ebook img

Orthogonal polynomials on the circle for the weight w satisfying conditions: w,1/w in BMO(T) PDF

0.2 MB·
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Orthogonal polynomials on the circle for the weight w satisfying conditions: w,1/w in BMO(T)

ORTHOGONAL POLYNOMIALS ON THE CIRCLE FOR THE WEIGHT w SATISFYING CONDITIONS w,w−1 BMO ∈ S.DENISOV,K.RUSH 6 1 0 2 Abstract. For the weight w satisfying w,w−1 ∈ BMO(T), we prove the asymptotics of {Φn(eiθ,w)} in Lp[−π,π],26p<p0where{Φn(z,w)}aremonicpolynomialsorthogonalwithrespecttowontheunitcircle v T. Immediate applications include the estimates on the uniform norm and asymptotics of the polynomial o entropies. TheestimatesonhigherordercommutatorsbetweentheCalderon-ZygmundoperatorsandBMO N functionsplaythekeyroleintheproofsofmainresults. MSC AMS classification: 42C05, 33D45; key words: orthogonal polynomials, weight, bounded mean 2 oscillation. ] A C . 1. Introduction h at Let σ be a probability measure on the unit circle. Define the monic orthogonal polynomials Φn(z,σ) { } m by requiring π [ degΦ =n, coeff(Φ ,n)=1, Φ (eiθ,σ)Φ (eiθ,σ)dσ =0, m<n, n n n m 2 Z−π v wherecoeff(Q,j)denotesthecoefficientinfrontofzj inthe polynomialQ. We canalsodefine theorthonor- 7 mal polynomials by the formula 6 Φ (z,σ) n 3 φn(z,σ)= Φ (eiθ,σ) 3 k n kL2σ 0 Later, we will need to use the following notation: for every polynomial Q (z)=q zn+...+q of degree at n n 0 1. most n, we introduce the ( )–operation: ∗ 0 6 Q (z) (∗) Q∗(z)=q¯ zn+...+q¯ n −→ n 0 n 1 v: TLphissta(∗n)dsdefporenLdps(Ton) onr. LInp[thπe,πp]a.peTrh,ewseyumsbeotlhseCs,hCortahraendreskefrkvped=fokrfkaLbps(oTl)u,tkefckoLnpwst=antsTw|fh(icθh)|pvwalduθe 1c/apn. Xi change from one formula to−another. 1 (cid:0)R (cid:1) r Thecurrentpaperismainlymotivatedbytwoproblems: theproblemofSteklovinthetheoryoforthogonal a polynomials [2] and the problem of the asymptotical behavior of the polynomial entropy [9]. The problem of Steklov [22] consists in obtaining the sharp estimates for φn(eiθ,σ) L∞[−π,π] assuming k k that the probability measure σ satisfies σ′ > δ/(2π) a.e. on [ π,π] and δ (0,1). This question attracted − ∈ a lot of attention [1, 11, 12, 13, 14, 18, 19, 23] and was recently resolved in [2]. In particular, the following stronger result was proved Theorem 1.1 ([2]). Assume that the measure is given by the weight w: dσ = wdθ. Let p [1, ) and ∈ ∞ C >C (p,δ), then 0 C (p,δ)√n6 sup φ (eiθ,w) 6C (p,δ)√n 1 n ∞ 2 k k w>δ/(2π),kwk1=1,kwkp6C Remark. If the measure σ satisfies the Szego˝ condition [24] π logσ′(θ)dθ > (1) −∞ Z−π 1 then Φ 1 and the polynomials φ and Φ are of the same size. In particular, φ can be replacedby n L2 n n n k k σ ∼ Φ in the previous Theorem. n Remark. In the formulation of the Steklov problem, the normalization that σ is a probability measure, i.e., π dσ =1 Z−π isnotrestrictivebecauseofthefollowingscalings: φ (z,σ)=α1/2φ (z,ασ)andΦ (z,ασ)=Φ (z,σ), α>0. n n n n The previous Theorem handles all p< but not the case p= . That turns out to be essential: if the ∞ ∞ weight w is bounded, we get an improvement in the exponent. Theorem 1.2. ([8], Denisov-Nazarov) If T 1, we have ≫ C sup kΦn(eiθ,w)kp0 6C(T), p0(T)=2+ T1, sup kΦn(eiθ,w)k∞ 6C(T)n12−CT 16w6T 16w6T and, if 0<ǫ 1, ≪ C sup Φ (eiθ,w) 6C(ǫ), p (ǫ)= 2, sup Φ (eiθ,w) 6C(ǫ)nCǫ k n kp0 0 ǫ k n k∞ 16w61+ǫ 16w61+ǫ The uniform bound on the Lp norm suggests that maybe a stronger result on the asymptotical behavior is true. It is well-known that for σ in the Szego˝ class (i.e., (1) holds), the following asymptotics is valid [10] φ∗(eiθ,σ) (∗) S(eiθ,σ), π φ∗n(eiθ,σ) 1 2dθ 0, n (2) n −→ S(eiθ,σ) − → →∞ Z−π(cid:12) (cid:12) (cid:12) (cid:12) where (∗) refers to weak-star convergence and S(z(cid:12)(cid:12),σ) is the Szeg(cid:12)(cid:12)o˝ function, i.e., the outer function in D −→ which satisfies S(eiθ,σ)−2 = 2πσ′(θ),S(0,σ) > 0. In particular, if σ′ > (2π)−1δ, then φ∗ S 0. | | k n − k2 → Recall that φ (z,σ)=znφ∗(z,σ),z T. n n ∈ The results stated above give rise to three questions: (a) What upper estimate can we get assuming w BMO(T) [21] instead of w L∞(T)? Recall that L∞(T) BMO(T) Lp(T). (b) Is it possible p<∞ ∈ ∈ ⊂ ⊂ ∩ to relax the Steklov condition w >1? (c) Can one obtain an asymptotics of φ∗ in Lp classes with p>2? { n} ThepartialanswerstothesequestionsarecontainedinthefollowingTheoremswhicharethemainresults of the paper. We start with a comment about some notation. If α is a positive parameter, we write α 1 ≪ to indicate the following: there is an absolute constant α (sufficiently small) such that α < α . Similarly, 0 0 we write α 1 as a substitute for: there is a constant α (sufficiently large) so that α > α . The symbol 0 0 ≫ α α (α α ) will mean α /α 1 (α /α 1). 1 2 1 2 1 2 1 2 ≪ ≫ ≪ ≫ Theorem 1.3. If w: w−1 BMO 6s, w BMO 6t, then there is Π Lp0[ π,π],p0 >2 such that k k k k ∈ − lim Φ∗ Π =0 n→∞k n− kp0 and we have for p : 0 C 1 2+ , ifst 1 (st)log2(st) ≫ p = (3) 0 C  (st)21/4, if0<st≪1 We also have the bound for the uniformnorm Φ∗ 6C n1/p0 (4) k nk∞ (st) where C denotes a function of u. (u) In the case when an additional information is given, e.g., w L∞ or w−1 L∞, this result can be ∈ ∈ improved. Theorem 1.4. Under the conditions of the previous Theorem, we have 2 If w >1, then p can be taken as 0 • C 1 2+ , ift 1 tlogt ≫ p = 0  C 2  √t, if0<t≪1 If w 61, then we have  • C 1 2+ , ifs 1 slogs ≫ p = 0  C 2  √s, if0<s≪1 We also have the bound for the uniform norm Φ∗ 6C n1/p0 (5) k nk∞ (t,s) where C depends on t or s. (t,s) Remark. It is clear that the allowed exponent p is decaying in s and t so it can be chosen larger than 0 2 for all values of s and t. Remark. As we have already mentioned, the following scaling invariance holds: Φ (z,σ) = Φ (z,ασ), n n α>0. TheBMOnormis1-homogeneous,e.g., αw =α w ,sotheestimatesintheTheorem1.3 BMO BMO k k k k are invariant under scaling w αw. → In the case when w =C, we get w = w−1 =0 and, although Φ∗(z,w)=1, we can not say k kBMO k kBMO n anything about the size of φ∗(z,w). The next Lemma explains how our results can be generalized to φ∗ . n { n} Lemma1.1. IntheTheorem1.3,ifonemakesanadditionalassumptionthat w =1,then φ∗ S 0 k k1 k n− kp0 → with p given by (3). 0 Proof. Indeed, Lemma 3.3 from Appendix shows that π logwdθ > −∞ Z−π and thus the sequence Φ has a finite positive limit [10, 20]. Therefore, φ∗ = Φ∗n has an {k nk2,w} { n} ||Φn||2,w Lp0 limit by Theorem 1.3. By (2), φ∗ converges weakly to S and therefore we have thenstatemeont of the { n} Lemma with Π being a multiple of S. (cid:3) The polynomial entropy is defined as E(n,σ)= φ 2log φ dσ n n T| | | | Z where φ are orthonormalwith respectto σ. In recentyears,a lotof efforts were made to understand the n { } asymptotics of E(n,σ) [3, 4, 5] as n . In [9], the sharp lower and upper bounds were obtained for σ in →∞ the Szego˝class. In[2], itwasshownthatE(n,w) cannotexceedClogn ifw >1andw Lp[ π,π],p< , ∈ − ∞ and that this bound saturates. This leaves us with very natural question: what are regularity assumptions onw thatguaranteeboundednessofE(n,w)? ThefollowingcorollaryofLemma1.1givesthepartialanswer. Corollary 1.1. If w :w,w−1 BMO(T) and w =1, then 1 ∈ k k 1 π lim E(n,w)= log(2πw)dθ n→∞ −4π Z−π So far, the only classes in which the E(n,w) was known to be bounded were the Baxter’s class [20]: dσ =wdθ, w W(T),w >0 (W(T) denotes the Wiener algebra)or the class givenby positive weights with ∈ a certain modulus of continuity [24]. Our conditions are obviously much weaker and, in a sense, sharp. The structure of the paper is as follows: the main results are proved in the next section, the Appendix contains auxiliary results from harmonic analysis. 3 We use the following notation: H refers to the Hilbert transform, P denotes the L2(dθ) projection to [i,j] the (i,...,j) Fourier modes. Given two non-negative functions f we write f .f is there is an absolute 1(2) 1 2 constant C such that f 6Cf 1 2 for all values of the arguments of f . We define & similarly and say that f f if f . f and f . f 1(2) 1 2 1 2 2 1 ∼ simultaneously. Given two operators A and B, we write [A,B]=AB BA for their commutator. If w is a − function, then in the expressionlike [w,A], the symbol w is identified with the operatorof multiplicationby w. The Hunt-Muckenhoupt-Wheeden characteristic of the weight w A will be denoted by [w] . For the ∈ p Ap basic facts about the BMO class, A and their relationship, we refer the reader to, e.g., the classical text p [21]. If A is a linear operator from Lp(T) space to Lp(T), then A denotes its operator norm. p,p k k 2. Proofs of main results Beforeprovingthemainresult,Theorem1.3,weneedsomeauxiliaryLemmas. Westartwiththefollowing observation which goes back to S. Bernstein [6, 24]. Lemma 2.1. For a monic polynomial Q of degree n, we have: Q(z)=Φ (z,w) if and only if P (wQ)=0. (6) n [0,n−1] Proof. It is sufficient to notice that (6) is equivalent to π Q(eiθ)e−ijθw(θ)dθ =0, j =0,...,n 1 − Z−π which is the orthogonality condition. (cid:3) Lemma 2.2. If f L2(T) is real-valued function, Q L∞(T), then ∈ ∈ znP (fznQ)=P (fQ) [0,n−1] [1,n] In particular, for a polynomial P of degree at most n with P(0)=1, we have: P(z)=Φ∗(z,w) if and only if P (wP)=0. n [1,n] Proof. The first statement is immediate. The second one follows from the Lemma above and the formula Φ =znΦ∗, z T. (cid:3) n n ∈ We have the following three identities for Φ∗(z,w); the first one was used in [8] recently. They are n immediately implied by the Lemma above. Φ∗ =1+P (1 αw)Φ∗ , α R (7) n [1,n] − n ∈ Φ∗ =1+(cid:16)w−1[w,P (cid:17)]Φ∗ (8) n [1,n] n Φ∗ =1 [w−1,P ](wΦ∗) (9) n − [1,n] n Denote the higher order commutators recursively: C =P , C =[w,P ], C =[w,C ], l =2,3,... 0 [1,n] 1 [1,n] l l−1 Define the multiple commutators of w−1 and P (in that order!) by C . [1,n] j Lemma 2.3. The following representations hold e j j 1 wjP Φ∗ = − C wj−lΦ∗ (10) [1,n] n l 1 l n l=1(cid:18) − (cid:19) X and j j w−jP Φ∗ = C w−(j−l)(wΦ∗) (11) [1,n] n − l l+1 n l=0(cid:18) (cid:19) X where j =1,2,.... e 4 Proof. We will prove (10), the other formula can be obtained in the similar way. The case j = 1 of this expressionis our familiar formula wP Φ∗ =[w,P ]Φ∗. Now the proof proceeds by induction. Suppose [1,n] n [1,n] n we have k−1 k 2 wk−1P Φ∗ = − C wk−1−lΦ∗ [1,n] n l 1 l n l=1(cid:18) − (cid:19) X Multiply both sides by w and write k−1 k−1 k 2 k 2 wkP Φ∗ = − wC wk−1−lΦ∗ = − C wk−1−lΦ∗ +C wk−lΦ∗ = [1,n] n l 1 l n l 1 l+1 n l n l=1(cid:18) − (cid:19) l=1(cid:18) − (cid:19) X X (cid:0) (cid:1) k−1 k k k 2 k 2 k 1 − C wk−lΦ∗ + − C wk−lΦ∗ = − C wk−lΦ∗ l 1 l n l 2 l n l 1 l n l=1(cid:18) − (cid:19) l=2(cid:18) − (cid:19) l=1(cid:18) − (cid:19) X X X because k 1 k 2 k 2 − = − + − l 1 l 2 l 1 (cid:18) − (cid:19) (cid:18) − (cid:19) (cid:18) − (cid:19) . (cid:3) Motivatedby the previous Lemma, we introduce certainoperators. Givenf Lp, define y recursively j ∈ { } by j−1 j 1 y =f, y =wj + − C y 0 j l+1 j−1−l l l=0(cid:18) (cid:19) X Then, we let j j z =w−j C z j l+1 j−l−1 − l l=0(cid:18) (cid:19) X where z−1 = y1,z0 = y0 = f. Notice that for fixed j botheyj and zj are affine linear transformations in f. We can write y =y′ +y′′ j j j where y′ =f, y′′ =0 0 0 and, recursively, j−1 j−1 j 1 j 1 y′ = − C y′ , y′′ =wj + − C y′′ j l l+1 j−1−l j l l+1 j−1−l l=0(cid:18) (cid:19) l=0(cid:18) (cid:19) X X Similarly, we write z =z′ +z′′ where j j j z′ =y′, z′′ =y′′, z′ =f, z′′ =0 −1 1 −1 1 0 0 and j j j j z′ = C z′ , z′′ =w−j C z′′ , j − l l+1 j−l−1 j − l l+1 j−l−1 l=0(cid:18) (cid:19) l=0(cid:18) (cid:19) X X Let us introduce linear operators: B ef =y′, D f =z′. We need an impeortant Lemma. j j j j Lemma 2.4. wjΦ∗ =y′′+B Φ∗, w−jΦ∗ =z′′+D Φ∗ n j j n n j j n Proof. This follows from wjΦ∗ =wj +wjP Φ∗, w−jΦ∗ =w−j +w−jP Φ∗ n [1,n] n n [1,n] n and the previous Lemma. (cid:3) The next Lemma, in particular, provides the bounds for B and D . j j 5 Lemma 2.5. Assume w>0, w =t, w−1 =s, w =1, and p [2,3]. Then, BMO BMO 1 k k k k k k ∈ B 6(Ctj)j, D 6(1+st)(Csj)j j p,p j p,p k k k k Moreover, y′′ 6(Ctj)j, z′′ 6st(Csj)j k jkp k jkp with e ee e t=max t,1 , s=max s,1 { } { } Proof. We will prove the estimates for B and y′′ only, the bounds for D , z′′ are shown e k jkp,p kejkp k jkp,p k jkp similarly. By John-Nirenberg inequality ([21], p.144), we get π ∞ w (2π)−1 jpdθ .j xjp−1exp( Cx/t)dx=j(Ct)jpΓ(jp)6(C tj)pj 1 | − | − Z−π Z0 where Stirling’s formula was used for the gamma function Γ. Since wjp 6(w (2π)−1 +(2π)−1)jp 6Cjp(w (2π)−1 jp+1) | | | − | | − | we have π wjpdθ 6Cjp(1+(tj)jp)6(C tj)jp 1 | | Z−π Lemma 3.2 yields e j−1 (j 1)! j−1 (Ct)−k y′ 6 − (C(l+1)t)l+1 y′ 6(Ct)jj! y′ k jkp l!(j 1 l)! k j−1−lkp k! k kkp l=0 − − k=0 X X e y′ Divide both sides by (Ct)jj! and denote β = k jkp . Then, j (Ct)jj! j−1 β 6 β j l l=0 X Since β = f , we have β 6 3j f by induction and thus y′ 6 (Ctj)j f . The estimates for 0 k kp j k kp k jkp k kp y′′ , z′ , z′′ can be obtained similarly. (cid:3) k jkp k jkp k jkp Lemma 2.6. If w =1, w =t, w−1 =s, and p [2,3], then 1 BMO BMO k k k k k k ∈ CΛ min Λ−l B 6exp l p,p l∈N k k − t (cid:16) (cid:17) (cid:18) (cid:19) and C min ǫj D (1+st)exp j p,p j∈N k k ≤ −ǫs (cid:16) (cid:17) (cid:18) (cid:19) provided that Λ t and ǫ s−1. ≫ ≪ Proof. By the previous Lemma, we have l Ctl Λ−l B 6 l p,p k k Λ (cid:16) (cid:17) (cid:18) (cid:19) Optimizing in l we get l∗ CΛ/(te) and it gives the first estimate. The proof for the second one is ∼ identical. (cid:3) Now we are ready to prove the main results of the paper. 6 Proof. (Theorem 1.3). Notice first that (4) follows from the Nikolskii inequality Q <Cn1/p0 Q , degQ=n, p >2 k k∞ k kp0 0 as long as the Lp0 norms are estimated. By scaling invariance, we can assume that w = 1. We consider two cases separately: st 1 and 1 k k ≫ st 1. The proofs will be different. ≪ 1. The case st 1. ≫ Let p =2+δ with δ <1. Take two n-independent parameters ǫ and Λ such that ǫs 1 and Λt−1 1. ≪ ≫ Consider the following sets Ω = x:w6ǫ , Ω = x:ǫ<w <Λ ,Ω = x:w>Λ . Notice that 1 2 3 { } { } { } ǫs 1,tΛ−1 1 = (ǫs)(tΛ−1) 1 = ǫΛ−1 (st)−1 1 = ǫ Λ ≪ ≪ ⇒ ≪ ⇒ ≪ ≪ ⇒ ≪ From (7), we have Φ∗ =1+P (1 w/Λ)Φ∗ n [1,n] − n The idea of our proof is to rewrite this identity in the form Φ∗ =f +O(n)Φ∗ n n n where f <C(s,t) and O(n) is a contraction in Lp for the suitable choice of p. To this end, we consider n p k k operators w j O (n)f =ǫjP (1 w/Λ)χ D f 1 [1,n] − Ω1 ǫ j O (n)f =P (1 (cid:16)w/Λ(cid:17))χ f 2 [1,n] − Ω2 O (n)f =Λ−lP (1 w/Λ)(Λ/w)lχ B f 3 [1,n] − Ω3 l where j and l will be fixed later, they will be n-independent. Let us estimate the (Lp,Lp) norms of these operators. Since P 61+Cδ (see Lemma 3.1), we choose j and l as in Lemma 2.6 to ensure [1,n] p,p k k C O (n) 6stexp 1 p,p k k −ǫs! b O (n) 6(1+Cδ)(1 ǫΛ−1) 2 p,p k k − CΛ O (n) 6exp 3 p,p k k − t ! b Lemma 2.4 now yields Φ∗ =1+f (n)+f (n)+(O (n)+O (n)+O (n))Φ∗ n 1 3 1 2 3 n where w j f (n)=ǫjP (1 w/Λ)χ z′′, f (n)=Λ−lP (1 w/Λ)(Λ/w)lχ y′′ 1 [1,n] − Ω1 ǫ j 3 [1,n] − Ω3 l Let (cid:16) f((cid:17)n)=1+f (n)+f (n) 1 3 Then Lemma 2.5 provides the bound f(n) 6C(s,t) (12) p k k uniforminn. DenoteO(n)=O (n)+O (n)+O (n)andselectparametersǫ,Λ,δsuchthat O(n) <1 Cδ. 1 2 3 p,p k k − To do so, we first let δ =cǫΛ−1 with small positive absolute constant c. Then, we consider C CΛ c ǫ 1 stexp +exp = −ǫs! − t ! Λ b b with c again being a small constant. Now, solving equations 1 stexp( C/(ǫs))=exp( CΛ/t), c ǫ/Λ=2exp( CΛ/t) 1 − − − we get the statement of the Theorem. Indeed, we have two equations: b b b Ct ǫ= s(CΛ+tlog(st)) b 7 b and Λ 1 Λ log(st) = C+log(sΛ)+log + t C t C (cid:18) (cid:16) (cid:17)(cid:19) Denote b u=CΛ/t b and then b log(st) u=C+log(st)+2logu+log 1+ u Tofindtherequiredroot,werestricttherangeofutoc log(st)(cid:16)<u<c lo(cid:17)g(st)forc 1,c 1. Rewrite 1 2 1 2 ≪ ≫ the equation above as log(st) u 2logu log 1+ =log(st)+C − − u (cid:18) (cid:19) Differentiating the left hand side in u, we see that l.h.s.′ 1 within the given range. Therefore, there is ∼ exactly one solution u and u log(st). Then, since ∼ log(st) log 1+ u (cid:18) (cid:19) is O(1), we get u=log(st)+2logu+O(1)=log(st)+2loglog(st)+O(1) by iteration. Thus, ǫ 1 =Ce−u Λ ∼ stlog2(st) and δ 1 . Now that we proved that O(n) 61 Cδ <1, we can rewrite ∼ stlog2(st) k kp,p − ∞ Φ∗ =f(n)+ Oj(n)f(n) n j=1 X and the series converges geometrically in Lp with tail being uniformly small in n due to (12). To show that Φ∗ convergesin Lp as n , it is sufficient to prove that Oj(n)f(n) converges for each j. n →∞ This, however,is immediate. Indeed, P f P f, as n [1,n] [1,∞] → →∞ in Lq for all f Lq,1 < q < . Since w,w−1 BMO Lp ([21], this again follows from John- p>1 ∈ ∞ ∈ ⊂ ∩ Nirenbergestimate),wesee thatmultiplicationby w±j mapsLp1 to Lp2 continuouslyby Ho¨lder’sinequality provided that p <p and j Z. Therefore, if µ L∞,j =1,...,k, then 2 1 j ∈ ∈ µ w±j1P µ w±j2...µ w±jk−1P µ w±jk (13) 1 [1,n] 2 k−1 [1,n] k has the limit in each Lp,p < when n . Notice that each f(n) and Oj(n)f(n) can be written as a ∞ → ∞ linear combination of expressions of type (13) ( µ taken as the characteristic functions). Now that δ is j { } chosen, we define p in the statement of the Theorem as p =2+δ. 0 0 2. The case st 1. ≪ The proof in this case is much easier. Let us start with two identities Φ∗ =1+w−1[w,P ]Φ∗, Φ∗ =1+[P ,w−1]wΦ∗ n [1,n] n n [1,n] n which can be recast as wΦ∗ =w+[w,P ]Φ∗, Φ∗ =1+[P ,w−1]wΦ∗ n [1,n] n n [1,n] n Substitution of the first formula into the second one gives Φ∗ =1+[P ,w−1]w+G Φ∗ n [1,n] n n where G =[P ,w−1][w,P ] n [1,n] [1,n] We have 1+[P ,w−1]w 6C(s,t,p) [1,n] p k k 8 and G .stp4 n p,p k k by Lemma 3.4. Taking p < p (st)−1/4 we have that G is a contraction. Now, the convergence of all 0 n ∼ terms in the geometric series can be proved as before. (cid:3) Let us give a sketch of how the arguments can be modified to prove Theorem 1.4. Proof. (Theorem 1.4). Consider the case w >1 first. 1. The case t 1. ≫ The proof is identical except that we can chose ǫ=1/2 so that Ω = . We get an equation for Λ 1 ∅ C CΛ =exp , Λ=C−1t(logΛ logC) Λ − t ! − b b Denote CΛ/t=u, then u=logt+logu+O(1), u=logt+loglogt+O(1) and δ b(tlogt)−1. ∼ 2. The case t 1. ≪ We have Φ∗ =1+L Φ∗, L f =w−1[w,P ]f n n n n [1,n] and Lemma 3.4 yields L .p2t<0.5 n p,p k k for p<p =O(t−1/2). 0 The case w 6 1 can be handled similarly. When s is large, we take Λ = 1 in the proof of the previous Theorem and get an equation for ǫ: C Cǫ=sexp −ǫs! b Its solution for large s gives the required asymptotics for ǫ and, correspondingly,for δ and p . If s is small, 0 it is enough to consider the equation Φ∗ =1 [w−1,P ]wΦ∗ n − [1,n] n where the operator [w−1,P[1,n]]w is contraction in Lp0 for the specified p0. (cid:3) Now we are ready to prove Corollary 1.1. Proof. (of Corollary 1.1). The following inequality follows from the Mean Value Formula x2logx y2logy 6C(1+x logx +y logy )x y , x,y >0 | − | | | | | | − | Since w Lp, the Theorem 1.3 yields p<∞ ∈∩ π π φ 2log φ S 2log S wdθ . (1+ φ logφ + SlogS ) φ∗ S wdθ 0, n || n| | n|−| | | || | n n| | | || n|−| || → →∞ Z−π Z−π by applying the trivial bound: u logu 6 C(δ)(1+u1+δ),δ > 0 and the generalized Ho¨lder’s inequality to | | φ 1+δ(or S 1+δ), φ∗ S , and w. To conclude the proof, it is sufficient to notice that | n| | | || n|−| || π 1 π S 2log S wdθ = log(2πw)dθ | | | | −4π Z−π Z−π because S −2 =2πw. (cid:3) | | 9 3. Appendix In this Appendix, we collect some auxiliary results used in the main text. Lemma 3.1. For every p [2, ), ∈ ∞ P 61+C(p 2). (14) [1,n] p,p k k − Proof. If P+ is the projection of L2(T) onto H2(T) (analytic Hardy space), then P+ =0.5(1+iH)+P , 0 where H is the Hilbert transform on the circle and P denotes the Fourier projection to the constants, i.e., 0 P f =(2π)−1 f(x)dx (15) 0 T Z We therefore have a representation P =zP+z−1 zn+1P+z−(n+1) =0.5i(zHz−1 zn+1Hz−(n+1))+zP z−1 zn+1P z−(n+1). (16) [1,n] 0 0 − − − Since H =cot(π/(2p)) [17], we have p,p k k π P 6cot +2 (17) k [1,n]kp,p 2p (cid:18) (cid:19) by triangle inequality. On the other hand, P = 1 so by Riesz-Thorin theorem, we can interpolate [1,n] 2,2 k k between p=2 and, e.g., p=3 to get P 61+C(p 2), p [2,3]. [1,n] p,p k k − ∈ Noticing that cot(π/(2p)) p, p>3, we get the statement of the Lemma. (cid:3) ∼ Remark. In the proof above,we could have used the expressionfor the norm P+ obtained in [15]. p,p k k The proof of the following Lemma uses some standard results of Harmonic Analysis. Lemma 3.2. If w =t and p [2,3], then we have BMO k k ∈ C 6(Cjt)j j p,p k k Proof. Consider the following operator-valued function F(z)=ezwP e−zw [1,n] If we can prove that F(z) is weakly analytic around the origin (i.e., analyticity of the scalar function F(z)f ,f with fixed f C∞), then 1 2 1(2) h i ∈ 1 F(ξ) F(z)= dξ, z B (0) ǫ 2πi ξ z ∈ Z|ξ|=ǫ − understood in a weak sense. By induction, one can then easily show the well-known formula j! F(ξ) C =∂jF(0)= dξ j 2πi ξj+1 Z|ξ|=ǫ which explains that we can control C by the size of F(ξ) on the circle of radius ǫ. Indeed, j p,p p,p k k k k C = sup C f ,f j p,p j 1 2 k k f1(2)∈C∞,kf1kp61,kf2kp′61|h i|≤ j! F(ξ)f ,f j! sup h 1 2idξ 6 max F(ξ) 2π f1(2)∈C∞,kf1kp61,kf2kp′61(cid:12)(cid:12)Z|ξ|=ǫ ξj+1 (cid:12)(cid:12) ǫj |ξ|=ǫk kp,p The weak analyticity of F(z) aroundthe origin(cid:12)follows immediately(cid:12)from, e.g., the John-Nirenbergestimate (cid:12) (cid:12) ([21],p.144). Tobound F ,weusethefollow(cid:12) ingwell-knownresul(cid:12)t(whichisagainanimmediatecorollary p,p k k from John-Nirenberg inequality, see, e.g., [21], p.218). There is ǫ such that 0 w <ǫ = [ewe] 6[ewe] <C, p>2 k kBMO 0 ⇒ Ap A2 e 10

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.