Download as pdf
Download as pdf
You are on page 1of 5
5.5. TRACES 257 7. We henceforth write Eu := @, and observe that the mapping ut+ Eu is linear. Recall that the construction so far assumed u € C®({). Suppose now u € WIP(U), and choose tm € C%°(T) converging to u in Wh?(U). Esti- mate (9) and the linearity of Z imply Bum — Euilly2qR) S Cllum — wllweewy- Thus {Eum}22_, is a Cauchy sequence and so converges to i =: Eu. This extension, which does not depend on the particular choice of the approxi- mating sequence {t,, }°°_), satisfies the conclusions of the theorem. a Remarks. (i) Assume now that QU is C?. Then the extension operator E constructed above is also a bounded linear operator from W??(U) to W??(R"). To see this, note first in steps 3, 4 of the proof that although & is not in general C?, it does belong to W??(B). We also have the bound llallwee(a) S Cllullw2ecet), which follows from the definition (3). As before, we consequently derive the estimate (10) |Eullw2ea) < Cllullw2eu), provided 8U is C?, the constants C depending only only on U,V,n and p. ‘We will need these observations later. (ii) The above construction does not provide us with an extension for the Sobolev spaces W*?(U), if k > 2. This requires a more complicated higher-order reflection technique. oo 5.5. TRACES Next we discuss the possibility of assigning “boundary values” along QU to a function u € W1?(U), assuming that OU is C!. Now if u € C(U), then clearly u has values on OU in the usual sense. The problem is that a typical function u € W}?(U/) is not in general continuous and, even worse, is only defined a.e. in U. Since @U has n-dimensional Lebesgue measure zero, there is no direct meaning we can give to the expression “u restricted to GU”. ‘The notion of a trace operator resolves this problem. For this section we take 1 < p < oo. 258 5. SOBOLEV SPACES THEOREM 1 (Trace Theorem). Assume U is bounded and QU is C'. Then there exists a bounded linear operator T : Wi?(U) = LP(aU) such that (i) Tu=ul if we WU) NCO) and (ii) [Tull zou) S$ Cllullwrewys for each u€ W1(U), with the constant C depending only on p and U. DEFINITION. We call Tu the trace of u on OU. Proof. 1. Assume first u € C'(Z). As in the first part of the proof of ‘Theorem 1 in §5.4 let us also initially suppose 2° € OU and QU is flat near x°, lying in the plane {zy = 0}. Choose an open ball B as in the previous proof and let B denote the concentric ball with radius r/2. Select ¢ € C2°(B), with ¢ > 0 in B, ¢ = 1 on B. Denote by I that portion of OU within B. Set 2’ = (21,...,¢n-1) € RR"! = {an = O}. Then firairde’ sf cure’ =~ f (cluP)e, de Tv {en=0} ‘B+ @) == ff uP, + plu Moen, de . < ef. \ul? + [Dul? de, where we employed Young’s inequality, from §B.2. 2. If 2° € OU, but OU is not flat near 2°, we as usual straighten out the boundary near <° to obtain the setting above. Applying estimate (1) and changing variables, we obtain the bound [iweas sc f {ul? + [Dul? dr, Ir ft where I’ is some open subset of OU containing 2°. 3. Since QU is compact, there exist finitely many points «9 ¢ OU and open subsets I C OU (i= 1,...,N) such that QU = UN, Ti and .N). llullueey SCllulwse@) 5.5. TRACES: 259 Consequently, if we write Tu =ulov, then (2) Tull cocav) S Cllullws.2v) for some appropriate constant C, which does not depend on u. 4, Inequality (2) holds for u € C1(0). Assume now u € W!?(U), Then there exist functions tim € C°(Z) converging to u in W1?(U). According to (2) we have (3) Tum — Puillze(avy S Clltm — wllweewys 50 that {Tum}22_, is a Cauchy sequence in L?(3U). We define oom lim Tum, mado the limit taken in L?(AU). According to (3) this definition does not depend on the particular choice of smooth functions approximating u. Finally if u € W)(U) N C(O), we note that the functions um, € C°(T) constructed in the proof of Theorem 3 in §5.3.3 converge uniformly to u on U. Hence Tu = ulov- o We next examine more closely what it means for a function to have zero trace. THEOREM 2 (‘Trace-zero functions in W1?). Assume U is bounded and OU is C'. Suppose furthermore that u€ W1?(U). Then (4) ueWy?(U) if and only if Tu=0 on dU. Proof*. 1. Suppose first u € Wg'(U). Then by definition there exist functions um € C°(U) such that tim win WH), As Tum = 0 on OU (m = 1,...) and T : W1?(U) > LP(AU) is a bounded linear operator, we deduce Tu = 0 on OU. 2. The converse statement is more difficult. Let us assume that (5) Tu=0 ond. “Omit on first reading. 260 5. SOBOLEV SPACES Using partitions of unity and flattening out OU as usual, we may as well assume - uéW'?(R"), uw has compact support in R", (6) * * Tu=0on dR? =R" Then since Tu = 0 on R"-1, there exist functions u, € C!(R’) such that (7) Um + u in W(R2) and (8) Tim = Umlgr-1 +0 in LP(R"="). Now if 2’ € R"“}, xp > 0, we have buna!) SMa 0)1+ fo” binant) a Thus if Ina! an) ae’ <0 f luna’, 0)|® da? i fens sort [ a [Dum(2',1)|? a'at) . 0 Re-1 Letting m — oo and recalling (7), (8), we deduce: (9) f lula", a)/? de! < Oxe? i " ii [Dul? deat R= 0 SR for a.e. ay > 0. 3. Next let ¢ € C®(R) satisfy ¢=1on (0,1), ¢=0 on R-(0,2], 0O<¢ <1, and write { Gn(z) = ¢(man) (@ RY) Wy = U(2)(1 = Gn). Then Wren = Ura (1 — Gm) — mug? { Dyrtiim = Dyti(1 — Cm). Consequently f [Dim — Dul? de < cf [Gnl?|Dul? dr Re, Re, (10) fim ' +Cn! f Jen ful? do'ae =A+B. 5.6. SOBOLEV INEQUALITIES 261 Now (1) A=0 asm—oo, since G, # 0 only if 0 < 2, < 2/m. To estimate the term B, we utilize inequality (9): 2/m 2/m B

You might also like