Download as pdf
Download as pdf
You are on page 1of 12
COMMUNICATIONS ON PURE AND APPLIED MATHEMATICS, VoL. XIV, 415-426 (1961) K. 0. Friedvichs anniversary issue On Functions of Bounded Mean Oscillation* F. JOHN and L. NIRENBERG § 1. We prove an inequality (Lemmas 1.1’) which has been applied by one of the authors and by J. Moser in their papers in this issue. The inequality expresses that a function, which in every subcube C of a cube Cy can be approximated in the L* mean by a constant ag with an error independ- ent of C, differs then also in the L” mean from ag in C by an error of the same order of magnitude. More precisely, the measure of the set of points in C, where the function difiers from ag by more than an amount ¢ decreases exponentially as o increases. In Section 2 we apply Lemma 1’ to derive a result of Weiss and Zyg- mund [8], and in Section 3 we present an extension of Lemma I’. EMMA 1. Let u(x) be an integrable function defined in a finite cube Co in n-dimensional space; x = (a, ***,,). Assume that there is a constant K such that for every parallel subcube C, and some constant ag, the inequality 1 7 de = K 0) mica holds. Here dx denotes element of volume and m(C) is the Lebesgue measure of C. Then, if u(0) is the measure of the set of points where \u—ae,| > , we have (2) Ha) where B, b are constants depending only on n. Since for every continuously differentiable function f(s), vanishing at the origin, Ber m(C,) for o> 0, Jo, Mu —ao,))ae = JF w(s)a(s), inequality (2) implies that » belongs to L? for every finite p = 1, and, in fact for b' < K-18 the function e!-#c,! is integrable and (1+ gg) mice) (3) f emaglde S eo “This paper represents results obtained under Contract Nonr-286(46), with the Oifice of Naval Research, United States Navy. Reproduction in whole or in partis permitted for any purpose of the United States Government. as 416 F, JOHN AND L. NIRENBERG A function satisfying (1) for every subcube C of Cp, for come constant ag, will be said to have “mean oscillation 0. Then I ju—aglde -[ ulo)do < 3 mc) so that the mean oscillation of « does not exceed B/b. Take now for # the function logle—yl, where y is fixed. Let C be any cube of side h, and let § and 7 be points of C for which [E-g| = Maxey], In—yl = Min Je—yl. ec Take a¢ = log|f—yl. Then for # in C ley [zl S, is the subset of C lying in the sphere |u(2)—ag| = log Iz—yl S |§—yle~*. If S, is not empty, it must contain 7, so that le-yle? & In—yl S lE-yl Fal S [E-yl Va A Thus Vnk -y| S FUNCTIONS OF BOUNDED MEAN OSCILLATION 417 It follows that S, is contained in the sphere _Vnh Roel rma and that its measure (a) does not exceed where the volume of the unit sphere in #-space is denoted by (w,)*. Since also (2) 0, H( where B and 8 do not depend on C. This proves that logla—y] is of bounded mean oscillation in every cube Cy. The same holds then for any function u(z) of the form u(z) =[£(y) logle—yldy with [|t@)ldy < 00. Lemma | will be derived from EMMA I’. Let u(x) be integrable in a cube Co and assume that there is a constant x such that for every parallel subcube C we have rt i ay TiO) glttele S where uo is the mean value of u in C. Then if S, ts the set of points where |u—to,| > 0, its measure m(S,) satisfies ey m(S.) S2f Inmates ese for Loa. edo, a Since m(S,) Sm(Cp) it follows that (2)" m(S,) S et#e**"*m(Cy) for o> 0. Here A E1, a, a are positive numbers depending only on the dimension n. By a standard type of argument we can derive, as consequences of (2)', the following inequalities: for 0 o, then a belongs to one of the J, (except for a set of measure zero). Hence, since the average 1, of « in I, is bounded by 2s in absolute value, we see that m(S_) = mallu (e)| > o} SX mfa||u(2)—m| > o—2"s in 1}. Now in the cube I, the function w—u, satisfies the hypotheses of Lemma V,, in particular it satisfies (1)’ for every cube in J,. Hence, using the definition of F(a), we have m{a|tu(@) ty] > 0-2" in 4} S F(o—2"5) {Ju —uglde S F(o—2"s)m(I,) Thus we find 420 F, JOHN AND L. NIRENBERG F(o—2"s) f lula % m(S,) S F(o—2"s) ¥ m(I,) © by iii), proving (5) Setting s = ¢ in (5) we see that if F(o) < Ae**, « = 1/(2*e) for some o, then 1 Flo+2%e) 57 ders = dersirst"o, From this it follows that if on some interval of length 2*e the inequality F(o) S Ae~* holds, then it holds for all larger o. But a calculation shows that 1m 2M, a, } satisiies 7) F(eth)+Fle—h)—2F (2) =0 (cen) uniformly in x, then F is the indefinite integral of an { belonging to every Ly They also give an example showing that the result does not hold for The proof of the theorem in [3] is rather short but it relies on a theorem of Littlewood and Paley, and it seems of interest to us to show how it may be derived from our Lemma 1’: we prove FUNCTIONS OF BOUNDED MEAN OSCILLATION 421 LEMMA 2. Let u(x) be an integrable function defined in a finite cube Cy in n-space. Assume that there is a constant K and a constant § > } such that if Cy and Co, are any two equal subcubes having a full (n—1)-dimensional face in common, then ee ) Mood SE Tog aE? ere Wg,, to, are the mean values of in the cubes Cy and Cy, and h is the common side length. Then u satisfies the conditions of Lemma V' with some constant x depending on K, B and n so that, consequently, u satisfies (2)" and (3)'. The preceding theorem follows easily from this lemma, By convolution of F with a smooth peaked kernel we may suppose that F is infinitely difter- entiable. It suffices merely to estimate the L, norm of the derivative f of F. Hypothesis (7) asserts simply that f satisfies (8) for n= 1. Applying Lemma 2 we obtain from (2) or (3)” an estimate for the Z, norm of f depending only on K and 8, proving the theorem. From (3)’' we find, furthermore, that eV" is integrable for some a’ > 0. Proof of Lemma 2: Consider a subcube C, of side length k subdivided into 2°¥ equal cubes C,, 7 =1,-++, 2°", obtained by dividing each edge into 2¥ equal parts, and let u, denote the mean value of « in C,. Then r | |u—ug|de = lim 2% S |u,—uel tee tela = fim 2-3 eH ‘Thus to prove (1)’ it suffices to show that 2-"™ 5 |u,—wel S x, with « depending only on K, f and ». By Schwarz inequality, 2-0 Ist, —ttel S [2-7 Sl, tHe PP = ay. We shall prove that the ay are uniformly bounded by showing that nk - k () ayes Say} (ar 2 wp) : ha oR Since dy = 0, it follows that S WKY (14+) log 2—log h/4)-* A ay, for some constant « independent of k, convergence being guaranteed by the fact that 6 > 4. ‘Thus to complete the proof we shall establish (9). We observe first that 2-"" Yu, =e so that using the general identity 422 F. JOHN AND L. NIRENBERG x ADU = (Zh)P+h Y (,—1,)2 for real 8,, we find (10) Nay — BY S hye? = 4S I, Now, on the next subdivision of C into 28/10 cubes each C, is divided into 2* equal cubes C,,, i= 1,-+-, 2, of side length h = 4/21, Tp ty, is the mean value of w'in C., we have (uy) Ha Zt Furthermore, since any two C,,, C,, can be connected by a chain of at most +1 cubes each having a fall face in common with the succeeding one, we find from (8) that nk “al Sey = + [log hi where M is so defined. This together with (11) implies Im, M, Mat] SM. According to formula (10) nv a 22M ayy = rg — ty, = E Lepet ah) — 20,15] = (thee —t)* + (ty) ®+ ty te, +24, —18 ayy] = lls —1)*+ ty —4)9] + 2 (uh + ut) 2248) 2.2% 0), by (11), <2MADINNI An L906 Sy, 8 = Bam gage gtarsng,, by (10), or Ay Say Mt This is the desired inequality (9) and the proof is complete. 0, FUNCTIONS OF BOUNDED MEAN OSCILLATION 423 § 8. In this section we present briefly a generalization of Lemma 1’. LEMMA 3. Let ube integrable in a finite cube Cy and consider a subdivision of Cy into a denumerable number of cubes C,, no two having a common interior point. Assume that jor fixed p, 1

9, satisfies K,/ mSe) SA «| for some constant A depending only on n and p. The result implies that the function » belongs to L* for every p' o—2"5}. < Applying the induetion hypothesis (13), for j—1, to the functions y, in I, we find 0-3 yey bee zag (2 wit) Ha oe ah gift s S01 Sayer Rye, by Holder's inequality, 2p(lae" ee ce oan 7 Gans CR fi ee Co y (17) and (15), so that C1 gute (1 uw mS.) SRY alle \wide) A slightly tedious calculation shows that this inequality is identical with the desired result (13). Having established (13) we may now express it in a more convenient form: if (12) holds, then, in virtue of (14), there is a constant & depending only on m and p such that aye!) m(Se) Sh (@) (Cy) FUNCTIONS OF BOUNDED MEAN OSCILLATION 425 or om (Ca) Ry | If now 2-"¢ = Kym (Cy) and we choose the largest integer j = 0 so that (12) is satisfied, we have the opposite inequality for j-+1: om (Co)? MME 2" (pgitt—2) +1) Samp Inserting into the previous inequality we find (Se) & &(%2)" arparpr™ @ K,\? SA @ for o 2 2"K,m(Cy)-”, o for some constant A depending only on n and p. Since m(S,) S m(C,), the same inequality holds for all ¢ > 0, with some other constant A, and the proof of the lemma is complete. § 4, Inequality (2) in Lemma 1’ can be replaced by the more general inequality (2)'"" m(Sgu2g) < Ae PR m(S,) for o>0 with A, B depending only on ». Proof: Let « = 1, #9, = 0. Fora fixed positive s the cubes J, shall be defined as in the proof of Lemma 1’. Put Hx(a) = m(z| |«(@)| > o in J). By definition, (co) is non-increasing and does not exceed m(Z,). By (2)"” applied to Ip, x(a) S m(2| u(z) a4] > o—2"5 im Ty) < ete t-20 (Ty) Then om(l,) S [, wld = [> axle) = [P mlode + [i malolde + J mlo)aa s 5 (ta) + (25) un (3) + = erm) It follows for s > at2"+#0*# that 426 F, JOHN AND L. NIRENBERG 1 at mp) Sr eee 4, (245) Then also mm(Sun) & Da (3) B Beem paler) “1e-99 8" (Syuis,) for s > nitgee Inequality (2) is an immediate consequence Bibliography [1] Calderon, A. P., and Zygmund, A., On the existence of certain singular integrals, Acta Math., Vol. 88, 1952, pp. 85-139, [2] Hormander, L., Estimates for translation invariant operators in Ly spaces, Acta Math., Vol. 104, 1960, pp. 93-140. [3] Weiss, M., and Zygmund, 8., A mole ox smooth functions, Nederl. Akad, Wetensch, Inda- gationes Math., Ser. A, Vol. 62, 1959, pp. 52-58. Received February, 1961.

You might also like