umkl∣∣q≤l1
for each l∈N. Finally, we consider the sequence {vj}j=1∞ such that vj=umjj, which is a subsequence of {um}. Moreover, for each N∈N, {vl}l=k∞ is a subsequence of {umjN}j=1∞ whenever k≥N so that
Since N is arbitrary, we conclude that ∣∣vj−vk∣∣j,k→∞0 so that {vj} is the desired subsequence of {um}.
Remark 5.8. Since p∗=n−pnp>p and p∗→∞ as p↗n, we have that the embedding W1,p(Ω)↪Lp(Ω) is compact for p∈[1,n]. Moreover, for p>n, we may use Morrey's inequality and the Arzelà-Ascoli theorem to conclude that this embedding is compact in the case where p>n. Using similar techniques, we also deduce that the embedding H01,p(Ω)↪Lp(Ω) is compact for all p.