oF}52. The Elements of Iteration sothat 4, 6X, dy © Ayy2sand X= dy According to Lemma4.s, an Jord < J toe fran. Since the sequence (4,) is monotone increasing and has union X, it follows from Lemmas 4.3(6) and 3.4(a) that Jods =tim [ode ‘Therefore, on taking the limit in (4.7) with respect to, we obtain fora 0, then of belongs toM™ and Jefde=e frau. (©) I hoe belong 10M, then f+ belongs to M* and Jus erde= frde+ fede The Iiegral 33 Proor. (a) Ife =O the result is immediate. If ¢ > 0, let (%) be ‘a monotone increasing sequence of simple functions in M* converging to fon X (see Lemma 2.11). Then (cp) is a monotone sequence converging to ¢f. If we apply Lemma 4.3(a) and the Monotone Convergence Theorem, we obtain Jere = tim fe ce = elim fonda = © f £4 () If () and (J) ate monotone increasing sequences of simple functions converging to f and g, respectively, then (py + ox) i8 ‘monotone ineteasing sequence converging to f + g. It follows from [Lemma 4,3{2) and the Monotone Convergence Theorem that for oe = tim fio, + 494 =n fonda + en fn ce ~ fra firds. ano. ‘The next result, A consequence of the Monotone Convergence ‘Theorem, is very important for it enables us to handle sequences of functions that are not monotone. 48 Farou’s Leva. 1/(f) belongs 10 M*(X, X), then 4s) Joi int) de < tim int f fae PROOF. Let fq inf an Sness:-} 89 that Bq < fq Whenever mn. Therefore Jute f[nm men so that Jed < timiat ff de ‘Since the sequence (g) is increasing and converges to lim inf f,, the34 The Elements of Integration Monotone Convergence Theorem implies that Joie ing A) du = tim fen de << timintf fd ae, 1 willbe seen in an exercise that Fatou's Lemma may fail ift is not assumed that f, > 0. 49 conse fe Ma ead oy as 0 = [7a then 4 is a measure. PROOF. Since f > 0 it follows that NE) > 0. If E= 0, then fe vanishes everywhere so that 4(0) = 0. To see that A is countably additive, let (Ey) be a disjoint seqyence of sets in X with union E and let f, be defined to be Son Shin Ie follows from Corollary 47(0) and induction that Jt 3 [srate= $69 Since (f,) is an increasing sequence in M* converging to je, the ‘Monotone Convergence Theorem implies that MB) = | Free = tim J fade 4.10 Conoutany. Suppose that f belongs to M*. Then f(x) = 0 soalmost everywhere on X if and only if Soe, ose (410) rae ROOF. If equation (4.10) olds, let B= {rex ste) > The Integral 35. s0 that f > (1/m) xe, from which 0m frau > 1aed > 0 It follows that w(E,) ence the set wer s9>o-0 & sho ta mea, Convene) = 0 palo eye E=(xeX:f)> 0), then u(E) = 0. Let fy = mxx- Since f < lim inf fy, it follows from Fatou's Lemma that 1 0< fran 0, then the mapping ¢-+ $= op is a ‘one-one correspondence between simple function @ in M* with » < f The Inert 37 and simple functions in M'* with y < cf. Use this observation to give a diferent proof of Corollary 4%). LE. Let fg belong to M'*, let be a simple function in M* with SJ, and let w be a simple function in MT with w 0, even in the presence of uniform convergence. 440. Fatou's Lemma has an extension toa case where the f tke fon negaive values. Let fi be in M*(X,X), and. suppose that Jide < 400, IU) a sequence in M(X, X) and if A < fos then iit omit Way doesn't Exercise 4.0 apply to Exercise 4N? Wye a(x, X)and Jfae< +0, then ice : fle) = $0) = 0, (Hts CE, = (ee 2 fle) > mh, then mye, 0) is eie (hat there exits a sequence (Fin Xsuch that © U and (,) < 42) 45. ye M*(X, X)and Jian < +, thn or ane» Oe eit at Fe ech ht () < Hand fra [sare AT. Suppose that (,) C M*(X,X), that (f) converges to f, and that [4s = tim [ode < 440 The Elements of Integration Prove that er eee 4.0. Show that the consusion of Exesie 4 may fui ifthe condition tn [ae 40 i net CHAPTER 5 Integrable Functions In Definition 44 we defined the integral of each function in M+ = M°(X,X) with respect to a measure and permitted this integral to be “+20. In this chapter we shall discuss the integration fof measurable functions which may take on both positive and negative real values. Here itis more convenient to require the values of the functions and the integral to be finite real numbers 54 Deriwinion. The collection L = L(X, X, ») of integrable (or summable) functions consists ofall real-valued X-measurable functions {defined on X, such that both the positive and negative parts [*./~+ ‘of fhave finite integrals with respect to In this case, we define the integral of f with respect t0 4 to be 6) frum [rr de forte ICE belongs to X, we define [tae [rae [rd [Aitiough the integral of fis defined to be the diference of the integrals of f°, i is easy to Soe that iff = fy ~ Ja where fifa are ‘any nonnegative measurable functions with finite integrals, then frden frie fide.42 The Blements of neration In fact, since f* ~ J = f= fi ~ fy, it follows that f* + fu =f + J-. Iwe apply Corollary 4.7(6), we infer that frre fridum frat fran. Since all hese terms ae rite, we obtain frae= [rte [r-dum fit fea 52 Linus. 1ffbelongs to L and dis defined on X to Ry 2) ME) = ff, ‘hen Nis a charge, PROOF. Since f* and f~ belong to M*, Corollary 4.9 implies that the functions A* and 2°, defined by wef fae e= fF des ‘are measures on X; they are finite because fe L. Since A= A* — A it follows that Ais charge. een. ‘The function A defined in (5.2) is frequently called the indefinite integral of / (with respect to n). Since Ais a charge, i (E,) i e disjoint sequence in X with union £, then free [4 We refer to this relation by saying thatthe indefinite incegral uf function ‘nL is countably additive, ‘The next result is sometimes referred to as the property of absolute integrability of the Lebesgue integral. The reader will recall that, although the absolute value of a (proper) Riemann integrable function js Riemann integrable, this may no longer be the case for a function Which has an improper Riemann integral (for example, consider ‘Sf(a) = 3°* sin x on the infinite interval 1-< x < +0), Intesrale Functions 43 5.3 Tweone, 4 measurable function f belongs to L if and only if Ly Belongs 10 L. In this case 63) | fra] < finde. Roor. By definition fbelongs to Lif and only iff and f~ belong. to M* and have finite integrals. Since [f|* = [f| = f* + and I/\- =, the assertion follows from Lemma 4.5(8) and Corollary 4.1). Moreover, [fral= [fra frra| = af", whence Ia > 0, then f) Sotde = fore de ~ fora {fre dn~ [7 de} = 0 free ‘The case « < 0 is handled similarly.The Elements of Integration If and g belong to L, then [f| and |g| belong to L. Since LF + 1 < L/| + |e! it follows from Corolaries 4.7 and 5.4 that f+ g belongs to L. To establish the desired relation, we observe that Sten +e U +8), Since f* + g* and f~ + g~ are nonnegative integrable functions, it follows from the observation made after Definition 5.1 that Jurade fir vende fy + eran. {we apply Corollary 47(6) and rearrange the tems, we obtain Jurarden fred frrdus for ae fer du = [ras fede. ep. We shall now establish the most important convergence theorem for integrable functions. 56 Lenescut Dominated Cosvencence Tuzone, Let (f,) be @ sequence of integrable funcions which converges almost everywhere 10 a rralsvalued measurable fnction f. If there exists an integrable funciton 4g such that | fal < g for all n, then fis integrable and 60 frac in fa noor. By redefining the functions f,fon a set of measure 0 we ‘nay asume thatthe convergence takes pace onal of X. It follows from Corollary 5.4 that fis integrable, Since g +f. > 0, we ean apply Fatou's Lemma 48 and Theorem 5 to obiain fetes [rie = fle + nae timiat fog + 4946 = tinior (fede [) = fete tim fa Iegrabe Factions 48 “Therefore, it follows that 65) Jd < timint fa Since g ~ f, > 0, another application of Fatou's Lemma and Theorem 5.5 yields Jet [rte fle—pae simi fe = s04 = fede —timsup fae, from which it follows that 6 uinap [J < fra ‘Combine (5.5) and (5.6) to infer that Jae =tin fa ano DEPENDENCE ON A PARAMETER Frequently one needs to consider integrals where the integrand depends on a real parameter. We shall show how the Lebesgue Dominated Convergence Theorem ean be used in this connestion. For the remainder of this chapter we shall let f denote a function defined on X x (a, 6] to R and shall assume that the function x—> ‘See, 1) is Xemeasurable for cach 1 (a, 6]. Additional hypotseses will be stated explicty 5.7 ConoLtaRy. Suppose that for some tein [a, 8] on L051) = lin fxs) Jor each x X, and that there exists an integrable function g en X such tha [fe O| < ax). Then J pes 10 46s) = tin fe. at.45 The Elements of Integration PROOF. Let (i) be a sequence in [a, 6] which converges to fo, and apply the Dominated Convergence Theorem to the sequence (/,) defined by fs) = fe, 1) for xe X. QED. 58 Conottany. If the function t—> f(x, 1) is continuous on (a, 8) {for each x€ X, and if there is an integrable function g on X such that [fee 0)| < el), then the function F defined by 6) #0) = | fos.) do) ‘és continuous for t in (a, 6). #00. This san immediate consequence of Corollary $7, 0. 59 Conouany. Suppose that Jor some ty [a,b], the function Xft, te) 18 integrable on X, that fer exists on X x {a,b}, and that there exists an integrable function g on X such that feo] f(x, 1) is integrable for each + in [a,6]. Hence, i # , then Flt) ~ FO) 81) = 8.9 yen), Since this integrand is dominated by (x), we may apply the Dominated Convergence Theorem to obtain the stated conclusion. ep. 5.10 Conottany. Under the hypotheses of Corollary 5.8, [rode f [fendi] a SU soe] ao, here the integrals with respect to are Riemann inte oor. Real ht is contnaos on fb thn Affnodnno, acres. Let h be defined on X x (a, 6] by mx, 0) [rene 1 follows that (@h@n\e,0) = f(x). Sine this Riemann integral exists, i isthe limit of a sequence of Riemann sums; hence the map +A, ismeasurabe foreach t, Moreover, since [/C, | < £0), we infer that [a |< (016 ~ a), a0 thatthe function x—> A isitegrable foreach 1¢[a, 8]. Let be defined om (, 6) by Ho = fice, antes ‘it fotlows from Corollary 5:9 that Teo = fF crane = [rend = FO.48 The Elements of Integration “Therefore we have [roam ne - He = Jeb) — Wr, a date) =f [fl ena] aco. ‘The interchange ofthe order of (Lebesgue) integrals will be considered in Chapter 10. ann. EXERCISES S.A. Ife L(X, X, panda > 0, show thatthe set (xe X : [/(2)| > 4} has finite measure. In addition, the set (xe X : f(s) #0) has finite measure (that is, the union of a sequence of measurable sets with finite measure). - S.B. Iffis an X-measurable real-valued function and if f(x) = 0 for almost all xin X, then fe LOX, Xp) and frano + SC. Ife L(X, X, ») and g is an X-measurable real-valued function such that (x) = g(x) almost everywhere on X, then g € L(X, X, ») and Srae~ fede. S.D. If feL(X,X, ») and e > 0, then there exists a measurable simple function » such that Jr-oluce SE. If feL and g is a bounded measurable function, then the product fg also belongs to L. S.P. If belongs to L, then it does not follow that f? belongs to L. Imegrabe Fusions 8 5G, Suppose that fis in L(X, X, ») and that its indefinite integral is na) [fa ex Show that A(E) > O forall Ee X if and only if /(x) > 0 for almost all x€X. Moreover, NE) = Oferall Fifand only if f(x) = Ofor almost axe xX. S.H. Suppose that f, and f, are in L(X, X, ») and let A, and Ay be their indefinite integrals. Show that A,(E) = 4,(E) for all Ee X if and only if f(x) = f(a) for almost all xin X. S.L If fis a complex-valued function on X such that Re fand Im belong to L(X, X, pi), we say that fis integrable and define [rae resin ims Let f be a compesalued measurable function. Show chat fis integrable if and only if | i integrable, in which case | fra] < fine. [Htin: HE fd = re with r @ real, consider g(x) = e°#/(0).1 54, Let (f) be # sequence of complex-valued measurable functions Which converges to f. If there exists an integrable function such that [fa] < gy show that Jae tim fan. % SK, Let X= IN, let X'be all subsets of NV, and let 1 be the counting measure on X. Show that f belongs to L(X, X, a) if and oaly if the series 5 f(r) is absolutely convergent, in which case SiL, IF (f) is a sequence in L(X, X, n) which converges uniformly fon X to. function f, and if w(X) < +, then fre = tim f fo dne$0 The Elements of Iteration S.M, Show that the conclusion in the Exercise $.L may fail if the hypothesis .X) < +20 is dropped. SIN, Let fa = mxo,amms Where X= R, X= B, and wis a Lebesgue measure, Show that the condition [f_| < g cannot be dropped in the Lebesgue Dominated Convergence Theorem. 5.0. If fee L(X, X, 1), and if 3 fina < +e, then the series 3 (2) converges almost everywhere to a function fn HX,X.0). Moreover, 5 Jaw fre= 5, Let feL(X,X, pw), and suppose that (f,) converges to a function f. Show that if tim’ ffi = 0, then fide = i ffl de. SQ. It1> 0,then L i: ot de Moreover, if #3 a> 0, then e-™ < e-®, Use this and Exercise n and to obtain the formula SiR, Suppose that f is defined on Xx a,b] to R and that the function x» flt,2) is Xemeasurable for each r¢[a, 6]. Suppose that for some fg and f in (a, 8] the function x» f(x, tis integrable fon X, that (@/20(x,) exists, and that there exists an integrable function g on X such that emery Th for xe X, and rela, 8], ¢# t. Then [fre dc0), fix, 1) is X-measurable for cach eR, and the function £> flx, 1) is continuous on R for each x © ¥ In addition, suppose that there are integrable functions g, h on X’sech that |, | < (2) and such that the improper Riemann integral [Ot niar <0) Show that J [feo deco] a= [ [77 servo at] doco, where the integrals with respect to # are improper Riemann integrals, 5.7. Let fbe an Xemeasurable function on Xto R. For n€N. let (Up be the sequence of truncates of f (see Exercise 2K). If fis integrable with respect to u, then [rae tiom fide. sup [ihe < +0, Conversely, if then fis integrableCHAPTER 6 The Lebesgue Spaces L, It is often useful to impose the structure of a Banach space on the set of all integrable functions on a measure space (X,X,4). In addition, we shall introduce the L,, 1 < p < co, spaces which occur frequently in analysis. Aside from the intrinsic importance of these spaces, we examine them here partly to indicate applications of some ofthe results in the earlier sections 6.1 Denixmon. If Visa real linear (~ vector) space, then a real: valued function on V is said to be @ norm for ¥ in cae it satisfies (NO) > Oforall ve Ys Gi) MO) = Oif and only if » = 0; Gi) N(we) = lalN() for all ve Vand real es (Muto) < NW + NO forallu, ve V. | condition (iis dropped, the function Wis said to bea semi-norm or a pseudo-nori for V. A normed linear space isa linear space V together ‘with @ norm for V, 62 ExaMPLes. (@) The absolute value function yields a norm for the real numbers. (©) The linear space R* of ntuples of real numbers can be normed by defining Malus sot) = fal +o lle Nalias st) = (lle bee ful, p21, Nels cyt) = sup {Ways ead ‘The Lebesgue Spaces L, $3 It is easy to check that N, and Ng are norms and that N, satisfies (), (id, (id. Ie is @ consecuence of Minkowski's Inequality, which will be proved subsequently, that, satisfies (iy. (©) The linear space ‘, of all real-valued sequences 1 = (u) such that Ns(W) = Z|uyl < #00 is a normed linear space under Nj Similarly, if 1

0 for fe Ly and that NUch) = flofl de = tol [flare = lela oreover it follows from the Triangle Inequality that nugtay= fire site s fr + ledae = funda field = MU + Nae Hence Ny isa semisnorm on L, and it follows from Corollary 410 that V.{f) = 0 if and only if f(x) = O for almost all x. GED. tn order to make L(Y, X,1) into a normed linear space, we shall identify two functions that are equal almost everywhere; that i, We Use equivalence classes of functions instead of funetions. 6.6 Denies. Two fonctions in L = LOX, X,p) are sid to be suite if they are equal pnalmost everywhere. The equivalence ‘lass determined by in Lis sometimes denoted by {1 and consists of tho set ofall finetions in L-which are wequivalent tf. The Lebesgue space Ly = LQ, X,p) consists ofall yequivalence eases in La If [7 belongs to L;, we define its norm by on unis = fuse 6.7 Tasonss. space ‘The Lebesgue space L(X,X,#) (8 a normed linear us wl | | EFT Shrrcreve Bhaces 2, | ROOF. It s understood, of course, that the vector operations in are defined by (N= (fl, U4 t= +a). ‘and that the zero element of Ly is (0). We shall check only 1 equation (6.1) gives a norm on Ly. Certainly [fl]: > O and j(0}. 0. Moreover, if [L/Ils = 0 then Jina =o, $0 f(x) = 0 for pealmost all x. Hence [f]= [0]. Fnsly itis ew seen that properties (i) and Gv) of Definition 6. are satisfied. Th fore | |. yields a norm on Ly, a TE should alvays be remembered thatthe elements of Ly are act cquivalence classes of functions in L. However, itis both conven land customary to regard these elements as being functions, and Shall subsequently do so. Thus we shall make reference to equivalence clas (/ by referring to “the element fof Ly,” and we vite [fin place of ILI} THE SPACES L,,1. 1 and (1p) (ig) = 1. Then fee, and | fats < Iflelte: moor. Let « be a real number satisfying O < « < 1, and consider the funetion g defined for ¢ > 0 by jaar wis easy to check that g(f) < OforO <1 < Land p(f) > Ofore > 1 It follows from the Mean Value Theorem of ealculus that ¢(?) > 9(1) and that 9(2) = g(I), if and only if = 1. Therefore we have Heat -o), 120. 1 a, 8 are nonnegative, and if we let = ab and multiply by 6, we ‘obtain the inequality eb gaat (lab, ‘where equality holds if and only ifa = 5. [Now let p and q satisfy 1 < p< eo and (Ip) + (I/g) = 1 and take «= Lip. It follows that if 4, B are any nonnegative real numbers, then ae oo aacek and thatthe equality holds if and only if 4” = BY Suppose that feL, and ge, and that [fly # 0 and gle # 0, ee eT le and (64) with 4 = |/(a)Iflp and ‘The product fis. measur B= |e(0|Male implies that deed ¢ Alt , Leonie ifs Lele alse * igh” Since both of the terms on the right are integrable, it follows from Corollary 5.4 and Theorem 'S.S that fe is integrable, Moreover, on integrating we obtain Viel 141 Visteh $2 * 4 which is Hélder’s Inequality oxo, Holder's Inequality implies thatthe produet of a funetion in Land 8 function in Z, is integrable when p > | and q satisfies the relation (Up) + (Wa) = 1 oF, equivalently, when p+ q = pq. Two numbers satisfying this elation are said to be conjugate indices. It will be noted that p = 2 is the only seconjagate index. Thus the product of two functions in Lis integrable 6.10 Caucay-Bunvaxovskit-Scuvwaaz INEQUALITY. belong t0 La, then fie integrable and 63) | fsa] < fide < ite tale 6.11 Mixkowski's INsquatity, then f +h belongs to L, and 66) Ut Ab < le + Mile: PROOF. ‘The case p = I has already been treated, so we suppose p> 1. The sum+ hisevidently measurable, Since If + AF < Rup (JAN < 24/1 + JAP) it follows from Corollary $4 and Theorem 5.5 that f+ he Ly GON AP = WF ALLE APE LST He Le P Since f+ heL,, then [f+ hl? eL,; since p = (p — 1g it follows that WS and g US and b belong t0 Ly, p> More-58 The Elements of Integration [f+ He-MeL,. Hence we can apply Holder's Inequality to infer that firiure mesa < Lnle{ fir areoe an) ~ le Is + Al, 1 ns treat the second term on the righ in (6.7) similarly, we obtain USF Hh? ULF + ALP + DLS L+H (Ufly + Va) A+ AL VA = [f+ hy =0, then equation (6.6) is vial. If A #0, we can divide the above inequality by 4°*; since p — plg = 1, we obtain Minkowski’s Inequality. geo. tis readily seen thatthe space L, isa Linear space and that formula (63) defines a norm on L,.. The only nontrivial thing to be checked here isthe inequality 6.1@9) and this is Minkowski's Inequality. We shall now show that L, is complete under this norm in the following 6.12 DeFINrrioN, A sequence (Jf) in Ly is a Cauchy sequence in 1, if for every positive number « there exists an. M(e) such that if min > M(), then [fy —fallp N(@), then [/—fely <¢. A normed linear space is complete if every Cauchy sequence converges to some element of the space. 6.13 Lenin. Uf the sequence (f.) converges 10 fin Ly, then itis @ Cauchy sequence. proor. Ifm,n > M(¢/2), then U-filb 0 there exists an M(@) such that if mm > Me) then 68) [ite = slr de Ua fbt M(@) and k is suficienty large, then finer cer Apply Fatou's Lemma to conclude that Jie - Aras < timin fifa = aura < e, wvheneverm > AM(@). This proves tha the sequence () converess to Fin the norm of Ly ano, ‘A complete normed linear space is usually called a Banach space ‘Thus the preceding theorem could be formulated: the space Ly is @ Banach space under the norm given in (63). THE SPACE ZL. We shall now introduce a space which is related to the L,-spaces 6.15 Derisrmios. The space Le = La(X, X,) consists of all the equivalence classes of Xemeasurable real-valued functions which are almost everywhere bounded, wo functions being equivalent when they are equal yealmost everywhere. If fe Le and Ne X with a(N) = 0, wwe define SW) = sup (L/G)] : x EN) and (10) fle y An clement of Lis called an essentially bounded function inf (SUM) : NX, aN) 1c follows (see Exercise 6.7) that if feLe, then [/(0)] & [fe for simost all x. Moreover, if 4 < [fla then there exists a set £ with positive measure such that [f(9)| > Afor x€ £, Iti algo clear that the norm in (6.10) is wellefined on Le The Lebeseue 6.16 THEOREM. The space Le is a complete normed linear space under the norm given by formula (6.10) PRoor. It is clear that Le is a linear space and that flu > 0, [Olle = 0, and Jafle = jal [fle- If [flo = 0, then there exists a sot Ny eX with w(N,) = Osuch that [/()| < Wk forxe Ne. If we put N=Ubs Mey then NEX, al) = 0, and [f(s)| = 0 for x eM. ‘Therefore, f(x) = 0 for almost all x. I f.geLes there exist sets Ni, Ny in X with w(N,) = w(N such that [fl < [fle for XEN, leGal < ele for x8, ‘Therefore f(x) + 9(8) < [fle + lela for x€ (MUN, from ‘which it follows that If+ gle < [fla + ete: It remains to prove ‘hat Ly is complete. Let (f—) be # Cauchy sequence in Ly, and let M be a set in X with w(M)~ 0, such that UAQ)| < Lule for [email protected], m= 1,2,..-5 and also such that |f,(x) — Fle) < fe Soke for all x€M, 1m m=l,2,.... Then the sequence (f,) i uniform.y convergent on X'\ M, and we let Se) = lim fe), x eM, xeM. It follows that fis measurable, and itis easily seen that fa — fle —> 0. Hence Ly is complete ‘aED. EXERCISES 6.A. Let C10, 1] be the linear space of continuous functions on (0, 1]t0.R. Define Ny or fin CLO, 1} by Nf) = LJ(O)I- Show that {Ng sa semi-norm on CP, 1) 6.B, Let C[0, 1] be as before and define N; for fin C10, t] 10 be the Riemann integral of [/| over (0, 1]. Show that Ny isa semi-norm on C10, 1]. If fis defined for m > I to be equal to 0 for O< x < (1 = Wn), to be equel to 1 for 4 0, then ihere exists a simple X- measurable function g such that [f gl; 0 as (we X: [A] # 0}, then 6G, Let Y=, and let u be the counting measure on N. I fis Uatined on N by fm) = in, then f does not belong to L, but it does Jong to L, fort

. [Alternatively let X= R, X= B, and Jet be Lebesgue measure and define g(x) = O for x < 1 and g(x) = Us tors > 1 6H, Let X=, and let \ be the measure on NV which has measure La atthe point a. (More precisely NE) = 3 {V/n®: ne E}.) Show That UX) < ten, Let f be defined on ¥ by f(a) = Vi. Show that fet, and only 1

0. Show that there ents tt eX with s(E) < 40 such that if FeX and FOE, = Ou then Ufvsly < 6 Let fre LyX, X.0), 1

O, then there exists a set E,e-X with u(E,) < too such that if FeXand FOE, = 0, then B,(F) < «forall meV.