Basic Properties of Sobolev Spaces
1.1.18 Removable Sets for Sobolev Functions
If such an operator exists for a domain then, by definition, this domain belongs tothe class EVpl.
Thus, domains with smooth boundaries are contained inEVpl. The bounded domains of the class C0,1 turn out to have the same property. The last as- sertion is proved in Stein [724], §3, Ch.VI (cf. also Comments to the present section).
In Sect. 1.5 we shall return to the problem of description of domains in EVpl.
1.1 The SpacesLp(Ω),Vp(Ω) andWp(Ω) 29 denote the projection of a pointx∈Rn on the coordinate hyperplane orthog- onal to thexi axis. By assumptions, each set pi(F) has (n−1)-dimensional Lebesgue measure zero. Thus, almost every straight line that is parallel to the xi axis is disjointed fromF. According to Fubini’s theorem, we have
Ω
η ∂u
∂xi dx=
Ω
dx
(x)
η∂u
∂xidxi, 1≤i≤n,
whereΩ =pi(Ω), x =pi(x) and (x) is the intersection ofΩ with the line x = const. Note that(x) is in Ω\F for almost allx∈Ω. An application of Theorem 1.1.3/2 leads to
Ω
dx
(x)
η ∂u
∂xi
dxi=−
Ω
dx
(x)
u∂η
∂xi
dxi−
Ω
u∂η
∂xi
dx.
Hence (1.1.28) follows for |α|=l= 1. The general case can be concluded by induction onl. Indeed, letl≥2 and let (1.1.28) hold foru∈Vpl−1(Ω\F) with
|α| ≤l−1. Ifu∈Vpl(Ω\F), |α|=l, andDα =DiDβ for some 1≤i≤n, the left-hand side of (1.1.28) equals
−
Ω
∂u
∂xi
Dβηdx, (1.1.29)
sinceu∈Vp1(Ω\F). By the induction hypothesis, expression (1.1.29) coincides with the right-hand side of (1.1.28) (because ∂u/∂xi ∈ Vpl−1(Ω\F)). This
completes the proof.
Let us show that the conditionHn−1(F) = 0 in the above theorem cannot be replaced by the finiteness ofHn−1(F).
Example.LetΩ={x∈R2:|x|<2} and letF be the segment{x∈Ω: x2 = 0,|x1| ≤1}. We introduce the set S ={x∈Ω :|x1| ≤ 1, x2 >0} and define the functionuonΩby
u(x) =
0 onΩ\S,
exp(−(1−x21)−1) onS.
We haveH1(F) = 2, u∈Vpl(Ω\S) for anyl, butu /∈Vpl(Ω).
1.1.19 Comments to Sect.1.1
The space Wpl(Ω) was introduced and studied in detail by Sobolev [711–
713]. (Note that as early as in 1935 he also developed a theory of distribu- tions in (C0l)∗ [710].) The definitions of the spaces Llp(Ω) and ˙Llp(Ω) are borrowed from the paper by Deny and Lions [234]. The proofs of Theo- rems 1.1.2, 1.1.5/1, 1.1.12, 1.1.13/1, and 1.1.13/2 follow the arguments of
this paper where similar results are obtained forl = 1. Theorem1.1.5/1 was also proved by Meyers and Serrin [599]. Concerning the contents of Sect.1.1.3 we note that the mollification operator Mε was used by Leray in 1934 [487]
and independently by Sobolev in 1935 [709]. A detailed exposition of proper- ties of mollifiers including those with a variable radius can be found in Chap. 2 of Burenkov’s book [155]. For a history of the mollifiers see Naumann [624].
The property of absolute continuity on almost all straight lines parallel to coordinate axes served as a foundation for the definition of spaces similar to L1p in papers by Levi [489], Nikod´ym [637], Morrey [612], and others. The example of a domain for whichW22(Ω)=V22(Ω) (see Sect.1.1.4) is due to the author. The example considered at the beginning of Sect. 1.1.6is borrowed from the paper by Gagliardo [299]. Remark1.1.6and the subsequent example are taken from the paper by Kolsrud [440]. Theorem 1.1.6/1 is given in the textbook by Smirnov [705] and Theorem 1.1.6/2 was proved in the above- mentioned paper by Gagliardo. The topic of Sect.1.1.6was also discussed by Fraenkel [285] and Amick [46]. The following deep approximation result was obtained by Hedberg (see [370]).
Theorem. Let Ω be an arbitrary open set in Rn and let f ∈W˚pl(Ω) for some positive integer l and somep,1< p <∞. Then there exists a sequence of functions{ων}ν≥1,0≤ων ≤1, such that suppων is a compact subset ofΩ, ωνf ∈(L∞∩W˚pl)(Ω), and
νlim→∞f−ωνfWpl(Ω)= 0.
An ingenious approximation construction for functions in a two-dimension- al bounded Jordan domain was proposed by Lewis in 1987 [492] (see also [491]). He proved that C∞( ¯Ω) is dense in Wp1(Ω) for 1< p < ∞. Unfortu- nately, this construction does not work for higher dimensions, for p= 1, and for the space Wpl(Ω) with l > 1. If a planar domain is not Jordan, it may happen that forl >1 bounded functions in the spaceLlp(Ω) are not dense in Llp(Ω) (Maz’ya and Netrusov [572], see Sect.1.7in the sequel).
The problem of the approximation of Sobolev functions on planar domains byC∞ functions, which together with all derivatives are bounded onΩ, was treated by Smith, Stanoyevitch, and Stegenga in [707], where some interesting counterexamples are given as well.
In [346] Hajlasz and Mal´y studied the approximation of mappingsu:Ω→ Rmfrom the Sobolev space [Wp1(Ω)]mby a sequence{uν}ν≥1of more regular mappings in the sense of convergence
Ω
f(x, uν,∇uν) dx→
Ω
f(x, u,∇u) dx for a large class of nonlinear integrands.
During the last two decades a theory of variable exponent Sobolev spaces W1,p(·)(Ω) was developed, where the Lebesgue Lp-space is replaced by the
1.1 The SpacesLp(Ω),Vp(Ω) andWp(Ω) 31 Lebesgue space with variable exponent p(x). This topic in not touched in the present book, but we refer to Kov´acˇik and R´akoˇsnik [461] where such spaces first appeared and to the surveying papers by Diening, H¨ast¨o and Nekvinda [236], Kokilashvili and Samko [439] and S. Samko [690]. In partic- ular, the density of C0∞(Rn) in W1,p(·)(Rn) was proved by S. Samko [688]
under the so-called log-condition onp(x), standard for the variable exponent analysis:
p(x)−p(y)≤ const
|log|x−y||
for small |x−y|. This question is nontrivial because of impossibility to use mollifiers directly. The Hardy type inequality in variable Lebesgue spaces are studied by Rafeiro and Samko [669].
In connection with Sect. 1.1.7 see the books by Morrey [613], Reshet- nyak [677], and the article by the author and Shaposhnikova [578], see also Sect. 9.4 in the book [588].
The condition of being starshaped with respect to a ball and having the cone property were introduced into the theory ofWpl spaces by Sobolev [711–
713]. Lemma 1.1.9/2 was proved by Glushko [312]. The example given in Sect.1.1.9is due to the author; another example of a Lipschitz domain that does not belong toC0,1can be found in the book by Morrey [613]. Properties of various classes of domains appearing in the theory of Sobolev spaces were investigated by Fraenkel in [285]. Fraenkel’s paper [286] contains a thorough study of the conditions on domains Ω guaranteeing the embedding of the spaceC1( ¯Ω) inC(0,α)( ¯Ω) whenα >0.
Integral representations (1.1.8) and (1.1.10) were obtained by Sobolev [712, 713] and used in his proof of embedding theorems. Various generalizations of such representations are due to Il’in [396, 397], Smith [706] (see also the book by Besov, Il’in, and Nikolsky [94]), and to Reshetnyak [675]. We follow Burenkov [154] in the proof of Theorem 1.1.10/1.
The Poincar´e inequality for bounded domains that are the unions of do- mains of the class C was proved by Courant and Hilbert [216]. Properties of functions in Llp(Ω) for a wider class of domains were studied by J.L. Li- ons [499].
Stanoyevitch showed [720] (see [721] for the proof) that the best constant in the one-dimensional Poincar´e inequality
u−u¯Lp(−1,1)≤C∇uLp(−1,1)
is equal to 1 ifp= 1 and
(p)1/pp1/p
Γ(1/p)Γ(1/p) = psin(π/p) π(p−1)1/p,
if 1 < p < ∞. Moreover, a unique extremal function exists if and only if 1< p≤ ∞.
Theorem1.1.16on equivalent norms inWpl(Ω) is due to Sobolev [713, 714].
The extension procedure described at the beginning of Sect.1.1.17was pro- posed by Hestenes [378] for the spaceCk( ¯Ω) (see also Lichtenstein [494]). The same method was used by Nikolsky [638] and Babich [59] forWpl(Ω). The fact that a space-preserving extension for functions in Wpl(Ω) (1< p <∞) onto Rn is possible for domains of the classC0,1was discovered by Calder´on [163].
His proof is based on the integral representation (1.1.8) along with the theo- rem on the continuity of the singular integral operator in Lp. A method that is appropriate forp= 1 andp=∞was given by Stein [724]. The main part of his proof was based on the extension of functions defined in a neighborhood of a boundary point. Then, using a partition of unity, he constructed a global extension. For the simple domain
Ω=
x= (x, xn) :x∈Rn−1, xn> f(x) ,
where f is a function onRn−1satisfying a Lipschitz condition, the extension ofuis defined by
u∗(x, xn) = ∞
1
u
x, xn+λδ(x, xn)
ψ(λ) dλ, xn < f(x).
Here δ is an infinitely differentiable function, equivalent to the distance to
∂Ω. The functionψis defined and continuous on [1,∞), decreases asλ→ ∞ more rapidly than any power of λ−1, and satisfies the conditions
∞
1
ψ(λ) dλ= 1,
∞
1
λkψ(λ) dλ= 0, k= 1,2, . . . .
More information on extension operators acting on Sobolev spaces can be found in Sect.1.5.
Theorem1.1.18on removable sets for Sobolev functions is due to V¨ais¨al¨a [771] (see also Reshetnyak [677, Chap. 1, 1.3]). Koskela showed thatWp1 re- movability of sets lying in a hyperplane depends on their thickness measured in terms of a so-called pporosity [455].