Since u,∂iu∈Lloc1, we may take limits above to obtain the relation
∫Ωu⋅∂iφ=−∫Ω∂iu⋅φ,
i.e. whenever p⋅(α+1)<n, we have that u∈W1,p(Ω).
Example 2.2. Let {rk}k=1∞⊂Ω:=B(0,1) be dense and define u:Ω→R such that x↦∑k=1∞2−k⋅∣x−rk∣−α. It may be shown that u∈W1,p(Ω) iff p⋅(α+1)<n⇔α<pn−p. If α>0 as well, then u is not bounded on any open subset of Ω.
Sobolev spaces via smooth functions
We now turn our attention to alternative characterisations of weak derivatives and Sobolev spaces.
Derivatives in the Lp sense (p<∞)
Definition 2.3. Suppose u∈Llocp(Ω). We say that Dαu exists in the Lp sense if there is a v∈Llocp(Ω) such that for every U⋐Ω, there exists a sequence {uj}j=1∞⊂C∣α∣(Ω) such that ujj→∞u and Dαujj→∞v in Lp(U). We then say v=Dαu in the Lp sense and call v the αth Lp derivative of u.
Exercise 2.4. If v=Dαu in the Lp sense, then v=Dαu in the weak sense.
As it turns out, the converse is also true for u,Dαu∈Llocp(Ω).
Lemma 2.5. If u∈Llocp(Ω) and Dαu exists in the weak sense and is in Llocp(Ω), then Dαu is the αth Lp derivative of u.
Proof. Let U⋐Ω be fixed. We know from Theorem 1.1 that Jεuε↘0u in Lp(U). We now show that for sufficiently small ε>0, DαJεu=JεDαu so that uj:=J1/ju is the desired sequence.
Now, by definition, for each x∈U, by the dominated convergence theorem,
In order to shift the derivative Dzα onto u, we need to verify that z↦ρ(εx−z) is compactly supported in Ω for sufficiently small ε>0. Now, we know that supp(z↦ρ(εx−z))⊂B(x,ε). Taking ε<dist(U,∂Ω), we see that B(x,ε)⋐Ω. Therefore, we may throw the derivative onto u:
⇒DαJεu(x)=ε−n∫Ωρ(εx−z)⋅Dαu(z)dz=JεDαu(x).
The claim now follows from our remarks above.
Remark 2.6. In analogy with Llocp, we say that u∈Wlock,p(Ω) if u∈Wlock,p(U) for every U⋐Ω, and given a sequence {ui}⊂Wlock,p(Ω), we say that uii→∞u in Wlock,p(Ω) if uii→∞u in Wk,p(U) for every U⋐Ω. In particular, the above lemma tells us that for every u∈Wlock,p(Ω), we may find a sequence {ui:=J1/iu}i=1∞⊂C∞(Ω) converging to u in Wlock,p(Ω).
The spaces Hk,p and H0k,p(Ω)
In light of the previous section, we know that we may approximate Sobolev functions locally by means of smooth functions. Recall from Theorem 1.1 that for p<∞, Lp(Ω)=C0∞(Ω). Motivated by this, we set
Ck,p(Ω):={f∈Ck(Ω):∣∣f∣∣k,p<∞}
and introduce the subspaces
Hk,p(Ω)H0k,p(Ω):=Ck,p(Ω):=C0∞(Ω)
of Wk,p(Ω). Note that Ck,p(Ω) is not closed in Wk,p(Ω).
[[Counterexample: Ω=]−1,1[, k=1, p=1, fn(x)=x2+n1]]
It is clear that H0k,p(Ω)⊂Hk,p(Ω)⊂Wk,p(Ω). We will show that the latter inclusion is actually an equality. We first recall the following technical lemma.
Lemma 2.7. Let {Uα} be an open cover of a set A⊂Rn. There exists a countable family of smooth functions {ψi:Rn→[0,∞[} such that
for each p∈A, there exists an open set Up∋p such that ψi∣Up≡0 for all but finitely many i;
∑iψi≡1; and
for each i, there exists an α such that suppψi⋐Uα.
If the cover {Uα} is countable and locally finite, then we may find a collection of functions {ψα} with the same index set as the cover satisfying the above conditions. In either case, we call the collection {ψi} a partition of unity.
Theorem 2.8. If p<∞, then Hk,p(Ω)=Wk,p(Ω).
Proof. We will show that any u∈Wk,p(Ω) may be approximated by a v∈C∞(Ω)∩Ck,p(Ω). This will then establish the claim.
For each k∈N, let Ωk=B(0,k)∩{x∈Ω:dist(x,∂Ω)>k1}. Clearly Ω=⋃kΩk and Ωk⋐Ωk+1⋐Ω for each k. Now set Ω0:=∅ and
Ui:=Ωi+1\Ωi−1
for each i∈N. We now have that ⋃i=1kUi=Ωk+1 so that {Ui}i=1∞ forms an open cover of Ω. Moreover, clearly Ui∩Ωj=∅ for all j≤i−1, which implies that {Ui}i=1∞ is a locally finite cover of Ω. Let {ψi}i=1∞ be a partition of unity subordinate to this cover (with the same index).
By Remark 2.4, Jδψi⋅uδ↘0ψi⋅u in Wk,p(Ω) so that for each ε>0 and i∈N, we may choose a δi<(i+1)(i+2)1<dist(Ωi+1,∂Ω) such that
∣∣Jδi(ψi⋅u)−ψi⋅u∣∣k,p<2i+1ε.
This choice of δi ensures that suppJδ(ψi⋅u)⊂Ωi+2\Ωi−2 so that for each j∈N we have the following equalities when restricting to Ωj:
By the monotone convergence theorem, we may take the limit of both sides as j→∞ to obtain ∣∣u−v∣∣k,p≤2ε<ε.
Remark 2.9. The above claim is false for p=∞ for much the same reason L∞ functions cannot be approximated by continuous functions.
Exercise 2.10. u∈W1,1(]a,b[) iff u′ exists almost everywhere, u′∈L1(]a,b[) and the fundamental theorem of calculus holds, i.e. for any x,y∈]a,b[ with x<y,
u(x)−u(y)=∫yxu′.
The case Ω=Rn is particularly special.
Theorem 2.11. If p<∞, then Wk,p(Rn)=H0k,p(Rn).
Proof. Let v∈C∞(Rn)∩Wk,p(Rn) be such that ∣∣u−v∣∣k,p<2ε. Let φ:Rn→[0,1] be a smooth function such that φ≡1 on B(0,1) and φ≡0 outside B(0,2). For each δ>0, define the function
vδ(x):=v(x)⋅φ(δ⋅x).
Clearly vδ∈C0∞(Rn). Set Vδ:={x∈Rn:∣x∣>δ1}. The triangle inequality implies that
where C is a constant depending only on α, where we assume that δ≤1. From the above and another application of the triangle inequality, we are left with the inequality ∣∣v−vδ∣∣k,p≤C0⋅∣∣v∣∣k,p,Vδ. However, if w∈Lp(Rn), then by the monotone convergence theorem,
∫Rn∣w∣p=δ↘0lim∫Rn\Vδ∣w∣p⇔δ↘0lim∫Vδ∣w∣p.
Therefore, we may choose δ sufficiently small so that ∣∣v∣∣k,p,Vδ<2C0ε. This then implies that