# Harmonic Analysis Lecture Notes University of Illinois at Urbana–Champaign

Richard S. Laugesen January 9, 2009
*

2

Preface
A textbook presents more than any professor can cover in class. In contrast, these lecture notes present exactly* what I covered in Harmonic Analysis (Math 545) at the University of Illinois, Urbana–Champaign, in Fall 2008. The ﬁrst part of the course emphasizes Fourier series, since so many aspects of harmonic analysis arise already in that classical context. The Hilbert transform is treated on the circle, for example, where it is used to prove Lp convergence of Fourier series. Maximal functions and Calder´n– o Zygmund decompositions are treated in Rd , so that they can be applied again in the second part of the course, where the Fourier transform is studied. Real methods are used throughout. In particular, complex methods such as Poisson integrals and conjugate functions are not used to prove boundedness of the Hilbert transform. Distribution functions and interpolation are covered in the Appendices. I inserted these topics at the appropriate places in my lectures (after Chapters 4 and 12, respectively). The references at the beginning of each chapter provide guidance to students who wish to delve more deeply, or roam more widely, in the subject. Those references do not necessarily contain all the material in the chapter. Finally, a word on personal taste: while I appreciate a good counterexample, I prefer spending class time on positive results. Thus I do not supply proofs of some prominent counterexamples (such as Kolmogorov’s integrable function whose Fourier series diverges at every point). I am grateful to Noel DeJarnette, Eunmi Kim, Aleksandra Kwiatkowska, Kostya Slutsky, Khang Tran and Ping Xu for TEXing parts of the document. Please email me with corrections, and with suggested improvements of any kind. Richard S. Laugesen Department of Mathematics University of Illinois Urbana, IL 61801 U.S.A. Email: Laugesen@illinois.edu

*

modulo some improvements after the fact

3

Introduction
Harmonic analysis began with Fourier’s eﬀort to analyze (extract information from) and synthesize (reconstruct) the solutions of the heat and wave equations, in terms of harmonics. Speciﬁcally, the computation of Fourier coeﬃcients is analysis, while writing down the Fourier series is synthesis, and the harmonics in one dimension are sin(nt) and cos(nt). Immediately one asks: does the Fourier series converge? to the original function? In what sense does it converge: pointwise? mean-square? Lp ? Do analogous results hold on Rd for the Fourier transform? We will answer these classical qualitative questions (and more!) using modern quantitative estimates, involving tools such as summability methods (convolution), maximal operators, singular integrals and interpolation. These topics, which we address for both Fourier series and transforms, constitute the theoretical core of the course. We further cover the sampling theorem, Poisson summation formula and uncertainty principles. This graduate course is theoretical in nature. Students who are intrigued by the fascinating applications of Fourier series and transforms are advised to browse [Dym and McKean], [K¨rner] and [Stein and Shakarchi], which are o all wonderfully engaging books. If more time (or a second semester) were available, I might cover additional topics such as: Littlewood–Paley theory for Fourier series and integrals, Fourier analysis on locally compact abelian groups [Rudin] (especially Bochner’s theorem on Fourier transforms of nonnegative functions), short-time Fourier transforms [Gr¨chenig], discrete Fourier transforms, the o Schwartz class and tempered distributions and applications in Fourier analysis [Strichartz], Fourier integral operators (including solutions of the wave and Schr¨dinger equations), Radon transforms, and some topics related to o signal processing, such as maximum entropy, spectral estimation and prediction [Benedetto]. I might also cover multiplier theorems, ergodic theorems, and almost periodic functions.

4 .

e. f ∈ L2 (T)) 2 . 8 Fourier series: convergence at a point 9 Fourier series: norm convergence 10 Hilbert transform on L2 (T) 11 Calder´n–Zygmund decompositions o 12 Hilbert transform on Lp (T) 13 Applications of interpolation 1 (Z) (or. f ∈ A(T)) (Z) (or.Contents I Fourier series 7 9 15 25 27 31 35 43 47 53 57 61 67 71 5 1 Fourier coeﬃcients: basic properties 2 Fourier series: summability in norm 3 Fourier series: summability at a point 4 Fourier coeﬃcients in 5 Fourier coeﬃcients in 6 Maximal functions 7 Fourier summability pointwise a.

6 CONTENTS II Fourier integrals 75 79 87 95 97 101 107 113 123 14 Fourier transforms: basic properties 15 Fourier integrals: summability in norm 16 Fourier inversion when f ∈ L1 (Rd ) 17 Fourier transforms in L2 (Rd ) 18 Fourier integrals: summability a.e. 19 Fourier integrals: norm convergence 20 Hilbert and Riesz transforms on L2 (Rd ) 21 Hilbert and Riesz transforms on Lp (Rd ) III Fourier series and integrals 127 129 135 141 22 Band limited functions 23 Periodization and Poisson summation 24 Uncertainty principles IV V Problems Appendices 147 159 161 163 165 A Minkowski’s integral inequality B Lp norms and the distribution function C Interpolation .

Part I Fourier series 7 .

.

Banach space with norm · L∞ (T) 1/p Trigonometric polynomial P (t) = Translation fτ (t) = f (t − τ ) N n=−N an eint 9 . continuous. 2π-periodic functions} 1 f Lp (T) = 2π T |f (t)|p dt where T can be taken over any interval of length 2π Nesting of Lp -spaces: L∞ (T) ⊂ L2 (T) ⊂ L1 (T) C(T) = {complex-valued. p-th power integrable. 2π-periodic functions}.1 Notation T = R/2πZ is the one dimensional torus Lp (T) = {complex-valued.Chapter 1 Fourier coeﬃcients: basic properties Goal Derive basic properties of Fourier coeﬃcients Reference [Katznelson] Section I.

τ ∈ T.1. fτ (n) = e−inτ f (n) takes modulation to translation. Exercise. f (n) = − 1 2π 1 =− 2π f (t)e−in(t+π/n) dt T 1 4π [f (t) − f (t − π/n)] e−int dt. T . FOURIER COEFFICIENTS: BASIC PROPERTIES Deﬁnition 1. Linearity (f + g)(n) = f (n) + g(n) and (cf )(n) = cf (n) Conjugation f (n) = f (−n) int Trigonometric polynomial P (t) = N has P (n) = an for |n| ≤ N n=−N an e and P (n) = 0 for |n| > N takes translation to modulation.10 CHAPTER 1. g = 2π T f (t)g(t) dt is that L2 inner product. T since e−iπ = −1 (1. c ∈ C. Thus f (n) =amplitude of f in direction of eint . eint where f. By (1. g ∈ L1 (T). 1 1 1 f (n) = f (n) + f (n) = 2 2 4π [f (t) − f (t − π/n)] e−int dt.2) and the deﬁnition (1. Let f. (1. Theorem 1. n ∈ Z. For f ∈ L2 (T).2 (Basic properties). See Chapter 5. deﬁne f (n) = n-th Fourier coeﬃcient of f 1 = f (t)e−int dt. with |f (n)| ≤ f L1 (T) Hence if fm → f in L1 (T) then fm (n) → f (n) (uniformly in n) as m → ∞.1). f (n) = Proof. For f ∈ L1 (T) and n ∈ Z. For n = 0. j.1) 1 Aside. Proof. note f (n) = f.3 (Diﬀerence formula). [f (t)eijt ] (n) = f (n − j) : L1 (T) → ∞ (Z) is bounded. Lemma 1. 2π T The formal series S[f ] = f (n)eint is the Fourier series of f .2) f (t − π/n)e−int dt T by t → t − π/n and periodicity.

Smoothness and decay The Riemann–Lebesgue lemma says f (n) = o(1). Let τ0 ∈ T. the diﬀerence f − g can be made arbitrarily small.6 (Less than one derivative). as desired. If f ∈ C α (T).4 (Continuity of translation). We show the smoother f is. Take g ∈ C(T) and observe f τ − f τ0 Lp (T) ≤ fτ − gτ Lp (T) + gτ − gτ0 = 2 f − g Lp (T) + gτ − gτ0 → 2 f − g Lp (T) Lp (T) Lp (T) + gτ0 − fτ0 Lp (T) as τ → τ0 . Proof. f (n) → 0 as |n| → ∞. 1 ≤ p < ∞.3 implies |f (n)| ≤ 1 f − fπ/n 2 L1 (T) . . The map φ : T → Lp (T) τ → fτ is continuous. Hence lim supτ →τ0 fτ − fτ0 Lp (T) = 0. Lemma 1. Proof. which tends to zero as |n| → ∞ by the L1 -continuity of translation in Lemma 1.4. 1 ≤ p < ∞. 0 < α ≤ 1. then f (n) = O(1/|n|α ). by uniform continuity of g.5 (Riemann–Lebesgue lemma).11 Lemma 1. the faster its Fourier coeﬃcients decay. since f = f0 .2. Theorem 1. Here C α (T) denotes the H¨lder continuous functions: f ∈ C α (T) if f ∈ o C(T) and there exists A > 0 such that |f (t) − f (τ )| ≤ A|t − τ |α whenever |t − τ | ≤ 2π. By density of continuous functions in Lp (T). Fix f ∈ Lp (T). Corollary 1. with f (n) = O(1) explicitly by Theorem 1.

CHAPTER 1. Absolute continuity of f says t f (t) = f (0) + 0 f (τ ) dτ. in By Riemann-Lebesgue applied to f .7 (One derivative).9. where f ∈ L1 (T). in in n 1 1 |f (n)| ≤ f |n| |n| L1 (T) . Theorem 1. Similar decay results hold for functions of bounded variation. . FOURIER COEFFICIENTS: BASIC PROPERTIES f (n) = 1 4π [f (t) − f (t − π/n)]e−int dt T by the Diﬀerence Formula in Lemma 1. Integrate by parts k times.8 (Higher derivatives).1 (T)) then f (n) = o(1/n) and |f (n)| ≤ f L1 (T) /|n|. If f is 2π-periodic and k times diﬀerentiable (f ∈ W k. provided one integrates by parts using the Lebesgue–Stieltjes measure df (t) instead of f (t) dt. Remark 1. |n|α Theorem 1. Proof. f (n) = with |f (n)| ≤ 1 1 1 f (n) = o(1) = o( ). Therefore |f (n)| ≤ 1 π A 4π n α 2π = const. Integrating by parts gives f (n) = 1 2π f (t)e−int dt = T 1 2π T e−int f (t) dt. . Proof.1 (T)) then f (n) = o(1/|n|k ) and |f (n)| ≤ f (k) L1 (T) /|n|k .12 Proof. If f is 2π-periodic and absolutely continuous (f ∈ W 1.3.

Further.8] 1.11 (Convolution and Fourier coeﬃcients). Convolution is commutative: (f ∗ g)(t) = 1 f (t − τ )g(τ ) dτ 2π T 1 = f (θ)g(t − θ) dθ 2π T = (g ∗ f )(t). Convolution facts [Katznelson. dτ = −dθ Convolution is also associative. if f ∈ C(T) and g ∈ L1 (T) then f ∗ g ∈ C(T). Finally. deﬁne their convolution (f ∗ g)(t) = 1 2π f (t − τ )g(τ ) dτ.13 Convolution Deﬁnition 1. Section I. 1 ≤ p ≤ ∞. n ∈ Z. T t ∈ T. Given f.3) is exactly Young’s Theorem A.3) T f (t − τ )e−in(t−τ ) dt g(τ )e−inτ dτ T T = f (n)g(n). Thus takes convolution to multiplication. then f ∗ g ∈ Lp (T) with f ∗g and (f ∗ g)(n) = f (n)g(n). That f ∗ g ∈ Lp (T) satisﬁes (1. and g ∈ L1 (T). Theorem 1. and linear with respect to f and g. (f ∗ g)(n) = 1 2π 1 = 2π 1 2π 1 2π f (t − τ )g(τ ) dτ e−int dt T Lp (T) ≤ f Lp (T) g L1 (T) (1. g ∈ L1 (T). Proof. where τ = t − θ. . if f ∈ C(T) and g ∈ L1 (T) then f ∗ g is continuous because (f ∗ g)(t + δ) → (f ∗ g)(t) as δ → 0 by uniform continuity of f .3.10. Then by Fubini’s theorem. If f ∈ Lp (T).1.

14 CHAPTER 1.4) Proof. FOURIER COEFFICIENTS: BASIC PROPERTIES 2. n (P ∗ f )(t) = j=−n n aj 1 2π eij(t−τ ) f (τ ) dτ T = j=−n aj eijt f (j).4) holds for P (t) = ∞ j=−∞ aj eijt provided {aj } ∈ 1 (Z). 3. Convolution is continuous on Lp (T): if fm → f ∈ Lp (T). to prove fm ∗ g → f ∗ g in Lp (T). Convolution with a trigonometric polynomial gives a trigonometric polyijt nomial: if f ∈ L1 (T) and P (t) = n then j=−n aj e n (P ∗ f )(t) = j=−n aj f (j)eijt . Use linearity and (1. (1. and g ∈ L1 (T) then fm ∗ g → f ∗ g in Lp (T). [Sanity check: (P ∗ f )(j) = aj f (j) = P (j)f (j) as expected. (1. Proof.] More generally. .3). 1 ≤ p ≤ ∞.

In this chapter we prove summability of Fourier series. when 1 < p < ∞. then the arithmetic means (s0 + · · · + sn )/(n + 1) also converge to s (Exercise). Aside.Chapter 2 Fourier series: summability in norm Goal Prove summability (averaged convergence) in norm of Fourier series Reference [Katznelson] Section I. In Chapter 9 we prove norm convergence of Fourier series: Sn (f ) → f in Lp (T). 15 n f (j) . where S0 (f ) + · · · + Sn (f ) |j| 1− σn (f ) = = n+1 n+1 j=−n = arithmetic mean of partial sums. if a sequence {sn } in a Banach space converges to s. Indeed.2 Write n (Sn f )(t) = j=−n f (j)eijt = n-th partial sum of Fourier series of f . meaning σn (f ) → f in Lp (T) when 1 ≤ p < ∞. Norm convergence is stronger than summability.

π). A summability kernel is a sequence {kn } in L1 (T) satisfying: 1 kn (t) dt = 1 2π T 1 sup |kn (t)| dt < ∞ n 2π T n→∞ (Normalization) (L1 bound) (L1 concentration) for each δ ∈ (0. You can show (optional exercise) that Dn (const. Example 2.2) (2.4) and (2.1). Deﬁne the Dirichlet kernel n Dn (t) = j=−n eijt by geometric series (2.3.16 CHAPTER 2. so that (S2) fails.4). Example 2.3) L1 (T) = ei(n+1)t − e−int eit − 1 1 sin (n + 2 )t = sin 1 t 2 (S1) holds by (2. (S1) (S2) (S3) lim |kn (t)| dt = 0 {δ<|t|<π} Some kernels satisfy a stronger concentration property: n→∞ δ<|t|<π lim sup |kn (t)| = 0 (L∞ concentration) for each δ ∈ (0. (2.4) (2.1) n+1 j=−n sin n+1 t 2 1 sin 2 t 2 ∼ (2. FOURIER SERIES: SUMMABILITY IN NORM Deﬁnition 2. (S4) Call the kernel positive if kn ≥ 0 for each n. π).) log n as n → ∞.1. Deﬁne the Fej´r kernel e Fn (t) = D0 (t) + · · · + Dn (t) n+1 n |j| = 1− eijt by (2.2) and geometric series (2.2.6) .1) (2.5) 1 = n+1 by (2. ∴ {Dn } is not a summability kernel.

1: Dirichlet kernel with n = 10 Π 10 Π Figure 2.2: Fej´r kernel with n = 10 e Π .17 20 10 Π Figure 2.

1).5). →0 ∴ {Pr } is a positive summability kernel. and Pr ≥ 0 by (2. For (S4).7) (2. and Fn ≥ 0 so that (S2) holds also.5.4. →0 ∴ {Fn } is a positive summability kernel. FOURIER SERIES: SUMMABILITY IN NORM (S1) holds by (2.18 CHAPTER 2.8) and simplifying. we see (S1) holds by (2.8) (2. with limiting process r 1. For (S4).9) as r 1.9) so that (S2) holds also.7 later in the course. Example 2.7). Deﬁne the Poisson kernel ∞ Pr (t) = 1 + 2 j=1 ∞ rj cos(jt) r|j| eijt (2. The Poisson kernel is indexed by r ∈ (0.9) = j=−∞ 1 − r2 = 1 − 2r cos t + r2 by summing two geometric series (j < 0 and j ≥ 0) in (2. Example 2. . sup |Pr (t)| ≤ δ<|t|<π 1 − r2 1 − 2r cos δ + r2 by (2. sup |Fn (t)| ≤ δ<|t|<π 1 1 2 1 n + 1 sin 2 δ by (2.6) as n → ∞.11) 2π 2 =√ e−(t+2πn) /4s 4πs n=−∞ by Example 23. Deﬁne the Gauss kernel ∞ Gs (t) = j=−∞ e−j s eijt ∞ 2 (2. After suitably modifying the deﬁnition of summability kernel.10) (2.

01 Π .19 20 10 Π Figure 2.9 Π 20 10 Π Figure 2.3: Poisson kernel with r = 0.4: Gauss kernel with s = 0.

4). with limiting process s 0. σn (f ) = Fn ∗ f Proof.4) n j=−n (1 − |j| )eijt n+1 n implies |j| f (j)eijt = σn (f ) n+1 (Fn ∗ f )(t) = j=−n 1− by Convolution Fact (1. we want Fn ∗ f → f . FOURIER SERIES: SUMMABILITY IN NORM The Gauss kernel is indexed by s ∈ (0. 1eijt implies n (Dn ∗ f )(t) = j=−n 1f (j)eijt = Sn (f ) by Convolution Fact (1.4). Alternatively.1) n j=−n by (2.11) as s 0. use that σn (f ) = [S0 (f ) + · · · + Sn (f )]/(n + 1) and Fn = [D0 + · · · + Dn ]/(n + 1). For (S4).20 CHAPTER 2. Fn (t) = (2. Connection to Fourier series Sn (f ) = Dn ∗ f Proof. Thus for summability of Fourier series. Dn (t) = (2.11) so that (S2) holds also.10). ∞). and Gs ≥ 0 by (2. 2π 2 2 e−δ /4s + e−(πn) /4s sup |Gs (t)| ≤ √ 4πs δ<|t|<π n=0 →0 ∴ {Gs } is a positive summability kernel. Abel mean of S[f ] = Pr ∗ f . The analogue of (S1) holds by (2.

6 (Summability in Lp (T) and C(T)).4) (with the series converging absolutely and uniformly).21 Proof. Let ε > 0. = 1 2π |τ |≤δ δ + −δ {δ<|τ |<π} Lp (T) |kn (τ )| fτ − f 1 2π 1 2π δ Lp (T) dτ ≤ max fτ − f |τ |≤π |kn (τ )| dτ −δ + max fτ − f <ε+ε Lp (T) |kn (τ )| dτ {δ<|τ |<π} . If {kn } is a summability kernel and f ∈ Lp (T).13) Then (kn ∗ f )(t) − f (t) = Lp (T) 1 kn (τ )[fτ (t) − f (t)] dτ Lp (T) by (S1) 2π T 1 ≤ |kn (τ )| fτ − f Lp (T) dτ 2π T by Minkowski’s Integral Inequality. Pr (t) = (2.8) ∞ j=−∞ r|j| eijt implies ∞ (Pr ∗ f )(t) = j=−∞ r|j| f (j)eijt (2. 1 ≤ p < ∞. as n → ∞. Proof. then kn ∗ f → f in Lp (T). By (S2) and continuity of translation on Lp (T) (Lemma 1.1. Similarly.12) by Convolution Fact (1. if f ∈ C(T) then kn ∗ f → f in C(T). Summability in norm Theorem 2. (2. we can choose 0 < δ < π such that max fτ − f |τ |≤δ Lp (T) · sup kn n L1 (T) < ε. and this last expression is the Abel mean of S[f ].4). Theorem A.

• Trigonometric polynomials are dense in C(T). • Uniqueness theorem: if f. 1 ≤ p < ∞. 1 ≤ p < ∞: σn (f ) → f in norm. using uniform continuity of f get that fτ → f in L∞ (T). (2. Connection to PDEs To ﬁnish the section. Proof. Choose kn = Fn = Fej´r kernel. Consequences • Summability of Fourier series in C(T).14) In other words. the map : L1 (T) → ∞ (Z) is injective. Fix f ∈ L1 (T). we connect our summability kernels to some important partial diﬀerential equations. Proof. σn (f ) is a trigonometric polynomial arbitrarily close to f . The Poisson kernel solves Laplace’s equation in a disk: v(reit ) = (Pr ∗ f )(t) = solves ∆v = vrr + r−1 vr + r−2 vtt = 0 on the unit disk {r < 1}. with boundary value v(1. Proof.6.4).6. t) = f (t) in the sense of Theorem 2. since f = g. Density of trigonometric polynomials in C(T) proves the Weierstrass Trigonometric Approximation Theorem. Lp (T).22 CHAPTER 2. FOURIER SERIES: SUMMABILITY IN NORM by (2. Fn ∗f = Fn ∗g by Convolution Fact (1. 1.13) and (S3). 1 2π 1 − r2 f (τ ) dτ 1 − 2r cos(t − τ ) + r2 T . Lp (T). Letting n → ∞ gives f = g. Then σn (f ) = Fn ∗ f → f in norm by e Theorem 2. g ∈ L1 (T) with f (n) = g(n) for all n. for all large n. If f ∈ C(T) then repeat the argument with p = ∞. then f = g in L1 (T). Aside.

23 That is. The Gauss kernel solves the diﬀusion (heat) equation: w(s. Gs (t) = (2. Proof. ∞) × T.12) for Pr ∗ f and note that ∂2 1∂ 1 ∂2 + + 2 2 (r|j| eijt ) = 0. Now diﬀerentiate through the sum and use that ∂ ∂2 − ∂s ∂t2 (e−j s eijt ) = 0. ∂r2 r ∂r r ∂t 2. 2 . Diﬀerentiate through formula (2. v is the harmonic extension of f from the boundary circle to the disk.4).6. t) = f (t) in the sense of Theorem 2. t) = (Gs ∗ f )(t) solves ws = wtt for (s. t) ∈ (0. with initial value w(0.10) ∞ −j 2 s ijt e j=−∞ e implies ∞ (Gs ∗ f )(t) = j=−∞ e−j s f (j)eijt 2 by Convolution Fact (1. Proof.

24 CHAPTER 2. FOURIER SERIES: SUMMABILITY IN NORM .

σn (f ) → f uniformly.Chapter 3 Fourier series: summability at a point Goal Prove a suﬃcient condition for summability at a point Reference [Katznelson] Section I. That is. Assume {kn } is a summability kernel. But what if f is merely continuous at a point? Theorem 3.1 (Summability at a point). (b) If in addition the summability kernel is even (kn (−t) = kn (t)) and L = lim exists (or equals ±∞). if f is continuous then σn (f ) → f in C(T). (a) If f is continuous at t0 then (kn ∗ f )(t0 ) → f (t0 ) as n → ∞. then (kn ∗ f )(t0 ) → L 25 as n → ∞. for each t ∈ T. σn (f )(t) → f (t). f (t0 + h) + f (t0 − h) h→0 2 .3 By Chapter 2.f ∈ L1 (T) and t0 ∈ T. Suppose either {kn } satisﬁes the L∞ concentration hypothesis (S4). or else f ∈ L∞ (T). In particular.

2. but uses symmetry of the kernel.6? 2. in Theorem 3. |τ |≤δ (3.1). (b) The proof is similar to (a). Then as n → ∞.1). for all large n.1 applies e in particular to summability at a point for σn (f ) = Fn ∗ f and for the Abel mean Pr ∗ f . The Fej´r and Poisson kernels satisfy (SR4). 1. Chapter 7 will prove kn ∗ f → f at almost every point. in Theorem 2. (S3) and (S1). Theorem 3. FOURIER SERIES: SUMMABILITY AT A POINT Note if f has limits from the left and right at t0 . . diﬀer from the proof of summability in norm. |(kn ∗ f )(t0 ) − f (t0 )| = {|τ |<δ} kn (τ )[f (t0 − τ ) − f (t0 )] dτ − {δ<|τ |<π} kn (τ ) dτ · f (t0 ) + {δ<|τ |<π} L1 (T) L∞ (T) kn (τ )f (t0 − τ ) dτ using (S1) < <ε ε + o(1) + o(1) · f ε + o(1) + o(1) · f by (3.1) using here (S2) and continuity of f at t0 . Proof. then the quantity L equals the average of those limits. or else by (3. for each integrable f . How does the proof of summability at a point. (a) Let ε > 0 and choose 0 < δ < π such that sup kn n L1 (T) · max |f (t0 − τ ) − f (t0 )| < ε.1(a).26 CHAPTER 3. Remark 3. and so Theorem 3.1 treats summability at a single point t0 at which f is continuous. (S3) and (S4).

The series for g converges uniformly since sup t∈T |n|>N cn eint ≤ |n|>N |cn | → 0 as N → ∞. let {cn } ∈ 1 (Z) and deﬁne g(t) = n∈Z cn eint . Injectivity follows from the uniqueness result (2.) We have g(m) = cm for every m.6 Deﬁne A(T) = {f ∈ L1 (T) : n∈Z |f (n)| < ∞} 1 = functions with Fourier coeﬃcients in (Z).Chapter 4 Fourier coeﬃcients in 1(Z) (or. Proof.14). and so g = {cm } as desired. (Hence g is continuous. The map : A(T) → 1 (Z) is a linear bijection. To prove surjectivity. 27 . f ∈ A(T)) Goal Establish the algebra structure of A(T) Reference [Katznelson] Section I.

meaning that if f. Deﬁnition 4.2 ( takes multiplication to convolution). g ∈ A(T) then f g ∈ A(T). Indeed fg = f ∗ g and f g A(T) ≤ f A(T) g A(T) .2).1) so that f is continuous (after redeﬁnition on a set of measure zero). (4. So (f g) 1 (Z) ≤ f 1 (Z) g 1 (Z) by (4. Clearly a∗b because |(a ∗ b)(n)| ≤ n m 1 (Z) ≤ a 1 (Z) b 1 (Z) (4. and hence integrable.e. This Fourier series converges absolutely and uniformly. Proof. b ∈ (a ∗ b)(n) = m∈Z 1 (Z) is one). (Z) by a(m)b(n − m).2) b . F ∈ A(T)) Our proof has shown each f ∈ A(T) is represented by its Fourier series: f (t) = n∈Z f (n)eint a. with (f g)(n) = = m 1 2π f (t)g(t)e−int dt T f (m) 1 2π g(t)e−i(n−m)t dt T by (4.28 CHAPTER 4.1) = m f (m)g(n − m) = (f ∗ g)(n). . Deﬁne a norm on A(T) by f A(T) = f 1 (Z) = n |f (n)|. FOURIER COEFFICIENTS IN 1 (Z) (OR. A(T) is an algebra. 1 A(T) is a Banach space under this norm (because Deﬁne the convolution of sequences a. f g is continuous.1. |a(m)| n |b(n − m)| = a 1 (Z) 1 (Z) Theorem 4.

H¨lder continuity: C α (T) ⊂ A(T) when α > 1 . We omit the proof.3 (Wiener’s Inversion Theorem). Section I. Clearly 1/f is continuous.6]. but it is not clear that (1/f ) belongs to 1 (Z). If f ∈ A(T) and f (t) = 0 for every t ∈ T then 1/f ∈ A(T). o 2 Theorem 4. for example. .29 Suﬃcient conditions for membership in A(T) are discussed in [Katznelson.

F ∈ A(T)) .30 CHAPTER 4. FOURIER COEFFICIENTS IN 1 (Z) (OR.

v and norm u = Given a sequence {un }n∈Z in H. 31 .) u 2 for all u ∈ H). and the series S({cn }) = cn un converges unconditionally. f ∈ L2(T)) Goal Study the Fourier ONB for L2 (T). If analysis is bounded ( n | u. (Z) → H c n un n {cn }n∈Z → and analysis operator T : H → 2 (Z) u → { u. un |2 ≤ (const. u .Chapter 5 Fourier coeﬃcients in 2(Z) (or. Theorem 5. un }n∈Z .1. deﬁne the synthesis operator S: 2 u. then so is synthesis. using analysis and synthesis operators Notation and deﬁnitions Let H be a Hilbert space with inner product u.

The limit as N → ∞ exists on the left side. Theorem 5. Convergence of the synthesis series is unconditional. u . F ∈ L2 (T)) (Z) → H is bounded. n=−N by deﬁnition of T u = N Hence T ∗ ({cn }n=−N ) = N n=−N cn un . and for 2 = n=−N N cn u. Remark 5.32 CHAPTER 5. we have N T ∗ ({cn }N n=−N ). which tends to 0 as A expands to ﬁll Z. g for all f.3 (Fourier coeﬃcients on L2 (T)). Proof. un c n un . because if A ⊂ Z then S({cn }n∈Z ) − S({cn }n∈A ) = S({cn }n∈Z\A ) ≤ S {cn } 2 (Z\A) . regardless of the order in which A expands. so that ∗ T = S. meaning analysis and synthesis are adjoint operations. First we prove Plancherel’s identity: since Pr ∗ f → f in L2 (T) by . The Fourier coeﬃcient (or analysis) operator : L2 (T) → 2 (Z) is an isometry. Hence S is bounded. therefore T ∗ ({cn }) = ∞ n=−∞ cn un . g L2 (T) L2 (T) = f 2 (Z) 2 (Z) (Plancherel) (Parseval) = f. The last proof shows S = T ∗ . u = {cn }n=−N . T u N 2 2 (Z) (OR. the adjoint T ∗ : each sequence {cn }. g ∈ L2 (T). with f f.2. u ∈ H. FOURIER COEFFICIENTS IN Proof. and hence on the right side. Since T is bounded. N ≥ 1.

Proof.6. analysis followed by synthesis equals the identity on L2 (T). That is.12) for Pr ∗ f r |f (n)|2 |f (n)|2 by monotone convergence.33 Theorem 2. (ˇ)ˆ = c. the Fourier synthesis operator ˇ: 2 (Z) → L2 (T) cn eint n {cn }n∈Z → Theorem 5. . (f )ˇ = f .4 (Fourier ONB). Part (a) says Fourier series converge in L2 (T). so is its adjoint. (b) If c = {cn } ∈ 2 (Z) then ( n∈Z cn eint ) (j) = cj . Since the Fourier analysis operator is bounded. Fourier analysis and synthesis are bounded operators. Parseval follows from Plancherel by polarization. That class is dense in L2 (T). we have 1 2π |f (t)|2 dt = lim T 1 r→1 2π 1 = lim r→1 2π = lim = n∈Z r→1 n∈Z f (t)(Pr ∗ f )(t) dt T f (t) T |n| n∈Z r|n| f (n)eint dt by (2. That is. g (and using dominated instead of monotone convergence). c int 2 (c) {e }n∈Z is an orthonormal basis of L (T). and so by continuity. or by repeating the argument for Plancherel with f. using the dense class of ﬁnite sequences in 2 (Z). Argue similarly for part (b). (a) If f ∈ L2 (T) then n f (n)eint = f with unconditional convergence in ˆ L2 (T). and analysis followed by synthesis equals the identity ( n f (n)eint = f ) on the class of trigonometric polynomials. Parts (a) and (b) together show that Fourier analysis and synthesis are inverse operations. f changed to f.

: L2 (T) → 2 (Z) is a linear bijection by Theorem 5. In Chapter 13 we will interpolate to show : Lp (T) → p (Z). noting f (n) = f. observe eint . 0 if n = m. 1 1 + = 1.5. The basis property follows from part (a). Fourier analysis satisﬁes : L1 (T) → : L2 (T) → ∞ 2 (Z) (Z) by Theorem 1.3. Further. eint Remark 5. eimt = 1 2π eint e−imt dt = T 2 (Z) (OR. whenever 1 ≤ p ≤ 2. F ∈ L2 (T)) 1 if n = m.2.4. L2 (T) . (isometrically) by Theorem 5.34 CHAPTER 5. p p . FOURIER COEFFICIENTS IN For orthonormality in part (c).

2 [Grafakos] Section 2. and 1 ≤ p.1 (Weak and strong operators). .) Call T strong (p. in next Chapter References [Duoandikoetxea] Section 2. f ∈ Lp (X). q ≤ ∞. Suppose T : Lp (X) → {measurable functions on Y }. we call T weak (p.e. µ) and (Y. Prove weak and strong bounds on the Hardy–Littlewood maximal function Prepare for sumability pointwise a.Chapter 6 Maximal functions Goals Connect abstract maximal functions to convergence a. Let (X. q) if C > 0 exists such that ν {y ∈ Y : |(T f )(y)| > λ} 1/q ≤ C f 35 Lp (X) λ ∀ λ > 0. meaning a constant C > 0 exists such that Tf Lq (Y ) ≤C f Lp (X) .1 [Stein] Section 1. (We do not assume T is linear. When q < ∞. q) if T is bounded from Lp (X) to Lq (Y ). f ∈ Lp (X). ν) be measure spaces.1 Deﬁnition 6.e.

MAXIMAL FUNCTIONS When q = ∞. When q = ∞ the result is immediate by deﬁnition. Suppose q < ∞. we call T weak (p. ∞): Tf L∞ (Y ) ≤C f Lp (X) . Lemma 6. ∞) if it is strong (p. q) <∞ since ν(Z) < ∞ and ∞ 1 λ−1−q+r dλ < ∞ (using that −q + r < 0). Write g = T f . Write E(λ) = {y ∈ Y : |(T f )(y)| > λ} for the level set of T f above height λ. Lemma 6. Strong (p. Then λq ν E(λ) = E(λ) λq dν(y) |(T f )(y)|q dν(y) E(λ) q Lq (Y ) ≤ ≤ Tf and so since λ < |T f | on E(λ) ν E(λ) 1/q ≤ ≤ Tf C f Lq (Y ) λ Lp (X) λ if T is strong (p. locally. q). loc Thus intuitively. q) then T f ∈ Lr (Y ) for all 0 < r < q. f ∈ Lp (X).2. q). We will show T f ∈ Lr (Z).3. Then |g(y)|r dν(y) Z ∞ 0 1 = ≤ 0 rλr−1 ν {y ∈ Z : |g(y)| > λ} dλ ∞ 1 by AppendixB Lp (X) q rλr−1 ν(Z) dλ + rλr−1 C f λ dλ by weak (p. Proof. q) ⇒ weak (p. Proof. If T is weak (p. Let f ∈ Lp (X) and suppose Z ⊂ Y with ν(Z) < ∞. .36 CHAPTER 6. T “almost” maps Lp into Lq .

so that f ∈ C. q) and each Tn is linear. Taking a countable sequence of λ 0. in general. Deﬁne T ∗ : Lp (X) → {measurable functions on X} by (T ∗ f )(x) = sup |(Tn f )|(x)|.37 Theorem 6. ≤ µ {x ∈ X : T ∗ (f − fk )(x) + |(f − fk )(x)| > 2λ} ∗ by triangle inequality ≤ µ {x ∈ X : T (f − fk )(x) > λ} + µ {x ∈ X : |(f − fk )(x)| > λ} C f − fk Lp (X) q f − fk Lp (X) p ≤ + by weak (p. Let fk ∈ C with fk → f in Lp (X). . 3.e} n is closed in Lp (X).e.). ∞]. We show f ∈ C. . Therefore lim supn |(Tn f )(x) − f (x)| ≤ 2λ a. T ∗ is called the maximal operator for the family {Tn }. q) on T ∗ λ λ →0 as k → ∞.).4 (Maximal functions and convergence a. Remark 6.e. we conclude lim supn |(Tn f )(x) − f (x)| = 0 a. In this theorem a quantitative hypothesis (weak (p. µ {x ∈ X : lim sup |(Tn f (x) − f (x)| > 2λ} n = µ {x ∈ X : lim sup |Tn (f − fk )(x) − (f − fk )(x)| > 2λ} n by linearity and the pointwise convergence Tn fk → fk a.e. . Therefore limn (Tn f )(x) = f (x) a. Suppose q < ∞. . If T ∗ is weak (p. Note T ∗ is not linear. 2.e. Proof.5. Clearly it takes values in [0. The case q = ∞ is left to the reader. n x ∈ X. then the collection C = {f ∈ Lp (X) : lim(Tn f )(x) = f (x) a.e. q)) implies a qualitative conclusion (closure of the collection C where Tn f → f a... Assume Tn : Lp (X) → {measurable functions on X} for n = 1.e. For any λ > 0.

Bij . so that radius (Bi ) ≤ radius(Bij ). Thus k l Bi ≤ i=1 l (3Bij ) j=1 ≤ j=1 |3Bij | l d j=1 l =3 =3 by disjointness of the Bij . . . . k}. Let i ∈ {1. If Bi is not one of the Bij chosen. . Re-label the balls in decreasing order of size: |B1 | ≥ |B2 | ≥ · · · ≥ |Bk |. choose ij+1 to be the smallest index i > ij such that Bi is disjoint from Bi1 . Proof. |Bij | Bij j=1 d . then Bi must intersect one of the Bij and be smaller than it. Thus the subcollection covers at least 1/3d of the total volume of the balls. . . The Bij are pairwise disjoint. Hence Bi ⊂ 3Bij (where we mean the ball with the same center and three times the radius).6 (Covering). After choosing ij . we will need: Lemma 6. Choose i1 = 1 and employ the following greedy algorithm. by construction. Then there exists a pairwise disjoint subcollection {Bij }l of balls such j=1 that k l l Bi ≤ 3 i=1 d j=1 Bij = 3 d j=1 |Bij |. . MAXIMAL FUNCTIONS To apply maximal functions on Rd and T. Continue until no such ball Bi exists. Let {Bi }k be a ﬁnite collection of open balls in i=1 Rd . .38 CHAPTER 6.

Each x ∈ F is the center of some ball B such that 1 |B| < |f (y)| dy. Let F ⊂ E(λ) be compact. For weak (1. . Properties Mf ≥ 0 |f | ≤ |g| ⇒ M f ≤ M g M (f + g) ≤ M f + M g (sub-linearity) M c = c if c = (const.) ≥ 0 Theorem 6. F is covered by ﬁnitely many such balls.r) for some r > 0.1) where E(λ) = {x ∈ Rd : M f (x) > λ}. and M f is lower semicontinuous (and measurable). r)| |f (y)| dy > λ B(x. 1) and strong (p.r) = “largest local average” of |f | around x.39 Deﬁnition 6. M is weak (1. . so that x ∈ E(λ). If x ∈ E(λ) then 1 |B(x. Thus E(λ) is open (and measurable). The Hardy–Littlewood (H-L) maximal function of a locally integrable function f on Rd is (M f )(x) = sup r>0 1 |B(x.8 (H-L maximal operator). The same inequality holds for all x close to x. . 1) we show |E(λ)| ≤ 3d f L1 (Rd ) λ (6.7. p) for 1 < p ≤ ∞. (6. Proof. . say B1 . . r)| |f (y)| dy B(x.2) λ B By compactness. Bk .

λ {x:|f (x)|>λ/2} by the above weak (1.3) . 0 otherwise f (x) if |f (x)| ≤ λ/2 = “small” part of f . . by deﬁnition of M f . . Hence |E(λ)| = {x : M f (x) > λ} ≤ {x : M g(x) > λ/2} 3d g L1 (Rd ) ≤ λ/2 2 · 3d = |f (x)| dx. .6 |f (y)| dy by (6. MAXIMAL FUNCTIONS The Covering Lemma 6. note M f (x) ≤ f L∞ (Rd ) for all x ∈ Rd . Hence M f L∞ (Rd ) ≤ f L∞ (Rd ) . For strong (p. 0 otherwise g(x) = h(x) = Then f = g + h and |h| ≤ λ/2. Then k |F | ≤ i=1 Bi l d j=1 ≤3 ≤ ≤ |Bij | l by Covering Lemma 6.2) by disjointness 3d λ d j=1 Bij 3 |f (y)| dy λ Rd 3d = f L1 (Rd ) . p) when 1 < p < ∞. . λ Taking the supremum over all compact F ⊂ E(λ) gives (6.6 yields a subcollection Bi1 . 1) result (6. Bil . For strong (∞.1).40 CHAPTER 6. so that M f ≤ M g + λ/2. ∞). let λ > 0 and deﬁne f (x) if |f (x)| > λ/2 = “large” part of f .

the indicator function f = 1[−1.3) ≤ 2 · 3d p = 2p 3d λp−2 {x:|f (x)|>λ/2} p p−1 |f (x)|p dx Rd by Lemma B. / The maximal function is locally integrable provided f ∈ L log L(Rd ).1 with r = 1. α = 2. Notice the constant in the strong (p. 1). p) bound. We have proved the strong (p. As this observation suggests. the Hardy–Littlewood maximal operator is not strong (1.41 Therefore |M f (x)|p dx Rd ∞ 0 = pλp−1 |E(λ)| dλ ∞ 0 by Appendix B |f (x)| dxdλ by (6.1] in 1 dimension has M f (x) ∼ c/|x| when |x| is large. so that M f ∈ L1 (R). p) bound blows up as p 1. . For example. see Problem 9.

42 CHAPTER 6. MAXIMAL FUNCTIONS .

Goal Prove summability a. ∗ Lemma 7. If k ∈ L1 (T) is nonnegative and symmetric (k(−t) = k(t)).h] (t). 43 . using Fej´r and Poisson maximal functions e Deﬁnition 7. Dirichlet maximal function (D∗ f )(t) = sup |(Dn ∗ f )(t)| = sup |Sn (f )(t)| n n Fej´r maximal function (F f )(t) = sup |(Fn ∗ f )(t)| = sup |σn (f )(t)| e n n ∗ Poisson maximal function (P f )(t) = sup |(Pr ∗ f )(t)| 0<r<1 ∗ ∗ Gauss maximal function (G f )(t) = sup |(Gs ∗ f )(t)| 0<s<∞ 0<h<π Lebesgue maximal function (L f )(t) = sup |(Lh ∗ f )(t)| 1 where the Lebesgue kernel is Lh (t) = 2π 2h 1[−h.e. π]. then |(k ∗ f )(t)| ≤ k ∗ L1 (T) (L f )(t) for all t ∈ T. t+h 1 Notice (Lh ∗ f )(t) = 2h t−h f (τ ) dτ is a local average of f around t. extended 2π-periodically.2 (Majorization). f ∈ L1 (T).e. and decreasing on [0.1.Chapter 7 Fourier series: summability pointwise a.

E. π] (exercise). if h ≥ |t|. FOURIER SUMMABILITY POINTWISE A. if h < |t|. e F ∗ f ≤ 2L∗ |f | P ∗ f ≤ L∗ f Proof. so that P ∗ f ≤ L∗ f . Assume k is absolutely continuous. representing it as a linear combination of kernels Lh : π k(t) = k(|t|) = k(π) − |t| k (h) dh 1 2π π = k(π) − since 1 2hLh (t) = 2π Hence 2hLh (t)k (h) dh. . Pr (t) is nonnegative. Theorem 7.44 CHAPTER 7. π 1 2h(Lh ∗ f )(t) − k (h) dh 2π 0 T π 1 2h − k (h) dh (L∗ f )(t) |(k ∗ f )(t)| ≤ k(π)|(Lπ ∗ f )(t)| + 2π 0 using k(π) ≥ 0 and k ≤ 0 π 2 ≤ k(h) dh (L∗ f )(t) by parts 2π 0 = k L1 (T) (L∗ f )(t) (k ∗ f )(t) = k(π) 1 2π f (τ ) dτ + by symmetry of k. Thus convolution with a symmetric decreasing kernel is majorized by the Hardy–Littlewood maximal function. and decreasing on [0. Proof.2. Hence |Pr ∗ f | ≤ L∗ f by Majorization Lemma 7. 0 1. We ﬁrst establish a “layer cake” decomposition of k. with Pr L1 (T) = 1. symmetric.3 (Lebesgue dominates Fej´r and Poisson). For all f ∈ L1 (T). 0. for simplicity.

1) for F ∗ f .e. Note the kernel k is nonnegative. so that G∗ f ≤ L∗ f . as n → ∞ (Fej´r summability) e Pr ∗ f → f a. 2π n+1 Hence |Fn ∗ f | ≤ k ∗ |f | ≤ 2L∗ |f | by Majorization Lemma 7. and decreasing on [0. F ∗ .2. as r 1 (Abel summability) Lh ∗ f → f a. Argue similarly for P ∗ f . Theorem 7. If f ∈ L1 (T) then σn (f ) = Fn ∗ f → f a. 1) estimates for L∗ f and L∗ |f | imply weak (1. as follows: 1 Fn (t) = n+1 ≤ k(t) = since sin(n + 1)θ ≤ (n + 1) sin θ. but it is bounded by a symmetric decreasing kernel. The Gauss kernel can be shown to be symmetric decreasing. if π/(n + 1) ≤ |t| ≤ π. Corollary 7. π].e.e. sin 1 t t ≥ . 1) on T. {t ∈ T : (L∗ f )(t) > 2πλ} ≤ {t ∈ T : (L∗ |f |)(t) > 2πλ} ≤ 3 λ |f (t)| dt T by repeating the weak (1. with k L1 (T) = 1 2π 4π − < 2. π]. 0 ≤ t ≤ π. 2 π 0≤θ≤ π .e. so that F ∗ f ≤ 2L∗ |f |. as h 0 (Lebesgue diﬀerentiation theorem) . 1) proof for the Hardy–Littlewood maximal function.4. 2 def sin n+1 t 2 sin 1 t 2 1 n+1 2 (n + 1)2 π 2 /t2 if |t| ≤ π/(n + 1). These weak (1.). symmetric.5 (Summability a. since if (F ∗ f )(t) > λ then (L∗ |f |)(t) > λ/2 by Theorem 7. but we omit the proof. Proof.3. P ∗ and L∗ are weak (1.45 The Dirichlet kernel is not decreasing on [0.

e.4. The result that Lh ∗ f → f a. for each e 1 f ∈ L (T). Argue similarly for Pr ∗ f and Lh ∗ f .e. Thus C is dense in L1 (T). .. since Fn ∗ f → f uniformly when f is continuous. Proof.4 and the abstract convergence result in Theorem 6. Obviously C contains the continuous functions on T. thus proving Fej´r summability a.E. the set C = {f ∈ L1 (T) : lim(Fn ∗ f )(t) = f (t) a. FOURIER SUMMABILITY POINTWISE A. By the weak (1. Because C is also closed.e.46 CHAPTER 7. means 1 2h t+h f (τ ) dτ → f (t) a. 1) estimate in Corollary 7.} n is closed in L1 (T). t−h which is the Lebesgue diﬀerentiation theorem on T. it must equal L1 (T).e.

Proof.1 (Kolmogorov). Section II.1 Fourier series can behave badly for integrable functions. Theorem 8. Recall Sn (f ) = Dn ∗ f and D∗ f is the maximal function for the Dirichlet kernel. 47 . [Katznelson. sup |Sn (f )(t)| = ∞ n for all t ∈ T.2.3]. II.3 [Duoandikoetxea] Section 1. so that D∗ f ≡ ∞. There exists f ∈ L1 (T) whose Fourier series diverges unboundedly at every point. That is.Chapter 8 Fourier series: convergence at a point Goals State divergence pointwise can occur for L1 (T) Show divergence pointwise can occur for C(T) Prove convergence pointwise for C α (T) and BV (T) References [Katznelson] Section II.

There exists a continuous function whose Fourier series diverges unboundedly at t = 0. Theorem 8.1) of Dn . π − using deﬁnition (2. Then |Tn (g)| = 1 2π 1 ≥ 2π 1 = 2π ≥ Dn Dn (τ )g(τ ) dτ T |Dn (τ )| dτ − T\{intervals} 1 2π |Dn (τ )| dτ {intervals} |Dn (τ )| dτ − T L1 (T) 2 2π |Dn (τ )| dτ {intervals} 1 ε (2n + 1) π 2n + 1 ε = Dn L1 (T) − π ε = Dn L1 (T) − g L∞ (T) . Deﬁne Tn : C(T) → C f → Sn (f )(0) = (nth partial sum of f at t = 0). FOURIER SERIES: CONVERGENCE AT A POINT Even continuous functions can behave badly. n Proof. We show Tn = Dn L1 (T) . That is. 1   −1 if D (t) < 0. Each Tn is bounded since |Tn (f )| = |Sn (f )(0)| = |(Dn ∗ f )(0)| 1 Dn (τ )f (0 − τ ) dτ = 2π T ≤ Dn L1 (T) f L∞ (T) . sup |Sn (f )(0)| = ∞.2.    with total length of those intervals < ε/(2n + 1). Thus Tn ≤ Dn L1 (T) . n g(t) =  except for small intervals around the zeros of Dn . Then Tn is linear. Let ε > 0 and choose g ∈ C(T) with g L∞ (T) = 1 and g even and  if Dn (t) > 0.48 CHAPTER 8.

II. Recalling that Dn L1 (T) → ∞ as n → ∞ (in fact. proving divergence not only at t = 0 but on a dense set of t-values. Proof. 2 Notice the factor {· · · } is integrable with respect to τ .3 (Dini’s Convergence Test). Ex.49 Thus Tn ≥ Dn L1 (T) − ε/π for all ε > 0.1) [f (t − τ ) − f (t)] cos(nτ ) dτ −π by expanding sin (n + 1 )τ with a trigonometric identity. then the Fourier series of f converges to f (t). Theorem 8. [Katznelson. And τ → [f (t − τ ) − f (t)] is integrable too. Sec II. as desired. Let f ∈ L1 (T).1) tend to 0 as n → ∞. by the Riemann–Lebesgue Corollary 1. Now we prove convergence results. Another proof. Dn ∼ c log n by [Katznelson. .4 (Convergence for H¨lder continuous f ). Corollary 8. by the Dini hypothesis.2] gives an explicit construction of f .1]) we conclude from the Uniform Bounded Principle (Banach–Steinhaus) that there exists f ∈ C(T) with supn |Tn (f )| = ∞. If π −π f (t − τ ) − f (t) dτ < ∞ τ then the Fourier series of f converges at t to f (t). t ∈ T. If f ∈ C α (T). 1 Sn (f )(t) − f (t) = 2π sin (n + 1 )τ 2 [f (t − τ ) − f (t)] dτ 1 sin 2 τ −π 1 π using that Dn (τ ) dτ = 1 2π −π π π −π = 1 2π 1 τ f (t − τ ) − f (t) cos τ 1 τ 2 sin 2 τ + 1 2π π sin(nτ ) dτ (8. 0 < o α ≤ 1. for every t ∈ T. Hence both integrals in (8.5 (after expressing sin(nτ ) and cos(nτ ) in terms of e±inτ ).1. and so Tn = Dn L1 (T) .

2) Sn (g)(t) − [g(t+) + g(t−)] = 2 2π 0 1 π + g(t + τ ) − g(t+) Dn (τ ) dτ (8. Write τ π Hn (τ ) = 0 Dn (σ) dσ so that Hn = Dn . Let 0 < δ < π. then the Fourier series of the other function converges at t to the same value. If f ∈ BV (T) then the 1 Fourier series converges everywhere to 2 [f (t+)+f (t−)].3.5 (Localization Principle). Thus Fourier series depend only on local information. Theorem 8. say f = g − h. On the interval (t − π.)|τ |α dτ < ∞.50 CHAPTER 8. so that G is increasing with G(0+) = 0. Let t ∈ T.3. FOURIER SERIES: CONVERGENCE AT A POINT Proof. (Exercise. π). if two functions agree on a neighborhood of t and the Fourier series of one of them converges at t. We have 1 1 π g(t − τ ) − g(t−) Dn (τ ) dτ (8. and hence converges to f (t) at every point of continuity.δ] .6 (Convergence for bounded variation f).3) 2π 0 1 since Dn (τ ) is even and 2π 0 Dn (τ ) dτ = 1 . 2 Let G(τ ) = g(t + τ ) − g(t+) for τ ∈ (0. Prove the Fourier series in fact converges uniformly. Then (8. It suﬃces to prove the theorem for g and h individually. t ∈ T. Proof.) Corollary 8. |τ | Now apply Dini’s Theorem 8. If f vanishes on a neighborhood of t. Put H¨lder into Dini: o π −π f (t − τ ) − f (t) dτ ≤ τ π −π (const. Apply Dini’s Theorem 8. In particular. then Sn (f )(t) → 0 as n → ∞. t + π).3) = 1 δ 1 π G(τ )Hn (τ ) dτ + G(τ )Dn (τ ) dτ 2π 0 2π δ 1 1 = G(δ)Hn (δ) − Hn (τ ) dG(τ ) + o(1) 2π 2π (0. Let f ∈ L1 (T). express f as the diﬀerence of two bounded increasing functions. Proof.

2).  0. Thus we are done.51 as n → ∞. Argue similarly for (8. 1 < p < ∞ then the Fourier series of f converges to f (t) for almost every t ∈ T. and for h. vanishes near the origin. δ < τ < π. since the function  G(τ ).3) → 0 as n → ∞. by parts in the ﬁrst term and by the Localization Principle in the last term. . −π < τ < −δ. Therefore (8. −δ < τ < δ. We have τ |Hn (τ )| ≤ 0 sin (n + 1 )σ 2 dσ + 1 σ 2 1 (n+ 2 )τ τ 0 1 sin (n + )σ 2 1 sin 1 σ 2 − 1 1 σ 2 dσ ≤2 0 ρ 0 ≤ 2 sup ρ>0 π sin σ σ3 dσ + (const.) σ by a change of variable <∞ since limρ→∞ ρ sin σ σ 0 dσ exists.δ] dG(τ ) since G(0+) = 0 L∞ (T) · 2G(δ) as δ → 0. A much deeper result is: Theorem 8. provided we show sup Hn n L∞ (T) < ∞.3)| ≤ n 1 sup Hn 2π n 1 sup Hn = 2π n →0 L∞ (T) G(δ) + (0.) 2 dσ σ σ 0 sin σ dσ + (const. Hence lim sup |(8. The convergence results so far in this chapter rely just on Riemann– Lebsgue and direct estimates.   G(−τ ).7 (Carleson–Hunt). If f ∈ Lp (T).

Thus one wants sup |Dn ∗ f | n Lp (T) ≤ Cp f Lp (T) for 1 < p < ∞.e. p) for 1 < p < ∞. follows from Chapter 6. Then it is weak (p. The idea is to prove that the Dirichlet maximal operator (D∗ f )(t) = supn |(Dn ∗ f )(t)| is strong (p. the result is spectacularly false by Kolmogorov’s Theorem 8. but that is not good enough to prove Carleson–Hunt! . Omitted. p). Proof. and so convergence a.1.52 CHAPTER 8. The next Chapters show sup Dn ∗ f n Lp (T) ≤ Cp f Lp (T) . FOURIER SERIES: CONVERGENCE AT A POINT For p = 1.

(b) This part follows immediately from the Uniform Boundedness Principle in functional analysis. 53 .1. Proof.Chapter 9 Fourier series: norm convergence Goals Characterize norm convergence in terms of uniform norm bounds Show norm divergence can occur for L1 (T) and C(T) Show norm convergence for Lp (T) follows from boundedness of the Hilbert transform Reference [Katznelson] Section II. 1 ≤ p < ∞. (b) If supn Sn B→B = ∞ then there exists f ∈ B whose Fourier series diverges unboundedly: supn Sn (f ) B = ∞. (a) If supn Sn B→B < ∞ then Fourier series converge in B: n→∞ lim Sn (f ) − f B =0 for each f ∈ B. Let B be one of the spaces C(T) or Lp (T).1 Theorem 9.

1 ≤ p < ∞. Consider a sequence fm ∈ C with fm → f . so that C is closed. We must show f ∈ C. Choose ε > 0 and ﬁx m such that fm − f < ε/2(A + 1). Then Tn f − f ≤ Tn f − Tn fm + Tn fm − fm + fm − f ≤ (A + 1) f − fm + Tn fm − fm < ε whenever n > N . Norm Estimates Sn B→B ≤ Dn L1 (T) when B is C(T) or Lp (T). Proof. . which proves part (a). Proposition 9. Hence the set C = {f ∈ B : lim Sn (f ) = f in B} n→∞ is dense in B. Since fm ∈ C there exists N such that Tn fm − fm < ε/2 whenever n > N . For B = C(T) we have Sn C(T)→C(T) = Dn L1 (T) . by the following proposition. Let B be any Banach space and assume the Tn : B → B are bounded linear operators.3 (Divergence in C(T)). This upper estimate is not useful. If supn Tn B→B < ∞ then C = {f ∈ B : lim Tn f = f in B} n→∞ is closed. Further. FOURIER SERIES: NORM CONVERGENCE (a) The collection of trigonometric polynomials is dense in B (as remarked after Theorem 2.2. and so C = B.6). Example 9.54 CHAPTER 9. The set C is also closed. since Sn (f ) B = Dn ∗ f B ≤ Dn L1 (T) f B. as desired. Let A = supn Tn B→B . if g is a trigonometric polynomial then Sn (g) = g whenever n exceeds the degree of g. since we know Dn L1 (T) → ∞.

Proof. see [Grafakos. 1 < p < ∞.) That is ∞ Hf ∼ (−i) sign(n)f (n)eint .4 (Divergence in L1 (T)).9]. (Thus H is a Fourier multiplier operator. Fix n. Of course. and so Dn L1 (T) = lim Sn (FN ) N →∞ L1 (T) ≤ Sn = Sn L1 (T)→L1 (T) L1 (T)→L1 (T) .1(b)) there exists an integrable function f ∈ L1 (T) whose Fourier series diverges unboundedly in the L1 norm: supn Sn (f ) L1 (T) = ∞. this result follows already from the pointwise divergence in Chapter 8. Therefore supn Sn C(T)→C(T) = ∞. For an explicit example of L1 divergence. For B = L1 (T) we have Sn L1 (T)→L1 (T) = Dn L1 (T) . FN L1 (T) Therefore supn Sn L1 (T)→L1 (T) = ∞. so that Sn (g) C(T) ≥ |Sn (g)(0)| ≥ Dn L1 (T) −ε g C(T) . so that (by Theorem 9. called the Hilbert transform on T. We shall prove (in Chapters 10–12) the existence of a bounded linear operator H : Lp (T) → Lp (T).55 Indeed. n=−∞ . with the property (Hf )(n) = −i sign(n)f (n).1(b)) there exists a continuous function f ∈ C(T) whose Fourier series diverges unboundedly in the uniform norm: supn Sn (f ) C(T) = ∞. for each ε > 0 one can construct g ∈ C(T) that approximates sign(Dn ) (like in Chapter 8). 1 < p < ∞ 1. Aside. Convergence in Lp (T).5. Example 9. Exercise 3. so that (by Theorem 9. Then Sn (FN ) = FN ∗ Dn → Dn in L1 (T) as N → ∞.

1) and boundedness of the Riesz projection it follows that sup Sm m Lp (T)→Lp (T) ≤2 P Lp (T)→Lp (T) <∞ when 1 < p < ∞.1) must equal Sm (f ). . Let 1 < p < ∞. 4. Then the Riesz projection P : Lp (T) → Lp (T) deﬁned by 1 1 P f = f (0) + (f + iHf ) 2 2 is also bounded. 3.) o Observe P projects onto the nonnegative frequencies: Pf ∼ n≥0 f (n)eint since i(−i sign(n)) = sign(n). when 1 < p < ∞. Proof. Then n→∞ lim Sn (f ) − f Lp (T) =0 for each f ∈ Lp (T). the left side of (9. and we conclude that the left side of (9.1) has the same Fourier coeﬃcients as Sm (f ).1 we conclude: Theorem 9.14). Hence from Theorem 9. by H¨lder’s inequality. By the uniqueness result (2. on the right side.56 CHAPTER 9. ∞ (9. (Note the constant term f (0) is bounded by f Lp (T) . From (9. FOURIER SERIES: NORM CONVERGENCE 2. The following formula expresses the Fourier partial sum operator in terms of the Riesz projection and some modulations: e−imt P (eimt f ) − ei(m+1)t P (e−i(m+1)t f ) = Sm (f ).5 (Fourier series converge in Lp (T)). It remains to prove Lp boundedness of the Hilbert transform.1) e imt f∼ n=−∞ f (n)ei(m+n)t f (n)ei(m+n)t n≥−m P (eimt f ) ∼ e e i(m+1)t −imt P (e imt f) ∼ n≥−m f (n)eint f (n)eint n≥m+1 P (e −i(m+1)t f) ∼ Subtracting the last two formulas gives Sm (f ).

1. Hence H L2 →L2 = 1. with Hf 2 L2 (T) = n∈Z |(Hf )(n)|2 = n=0 |f (n)|2 ≤ f 2 2 (Z) = f 2 L2 (T) by Plancherel in Chapter 5.2 (Adjoint of Hilbert transform). We call {−i sign(n)} the multiplier sequence of H. H ∗ = −H 57 . the deﬁnition indeed yields Hf ∈ L2 (T).3 Deﬁnition 10.Chapter 10 Hilbert transform on L2(T) Goal Obtain time and frequency representations of the Hilbert transform Reference [Edwards and Gaudry] Section 6. Since | sign(n)| ≤ 1. The Hilbert transform on L2 (T) is H : L2 (T) → L2 (T) ∞ f→ n=−∞ − i sign(n)f (n) eint . Lemma 10. Observe also H 2 (f ) = H(Hf ) = − n=0 f (n)eint = −f + f (0).

58 CHAPTER 10.3. −Hg 2 (Z) 2 (Z) L2 (T) . g 2 (Z) 2 (Z) = −i sign(n)f (n). Proof. 2 But the convolution is ill-deﬁned because the Hilbert kernel cot(t/2) is not integrable. geometric series calculations show that N −1 N − i sign(n)einτ = i n=−N n=−N einτ − i n=1 −iτ einτ =i −e ei(N +1)τ − eiτ −i e−iτ − 1 eiτ − 1 e−i(N +1/2)τ − e−iτ /2 + ei(N +1/2)τ − eiτ /2 =i e−iτ /2 − eiτ /2 1 cos τ − cos (N + 2 )τ 2 = . −Hg = f. g(n) = f (n).1) (10. g ∈ L2 (T). 1 ε→0 2π f (t − τ ) cot ε<|τ |<π τ dτ 2 Remark 10. ε]. Hf.2) evaluates the convolution in the principal valued sense. If f ∈ L2 (T) is C 1 -smooth on an open interval I ⊂ T. then (Hf )(t) = 1 2π π [f (t − τ ) − f (t + τ )] cot 0 τ dτ 2 (10. i sign(n)g(n) = f . HILBERT TRANSFORM ON L2 (T) Proof. taking the limit of integrals over T \ [−ε.4. Formally (10. (10. First.2) says that Hf = f ∗ cot t . That is why (10. For f.3) sin( τ ) 2 e −i(N +1)τ .2) = lim for almost every t ∈ I. g L2 (T) = Hf . Proposition 10.

the N th partial sum of Hf is N − i sign(n)f (n) einτ n=−N 1 = 2π 1 = 2π = 1 2π 1 = 2π − N f (τ ) T π n=−N (−i) sign(n)ein(t−τ ) dτ N f (t − τ ) −π π n=−N (−i) sign(n)einτ dτ cos τ 2 by τ → t − τ [f (t − τ ) − f (t + τ )] 0 π − cos (N + 1 )τ 2 dτ sin( τ ) 2 dτ by (10.3) [f (t − τ ) − f (t + τ )] cot 0 τ 2 1 2π π 0 f (t − τ ) − f (t + τ ) 1 cos (N + )τ dτ.2). by the C 1 -smoothness of f . Now write (10.59 Second. Hence the second integral tends to 0 as N → ∞ by the Riemann-Lebesgue Corollary 1. τ sin( 2 ) 2 If t ∈ I then the second integrand belongs to L1 (T) since it is bounded for τ near 0.1) now follows.5. Formula (10.e.1) as (Hf )(t) = lim 1 ε→0 2π π [f (t − τ ) − f (t + τ )] cot ε τ 2 dτ and use oddness of cot(τ /2) to obtain (10. . because the partial sum N −i sign(n)f (n) einτ n=−N converges to Hf (t) in L2 (T) and hence some subsequence of the partial sums converges to (Hf )(t) a.

HILBERT TRANSFORM ON L2 (T) .60 CHAPTER 10.

there exists a unique m ∈ Zd with x ∈ 2−k [0. Given Q ∈ Qk and j < k. 2.1. That is. 61 .5 [Grafakos] Section 4. 1)d + m : m ∈ Zd }. Call ∪k Qk the collection of dyadic cubes. preparing for a weak (1. For k ∈ Z. 1)d + m .Chapter 11 Calder´n–Zygmund o decompositions Goal Decompose a function into good and bad parts. Facts (exercise) 1. For all x ∈ Rd and k ∈ Z. Notice the cubes in Qk are small when k is large. there exists a unique Q ∈ Qk such that x ∈ Q.3 Deﬁnition 11. 1) estimate on the Hilbert transform References [Duoandikoetxea] Section 2. let Qk = {2−k [0. there exists a unique Q ∈ Qj with Q ⊂ Q.

e. λ > 0. let loc (Ek f )(x) = Q∈Qk 1 |Q| f (y) dy 1Q (x). Q Then Ek f is constant on each cube in Qk (equalling there the average of f over that cube). or else the cubes are disjoint. let f ∈ L1 (Rd ). Each cube in Qk contains exactly 2d cubes in Qk+1 . Given two dyadic cubes. Since Md f ≤ Md |f |. 4. (a) Md is weak (1. Let Ω = {x ∈ Rd : (Md f )(x) > λ}. either one of them is contained in the other. Deﬁnition 11. 1 (2−j )d Rd j→−∞ f (y) dy . Ωk = {x ∈ Rd : (Ek f )(x) > λ and (Ej f )(x) ≤ λ for all j < k}. For part (a). CALDERON–ZYGMUND DECOMPOSITIONS 3.1) whenever Ω is a ﬁnite union of cubes in Qk . And if x ∈ Ω then (Ek f )(x) > λ for some k. and Ek f dx = Ω Ω f dx (11. Clearly Ωk ⊂ Ω. 1). Deﬁne the dyadic maximal function (Md f )(x) = sup |(Ek f )(x)| k = sup 1 |Q| f (y) dy : Q is a dyadic cube containing x . (b) If f ∈ L1 (Rd ) then limk→∞ (Ek f )(x) = f (x) a.2.3. We employ a “stopping time” argument like in probability theory for martingales.62 ´ CHAPTER 11. For f ∈ L1 (Rd ). loc Proof. because j→−∞ lim (Ej f )(x) ≤ lim =0 < λ. a smallest such k exists. we can assume f ≥ 0. Q Theorem 11.

just Q(l) ∈ Qk for some k. bl vi. 1). using that the dyadic maximal operator Md is weak (1. l ≤ 2d+1 λ|Q(l)| 1 λ |Q(l)| ≤ f L1 (Rd ) .4 (Calder´n–Zygmund decomposition at level λ). so that |Ω| = k |Ωk | by disjointness of the Ωk Ek f dx since Ek f > λ on Ωk 1 ≤ λ 1 = λ 1 λ k Ωk f dx k Ωk by (11. and so x ∈ Ωk . when proving the dyadic maximal function is weak (1. Rd Therefore Md is weak (1. g L1 (Rd ) ≤ f L1 (Rd ) . since Ωk equals a union of cubes in Qk (recall Ek f is constant on each cube in Qk ) ≤ f dx.63 Choosing the smallest k implies (Ej f )(x) ≤ λ for all j < k. Hence Ω = ∪k Ωk .4 loc (exercise). Part (b) holds if f is continuous. and hence if f ∈ L1 (Rd ) by Theorem 6.1). λ > 0. g L∞ (Rd ) ≤ 2d λ. Theorem 11. 1). f = g + b ii. 1). b L1 (Rd ) ≤2 f L1 (Rd ) iii. iv. Note we did not need a covering lemma. because disjointness of the cubes is built into the construction. b (x) dx Q(l) l L1 (Rd ) =0 v. b = l bl where bl is supported in a dyadic cube Q(l) and the {Q(l)} are disjoint. Let f ∈ o 1 d L (R ). we do not assume Q(l) ∈ Ql . Then there exists a “good’ function g ∈ L1 ∩ L∞ (Rd ) and a “bad” function b ∈ L1 (Rd ) such that i.

so that x ∈ Ωk for some k. Next we show g L∞ (Rd ) ≤ 2d λ. 1) estimate that we proved. Q(l) |g(x)| dx ≤ Q(l) Hence b L1 (Rd ) = f − g L1 (Rd ) ≤ 2 f L1 (Rd ) . Then (Ek−1 |f |)(x) ≤ λ. Together. Deﬁne b(x) = l 1 |Q(l)| f (y) dy 1Q(l) (x) Q(l) bl (x) = f (x) − 0 1 |Q(l)| Q(l) f (y) dy on Q(l). Therefore |g(x)| ≤ 2d λ. Then let g =f −b = For (ii). Since x ∈ Ωk for all k we have (Ek |f |)(x) ≤ λ for all k. For (i). on Rd \ ∪l Q(l). Apply the proof of Theorem 11. Next suppose x ∈ Q(l) for some l. f (x) L1 (Rd ) ≤ f L1 (Rd ) . Then g(x) = f (x). argue as follows. Hence / |f (x)| ≤ λ (for almost every such x) by Theorem 11. for each l.3(b). since g = f oﬀ ∪l Q(l) and on Q(l) |f (x)| dx. which means 1 |Q| |f (y)| dy ≤ λ Q for some cube Q ∈ Qk−1 with x ∈ Q(l) ⊂ Q. CALDERON–ZYGMUND DECOMPOSITIONS Proof. so that |g(x)| ≤ λ. Let bl (x) = f (x) − so that bl integrates to 0. note g we have 1 |Q(l)| Q(l) f (y) dy on Q(l).2) since Q(l) ⊂ Q and side(Q) = 2 side(Q(l)). Hence 1 2d |Q(l)| Q(l) |f (y)| dy ≤ λ (11. . Property (vi) is just the weak (1. (iii).64 ´ CHAPTER 11.3 to |f |. on Rd \ ∪l Q(l). Suppose x ∈ Rd \ ∪l Q(l). for each l. by deﬁnition of g. these cubes form the collection {Q(l)}. (iv). and decompose the disjoint sets Ωk into dyadic cubes in Qk .

Corollary 11. 1) + m where k ≥ 1. g L1 (T) ≤ f L1 (T) . to get f = g + b. bl vi. 0. Then there exists a “good’ function g ∈ L∞ (T) and a “bad” function b ∈ L1 (T) such that i. b (t) dt I(l) l L1 (T) =0 4 λ|I(l)| 2π 2π λ v. 0 ≤ t < 1. so that the dyadic intervals have length at most 2π and thus ﬁt into T. and where the {I(l)} are disjoint. 0 ≤ m ≤ 2k − 1. Apply the Calder´n–Zygmund Theorem 11.4 to o f (t) = f (2πt). b = l bl where bl is supported in some interval I(l) of the form 2π · 2−k [0. b L1 (T) ≤2 f L1 (T) iii. We will restrict to “large” λ values. Let f ∈ L1 (T). just note |bl (x)| dx ≤ 2 Q(l) |f (x)| dx Q(l) d+1 by deﬁnition of bl ≤2 by (11. λ|Q(l)| Now we adapt the theorem to T. λ > o f L1 (T) . otherwise. since (Ek |f |)(t) ≤ 1 1 |f (τ )| dτ 2−k 0 2π 1 = 2k |f (τ )| dτ 2π 0 ≤ f L1 (T) since k ≤ 0 <λ .5 (Calder´n–Zygmund decomposition on T). g L∞ (T) ≤ 2λ.2). Let d = 1. l ≤ |I(l)| ≤ f L1 (T) Proof. Note Ωk is empty for k ≤ 0.65 For (v). f = g + b ii. iv.

4. 1]. CALDERON–ZYGMUND DECOMPOSITIONS by assumption on λ. Thus I(l) = 2πQ(l) has the form stated in the Corollary. 1] for k ≥ 1. with f = g + b yielding f = g + b. Ωk ⊂ [0. Further. since Ek |f | = 0 outside [0.66 ´ CHAPTER 11. . The Corollary now follows from Theorem 11.

1 (weak (1. we have |{t ∈ T : |(Hf )(t)| > λ}| ≤ |{t ∈ T : |(Hg)(t)| > λ/2}| + |{t ∈ T : |(Hb)(t)| > λ/2}| = γ + β. 1) on L2 (T)). So suppose λ > f L1 (T) .5 to get f = g + b. If λ ≤ f L1 (T) then A = 2π works. p) estimates by interpolation and duality Reference [Duoandikoetxea] Section 3. And b = f − g ∈ L2 (T) so that Hb ∈ L2 (T). Further. Hence Hb = l Hbl with convergence in L2 (T).Chapter 12 Hilbert transform on Lp(T) Goals Prove a weak (1. There exists A > 0 such that |{t ∈ T : |(Hf )(t)| > λ}| ≤ for all λ > 0 and f ∈ L2 (T). bl ∈ L2 (T) and b = l bl with convergence in L2 (T). 67 A f λ L1 (T) . Note g ∈ L∞ (T) o and so g ∈ L2 (T). using disjointness of the supports of the bl . Proof. hence Hg ∈ L2 (T) by Chapter 10. Apply the Calder´n–Zygmund Corollary 11. Since Hf = Hg + Hb.3 Theorem 12. 1) estimate on the Hilbert transform on T Deduce strong (p.

68

CHAPTER 12. HILBERT TRANSFORM ON LP (T)

say. First, use the L2 theory on g:

γ≤
T

|(Hg)(t)|2 dt (λ/2)2 since H since g since g
L2 (T)→L2 (T)

4 ≤ 2 |g(t)|2 dt λ T 8 ≤ |g(t)| dt λ T 8 · 2π ≤ f L1 (T) λ

= 1 by Chapter 10

L∞ (T)

≤ 2λ ≤ f
L1 (T) .

L1 (T)

Second, use L1 estimates on b, as follows:

β≤
l

2I(l) + |{t ∈ T \
l L1 (T)

2I(l) : |(Hb)(t)| > λ/2}| |(Hb)(t)| dt λ/2

4π ≤ f λ

+
T\∪l 2I(l)

by the Calder´n–Zygmund Corollary 11.5(vi) o 4π 2 |(Hbl )(t)| dt ≤ f L1 (T) + λ λ l T\2I(l)

since |Hb| ≤

l

|Hbl | a.e.

To ﬁnish the proof, we show

|(Hbl )(t)| dt ≤ (const.) f
l T\2I(l)

L1 (T) .

(12.1)

69 By Proposition 10.3 on the interval T \ 2I(l), we have |Hbl (t)| dt
T\2I(l)

=
T\2I(l)

1 2π

bl (τ ) cot
I(l)

1 (t − τ ) dτ dt 2

noting t − τ is bounded away from 0, since τ ∈ I(l) and t ∈ 2I(l), / 1 1 1 bl (τ ) cot (t − τ ) − cot (t − cl ) dτ dt = 2 2 T\2I(l) 2π I(l) where cl is the center of I(l), using here that
I(l)

bl (τ ) dτ = 0,

sin 1 (τ − cl ) 2 dτ dt 1 sin 2 (t − τ ) sin 1 (t − cl ) T\2I(l) I(l) 2 |I(l)| ≤ (const.) |bl (τ )| dtdτ. I(l) R\2I(l) |t − τ ||t − cl | = 1 2π bl (τ ) Note that |t − cl | ≤ |t − τ | + |τ − cl | 1 ≤ |t − τ | + |I(l)| 2 ≤ 2|t − τ | Hence
R\2I(l)

when τ ∈ I(l) when t ∈ R \ 2I(l).

|I(l)| dt ≤ 2 |t − τ ||t − cl | =4

R\2I(l) ∞ 2r

|I(l)| dt |t − cl |2 where 2r = |I(l)|

2r dt t2

= 4. Thus the left side of (12.1) ≤ (const.)
l I(l) L1 (T) L1 (T)

|bl (τ )| dτ

= (const.) b ≤ (const.) f

70

CHAPTER 12. HILBERT TRANSFORM ON LP (T)

by the Calder´n–Zygmund Corollary 11.5. o We have proved (12.1), and thus the theorem. Corollary 12.2. The Hilbert transform is strong (p, p) for 1 < p < ∞, with (Hf )(n) = −i sign(n)f (n) for all f ∈ Lp (T), n ∈ Z. Proof. H is strong (2, 2) and linear, by deﬁnition in Chapter 10, and H is weak (1, 1) on L2 (T) (and hence on the simple functions on T) by Theorem 12.1. So H is strong (p, p) for 1 < p < 2 by Remark C.4 after Marcinkiewicz Interpolation (in Appendix C). That is, H : Lp (T) → Lp (T) is bounded and linear for 1 < p < 2. For 2 < p < ∞ we will use duality and anti-selfadjointness H ∗ = −H on 1 1 2 L (T) (see Lemma 10.2) to reduce to the case 1 < p < 2. Write p + p = 1. If f ∈ Lp ∩ L2 (T) then Hf
p

= sup

1 (Hf )g dt : g ∈ Lp (T) with norm 1 2π T 1 = sup{ (Hf )g dt : g ∈ Lp ∩ L2 (T) with norm 1} 2π T by density of Lp ∩ L2 in Lp 1 = sup{ f (Hg) dt : g ∈ Lp ∩ L2 (T) with norm 1} 2π T since H ∗ = −H on L2 (T) ≤ f
Lp (T)

sup{ Hg
Lp (T)

Lp (T)

: g ∈ Lp ∩ L2 (T) with norm 1} by Holder

≤ (const.)p f

by the strong (p , p ) bound proved above, noting 1 < p < 2. Thus H is bounded and linear on the dense subset Lp ∩ L2 (T) of Lp (T). Hence H extends to a bounded operator on Lp (T). Finally, for f ∈ Lp (T), 1 < p < ∞, let fm ∈ Lp ∩ L2 (T) with fm → f in Lp (T). Boundedness of H on Lp implies Hfm → Hf in Lp . Hence fm → f and Hfm → Hf in L1 (T). Thus passing to the limit in (Hfm )(n) = −i sign(n)fm (n) yields (Hf )(n) = −i sign(n)f (n), as desired.

by the Marcinkiewicz interpolation and duality argument in Corollary 12.2. 1 1 + = 1. and hence M is strong (p.3]. Some important applications are: Hilbert transform. p) for 1 < p < ∞ by the Marcinkiewicz Interpolation Theorem C. H : Lp (T) → Lp (T) is bounded.Chapter 13 Applications of interpolation Goal Apply Marcinkiewicz and Riesz–Thorin interpolation to the Hilbert transform. for 1 < p < ∞. in Chapter 6. The Hausdorﬀ–Young theorem says : Lp (T) → p (Z).2. ∞) by Chapter 6. 71 . p p It fails for p > 2 [Katznelson. Fourier analysis. 1) and strong (∞. maximal operator.) Strong (p. 1 ≤ p ≤ 2. already. (Note M is sublinear. M is weak (1. Hardy–Littlewood maximal operator.2. Section IV. p) was proved directly. Fourier analysis and convolution The Marcinkiewicz and Riesz–Thorin interpolation theorems are covered in Appendix C.

72

CHAPTER 13. APPLICATIONS OF INTERPOLATION

To interpret the theorem, note Lp (T) gets smaller as p increases, and so does p (Z). Proof. The analysis operators : L1 (T) → ∞ (Z) and : L2 (T) → 2 (Z) are bounded. Observe 1 1−θ θ = + p 1 2 ⇐⇒ θ 1 =1− 2 p ⇐⇒ 1 1−θ θ = + . p ∞ 2

Now apply the Riesz–Thorin Interpolation Theorem C.6. Convolution. The Generalized Young’s theorem says f ∗g
Lr (Rd )

≤ f

Lp (Rd )

g

Lq (Rd )

when

1 1 1 + = +1, p q r

1 ≤ p, q, r ≤ ∞.

Proof. Fix g ∈ Lq (Rd ) and deﬁne T f = f ∗ g. Then T is strong (1, q) since f ∗g
Lq (Rd )

≤ f

L1 (Rd )

g

Lq (Rd )

by Young’s Theorem A.3, and T is strong (q , ∞) since f ∗g
L∞ (Rd )

≤ f

Lq (Rd )

g

Lq (Rd ) Lq (Rd ) .

by H¨lder’s inequality. In both cases, T ≤ g o 1 1−θ θ = + p 1 q ⇐⇒ θ 1 1 1 =1− = − q p q r

Observe 1 1−θ θ = + . r q ∞

⇐⇒

Now apply the Riesz–Thorin Interpolation Theorem C.6.

Epilogue: Fourier series in higher dimensions
We have studied Fourier series only on the one dimensional torus T = R/2πZ. The theory extends readily to the higher dimensional torus Td = Rd /2πZd . Summability kernels can be obtained by taking products of one dimensional kernels. Thus the higher dimensional Dirichlet kernel is Dn (t) = Dn (t1 ) · · · Dn (td )
n

=
j1 ,...,jd =−n

eijt ,

where j = (j1 , . . . , jd ), t = (t1 , . . . , td )† and † denotes the transpose operation. The Dirichlet kernel corresponds to “cubical” partial sums of multiple Fourier series, because (Dn ∗ f )(t) = =
j1 ,...,jd =−n

1 (2π)d
n

···
T T

Dn (t − τ )f (τ ) dτ1 · · · dτd

f (j)eijt .

“Spherical” partial sums of the form |j|≤n f (j)eijt can be badly behaved. For example, they can fail to converge for f ∈ Lp (Td ) when p = 2. See [Grafakos] for this theorem and more on Fourier series in higher dimensions.

73

74

CHAPTER 13. APPLICATIONS OF INTERPOLATION

Part II Fourier integrals 75 .

.

π). 77 . by using Riemann sums on the ξ-integral.Prologue: Fourier series converge to Fourier integrals Fourier series do not apply to a function g ∈ L1 (R). and extend f to be 2π-periodic. for |x| < ρπ we have ∞ g(x) = f (ρ x) = j=−∞ −1 f (j)eij(ρ ∞ −1 x) 1 = 2π → 1 2π ρπ −ρπ ∞ −∞ g(y)e−i(j/ρ)y dy ei(j/ρ)x · g(y)e−iξy dy eiξx dξ j=−∞ ∞ −∞ 1 ρ as ρ → ∞. The outer integral (“Fourier inverse”) is analogous to a Fourier series. We aim to develop a Fourier integral theory that is analogous to the theory of Fourier series. The inner integral (“Fourier transform”) is analogous to a Fourier coeﬃcient. Instead we take a large piece of g and look at its Fourier series: for ρ > 0. by changing variable. Then f (j) = 1 π g(ρt)e−ijt dt 2π −π ρπ 1 g(y)e−i(j/ρ)y dy = 2πρ −ρπ t ∈ [−π. Formally. since g is not periodic. let f (t) = g(ρt).

78 .

Chapter 14 Fourier transforms: basic properties Goal Derive basic properties of Fourier transforms Reference [Katznelson] Section VI. deﬁne f (ξ) = Fourier transform of f = Rd 1/p f (x)e−iξx dx. 1/(1 + |x|) is in L2 (R) but not L1 (R) Cc (Rd ) = {complex-valued. and so ξx = ξ1 x1 + · · · + ξd xd equals the dot product. continuous functions with compact support} C0 (Rd ) = {complex-valued.1. For f ∈ L1 (Rd ) and ξ ∈ Rd . x is a column vector. Banach space with norm · L∞ (Rd ) Translation fy (x) = f (x − y) Deﬁnition 14. 79 .1 Notation f Lp (Rd ) = Rd |f (x)|p dx Nesting of Lp -spaces fails: L∞ (Rd ) ⊂ L2 (Rd ) ⊂ L1 (Rd ) due to behavior at inﬁnity e.1) Here ξ is a row vector. continuous functions with f (x) → 0 as |x| → ∞}.g. (14.

fy (ξ) = e−iξy f (ξ) takes modulation to translation. g ∈ L1 (Rd ). j=1 fj (xj ) has transform f (ξ) = Proof. If f1 . . fd ∈ L1 (R) then f (x) = d d j=1 fj (ξj ). Proof. and so f is uniformly continuous. . j=1 e . ξ. observe |f (ξ + ω) − f (ξ)| ≤ Rd |f (x)||e−iξx ||e−iωx − 1| dx →0 as ω → 0. with f L∞ (Rd ) ≤ f L1 (Rd ) f is uniformly continuous If fm → f in L1 (Rd ) then fm → f in L∞ (Rd ). Exercise. Suppose A is orthogonal. f is radial if f (Ax) = f (x) for every x and every orthogonal (“rotation and reﬂection”) matrix A. Use Fubini and the homomorphism property of the exponential: e−iξx = d −iξj xj . c ∈ C. d). Recall that f is radial if it depends only on the distance to the origin: f (x) = F (|x|) for some function F . A ∈ GL(R. Corollary 14. Equivalently. [f (x)eiωx ] (ξ) = f (ξ − ω) takes matrix dilation to its inverse. ω ∈ Rd .2 (Basic properties). If f ∈ L1 (Rd ) is radial then f is radial.4 (Transform of a product). y ∈ Rd . For continuity. FOURIER TRANSFORMS: BASIC PROPERTIES Theorem 14. using Theorem 14.2 and that | det A| = 1. . Then f (Ax) = f (x) (since f is radial) and so f (ξA−1 ) = [ | det A|f (Ax) ] (ξ) = f (ξ). The convergence is independent of ξ. Lemma 14. . Linearity (f + g)(ξ) = f (ξ) + g(ξ) and (cf )(ξ) = cf (ξ) Conjugation f (ξ) = f (−ξ) takes translation to modulation. Proof. [ | det A|f (Ax) ] (ξ) = f (ξA−1 ) : L1 (Rd ) → L∞ (Rd ) is bounded.80 CHAPTER 14. Let f. by dominated convergence.3 (Transform of a radial function).

1. since ξ † /|ξ|2 has magnitude 1/|ξ| → 0. f (ξ) → 0 as |ξ| → ∞.1] (x)e−iξx dx = 2 and 1 − cos ξ = 2 sin2 (ξ/2) − x) cos(ξx) dx = 2ξ −2 (1 − cos ξ). Proof. Fix f ∈ Lp (Rd ).1. Rd where ξ † is the column vector transpose of ξ. We compute the Fourier transforms in Table 14.5 implies |f (ξ)| ≤ 1 f − fπξ† /|ξ|2 2 L1 (Rd ) .7 (Riemann–Lebesgue lemma). Example 14. . Lemma 14.1] (x)e−iξx dx = dx = 2 sin(ξ)/ξ 1 (1 0 2. the Gaussian e−|x| /2 . Corollary 14. g (ξ) = R e−x 2 /2 (−ix)e−iξx dx. For ξ = 0. Proof.81 Lemma 14.3. Like Lemma 1. Lemma 14. Diﬀerentiating. Like Lemma 1. 1 ≤ p < ∞. Note √ g(0) = 2π. 2 For d = 1. let g(ξ) = R e−x /2 e−iξx dx be the transform we want.4 except using Cc (Rd ). so that we can use it later for the third example e−|x| . which tends to zero as |ξ| → ∞ by the L1 -continuity of translation in Lemma 14.8. f (ξ) = 1 2 [f (x) − f (x − πξ † /|ξ|2 )] e−iξx dx.5 (Diﬀerence formula). Next we compute for the fourth example. 2 4.6. The map φ : Rd → Lp (Rd ) y → fy is continuous. Thus f ∈ C0 (Rd ).6 (Continuity of translation). R (1 − |x|)1[−1. R 1 e−iξx −1 1[−1. which is dense in Lp (Rd ). Proof.

4. cd = Γ d+1 π (d+1)/2 . FOURIER TRANSFORMS: BASIC PROPERTIES dimension 1 1 d d f (x) 1[−1. we need a calculus lemma that expresses a decaying exponential as a superposition of Gaussians. 3. which simpliﬁes to the desired result.1] (x) f (ξ) 2 sin ξ = 2 sinc ξ ξ sin(ξ/2) ξ/2 2 (1 − |x|)1[−1.1] (x) e−|x| e−|x| 2 /2 = sinc2 (ξ/2) (2π)d cd (1+|ξ|2 )(d+1)/2 (2π)d/2 e−|ξ| 2 /2 Table 14. R e−|x| e−iξx dx = 0 e−(1+iξ)x dx + −∞ e(1−iξ)x dx = 1/(1 + iξ) + 1/(1 − iξ). √ 2 Solving the diﬀerential equation yields g(ξ) = 2πe−ξ /2 .1: Fourier transforms from Example 14. ∞ 0 . note the product structure e−|x| /2 = d e−xj /2 and apply j=1 Lemma 14.82 CHAPTER 14.8. The fourth example says the 2 Fourier transform of a Gaussian is a Gaussian. Hence g (ξ) = i R e−x 2 /2 e−iξx dx e−iξx dx e−iξx dx by parts = −i R e−x 2 /2 = −ξ R e−x 2 /2 = −ξg(ξ). with the diﬀerentiation through the integral justiﬁed by using diﬀerence quotients and dominated convergence (Exercise). 2 2 For d > 1. In the third example. so that c1 = 1/π. To handle d > 1. For d = 1.

e Proof.1 0 = (2π)(d−1)/2 (1 + |ξ|2 )/2 ∞ 0 u(d−1)/2 e−u du where u = a(1+|ξ|2 )/2.9 and x → √ ax ∞ a(d−1)/2 e−a/2 (2π)d/2 e−|ξ −(d+1)/2 a|2 /2 by the Gaussian in Table 14. so that the transform equals (2π)d cd (1 + |ξ|2 )−(d+1)/2 as claimed in the Table. a e−a/2 −b2 /2a √ e da a e−b(c−1/c) e−b(c−1/c) e−b(c−1/c) e−bu 2 /2 2 /2 dc c−2 dc by letting a = bc2 by c → 1/c 2 /2 2 /2 (1 + c−2 ) dc by averaging the last two formulas where u = c − 1/c du Now we compute the Fourier transform of e−|x| as e−|x| e−iξx dx Rd 1 =√ 2π 1 =√ 2π ∞ 0 e−a/2 √ a e−|x| Rd 2 /2 e−i(ξ √ a)x √ dx ad/2 da da by Lemma 14. The last integral is Γ((d+1)/2). b ∞ 0 ∞ 0 ∞ 0 ∞ 0 ∞ −∞ −b 1 =√ 2π ∞ 0 e−a/2 −b2 /2a √ e da.10 (Diﬀerentiation and Fourier transforms). For b > 0. .9.83 Lemma 14. 1 e √ 2π √ 2 b =√ 2π √ 2 b =√ 2π √ b =√ 2π √ b =√ 2π = 1. Smoothness and decay Theorem 14.

(b) If (1 + |x|)f (x) ∈ L1 (Rd ) then f is continuously diﬀerentiable. . d.1 (Rd )) then (∂j f )(ξ) = iξj f (ξ). . Thus: takes multiplication by −ixj to diﬀerentiation. f ∈ W 1. and |f (ξ)| ≤ f L1 (Rd ) = O(1). which is continuous by Theorem 14. where ∂j = ∂/∂xj for j = 1.11 (Smoothness of f and decay of f ). For (b) we compute a diﬀerence quotient. with (−ixj f )(ξ) = (∂j f )(ξ). . For (a) (∂j f )(x)e−iξx dx = Rd Rd f (x)(iξj )e−iξx dx by parts = iξj f (ξ).84 CHAPTER 14. Thus: takes diﬀerentiation to multiplication by iξj . .2. Theorem 14. with δ ∈ R and ej = unit vector in the j-th direction: f (ξ + δej ) − f (ξ) = δ → Rd f (x)e−iξx Rd e−iδxj − 1 dx δ f (x)e−iξx (−ixj ) dx = (−ixj f )(ξ) as δ → 0. d. Hence f (ξ) has partial derivative (−ixj f )(ξ). by dominated convergence with dominating function f (x)|x| ∈ L1 (Rd ). 1 (b) If f ∈ Cc (Rd ) then f (ξ) = o(1/|ξ|) as |ξ| → ∞. . (a) If f ∈ L1 (Rd ) then f (ξ) = o(1) as |ξ| → ∞. . . Proof. and |f (ξ)| ≤ d maxj ∂j f |ξ| L1 (Rd ) = O(1/|ξ|). . where ∂j = ∂/∂ξj for j = 1. FOURIER TRANSFORMS: BASIC PROPERTIES 1 (a) If f ∈ Cc (Rd ) (or more generally. .

12.7) and Theorem 14. Given f.13 (Convolution and Fourier transforms). ξ ∈ Rd . deﬁne their convolution (f ∗ g)(x) = Rd f (x − y)g(y) dy. x ∈ Rd .2 Convolution Deﬁnition 14.85 Proof. (b) For each ξ there exists j such that |ξj | ≥ |ξ|/d (since |ξ1 | + · · · + |ξd | ≥ |ξ|). by Riemann–Lebesgue by Theorem 14. If f.2 Or one could argue more directly using the gradient vector: |f (ξ)| = |( f )(ξ)| = o(1/|ξ|) |iξ| f L1 (Rd ) ≤ |ξ| = O(1/|ξ|). (a) Use Riemann–Lebesgue (Corollary 14. ≤ by Riemann–Lebesgue by Theorem 14.2. Then |f (ξ)| = |(∂j f )(ξ)| (∂j f )(ξ) ≤ iξj |ξ|/d d maxj |(∂j f )(ξ)| |ξ| = o(1/|ξ|) d maxj ∂j f L1 (Rd ) ≤ |ξ| = O(1/|ξ|). g ∈ L1 (Rd ) then f ∗ g ∈ L1 (Rd ) with f ∗g and (f ∗ g)(ξ) = f (ξ)g(ξ). Theorem 14. g ∈ L1 (Rd ). L1 (Rd ) ≤ f L1 (Rd ) g L1 (Rd ) .

3. 1 ≤ p ≤ ∞. convolution is a smoothing operation. Example 14. For the second. FOURIER TRANSFORMS: BASIC PROPERTIES Thus the Fourier transform takes convolution to multiplication. We ﬁnd f (ξ) = sinc(ξ/2) like example 1 of Table 14.1/2] . Like Theorem 1. If f ∈ Lp (Rd ). if f ∈ C0 (Rd ) and g ∈ L1 (Rd ) then f ∗ g ∈ C0 (Rd ).1] (x) by direct calculation. Proof. It is also associative.86 CHAPTER 14. Proof. then f ∗ g ∈ Lp (Rd ) with f ∗g Lp (Rd ) ≤ f Lp (Rd ) g L1 (Rd ) . then Q(ξ)f (ξ)eiξx dξ. and hence improves the decay of the transform: sinc(ξ/2) decays like 1/ξ while sinc2 (ξ/2) decays like 1/ξ 2 . (P ∗ f )(x) = Rd (14.13 predicts.1. If f ∈ L1 (Rd ) and P (x) = Rd Q(ξ)eiξx dξ for some Q ∈ L1 (Rd ). 1 ≤ p ≤ ∞.11. and (f ∗ f )(ξ) = sinc2 (ξ/2) by example 2 of Table 14. 2. so that (f ∗ f )(x) = (1 − |x|)1[−1. 3. Proof. Use linearity and Fact 2. Convolution is commutative: f ∗ g = g ∗ f . Rd (P ∗ f )(x) = Proof. 4.14.2) Q(ξ) Rd eiξ(x−y) f (y) dydξ by Fubini = Rd Q(ξ)eiξx f (ξ) dξ. if f ∈ C0 (Rd ) and g ∈ L1 (Rd ) then f ∗ g is continuous because (f ∗ g)(x + z) → (f ∗ g)(x) as z → 0 by uniform continuity of f (exercise). Further. . then fm ∗ g → f ∗ g in Lp (Rd ). Convolution is continuous on Lp (Rd ): if fm → f in Lp (Rd ). Hence (f ∗ f ) = (f )2 . Let f = 1[−1/2. use Young’s Theorem A. As this example illustrates. as Theorem 14. For the ﬁrst claim. Convolution facts (similar to Chapter 2) 1.1. And (f ∗ g)(x) → 0 as |x| → ∞ by dominated convergence. and linear with respect to f and g. since f (x − y) → 0 as |x| → ∞. and g ∈ L1 (Rd ). and g ∈ L1 (Rd ).

(SR1) (SR2) (SR3) sup ω Rd ω→∞ |kω (x)| dx < ∞ |kω (x)| dx = 0 lim {x:|x|>δ} Some kernels further satisfy ω→∞ |x|>δ lim sup |kω (x)| = 0 (L∞ concentration) for each δ > 0. A summability kernel on Rd is a family {kω } of integrable functions such that kω (x) dx = 1 Rd (Normalization) (L1 bound) (L1 concentration) for each δ > 0. (SR4) (Notation.1.1 Deﬁnition 15.) 87 .Chapter 15 Fourier integrals: summability in norm Goal Develop summability kernels in Lp (Rd ) Reference [Katznelson] Section VI. Here kω (x) does not mean the translation k(x − ω).

2.3.1] (ξ)eiξx dξ 2π R 1 sin x = = sinc x.2) The Dirichlet kernel is Dω (x) = ωD(ωx) = (15.1) (15. Example 15. Show (SR1) and (SR2) by changing variable with y = ωx. let D(x) = 1 1[−1. πx π ω 1 eiξx dξ 2π −ω sin(ωx) = . For (SR3). πx (15.4) . dy = ω d dx.1: Dirichlet kernel with ω = 10 2 Example 15. Suppose k ∈ L1 (Rd ) is continuous with Put kω (x) = ω d k(ωx) Rd k(x) dx = 1. For d = 1. Proof. for ω > 0. |kω (x)| dx = {x:|x|>δ} {y:|y|>ωδ} |k(y)| dy →0 as ω → ∞. by dominated convergence. FOURIER INTEGRALS: SUMMABILITY IN NORM 4 2 Figure 15.88 CHAPTER 15. Then {kω } is a summability kernel.3) (15.

let F (x) = 1 2π (1 − |ξ|)1[−1. Example 15. ∴ {Dω } is not a summability kernel. And . For d = 1.1] (ξ)eiξx dξ R (15.5) by Table 14. D is not integrable since |D(x)| ∼ |x|−1 at inﬁnity.1. (15.6) 1 = 2π sin 1 x 2 1 x 2 2 The Fej´r kernel is e 1 Fω (x) = ωF (ωx) = 2π ω = 2π ω −ω (1 − |ξ|/ω)eiξx dξ 1 ωx 2 1 ωx 2 2 (15. See Figure 15. with associated j=1 d kernel Dω (x) = j=1 Dω (xj ).4.8) sin .89 4 2 Figure 15.2: Fej´r kernel with ω = 10 e 2 See Figure 15.1. F is integrable since F (x) ∼ x−2 at inﬁnity.2. In higher dimensions the Dirichlet function is d D(xj ).7) (15.

13) (15. (15.90 CHAPTER 15.3). alRd ternatively. one can integrate (15. Example 15. ∴ {Pω } is a summability kernel. And P (x) dx = P (0) = 1 because P (ξ) = e−|ξ| by Example 16.6. d j=1 F (xj ). P (x) = = 1 (2π)d (1 + e−|ξ| eiξx dξ by Table 14.5.3 below. by integrating (15. e 1 F (x) = 0 Dω (x) dω in 1 dimension.1.12) = cd ω −1 |x|2 + ω −2 (d+1)/2 . for example.10) directly (see [Stein and Weiss. Example 15. FOURIER INTEGRALS: SUMMABILITY IN NORM F (x) dx = R 2 lim π ρ→∞ 2 = lim π ρ→∞ 1 = lim π ρ→∞ = 1. (15.9) (15. G(x) = 1 (2π)d e−|ξ| Rd 2 /2 2 /2 eiξx dξ by Table 14. ρ −ρ ρ −ρ ρ −ρ sin2 (x/2) dx x2 2 sin(x/2) cos(x/2) · (1/2) dx x sin x dx x by parts ∴ {Fω } is a summability kernel.11) (15. with associated The Fej´r kernel is an arithmetic mean of Dirichlet kernels. P is integrable since P (x) ∼ |x|−(d+1) at inﬁnity.1. p. See Figure 15.3. 9] for d > 1).14) = (2π)−d/2 e−|x| . In higher dimensions the Fej´r function is e d kernel Fω (x) = j=1 Fω (xj ).10) Rd cd |x|2 )(d+1)/2 The Poisson kernel is Pω (x) = ω d P (ωx) = 1 (2π)d e−|ξ|/ω eiξx dξ Rd (15.

14). G is clearly integrable.3: Poisson kernel with ω = 10 2 The Gauss kernel is Gω (x) = ω d G(ωx) = 1 2 e−|ξ/ω| /2 eiξx dξ d (2π) Rd ωd 2 e−|ωx| /2 .16) See Figure 15. Connection to Fourier integrals For f ∈ L1 (Rd ): (Dω ∗ f )(x) = 1 (2π)d f (ξ)eiξx dξ [−ω. (15.19) (15.ω]d d G(x) dx = 1 from (15. = (2π)d/2 Rd (15.20) 1 (Fω ∗ f )(x) = (2π)d 1 (Pω ∗ f )(x) = (2π)d 1 (Gω ∗ f )(x) = (2π)d (1 − |ξj |/ω) f (ξ)eiξx dξ [−ω.4. and ∴ {Gω } is a summability kernel.91 4 2 Figure 15.18) (15.ω]d j=1 e−|ξ|/ω f (ξ)eiξx dξ Rd e−|ξ/ω| Rd 2 /2 f (ξ)eiξx dξ .17) (15.15) (15.

Caution.5).6. Use Convolution Fact (14. Recall that C0 (Rd ) uses the L∞ norm. but the right side does not: so far we have deﬁned the Fourier transform only for f ∈ L1 (Rd ).2) and deﬁnitions (15.4: Gauss kernel with ω = 10 2 Proof. respectively. Argue as for Theorem 2. (15. Use that if f ∈ C0 (Rd ) then f is uniformly continuous. 1 ≤ p < ∞. Assume {kω } is a summability kernel. (b) If f ∈ C0 (Rd ) then kω ∗ f → f in C0 (Rd ) as ω → ∞. Proof.13).7 (Summability in Lp (Rd ) and C0 (Rd )). FOURIER INTEGRALS: SUMMABILITY IN NORM 4 2 Figure 15. then kω ∗ f → f in Lp (Rd ) as ω → ∞. Summability in norm Theorem 15.1). (15. The left sides of the above formulas make sense for f ∈ Lp (Rd ). (15.92 CHAPTER 15. (a) If f ∈ Lp (Rd ). .9).

• Uniqueness theorem: if f.7 and formulas (15.ω]d j=1 in L1 (Rd ). .7. 1. ∞). v is the harmonic extension of f from Rd to the halfspace Rd × (0.18)–(15. xd+1 ) = (P1/xd+1 ∗ f )(x) = cd Rd (15.19) and diﬀerentiate through the integral.93 Consequences • Fej´r summability for f ∈ L1 (Rd ): e 1 (2π)d d (1 − |ξj |/ω) f (ξ)eiξx dξ → f (x) [−ω. with boundary value v(x. e Connection to PDEs Fix f ∈ L1 (Rd ). Proof. Take ω = 1/xd+1 in (15. 0) = f (x) in the sense of Theorem 15. the Fourier transform : L1 (Rd ) → L∞ (Rd ) is injective. Use Fej´r summability (15. (15. Proof.20). That is. That is.21) on f and g. Proof. using d+1 j=1 ∂ 2 −|ξ|xd+1 iξx (e e ) = (iξ1 )2 + · · · + (iξd )2 + (−|ξ|)2 e−|ξ|xd+1 eiξx ∂x2 j = 0.22) xd+1 |x − y|2 + x2 d+1 (d+1)/2 f (y) dy solves 2 2 2 (∂1 + · · · + ∂d + ∂d+1 )v = 0 on Rd × (0. g ∈ L1 (Rd ) with f = g then f = g.21) Similarly for Poisson and Gauss summability. The Poisson kernel solves Laplace’s equation in a half-space: v(x. Use Theorem 15. ∞).

Take ω = 1/ 2t in (15. x) = (G1/√2t ∗ f )(x) 1 2 e−|x−y| /4t f (y) dy = d/2 (4πt) Rd solves wt = ∆w for (t. note ω = 1/ 2t → ∞ as t → 0. FOURIER INTEGRALS: SUMMABILITY IN NORM For the boundary value.) √ Proof.7. (Here ∆ = ∂1 + · · · + ∂d . x) ∈ (0. ∞) × Rd .94 CHAPTER 15. The Gauss kernel solves the diﬀusion (heat) equation: w(t. 2 2 √ For the boundary value. 2. . x) = f (x) in the sense of 2 2 Theorem 15. with initial value w(0. note ω = 1/xd+1 → ∞ as xd+1 → 0. using ∂ − ∂t d j=1 ∂2 ∂x2 j (e−|ξ| t eiξx ) = − |ξ|2 − (iξ1 )2 − · · · − (iξd )2 e−|ξ| t eiξx = 0.20) and diﬀerentiate through the integral.

1] Deﬁnition 16. (Fourier inversion) (a) If f. f ∈ L1 (Rd ) then f is continuous and f (x) = 1 (2π)d f (ξ)eiξx dξ. (2π)d We call ˇ the inverse Fourier transform. g ∈ L1 (Rd ) then g is continuous and ˇ g(ξ) = Rd g (x)e−iξx dx. (b) If g.1.2. ˇ 95 ξ ∈ Rd .Chapter 16 Fourier transforms in L1(Rd). Theorem 16. Rd x ∈ Rd . and Fourier inversion Goal Fourier inversion when f is integrable Reference [Katznelson. . in view of the next theorem. Section VI. Deﬁne g (x) = ˇ 1 g(ξ)eiξx dξ (2π)d Rd 1 = g(−x).

because deﬁnitions (15. FOURIER INVERSION WHEN F ∈ L1 (RD ) dimension d d d f (x) F (x) = P (x) = 1 (2π)d d j=1 sin(xj /2) xj /2 2 f (ξ) F (ξ) = 1[−1.21) implies pointwise e convergence a.96 CHAPTER 16. so that G = g by Theorem 16. (15. e from Example 16. ˇ Table 16.1]d (ξ) P (ξ) = e−|ξ| 2 /2 d j=1 (1 − |ξj |) cd (1+|x|2 )(d+1)/2 G(x) = (2π)−d/2 e−|x| G(ξ) = e−|ξ| 2 /2 Table 16. (a) The L1 convergence in Fej´r summability (15.13) says G(x) = g (x).13) express those kernels as inverse Fourier 2 transforms.9) and (15. if we choose g(ξ) = e−|ξ| /2 then deﬁnition (15. The Fourier transforms of the Fej´r. for some subsequence of ω-values: f (x) = lim = 1 ω→∞ (2π)d 1 (2π)d Rd d Rd 1[−ω. ˆ The theorem says (f )ˇ = f and (ˇ)ˆ = g . g Proof.ω]d (ξ) j=1 (1 − |ξj |/ω) f (ξ)eiξx dξ f (ξ)eiξx dξ by dominated convergence. and then swap x and ξ. change ξ → −ξ.1: Fourier transforms of the Fej´r. using that f ∈ L1 (Rd ). . For example. Poisson and Gauss e functions can be computed by Fourier Inversion Theorem 16.1 displays the results.3.3.2(b). Example 16. (b) Apply part (a) to g.2(b).e.5). Poisson and Gauss functions.

Then by density of L1 ∩ L2 in L2 . g ˆ (f )ˇ = f. 97 .3 Notation Inner product on L2 (Rd ) is f.Chapter 17 Fourier transforms in L2(Rd) Goal Extend the Fourier transform to an isometric bijection of L2 (Rd ) to itself Reference [Katznelson] Section VI. The proof will show : L1 ∩ L2 (Rd ) → L2 (Rd ) is bounded with respect to the L2 norm. : Theorem 17. g = Rd f (x)g(x) dx. for all f. g = (2π)−d f . The Fourier transform L2 (Rd ) → L2 (Rd ) is a bijective isometry (up to a constant factor) with f L2 (Rd ) = (2π)−d/2 f (ˇ)ˆ = g g L2 (Rd ) (Plancherel) (Parseval) (Inversion) f.1 (Fourier transform on L2 (Rd )). g ∈ L2 (Rd ). we conclude the Fourier transform extends to a bounded operator from L2 to itself.

Parseval follows from Plancherel by polarization. with (f )ˇ = f by Inversion Theorem 16.2. and surjective by Inversion. If f is smooth with compact support then f is bounded and decays rapidly at inﬁnity.1] )ˇ by deﬁnition in (15. the Fourier transform extends to a bounded operator from L2 (Rd ) to itself. For Inversion. Thus the Fourier transform is an isometry. g (and using dominated instead of monotone convergence).1] by Theorem 17.7 = lim 1 2 by (15.1). using f ∈ L∞ (Rd ).1] (ξ). the Dirichlet function D(x) = belongs to L2 (R) and has D(ξ) = 1[−1.1 Inversion. and so D = 1[−1. Similarly (ˇ)ˆ = g for all g ∈ L2 (Rd ). = lim ω→∞ (2π)d Rd 1 = |f (ξ)|2 dξ by monotone convergence (2π)d Rd 1 f 2 2 (Rd ) . by repeated use of Theoˆ rem 14. g Finally. sin x πx . since the Fourier transform is bounded. note ˇ : L2 (Rd ) → L2 (Rd ) is bounded by Deﬁnition 16. For f ∈ L1 ∩ L2 (Rd ). So the Fourier transform followed by the inverse transform gives the identity on the dense set L1 ∩L2 (Rd ).1) = L (2π)d By density of L1 ∩ L2 in L2 . D = (1[−1. the Fourier transform is injective by Plancherel. Example 17.1. f 2 L2 (Rd ) = lim ω→∞ f (x)(Gω ∗ f )(x) dx Rd since Gω ∗ f → f in L2 (Rd ) by Theorem 15. (17. FOURIER TRANSFORMS IN L2 (RD ) Proof.20) f (x)e−iξx f (ξ)e−|ξ/ω| /2 dξdx d ω→∞ (2π) Rd Rd 1 2 |f (ξ)|2 e−|ξ/ω| /2 dξ by Fubini. or by repeating the argument for Plancherel with f.2. Hence f ∈ L1 (Rd ). and hence on all of L2 (Rd ) by continuity. f changed to f. up to a constant factor. Plancherel follows from (17.11. Proof.1) by density. In 1 dimension.98 CHAPTER 17.

= lim n→∞ f (x)e−iξx dx. The ﬁrst ﬁve Basic Properties in Theorem 14.3. Note the Fourier transform is well deﬁned on L1 + L2 (Rd ). and in Theorem 17.3 (radial functions) and Lemma 14. Apply the Riesz–Thorin Interpolation Theorem C. Proof.4 (Hausdorﬀ–Young for Fourier transform). 1 ≤ p ≤ 2.2. Hence f (ξ) = lim (f 1B(n) )(ξ) n→∞ in L2 (Rd ).1.17)–(15.20) (connection to Fourier integrals). The Fourier transform : Lp (Rd ) → Lp (Rd ) is bounded for 1 ≤ p ≤ 2. Remark 17.1: Fourier transform of the Dirichlet function. and so do Corollary 14. .5. If f ∈ L2 (Rd ) then f 1B(n) ∈ L1 ∩ L2 (Rd ) and f 1B(n) → f in L2 (Rd ).2 still hold for the Fourier transform on Lp (Rd ). Remark 17.2.4 (product functions) and (15. B(n) How can this limit exist. as above.99 dimension f (x) d D(x) = 1 πd sin xj d j=1 xj f (ξ) D(ξ) = 1[−1. since the L1 and L2 Fourier transforms agree on L1 ∩ L2 (Rd ). Theorem 17. when f need not be integrable? The answer must be that oscillations of e−iξx yield cancelations that allow f (x)e−iξx to be integrated improperly. using boundedness of : L1 (Rd ) → L∞ (Rd ) : L2 (Rd ) → L2 (Rd ) in Theorem 14.6.1. for almost every ξ.1]d (ξ) Table 17. where 1 p + 1 p = 1. from Example 17. by Theorem 17.

20) all apply to fm .3. and (Fω ∗ f ) = Fω f has compact support (because Fω has compact support by Table 16. Let m → ∞ and use the Hausdorﬀ–Young Theorem 17.7.4 and (15. Given f ∈ Lp (Rd ). Now let m → ∞ in those results.1). take fm ∈ L1 ∩ Lp (Rd ) with fm → f in Lp (Rd ). Then (f ∗ gm ) = f gm by Theorem 17. Fω ∗ f → f in Lp (Rd ) by Theorem 15. Proof. g ∈ Lp (Rd ). so that Theorem 14.4. If f ∈ L1 (Rd ).6 (Convolution and Fourier transforms).4. Corollary 14. FOURIER TRANSFORMS IN L2 (RD ) Proof. Lemma 14. 1 ≤ p ≤ 2. Then fm → f in Lp (Rd ) by the Hausdorﬀ–Young Theorem 17. Corollary 17. Take gm ∈ L1 ∩ Lp (Rd ) with gm → g in Lp (Rd ).2. then f ∗ g ∈ Lp (Rd ) and (f ∗ g) = f g. Proof.100 CHAPTER 17.17)–(15. noting f is bounded. 1 ≤ p ≤ 2. .1. Consequence Analogue of Weierstrass trigonometric approximation: functions with compactly supported Fourier transform are dense in Lp (Rd ). Here fm is the usual Fourier transform of fm ∈ L1 (Rd ).

Chapter 18 Fourier integrals: summability pointwise Goal Prove suﬃcient conditions for summability at a single point. Adapt the corresponding result on the torus. by using maximal functions like we did for Fourier series in Chapter 7.3b If f ∈ C0 (Rd ) then kω ∗ f → f uniformly by Theorem 15. and so does the Fej´r kernel e d in 1 dimension. Proof.1b.7(b).1 (Summability at a point). and a. Theorem 3. Assume {kω } is a summability kernel. Reference [Grafakos] Sections 2. and hence convergence holds at every x.. 3. Suppose either f ∈ L1 (Rd ) and {kω } satisﬁes the L∞ concentration hypothesis (SR4). More generally. The Poisson and Gauss kernels satisfy (SR4).1(a).e.e. Next we aim at summability a. if k(x) = o(1/|x| ) as |x| → ∞ then kω (x) = ω d k(ωx) satisﬁes (SR4) (Exercise). or else f ∈ L∞ (Rd ). 101 . But what if f is merely continuous at a point? Theorem 18. If f is continuous at x0 ∈ Rd then (kω ∗ f )(x0 ) → f (x0 ) as ω → ∞.

If k ∈ L1 (Rd ) is nonnegative and radially symmetric decreasing. L∗ f ≤ L∗ |f | ≤ M f where M is the Hardy–Littlewood maximal operator from Chapter 6. f ∈ L1 (Rd ).4 (Majorization).3. L1/ω (y) = (1/ω)d L(y/ω) = Hence |(L1/ω ∗ f )(x)| ≤ 1 |B(ω)| |f (x − y)| dy B(ω) 1 1B(ω) (y). Lemma 18. . Deﬁnition 18. FOURIER INTEGRALS: SUMMABILITY A. First.E. Deﬁne the Dirichlet maximal function (D∗ f )(x) = sup |(Dω ∗ f )(x)| ω ω ω Fej´r maximal function (F f )(x) = sup |(Fω ∗ f )(x)| e Poisson maximal function (P f )(x) = sup |(Pω ∗ f )(x)| Gauss maximal function (G f )(x) = sup |(Gω ∗ f )(x)| ω ω ∗ ∗ ∗ Lebesgue maximal function (L∗ f )(x) = sup |(Lω ∗ f )(x)| where L(x) = 1 1B(1) (x) |B(1)| is the normalized indicator function of the unit ball. Proof. then |(k ∗ f )(x)| ≤ k L1 (Rd ) (L ∗ f )(x) for all x ∈ Rd .1) ≤ (M f )(x). Lemma 18.102 CHAPTER 18. |B(ω)| (18.2.

like we did on the torus in Lemma 7. 0 if ω ≤ |y|.2: ∞ k(y) = ρ(|y|) = − |y| ∞ ρ (ω) dω since ρ(∞) = 0 by integrability of k =− 0 |B(ω)|L1/ω (y)ρ (ω) dω. Write k(x) = ρ(|x|) where ρ : [0.103 Proof. ∞) → R is nonnegative and decreasing. Assume ρ is absolutely continuous. (when d = 1) . L1/ω (y) = Hence (k ∗ f )(x) = 0 ∞ 1/|B(ω)| if ω > |y|. 1 ≤ p ≤ ∞. for simplicity. |B(ω)|(L1/ω ∗ f )(x) − ρ (ω) dω and so ∞ |(k ∗ f )(x)| ≤ 0 ∞ |B(ω)| − ρ (ω) dω · (L∗ f )(x) ω 0 since ρ ≤ 0 = 0 |∂B(1)|rd−1 dr − ρ (ω) dω · (L∗ f )(x) by spherical coordinates for |B(ω)| = B(ω) dy ∞ = 0 |∂B(1)|ω d−1 ρ(ω) dω · (L∗ f )(x) by parts with respect to ω (why does the ω = ∞ term vanish?) = Rd k(y) dy · (L∗ f )(x) by using spherical coordinates again. e 4 ∗ L |f | π P ∗ f ≤ L∗ f G∗ f ≤ L∗ f F ∗f ≤ for all f ∈ Lp (Rd ).1). because by (18.5 (Lebesgue dominates Poisson and Gauss in all dimensions. We ﬁrst establish a layer-cake decomposition of k. and Fej´r in 1 dimension). Theorem 18.

1 ≤ p ≤ ∞. Corollary 18.3 with the weak and strong bounds on the Hardy–Littlewood maximal operator in Chapter 6. Hence |Fω ∗ f | ≤ k ∗ |f | ≤ (4/π)L∗ |f | by Majorization Lemma 18.). (The last statement is the Lebesgue diﬀerentiation theorem. which is not integrable at inﬁnity in dimensions d ≥ 2. For example. 1/ |x| ≤ 2/ω. FOURIER INTEGRALS: SUMMABILITY A.5 and Lemma 18. Similarly G∗ f ≤ L∗ f . P ∗ f ≤ L∗ f by the Majorization Lemma 18. for 1 < p ≤ ∞. 2 Note k is nonnegative.6. The Fej´r kernel is not majorized by a radially symmetric e decreasing integrable function. → ∞. |x| > 2/ω.4. since Pω is nonnegative and radially symmetric decreasing. If f ∈ Lp (Rd ).e.104 CHAPTER 18. G∗ and L∗ are weak (1. which decays like x−2 along the x1 -axis. Proof.3. → ∞. Remark 18.e. F ∗ . a. see [Grafakos. Theorem 3.7. Combine Theorem 18.) .e. ω Fω (x) = 2π sin ω x 2 2 ω x 2 def by (15. → ∞. p) on Lp (Rd ). ≤ k(x) = ω 2π 2 ω x . e Theorem 18. taking ω = 2 gives 1 F2 (x) = π j=1 d sin xj xj 2 .e. a. Proof. with Pω L1 (Rd ) = 1.3].8) 1. then Fω ∗ f Pω ∗ f Gω ∗ f Lω ∗ f →f →f →f →f a. a. with k L1 (R) = 4/π. 1) and strong (p. Thus the best possible radial bound 1 would be O(|x|−2 ).4.E. as as as as ω ω ω ω → ∞. even and decreasing.e. P ∗ . For the Fej´r kernel in dimensions d ≥ 2. When d = 1.8 (Summability a. when d ≥ 2.

Assume 1 ≤ p < ∞. Hence Fω ∗ g → f a.} ω→∞ is closed in Lp (Rd ). .1. with h = 0 there. on B(m).7. p) by Corollary 18. Letting m → ∞ proves the result. it must equal Lp (Rd ). by the part of the theorem already proved.e. Obviously C contains every f ∈ Cc (Rd ).105 Proof..e. Thus C is dense in Lp (Rd ) (using here that p < ∞). and so Fω ∗ g → g a. put g = 1B(m) f and h = f − g. because Fω ∗f → f uniformly by Theorem 15. F ∗ is weak (p. Argue similarly for the other kernels. on B(m). Since f = g + h we deduce Fω ∗ f → f a. When p = ∞. Then g ∈ L1 (Rd ). and so Fω ∗ h → h = 0 on B(m) by Theorem 18.e. For m ∈ N. Because C is closed. Next h ∈ L∞ (Rd ) is continuous on B(m). which proves the result. consider f ∈ L∞ (Rd ).7. Hence the Theorem in Chapter 6 says C = {f ∈ Lp (Rd ) : lim Fω ∗ f = f a.e.

. FOURIER INTEGRALS: SUMMABILITY A.E.106 CHAPTER 18.

Sω is the “partial sum” operator for the Fourier integral.5. Suggestions are welcome! Deﬁnition 19. because if f ∈ ˆ L (Rd ). then Sω (f ) = (1[−ω. In particular.Chapter 19 Fourier integrals: norm convergence Goal Show norm convergence for Lp (Rd ) follows from boundedness of the Hilbert transform on R Reference I do not know a fully satisfactory reference for this material.17) and Remark 17. 1 ≤ p ≤ 2. Sω : L2 (Rd ) → L2 (Rd ) p 107 .ω]d f )ˇ by (15. Write Sω (f ) = Dω ∗ f where Dω (x) = j=1 d d ωD(ωxj ) = sin(ωxj ) πxj j=1 d is the Dirichlet kernel on R and D(z) = (sin z)/πz is the Dirichlet function in 1 dimension.1.

7) to Lp convergence (Dω ∗f = Sω (f ) → f in Theorem 19.7.2. Let 1 < p < ∞ and suppose supω Sω Lp (Rd )→Lp (Rd ) < ∞. Because C is closed by Proposition 9. We conclude A ⊂ {f ∈ Lp (Rd ) : lim Sω (f ) = f in Lp (Rd )} = C. Proof. First we reduce norm convergence to norm boundedness. we have the result already for p = 2. thus Sω (g) = Dω g = 1[−ω. by Table 17. ∞]. by the Generalized Young’s Theorem in Chapter 13.ω]d f → f and so Sω (f ) → f in L2 (Rd ). As remarked above. note Sω (g) ∈ L2 (Rd ) because Dω ∈ L2 (Rd ) and g ∈ L1 (Rd ). Sω (f ) is well deﬁned whenever f ∈ Lp (Rd ). We will prove below that Sω (f ) ∈ Lp (Rd ) when f ∈ Lp (Rd ). provided ω is large enough that [−ω. Theorem 19. Let A = {g ∈ L1 ∩ Lp (Rd ) : g has compact support}. Indeed. we see A is dense in Lp (Rd ). we conclude C = Lp (Rd ). because Dω ∈ q L (Rd ) for each q > 1 and so Dω ∗ f ∈ Lr (Rd ) for each r ∈ (p. To see this fact.108 CHAPTER 19. Thus Fω ∗ f ∈ A.2 (using the assumption that supω Sω Lp (Rd )→Lp (Rd ) < ∞). . and L1 ∩ Lp is dense in Lp .ω]d g = g. ω→∞ def so that C is dense in Lp (Rd ). Further. We further show Sω (g) = g. ω]d contains the support of g. 1 ≤ p < ∞.1 Applying Fourier inversion in L2 gives Sω (g) = g. FOURIER INTEGRALS: NORM CONVERGENCE is bounded. Our goal in this Chapter is to improve the Lp summability for Fourier integrals (Fω ∗f → f in Theorem 15.4 below). by boundedness of the Fourier transform and its inverse on L2 . if f ∈ L1 ∩ Lp (Rd ) then Fω ∗ f ∈ L1 ∩ Lp (Rd ) and (Fω ∗ f ) = Fω f has compact support by Table 16.1. But Sω (f ) need not belong to L1 (Rd ) when f ∈ L1 (Rd ) (Exercise). We claim A is dense in Lp (Rd ). which proves the theorem. 1[−ω. 1 < p < ∞. Since Fω ∗ f → f in Lp (Rd ) by Theorem 15. Then Fourier integrals converge in Lp (Rd ): limω→∞ Sω (f ) − f Lp (Rd ) = 0 for each f ∈ Lp (Rd ). when g ∈ A. as ω → ∞.

. proof omitted. (Thus our “square partial sums” for convergence of Fourier integrals can be relaxed to “rectangular partial sums”. ωd ) of positive numbers.1) |g(y)|p dy. Cp.ω1 ]×···×[−ωd .109 Next we reduce to norm boundedness in 1 dimension.ωd ] is the indicator function of a rectangular box. . deﬁne d Dω (x) = j=1 ωj D(ωj xj ). We have not yet shown that this constant is ﬁnite. Proof.d ≤ (Cp. The Fourier multiplier Dω = 1[−ω1 .d = sup Sω ω Lp (Rd )→Lp (Rd ) for the norm bound on the partial sum operators.) Given a vector ω = (ω1 . For the sake of generality we allow diﬀerent ω-values in each coordinate direction. . Write Cp.1 (19.1 p Lp (R) by deﬁnition of Cp. . Theorem 19.1 g p = Cp. p p Lp (R) ωD(ω(x − y))g(y) dy dx = Dω ∗ g R R p ≤ Cp. R . First observe that for g ∈ Lp (R) and ω > 0.3 (Reduction to 1 dimension).1 )d .

Boundedness in Lp (R) 1. y2 ) dy2 dy1 dx2 R R R by (19. and not “spherical” sums.1) with g(y1 ) = R 2p ≤ Cp.1 ω2 D(ω2 (x2 − y2 ))f (y1 . y2 ) dy2 dy1 dx1 dx2 p p ω2 D(ω2 (x2 − y2 ))f (y1 . when p = 2.1) with g(y2 ) = f (y1 . (Thus H is a Fourier multiplier operator. Aside. 2 Taking p-th roots gives Sω Lp (R2 )→Lp (R2 ) ≤ Cp.1 . called the Hilbert transform on R. FOURIER INTEGRALS: NORM CONVERGENCE Hence for f ∈ Lp (R2 ) and ω = (ω1 . Then the Riesz projection P : Lp (R) → Lp (R) deﬁned by 1 P f = (f + iHf ) 2 . We shall prove (in Chapters 20 and 21) the existence of a bounded linear operator H : Lp (R) → Lp (R). with the property that (Hf )(ξ) = −i sign(ξ)f (ξ) when f ∈ Lp ∩ L2 (R). x2 )|p dx1 dx2 R2 = R p ≤ Cp. y2 )|p dy2 dy1 R R by (19.) 2.1]. |(Dω ∗ f )(x1 . Section 10. The “ball” multiplier 1B(1) (ξ) does not yield a partial sum operator with uniform norm bounds. ω2 ). 1 < p < ∞. y2 ) dy2 |f (y1 .1 R R ω1 D(ω1 (x1 − y1 )) R ω2 D(ω2 (x2 − y2 ))f (y1 . see [Grafakos. y2 ). Therefore Fourier integrals and series in higher dimensions should be evaluated with “rectangular” partial sums.110 CHAPTER 19. Argue similarly for d ≥ 3. which proves the theorem when d = 2. when working in Lp for p = 2.

ω→∞ It remains to prove Lp boundedness of the Hilbert transform on R.∞) (ξ)f (ξ) Subtracting the last two formulas gives 1(−ω. when 1 < p < ∞. Fourier inversion now completes the proof.1 = sup Sω ω Lp (R)→Lp (R) f ∈ L2 (R). f ∈ L2 (R). e−iωx P (eiωx f ) − eiωx P (e−iωx f ) = Sω (f ). which equals Sω (f ). (19.∞) (ω + ξ)f (ξ) = 1(−ω.3 we conclude: Theorem 19. From (19.111 is also bounded. The following formula expresses the Fourier partial sum operator in terms of the Riesz projection and some modulations: for ω > 0. Observe P projects onto the positive frequencies: (P f )(ξ) = 1(0.∞) (ξ)f (ξ − ω) [e−iωx P (eiωx f )] (ξ) = 1(0. Then lim Sω (f ) − f Lp (Rd ) = 0 for each f ∈ Lp (Rd ).4 (Fourier integrals converge in Lp (Rd )).2) ≤2 P Lp (R)→Lp (R) <∞ when 1 < p < ∞.ω] f . Proof.2 and 19. since i(−i sign(ξ)) = sign(ξ). 3. Hence from Theorems 19.∞) (ξ)f (ξ).2) applied to the dense class of f ∈ Lp ∩ L2 (R).∞) (ξ)f (ξ) [eiωx P (e−iωx f )] (ξ) = 1(ω. . Let 1 < p < ∞. it follows that Cp. [eiωx f ] (ξ) = f (ξ − ω) [P (eiωx f )] (ξ) = 1(0. and from boundedness of the Riesz projection. 4.

112 CHAPTER 19. FOURIER INTEGRALS: NORM CONVERGENCE .

.3 [Grafakos] Section 4. the Riesz transform equals the Hilbert transform on R. .1 Deﬁnition 20. In dimension d = 1. 113 .1. . d.Chapter 20 Hilbert and Riesz transforms on L2(Rd) Goal Develop spatial and frequency representations of Hilbert and Riesz transforms Reference [Duoandikoetxea] Section 4. . The Riesz transforms on Rd are Rj : L2 (Rd ) → L2 (Rd ) f → (−i(ξj /|ξ|)f )ˇ for j = 1. deﬁned by H : L2 (R) → L2 (R) f → (−i sign(ξ)f )ˇ because sign(ξ) = ξ/|ξ|.

[This proof is similar to Proposition 10. 1 2π (−i) sign(ξ)eiξy dξ = [−ω.] For ω > 0. d since j=1 (−iξj /|ξ|)2 = −1. R f (x − y) 1 dy πy (20.3 on T. HILBERT AND RIESZ TRANSFORMS ON L2 (RD ) Rj is bounded since the Fourier multiplier −iξj /|ξ| is a bounded function (in fact.1) for almost every x in the interval. πy ω 0 eiξy dξ (20. since the convolution kernel 1/πy is not integrable.2 (Spatial representation of Hilbert transform). The right side of (20.114 CHAPTER 20.1) is a singular integral.ω] 0 i i eiξy dξ − 2π −ω 2π 1 − cos(ωy) = .2) . 1 πx Later we will justify these formulas in terms of distributions.v. Proposition 20. The proposition says formally that Hf = f ∗ or p. bounded by 1). If f ∈ L2 (R) is C 1 -smooth on an interval then (Hf )(x) = p. Clearly Rj L2 (Rd )→L2 (Rd ) d 2 Rj j=1 ∗ Rj ≤1 = −I = −Rj by Plancherel. and so was skimmed only lightly in class. by Parseval.v. 1 πx = −i sign(ξ). Proof.

since (1/πy) dy = 0.2) πy R 1 − cos(ωy) dy = [f (x − y) − f (x)] πy |y|<1 1 − cos(ωy) + f (x − y) dy πy |y|>1 = f (x − y) by oddness of 1 − cos(ωy) /πy. 1[−ω. for some subsequence of ω-values. Formula (20. by the Riemann–Lebesgue Corollary 14.4).ω] Hf )ˇ(x) dξ = = R 1 2π (−i) sign(ξ)f (ξ)eiξx dξ [−ω. πy (20. ε<|y|<1 Finally.3) as ω → ∞. so that (1[−ω. d.ω] by Fubini 1 − cos(ωy) dy by y → x − y and (20.3) and (20.3 (Spatial representation of Riesz transform). . Meanwhile. .v.ω] f (y) 1 2π (−i) sign(ξ)eiξ(x−y) dξdy [−ω.e. .1) in full generality by approximating f oﬀ a neighborhood of x using functions in L1 ∩L2 . If f ∈ L2 (Rd ) is C 1 -smooth on an open set U ⊂ Rd then (Rj f )(x) = p. (Obviously f belongs to L1 ∩L2 already on each neighborhood of x. for each j = 1. Convergence holds a.) Proposition 20. The ﬁrst integral similarly converges to [f (x − y) − f (x)] |y|<1 1 dy. .1) therefore follows from (20.4) assuming f is C 1 -smooth on a neighborhood of x (which ensures integrability of y → [f (x − y) − f (x)]/πy on |y| < 1). Rd f (x − y) cd yj dy |y|d+1 for almost every x ∈ U .115 If f ∈ L1 ∩ L2 (R) then (1[−ω. The second integral converges to f (x − y) |y|>1 1 dy πy (20. .7.ω] Hf converges to Hf in L2 (R) as ω → ∞.ω] Hf )ˇ converges to Hf . one deduces (20.

and hence pointwise a. |ξ| Proof.e. (Obviously f belongs to L1 ∩ L2 already on each neighborhood of x.6) in the proof of Lemma 20. Finally.6) In class we proceeded formally. observe 1 = |ξ| ∞ 0 e−|ξ|z dz (20. Our proof will use a truncated version of this identity: e−|ξ|δ − e−|ξ|/δ = |ξ| 1/δ δ e−|ξ|z dz. (20. for some subsequence of δ values. one deduces the theorem for f ∈ L2 (Rd ) by approximating f oﬀ a neighborhood of x using functions in L1 ∩ L2 .116 CHAPTER 20.4 below. To motivate the following proof. Applying L2 Fourier inversion yields (Rj f )(x) = lim −iξj δ→0 e−|ξ|δ − e−|ξ|/δ ˆ f ˇ(x) |ξ| in L2 (Rd ).5) instead of (20. The proposition says formally that Rj f = f ∗ or p. (Rj f )(ξ) = −i ξj f (ξ) |ξ| e−|ξ|δ − e−|ξ|/δ = lim(−iξj ) f (ξ) δ→0 |ξ| with convergence in L2 (Rd ) (by dominated convergence). cd yj |y|d+1 cd yj |y|d+1 = −i ξj . Thus the theorem is proved when f ∈ L1 ∩ L2 (Rd ).v. HILBERT AND RIESZ TRANSFORMS ON L2 (RD ) Here cd = Γ (d + 1)/2 /π (d+1)/2 > 0.) . For example. For f ∈ L2 (Rd ).4 below. by Lemma 20. c1 = 1/π. skipping the rest of this proof and using (20.5) and that e−|ξ|z is the Fourier transform of the Poisson kernel P1/z .

and the exponentials e−|ξ|δ and e−|ξ|/δ are square integrable.4. Further.2 (and the deﬁnition of f for f ∈ L1 (Rd )). ξj /|ξ| is bounded by 1. Thus their product is integrable. we express it using . so that by the L1 Fourier Inversion Theorem 16. First. then e−|ξ|δ − e−|ξ|/δ ˆ f ˇ(x) δ→0 |ξ| cd yj = [f (x − y) − f (x)] d+1 dy + |y| |y|<1 lim −iξj f (x − y) |y|>1 cd yj dy |y|d+1 (20. To evaluate the inner integral. If f ∈ L1 ∩ L2 (Rd ) is C 1 -smooth on an open set U ⊂ Rd . and so is f .7) equals ε→0 lim f (x − y) ε<|y|<1 cd yj dy. the ﬁrst integral in (20. |y|d+1 Proof. e−|ξ|δ − e−|ξ|/δ ˆ f ˇ(x) |ξ| 1 e−|ξ|δ − e−|ξ|/δ =− iξj f (y)e−iξy dy eiξx dξ d (2π) Rd |ξ| Rd −|ξ|δ e − e−|ξ|/δ iξy 1 iξj e dξdy =− f (x − y) (2π)d Rd |ξ| Rd −iξj after changing y → x − y.117 Lemma 20.7) for almost every x ∈ U .

we ﬁnd e−|ξ|δ − e−|ξ|/δ ˆ f ˇ(x) −iξj |ξ| =− Rd f (x − y) cd yj (|y|2 + z 2 )(d+1)/2 z=1/δ dy z=δ z=1/δ =− |y|<1 [f (x − y) − f (x)] cd yj (|y|2 + z 2 )(d+1)/2 z=1/δ dy z=δ (20. HILBERT AND RIESZ TRANSFORMS ON L2 (RD ) Poisson kernels: e−|ξ|δ − e−|ξ|/δ iξy 1 iξj e dξ (2π)d Rd |ξ| ∂ 1 e−|ξ|δ − e−|ξ|/δ iξy = e dξ ∂yj (2π)d Rd |ξ| 1/δ = δ 1/δ 1 ∂ ∂yj (2π)d e−|ξ|z eiξy dξdz Rd by identity (20.9) cd yj − f (x − y) 2 + z 2 )(d+1)/2 (|y| |y|>1 dy z=δ where we used the oddness of yj to insert f (x) in (20.8) converges to [f (x − y) − f (x)] |y|<1 cd yj dy |y|d+1 by dominated convergence (noting the C 1 -smoothness ensures the integrand is O(|y|) · O(1/|y|d ) = O(1/|y|d−1 ) near the origin. Now ﬁx a point x ∈ U . expression (20.12) (why?!) = δ 1/δ ∂ P1/z (y) dz ∂yj cd z ∂ 1 dz 2 + z 2 )(d+1)/2 ∂yj (|y| ∂ 1 dz 2 + z 2 )(d+1)/2 ∂z (|y| z=1/δ = δ 1/δ = δ cd yj cd yj = 2 + z 2 )(d+1)/2 (|y| . And . z=δ By substituting this expression into the above.6) by (15. which is integrable).8) (20.118 CHAPTER 20.8). As δ → 0.11) by (15.

∂xj j = 1. xd+1 ) ∂xd+1 v(x) = = normal derivative of u at the boundary. ∞) with boundary value u = f when xd+1 = 0 (see Chapter 15). Put ∂ u(x. ε<|y|<1 and use Connections to PDEs 1. x ∈ Rd . Formal Proof. using dominated convergence. (Exercise: explain why the terms with z = 1/δ in (20. let u(x. write |y|<1 = limε→0 the oddness of yj to remove the term with f (x).9) converges to cd yj dy |y|d+1 f (x − y) |y|>1 by dominated convergence (noting f ∈ L2 (Rd ) and yj /|y|d+1 = O(1/|y|d ) is square integrable for |y| > 1). . d.9) vanish as δ → 0. so that u is harmonic on the upper halfspace Rd × (0. xd+1 =0 Then Rj v = ∂f . . .119 as δ → 0. Given a function f . . xd+1 ) = (P1/xd+1 ∗ f )(x). xd+1 > 0.) For the ﬁnal claim in the lemma. The Riesz transforms map the normal derivative of a harmonic function to its tangential derivatives. expression (20. .8) and (20.

Hence (Rj Rk ∆f ) (ξ) = (−iξj ) (−iξk ) (−|ξ|2 )f (ξ) |ξ| |ξ| = −(iξj )(iξk )f (ξ) ∂ 2f =− (ξ) ∂xj ∂xk so that Rj Rk ∆f = − ∂2f . ∂ 2f ∂x2 j 2 (ξ) = (iξj )2 f (ξ) = −ξj f (ξ) and so summing over j gives (∆f ) (ξ) = −|ξ|2 f (ξ). HILBERT AND RIESZ TRANSFORMS ON L2 (RD ) because (Rj v)(ξ) = −i ξj v(ξ) |ξ| ξj ∂ u(ξ. ∂xj ∂xk That is. ∂xj Thus we have shown the jth Riesz transform maps the normal derivative of u to its jth tangential derivative. . xd+1 ) = −i |ξ| ∂xd+1 = −i = −i xd+1 =0 ξj ∂ e−|ξ|xd+1 f (ξ) |ξ| ∂xd+1 ξj (−|ξ|)f (ξ) |ξ| xd+1 =0 = iξj f (ξ) ∂f = (ξ).120 CHAPTER 20. Mixed Riesz transforms map the Laplacian to mixed partial derivatives. mixed Riesz transforms map the Laplacian to mixed partial derivatives. Formal Proof. 2. on the boundary.

121 The above formal derivation is rigorous if. with ∂ 2f ∂xj ∂xk ≤ ∆f L2 (Rd ) L2 (Rd ) since each Riesz transform has norm 1 on L2 (Rd ). . Similar estimates hold on Lp (Rd ). 1 < p < ∞. by the Lp boundedness of the Riesz transform proved in the next chapter. the norm of a mixed second derivative is controlled by the norms of the pure second derivatives in the Laplacian. Consequently. f is C 2 -smooth with compact support. for example.

122 CHAPTER 20. HILBERT AND RIESZ TRANSFORMS ON L2 (RD ) .

Note o 1 ∞ d g ∈ L ∩ L (R ) and so g ∈ L2 (Rd ).1.1 (weak (1. we want to show √ Rd \2 dQ(l) A f ω L1 (Rd ) |(Rj bl )(x)| dx ≤ (const. Apply the Calder´n–Zygmund Theorem 11. j = 1. hence Rj g ∈ L2 (Rd ) by Chapter 20.Chapter 21 Hilbert and Riesz transforms on Lp(Rd) Goal Prove weak (1. Now proceed like in the proof of Theorem 12. p) by interpolation and duality Reference [Duoandikoetxea] Section 5. To ﬁnish the proof. 1) on L1 ∩ L2 (Rd )).) f 123 L1 (Rd ) . . (21. and deduce strong (p.4 to get f = g + b.1 Theorem 21. just changing T to Rd and the interval I(l) to the cube Q(l). . And b = f − g ∈ L2 (Rd ) so that Rj b ∈ L2 (Rd ). There exists A > 0 such that |{x ∈ Rd : |(Rj f )(x)| > ω}| ≤ for all ω > 0.1) l . . d and f ∈ L1 ∩ L2 (Rd ). 1) for Riesz transform. . Proof.

2) ≤ (const.2 below. here we used that b (y) dy = 0.1) follows by summing (21.) Q(l) |bl (y)| dy by Lemma 21.4. Hence Q(l) l √ Rd \2 dQ(l) |Rj bl (x)| dx √ Rd \2 dQ(l) ≤ Q(l) |bl (y)| |ρj (x − y) − ρj (x − c(l))| dxdy (21.2) over l and recalling that b 2 f L1 (Rd ) by the Calder´n–Zygmund Theorem 11. / bl (y) ρj (x − y) − ρj (x − c(l)) dy dx Q(l) cd (xj − yj ) dy dx |x − y|d+1 = √ Rd \2 dQ(l) where ρj (x) = cd xj |x|d+1 is the jth Riesz kernel and c(l) is the center of Q(l). we have √ Rd \2 dQ(l) |Rj bl (x)| dx bl (y) Q(l) = √ Rd \2 dQ(l) √ noting x − y is bounded away from 0.3 applied on the open set U = Rd \2 dQ(l) (where bl = 0). HILBERT AND RIESZ TRANSFORMS ON LP (RD ) √ By Proposition 20.) |( ρj )(x)| ≤ |x|d+1 √ and if x ∈ Rd \ 2 dQ(l) and y ∈ Q(l) then |x − c(l)| ≥ √ 1 side 2 dQ(l) 2 ≥ 2|y − c(l)|. L1 (Rd ) Now (21. since y ∈ Q(l) and x ∈ 2 dQ(l).124 CHAPTER 21. o ≤ . the hypotheses of that lemma are satisﬁed here because (const.

z∈Rd {x:|x−z|≥2|y−z|} (const. by deﬁnition in Chapter 20.)|y| d+1 2|y| r = (const. Hence |ρ(x − y) − ρ(x)| dx {x:|x|≥2|y|} 1 ≤ |y| 0 |x|≥2|y| |( ρ)(x − sy)| dxds 1 dx (|x|/2)d+1 ≤ (const. That is. |ρ(x − y) − ρ(x − z)| dx < ∞. p) estimates. Proof.1. If ρ ∈ C 1 (Rd \ {0}) with o |( ρ)(x)| ≤ then sup y. Rj : Lp (Rd ) → Lp (Rd ) is bounded and linear for 1 < p < 2. 1 ρ(x − y) − ρ(x) = 0 ∂ ρ(x − sy) ds ∂s 1 =− 0 y · ( ρ)(x − sy) ds. by a translation. since |x − sy| ≥ |x| − |y| ≥ |x|/2. 2) and linear. So Rj is strong (p.) Now we deduce strong (p. Corollary 21. ∞ 1 d−1 r dr = (const. and Rj is weak (1. p) for 1 < p < ∞. We can take z = 0.3. p) for 1 < p < 2 by Remark C. |x|d+1 x ∈ Rd . Proof.2 (H¨rmander condition).)|y| |x|≥2|y| by using the hypothesis.4 after Marcinkiewicz Interpolation (in Appendix C). By the Fundamental Theorem.) . Rj is strong (2. The Riesz transforms are strong (p. 1) on L1 ∩ L2 (Rd ) (and hence on all simple functions with support of ﬁnite measure) by Theorem 21.125 Lemma 21. .

1]. HILBERT AND RIESZ TRANSFORMS ON LP (RD ) ∗ For 2 < p < ∞ we use duality and anti-selfadjointness Rj = −Rj on L2 (Rd ) to reduce to the case 1 < p < 2. p) bound on the Riesz transform can be generalized to a whole class of convolution-type singular integral operators [Duoandikoetxea. for singular integral kernels of the form O(x/|x|) |x|d where O is an odd function on the unit sphere. Alternatively. . The Riesz kernel cd (xj /|x|)/|x|d ﬁts this form.2c]. The idea is to express convolution with this kernel as an average of Hilbert transforms taken in all possible directions in Rd . The strong (p.2. Section 5. Section 4.126 CHAPTER 21. since O(y) = yj is odd. just like in the proof of Corollary 12. one can instead use the method of rotations [Grafakos.

Part III Fourier series and integrals 127 .

Chapter 22 Compactly supported Fourier transforms, and the sampling theorem
Goal Show band limited functions are holomorphic Prove the Kotelnikov–Shannon–Whittaker sampling theorem Reference [Katznelson] Section VI.7

ˇ Deﬁnition 22.1. We say f = g is band limited if g ∈ L1 (Rd ) has compact support. Theorem 22.2 (Band limited functions are holomorphic). Assume g ∈ L1 (Rd ) is supported in a ball B(R), and deﬁne f (z) = g (z) = ˇ 1 (2π)d g(ξ)eiξz dξ
B(R)

for z = x + iy ∈ Cd , x, y ∈ Rd . (Here ξz = ξ1 z1 + · · · + ξd zd .) Then f is holomorphic, and |f (z)| = O(eR|y| ). If in addition g ∈ L2 (Rd ) then |f (z)| = O(eR|y| / |y|). 129

130

CHAPTER 22. BAND LIMITED FUNCTIONS

Thus once more, decay of the Fourier transform (here, compact support) implies smoothness of the function (here, holomorphicity). The theorem also bounds the rate of growth of the function in the complex directions. (The function must vanish at inﬁnity in the real directions, by the Riemann– Lebesgue corollary, since g is integrable.) For example, the Dirichlet kernel D(x) = sin(x)/πx = (1[−1,1] )ˇ(x) is band limited with R = 1, in 1 dimension. Taking z = 0 + iy, we calculate ey − e−y = O(e|y| /|y|), 2πy

D(iy) =

which is better (by a factor of

|y|) than is guaranteed by the theorem.

Proof. f is well deﬁned because ξ → eiξz is bounded on B(R), for each z. And f is holomorphic because eiξz is holomorphic and f can be diﬀerentiated through the integral with respect to the complex variable z. (Exercise. Justify these claims in detail.) Clearly 1 |g(ξ)|e−ξy dξ d (2π) B(R) 1 g L1 (Rd ) eR|y| . ≤ (2π)d

|f (z)| ≤

since eiξz = eiξx e−ξy

If in addition g ∈ L2 (Rd ), then 1 g (2π)d

|f (z)| ≤

L2 (Rd ) B(R)

e−2ξy dξ

1/2

The sampling rate ω/π is proportional to the bandwidth ω. 2 |y| Hence |f (z)| ≤ (const.)eR|y| / |y|. Remark 22. Assume f ∈ L2 (Rd ) is band limited. It vanishes at all the other sampling locations (π/ω)mj . with f supported in the cube [−ω.4. Holomorphic functions are known to be determined by their values on lower dimensional sets in Cd . Then d π f (x) = f n sinc(ωxj − πnj ) ω j=1 d n∈Z with the series converging in L2 (Rd ). sinc(ωxj − πnj ) is centered at the sampling location (π/ω)nj and rescaled to have bandwidth ω. Theorem 22. and also uniformly (in L∞ (Rd )). 1. to the highest frequency contained in the signal f . that “sampling set” can be a lattice in Rd .3 (Sampling theorem for band limited functions). For a band limited function. because many samples are needed to determine a highly oscillatory function. .R]d ≤ = (2R)d−1 e2R|y| − e−2R|y| 2|y| d−1 2R|y| (2R) e ≤ . ω]d for some ω > 0. that is. the sampling rate must be high when the frequencies are high.131 and e−2ξy dξ = B(R) B(R) e−2ξ|y|e1 dξ by ξ → ξA for some orthogonal matrix A with Ay = |y|e1 = B(R) e−2ξ1 |y| dξ e−2ξ1 |y| dξ [−R. Intuitively. 2. since sinc ω(π/ω)mj − πnj = sinc π(mj − nj ) = 0.

ξ ∈ [−π.3.1. f is continuous (after redeﬁning it on some set of measure zero) with f (x) = 1 (2π)d f (ξ)eiξx dξ. π]d . Proof of Sampling Theorem. 3.2) . Hence by L1 Fourier inversion. We will prove f (ξ) = n∈Zd f (−n)eiξn .132 CHAPTER 22. by replacing x with (π/ω)x (Exercise).1) Thus the pointwise sampled values f (π/ω)n in the theorem are well deﬁned. and so is integrable. The ﬁgure shows f with a solid curve. Next. BAND LIMITED FUNCTIONS 3 2 1 2 3 2 1 1 2 1 1 2 1 3 2 2 Figure 22. with ω = 2π and sampling rate ω/π = 2. A graphical example of the sampling formula is shown in Figure 22. We can assume ω = π. Rd x ∈ Rd . The dashed curves are two of the sinc functions making up the signal. (22.4. for ω = 2π and f (x) = − sinc 2π(x + 1) + 2 sinc 2π(x + 1/2) + 3 sinc 2πx + 2 sinc 2π(x − 1/2) + 1 sinc 2π(x − 1) .1: Example of sampling formula in Theorem 22. (22. f is square integrable and compactly supported. and 3 sinc 2πx and 2 sinc 2π(x − 1/2) with dashed curves. See Remark 22.

Hence P W (ω) is a Hilbert space with the L2 inner product. . ω]d ). · L2 .3.π]d (ξ). Indeed. Applying L2 inversion gives f (x) = n∈Zd f (n) e−iξn 1[−π. ω]d . It is isometric. ξ ∈ Rd . if we regard f as a square integrable function on the cube Td = [−π. consider the Paley– Wiener space P W (ω) = {f ∈ L2 (Rd ) : f is supported in [−ω.π]d ˇ(x) f (n) n∈Zd = sin(π(xj − nj )) π(xj − nj ) j=1 d with convergence in L2 (Rd ). and it is a closed subspace (since if f = limm fm in L2 (Rd ) and fm is supported in [−ω. ω]d }. π]d ). After changing n → −n in (22.133 with convergence in L2 ([−π. Thus (22. under the Fourier transform.π]d 1 f (ξ)e−iξn dξ = (2π)d Rd = f (−n) since f is supported in [−π. ω]d ) with inner product (2π)−d ·. Paley–Wiener space For a deeper perspective on Sampling Theorem 22.2). we have f (ξ) = n∈Zd f (n)e−iξn 1[−π.ω]d (ξ)e−iξ(π/ω)n n∈Zd . with convergence in L2 (Rd ) and in L1 (Rd ). Applying L1 inversion gives convergence in L∞ . π]d by the inversion formula (22. then f = limm fm is also supported in [−ω. π]d .1). then its Fourier coeﬃcients are 1 f (ξ)e−iξn dξ (2π)d [−π. That space has orthonormal Fourier basis (π/ω)d/2 1[−ω.2) simply expresses the Fourier series of f on the cube. Clearly P W (ω) is a subspace of L2 (Rd ). to L2 ([−ω.

134 CHAPTER 22. Our calculations have.3) where the coeﬃcient is f. (22. of course. (π/ω)d/2 1[−ω. gn L2 = 1 f .3. we expand f= n∈Zd f. Thus the orthonormal expansion (22. gn L2 gn . Using this orthonormal basis.3) simply restates the Sampling Theorem 22. Taking the inverse Fourier transform gives an orthonormal basis of sinc functions for the Paley–Wiener space: d {gn }n∈Zd = (ω/π) d/2 j=1 sinc(ωxj − πnj ) n∈Zd . . essentially repeated the proof of the Sampling Theorem.ω]d e−iξ(π/ω)n (2π)d = (π/ω)d/2 f (π/ω)n L2 by Parseval by Fourier inversion. for all f ∈ P W (ω). BAND LIMITED FUNCTIONS where the indicator function simply reminds us that we are working on the cube.

1 Deﬁnition 23. Lemma 23. x ∈ Rd .3.2. its periodization is the function Pe(f )(x) = (2π)d n∈Zd f (x + 2πn).Chapter 23 Periodization and Poisson summation Goal Periodize functions on Rd to functions on Td Show the Fourier series of periodization gives the Poisson summation formula References [Folland] Section 8. then Pe(f ) = 2π(21[−π.2π) . if f = 1[−π. In 1 dimension. Pe : L1 (Rd ) → L1 (Td ) is bounded. 135 .3 [Katznelson] Section VI. π). and Pe(f ) is 2πZd -periodic.0) + 1[0.π) ) for x ∈ [−π. Example 23. with Pe(f ) extending 2π-periodically to R. Given f ∈ L1 (Rd ). with Pe(f ) L1 (Td ) ≤ f L1 (Rd ) . If f ∈ L1 (Rd ) then the series for Pe(f )(x) converges absolutely for almost every x. Further.1.

If f.) . below. g ∈ L1 (Rd ) then Pe(f ∗ g) = Pe(f ) ∗ Pe(g). Proof. Proof. suppose f and g are bounded with compact support.4 (Periodization of a convolution). j ∈ Zd . PERIODIZATION AND POISSON SUMMATION The periodization has Fourier coeﬃcients Pe(f )(j) = f (j).3 again and so Pe(f ∗g) = Pe(f )∗Pe(g) by the uniqueness theorem for Fourier series. Lemma 23. That is. We have Pe(f ∗ g) (j) = (f ∗ g) (j) = f (j) g(j) = Pe(f )(j) Pe(g)(j) = Pe(f ) ∗ Pe(g) (j) by Lemma 23. so that the sums in the following argument are all ﬁnite rather than inﬁnite. the jth Fourier coeﬃcient of Pe(f ) equals the Fourier transform of f at j.3 by Lemma 23. See Problem 19 in Assignment 3. For a more direct proof. (Thus sums and integrals can be interchanged.136 CHAPTER 23.

1) (1 + |x|)d+ε C . Suppose f ∈ L1 (Rd ) is continuous and decays in space and frequency according to: C |f (x)| ≤ . pass to the general case by a limiting argument. Finally. Theorem 23.2) |f (ξ)| ≤ (1 + |ξ|)d+ε for some constants C. remembering that our deﬁnition of convolution on Td has a prefactor of (2π)−d . ε > 0. using that if fm → f in L1 (Rd ) then Pe(fm ) → Pe(f ) in L1 (Td ) by Lemma 23.5 (Poisson summation formula). taking x = 0 gives (2π)d n∈Zd f (2πn) = j∈Zd f (j).3. ξ ∈ Rd . In particular.137 For each x ∈ Rd . (23. x ∈ Rd . (23. Then the periodization Pe(f ) equals its Fourier series at every point: (2π)d n∈Zd f (x + 2πn) = j∈Zd f (j)eijx . Pe(f ∗ g)(x) = (2π)d n∈Zd (f ∗ g)(x + 2πn) f (x + 2πn − y)g(y) dy n∈Zd Rd = (2π)d = Rd Pe(f )(x − y)g(y) dy by deﬁnition of Pe(f ) (Td + 2πm) m = m∈Zd Td Pe(f )(x − y − 2πm)g(y + 2πm) dy since Rd = Pe(f )(x − y)g(y + 2πm) dy m∈Zd Td = = using periodicity of Pe(f ) 1 Pe(f )(x − y) Pe(g)(y) dy (2π)d Td = Pe(f ) ∗ Pe(g) (x). x ∈ Rd . .

138 CHAPTER 23. That continuous function has the same Fourier coeﬃcients as Pe(f ). The Poisson kernel Pr on T equals the periodization of the Poisson kernel Pω on R: 1 − r2 1 ω −1 = 2π . Proof.3) provided r = e−1/ω .) dξ (1 + |ξ|)d+ε ≤ j∈Zd ≤ Rd <∞ by spherical coordinates. Example 23.4. (23. exercise). Hence we obtain a series expansion for the square of the cosecant: 1 π2 = .3 by (23. and so Pe(f ) is continuous. (just like in 1 dimension. and so it equals Pe(f ) a. . Pe(f ) has Fourier coeﬃcients in Pe(f )(j) = j∈Zd j∈Zd 1 (Zd ). x ∈ R \ Z. First. so that it is plausible Pω periodizes to Pr for some r. Hence the Fourier series of Pe(f ) converges absolutely and uniformly to a continuous function.2) |f (j)| C (1 + |j|)d+ε (const. see Chapter 4). Notice that Pe(f )(x) = (2π)d n∈Zd f (x + 2πn) is a series of continuous functions.1). 2 sin πx n∈Z (x + n)2 Proof. PERIODIZATION AND POISSON SUMMATION This Poisson summation formula relates a lattice sum of values of the function to a lattice sum of values of its Fourier transform. The series converges absolutely and uniformly on each ball in Rd (by using (23.e. 1 − 2r cos x + r2 π (x + 2πn)2 + ω −2 n∈Z x ∈ R. To complete the proof we will show Pe(f ) is continuous. to partially motivate these results we note Pe(Pω ∗f ) = Pe(Pω )∗ Pe(f ) by Lemma 23. since by Lemma 23.6 (Periodizing the Poisson kernel). for then Pe(f ) equals its Fourier series everywhere (and not just almost everywhere).

Changing x to 2πx in (23.139 To prove (23. by (15. Hence the Poisson Summation Formula says that Pe(Pω )(x) = j∈Z Pω (j)eijx e−|j|/ω eijx j∈Z = = j∈Z r|j| eijx since r = e−1/ω = Pr (x) by (2. Example 23.1) and (23.4) . Pω (x) = π x2 + ω −2 Pω (ξ) = e−|ξ|/ω . (x + n)2 + (2πω)−2 1 − 2r cos(2πx) + r2 1 1 +O 2 .12) and Table 16. which proves (23.7 (Periodizing the Gauss kernel).3). (23. (x + n)2 2 − 2 cos(2πx) (sin πx)2 where we used monotone convergence on the left side.3). x ∈ R. observe that Pω satisﬁes decay hypotheses (23. ω ω n∈Z Since r = e−1/ω = 1 − letting ω → ∞ implies that n∈Z 1 4π 2 π2 = = .2) because 1 ω −1 .3) gives 1 1 − r2 = 2π 2 ω . 2π x ∈ R.1.8). The Gauss kernel Gs (t) = −j 2 s ijt e on T equals the periodization of the Gauss kernel Gω on R: j∈Z e e−j s eijx = 2π j∈Z n∈Z 2 ω 2 2 √ e−ω (x+2πn) /2 . ξ ∈ R.

Decay hypotheses (23.16) and Table 16.4). x ∈ R. by (15. 2 πs In terms of the theta function ϑ(s) = presses the functional equation n∈Z . Hence the Poisson Summation Formula says that Pe(Gω )(x) = j∈Z Gω (j)eijx e−(j/ω) j∈Z 2 2 /2 = = j∈Z eijx √ since ω = 1/ 2s e−j s eijx = Gs (x). ξ ∈ R.2) hold for Gω because ω 2 Gω (x) = √ e−(ωx) /2 . e−n s > 0. 2π −(ξ/ω)2 /2 Gω (ξ) = e .140 CHAPTER 23. Hence e−n n∈Z 2 πs = s−1/2 n∈Z e−n 2 π/s . which proves (23. PERIODIZATION AND POISSON SUMMATION √ provided s > 0 and ω = 1/ 2s.1. Proof. Taking x = 0 in (23.1) and (23. .4) and changing s to πs yields the functional equation for the theta function. the last formula ex- ϑ(s) = s−1/2 ϑ(s−1 ).

f vanishes on some open set. Proof. Consequently. Proposition 24. f has inﬁnitely many zeros in T.1 (Qualitative uncertainty principles). and f is ﬁnitely supported. 141 . Thus part (a) says: a trigonometric polynomial that vanishes inﬁnitely often in T must vanish identically. then f ≡ 0. (a) f is a trigonometric polynomial since it has only ﬁnitely many nonzero Fourier coeﬃcients. (b) If f ∈ L2 (Rd ) is continuous. and so we are “uncertain” of the value of f . if f is well localized then f is not. and f is compactly supported. then f ≡ 0.Chapter 24 Uncertainty principles Goal Establish qualitative and quantitative uncertainty principles References [Goh and Micchelli] Section 2 [Jaming] Section 1 Uncertainty principles say that f and f cannot both be too localized. (a) If f ∈ L2 (T) is continuous.

Then f (t) = p(eit )/eiN t p(z) = n=0 an−N z n . If f ∈ L2 (Rd ) is continuous and f and f are supported on sets of ﬁnite measure. (b) f is band limited.1. That Taylor series equals f on Rd . Let A = {x ∈ R : f (x) = 0} and B = {ξ ∈ R : f (ξ) = 0}. In particular. Therefore the complementary set E = {x ∈ T : f (x + 2πn) = 0 for all n ∈ Z} has positive measure. Choose x0 ∈ Rd such that f ≡ 0 on a neighborhood of x0 . The Fundamental Theorem of Algebra implies p ≡ 0. Proof. here the support of f need not be compact. then the Taylor series of f centered at x0 is identically zero. then f ≡ 0. Theorem 24.2. We prove only the 1 dimensional case. we see p has inﬁnitely many zeros eit on the unit circle. In contrast to Proposition 24. and hence is holomorphic on Cd by Theorem 22.2 (Benedicks’ qualitative uncertainty principle). and so f ≡ 0. UNCERTAINTY PRINCIPLES N n=−N To prove this claim. . f is real analytic on Rd . write f (t) = where p is the polynomial 2N an eint . Since f has inﬁnitely many zeros t ∈ T. z ∈ C.142 CHAPTER 24. By dilating f we can suppose |A| < 2π. Then {x ∈ T : f (x + 2πn) = 0 for some n ∈ Z} = {x ∈ T : n∈Z 1A (x + 2πn) ≥ 1} ≤ T n∈Z 1A (x + 2πn) dx 1A (x) dx R = = |A| < |T| = 2π.

Fix ξ ∈ F and consider the periodization Pe(f e−iξx )(x) = 2π n∈Z f (x + 2πn)e−iξ(x+2πn) .3 (Nazarov’s quantitative uncertainty principle). But Pe(f e−iξx )(x) = 0 for all x ∈ E. Hence when ξ ∈ F . |[0.1) j∈Z R 1B (ξ) dξ = |B| < ∞. Q vanishes at inﬁnitely many points in T (using here that E has positive measure). The Fourier coeﬃcient f (ξ + j) of Q therefore vanishes for all j.143 Next. . on E. the set {j ∈ Z : f (ξ + j) = 0} is ﬁnite. we deduce f (ξ) = 0 a. so that j∈Z 1B (ξ + j) is ﬁnite for almost every ξ ∈ [0. We omit the proof. 1) where F has full measure.1(a). The jth Fourier coeﬃcient of the periodization equals (f e−iξx ) (j) = f (ξ + j). Nazarov’s theorem implies Benedicks’ theorem. say for all ξ ∈ F ⊂ [0. 1B (ξ + j) dξ = [0. In particular. then the right side is zero and so f ≡ 0. which equals zero for but ﬁnitely many j. Theorem 24. Since f (ξ + j) = 0 for all j ∈ Z and almost every ξ ∈ [0. and so Q vanishes a. because if f is supported in A and f is supported in B. B ⊂ Rd of ﬁnite measure and all f ∈ L2 (Rd ). 1).e.e. which is well deﬁned since f ∈ L1 (R). 1).e. since ξ ∈ F . Thus Pe(f e−iξx ) equals some trigonometric polynomial Q(x) a. and so f ≡ 0. 1) \ F | = 0.. A constant Cd > 0 exists such that f 2 L2 (Rd ) = f 2 L2 (Rd ) ≤ Cd |A||B|+1 Rd \A |f (x)|2 dx + Rd \B |f (ξ)|2 dξ for all sets A. Hence Q ≡ 0 by Proposition 24.

β ∈ C.144 CHAPTER 24. [T. f ≤ U f − βf T ∗ f − αf + T f − αf U ∗ f − βf . g ∈ D(T ∗ ). noting for the adjoints that α= T f. T ∗ g Deﬁne ∆f (T ) = min T f − αf α∈C whenever f ∈ D(T ). f = T U f. U ] = T U − U T is the commutator of T and U . f f 2 ⇐⇒ α= T ∗ f. g = f. f 2 . Note that [T − αI. f ≤ ∆f (T ∗ )∆f (U ) + ∆f (T )∆f (U ∗ ) for all f ∈ D(T U ) ∩ D(U T ) ∩ D(T ∗ ) ∩ D(U ∗ ).1) we ﬁnd [T. Minimizing over α and β proves the theorem.4 (Commutator estimate). U − βI] = [T. U ]f. The minimum is attained for α = T f. Suppose T is a linear operator from a subspace D(T ) into H. = norm of component of T f perpendicular to f . f (24. f . deﬁned on a subspace D(T ∗ ). U ]. f / f 2 . UNCERTAINTY PRINCIPLES Next we develop an abstract commutator inequality that leads to the Heisenberg Uncertainty Principle. Let H be a Hilbert space. U ∗ f ≤ U f T ∗f + T f U ∗f . meaning T ∗ is linear and T f. U ]f. Then [T. T ∗ f − T f. Theorem 24. f − U T f. Let T and U be linear operators like above.1) = U f. U ]f. Let α. Write T ∗ for its adjoint. Proof. Hence by replacing T with T − αI and U with U − βI in (24. Here [T.

The Commutator Theorem 24. We interpret (24. Remark 24.6. In quantum mechanics. with D(U ) = {f ∈ L2 (R) : f ∈ L2 (R)}. γ > 0). U ]f = T U f − U T f d d = x · − i f (x) + i xf (x) dx dx = if (x). the less precisely one knows its momentum. (T f )(x) = xf (x) (U f )(x) = −if (x) with D(T ) = {f ∈ L2 (R) : xf (x) ∈ L2 (R)}. and vice versa.2) if and only if f (x) = iβx −γ(x−α)2 Ce e is a β-modulated Gaussian at α (with C ∈ C. Here T is the position operator and U is the momentum operator. β ∈ C. Thus the Heisenberg Uncertainty Principle implies that the variance (or uncertainty) in position multiplied by the variance in momentum is at least 1/4. the Principle says that the more precisely one knows the position of a quantum particle. U ∗ = U and [T.5 (Heisenberg Uncertainty Principle). and regard |f |2 /2π as the probability density for the momentum ξ.2) for all α. Equality holds in the Heisenberg Principle (24. 1.2) as restricting how localized f and f can be. . Roughly. we normalize f L2 (R) = 1 and interpret |f (x)|2 as the probability density for the position x of some particle. Squaring yields the Heisenberg Uncertainty Principle: 1 f 4 4 L2 (R) ≤ R |x − α|2 |f (x)|2 dx · 1 2π |ξ − β|2 |f (ξ)|2 dξ R (24. around the locations α and β. Observe T ∗ = T.4 implies f 2 L2 (R) ≤ 2∆f (T )∆f (U ) ≤ 2 xf − αf = 2 (x − α)f L2 (R) L2 (R) −if − βf L2 (R) 1 √ (ξ − β)f L2 (R) 2π by Plancherel. Take H = L2 (R).145 Example 24.

4. β ∈ C. .146 CHAPTER 24. in terms of Fourier coeﬃcients of the position density |f |2 : 1 f 4 −2 L2 (T) sup m2 (|f |2 ) (m) m∈Z 2 ≤ n∈Z |n − β|2 |f (n)|2 . One considers here a quantum particle at position eit on the unit circle. m ∈ Z (exercise). with momentum n ∈ Z. On T. The Heisenberg Uncertainty Principle extends naturally to higher dimensions. A more direct proof of (24. UNCERTAINTY PRINCIPLES 2. the analogous uncertainty principle says 1 2 1 m 4 2π eimt |f (t)|2 dt T 2 ≤ 1 2π |eimt − α|2 |f (t)|2 dt · T n∈Z |n − β|2 |f (n)|2 for all α.2) can be given by integrating by parts in f 2 L2 (R) = R f (x)f (x)(x − α) dx and then applying Cauchy–Schwarz. When α = 0 we deduce a lower bound on the localization of momentum. 3.

Part IV Problems 147 .

.

([Katznelson] Ex. where 1 < p ≤ ∞. Problem 2.1. 1. Remark. (b) Let f ∈ L2 (T).1.8: Fej´r’s Lemma) p q Let f ∈ L (T) and g ∈ L (T). 1. as m → ∞. 149 .1.2. We say un converges weakly to u. but do not hand them in: [Katznelson] Ex. Prove that p q 1 m→∞ 2π lim f (mt)g(t) dt = f (0)g(0). Use only the deﬁnition of the Fourier coeﬃcients. as m → ∞. ([Katznelson] Ex. Thus rapid oscillation yields weak convergence to the mean. Show f(m) f (0) = (mean value of f ) weakly in L2 (T).5: downsampling) Let f ∈ L1 (T).4. written un u weakly. Use that trigonometric polynomials are dense in Lq (T). and deﬁne f(m) (t) = f (mt).Assignment 1 Problem 1. and elementary manipulations. (a) Prove that f(m) (n) = f (n/m) if m|n and f(m) (n) = 0 otherwise. Do the following problems. v → u. for each v ∈ H. (Weak convergence and oscillation) Let H be a Hilbert space. (b) Then give a quick. if un . 1. v as n → ∞. e Problem 3. T Hint.2. Clearly if un → u in norm (meaning un − u → 0) then un u weakly. 1 ≤ q < ∞ and 1 + 1 = 1. imt 2 (a) Show that e 0 weakly in L (T). Problem 4. m ∈ N. formal (nonrigorous) proof using the Fourier series of f . 1.

the faster σN (f ) converges to f as N → ∞. (b) Show that if f is absolutely continuous and f has bounded variation. 1. Remark. Remark. And functions that are smooth except for ﬁnitely many corners (such as the triangular wave f (t) = |t|. Problem 8. For example. Hint. (a) Prove that ∞ f 1 (Z) ≤ f L1 (T) + 2 n=1 1 n2 1/2 f L2 .4) Let f be absolutely continuous on T with f ∈ L2 (T). then f (n) = O(|n|−1 ). π]). 1 − α Nα N ∈ N. if f is smooth except for ﬁnite many corners (such as the triangular wave f (t) = |t| for t ∈ (−π. then the Fourier series converges uniformly to f . ([Katznelson] Ex.150 Problem 5. so that Sn (f ) → f in L∞ (T). f ∈ W 1. Hence the Fourier series of f converges uniformly by Chapter 4.5. with f ∈ C α (T) for some 0 < α < 1. First evaluate f 2 2 . o Prove there exists C > 0 (depending on the H¨lder constant of f ) such that o σN (f ) − f L∞ ≤ 1 C . (Smoothness of f implies rate of decay of f ) (a) Show that if f has bounded variation. t ∈ (−π. then f (n) = O(|n|−2 ).2 (T). π]) have bounded variation.2: rate of uniform summability) Assume f is H¨lder continuous.3. ([Katznelson] Ex. functions that are smooth expect for ﬁnitely many jumps (such as the sawtooth f (t) = t. These results cover most of the functions encountered in elementary courses. Remark. t ∈ (−π. . L (b) Deduce that f ∈ A(T). That is why one encounters so many functions with Fourier coefﬁcients decaying like 1/n or 1/n2 . (A lacunary series) Assume 0 < α < 1. In other words. 1. Problem 7. Problem 6. π]) have ﬁrst derivative with bounded variation. Thus the “smoother” f is. In particular.

Sums of periodic functions having diﬀerent periods are called almost periodic functions. Read Chapter 8 “Compass and Tides” from [K¨rner]. Their theory was developed by the Danish mathematician Harald Bohr. which shows how o sums of Fourier series having diﬀerent underlying periods can be used to model the heights of tides. when β > α. brother of physicist Niels Bohr. loc Remark. Deduce that the rate n=0 of decay f (n) = O(|n|−α ) proved in Theorem 1. in soccer.) Problem 9.151 (a) Suppose that f is continuous on T and that |f (j)| ≤ C2−nα 2n ≤|j|<2n+1 for each n ≥ 0. Prove that Rd f ∈ L log L(Rd ) =⇒ M f ∈ L1 (Rd ). (That is. Problem 10. Harald Bohr won a silver medal at the 1908 Olympics. and then f ∈ C α (T). Enjoyable reading (nothing to hand in). Thus if the singularities of f are “logarithmically better than L1 ” then the Hardy–Littlewood maximal function belongs to L1 (at least locally). show f (n) = O(|n|−β ) fails for some f ∈ C α (T). (Maximal function when p = 1) Deﬁne L log L(Rd ) to be the class of measurable functions for which |f (x)| log 1 + |f (x)| dx < ∞. . n (b) Let f (t) = ∞ 2−nα ei2 t .6 for C α (T) is sharp. Prove f ∈ A(T). Show f ∈ C α (T).

152 .

(a) Let 1 ≤ p ≤ 2. b] ⊂ (−π. Problem 12 (Fourier synthesis on p ). explain why the integral and sum are absolutely convergent. Do part (a) or part (b). Prove that g ∈ Lp (T). (a) Evaluate (H 1[a. π]. Estimate the norm of T . π) is a closed interval. Take f ∈ Lp (T) and g ∈ Lp (T). Prove the Parseval identity 1 f (t)g(t) dt = lim f (n)g(n). Show the series converges unconditionally. ∞). (In your solution. for t ∈ [−π. deﬁned by (T {cn })(t) = n∈Z cn eint . is bounded from p (Z) to Lp (T).b] )(t). Problem 13 (Parseval on Lp ).) (b) Let 1 < p < ∞. Take f ∈ Lp (T) and g ∈ L1 (T) with {g(n)} ∈ p (Z). and establish the Parseval identity 1 2π f (t)g(t) dt = T n∈Z f (n)g(n). N →∞ 2π T |n|≤N 153 . Prove that the Fourier synthesis operator T . You may do both parts if you wish. Sketch the graph. Let 1 ≤ p ≤ 2. in Lp (T). Extra credit. where [a. (b) Conclude that the Hilbert transform on T is not strong (∞.Assignment 2 Problem 11 (Hilbert transform of indicator function).

Problem 16 (Boundary values lose half a derivative). and write D for the open unit disk in the complex plane. Assume u is a smooth. (Pr ∗f )(t) is called the harmonic extension to the disk of the boundary function f . Hence f ∈ C ∞ (T). Hint. v(reit ) = f (t) for r = 1. Problem 15 (Poisson extension).). Y = Z \ {0} with ν = counting measure weighted by n−2 . . t ∈ T. Thus the task is to prove each partial derivative of v on D extends continuously to D. no credit] Assume f ∈ C ∞ (T) and show v ∈ C ∞ (D). (a) Prove 1 | v|2 dA = |n||f (n)|2 . Use one of Green’s formulas. and deﬁne f (t) = u(eit ) for the boundary value function of u. (c) [Optional. Aside. Recall Pr denotes the Poisson kernel on T. real-valued function on a neighborhood of D. Let 1 < p ≤ 2. (b) Show that combining the H¨lder and Hausdorﬀ–Young inequalities in o the obvious way does not prove part (a).154 Problem 14 (Fourier analysis into a weighted space). (b) Show v is continuous on D. (a) Show v is C ∞ smooth and harmonic (∆v = 0) in D. (Parts (a) and (b) show v is smooth on D and continuous on D. Suppose f ∈ C(T) and deﬁne (Pr ∗ f )(t) for 0 ≤ r < 1. so that v is deﬁned on the closed disk D. 2π D n∈Z Hint. and so the Poisson extension v belongs to C ∞ (D) by Problem 15(c). and remember v = v since f and v are real-valued. t ∈ T. (a) Show 1/p |f (n)| |n| n=0 p p−2 ≤ Cp f Lp (T) for all f ∈ Lp (T).

since {|n|1/2 f (n)} ∈ 2 (Z). which implies {nf (n)} ∈ 2 (Z). Hint. But actually you prove more in part (c): you obtain a norm estimate on {|n|1/2 f (n)} ∈ 2 (Z) in terms of the L2 norm of u. this result holds on arbitrary domains.155 (b) Prove | v|2 dA ≤ D D | u|2 dA.) This norm estimate means that the restriction map H 1 (D) → H 1/2 (∂D) u→f is bounded from the Sobolev space H 1 (D) on the disk with one derivative in L2 to the Sobolev space H 1/2 (∂D) on the boundary circle with half a derivative in L2 . (We do not have such a norm estimate on {nf (n)}. Write u = v + (u − v) and use one of Green’s formulas. then u has half a derivative in L2 on the boundary. Aside. It asserts that among all functions having the same boundary values. f ∈ C ∞ (T) and so certainly f ∈ L2 (T). We say f has “half a derivative” in L2 . that if a function u has one derivative u belonging to L2 on a domain. Halfway inbetween lies the condition {|n|1/2 f (n)} ∈ 2 (Z). Aside. You might wonder. As your proof reveals. compared to the original function. This result is known as “Dirichlet’s principle”. Boundary trace inequalities like in part (c) are important for partial differential equations and Sobolev space theory. (c) Conclude 1 |n||f (n)|2 ≤ | u|2 dA. Justiﬁcation: if f has zero derivatives (f ∈ L2 (T)) then {f (n)} ∈ 2 (Z). and if f has one derivative (f ∈ L2 (T)) then {nf (n)} ∈ 2 (Z). . 2π D n∈Z Discussion. by using Fourier transforms. why you should bother proving the weaker result {|n|1/2 f (n)} ∈ 2 (Z) in part (c). the harmonic function has smallest Dirichlet integral. Note that in this problem. basically. Thus the boundary value loses half a derivative. The inequality says. The notion of fractional derivatives deﬁned via Fourier coeﬃcients can be extended to fractional derivatives in Rd . then.

156 Problem 17 (Measuring diameters of stars). Enjoyable reading; nothing to hand in. Read Chapter 95 “The Diameter of Stars” from [K¨rner], which shows o how the diameters of stars can be estimated using Fourier transforms of radial functions, and convolutions.

Assignment 3
Problem 18 (Adjoint of Fourier transform). Find the adjoint of the Fourier transform on L2 (Rd ). Problem 19 (Periodization, and Fourier coeﬃcients and transforms). Suppose f ∈ L1 (Rd ). (a) Prove that the periodization Pe(f )(x) = (2π)d
n∈Zd

f (x + 2πn)

of f satisﬁes Pe(f )
L1 (Td )

≤ f

L1 (Rd ) .

(b) Deduce from your argument that the series for Pe(f )(x) converges absolutely for almost every x, and that Pe(f ) is 2πZd -periodic. (c) Show that the jth Fourier coeﬃcient of Pe(f ) equals the Fourier transform of f at j: Pe(f )(j) = f (j), j ∈ Zd Problem 20 (Course summary). Write a one page description of the most important and memorable results and general techniques from this course. Be brief, but thoughtful; explain how these main results ﬁt together. You need not state the results technically — intuition is more helpful than rigor, at this stage.

157

158

Part V Appendices 159 .

.

ν) are σ-ﬁnite measure spaces. µ) is simply the triangle inequality for Lp (X). with 161 1 p + 1 q = 1. y) Lp (X) dν(y) whenever the right side is ﬁnite.1. Then for all . Proof. the norm of an integral is bounded by the integral of the norms: Theorem A. Suppose (X. Similarly. Take q to be the conjugate exponent. saying that the norm of a sum is bounded by the sum of the norms: fj νj j Lp (X) ≤ j fj Lp (X) νj whenever fj ∈ Lp (X) and the constants νj are nonnegative. If 1 ≤ p ≤ ∞ then f (x. y) dν(y) Y Lp (X) ≤ Y f (x. µ) and (Y.Appendix A Minkowski’s integral inequality Goal State Minkowski’s integral inequality. y) is measurable on the product space X × Y . and apply it to norms of convolutions Minkowski’s inequality on a measure space (X. and that f (x.

whenever the integral makes sense. APPENDIX A. f ∗g Lp (Rd ) = Rd f (· − y)g(y) dy f (· − y) Rd Lp (Rd ) ≤ = f Lp (Rd ) |g(y)| dy Lp (Rd ) by Minkowski’s integral inequality. Deﬁne the convolution of functions f and g on Rd by (f ∗ g)(x) = Rd f (x − y)g(y) dy. (f ∗ g)(t) = 2π T whenever the integral makes sense. Theorem 6. y) dν(y) g(x) dµ(x) X Y ≤ Y X |f (x.2. Lp (T) whenever the right sides are ﬁnite. whenever f ∈ Lp and g ∈ L1 . the convolution f ∗ g is well deﬁned a.14]). Deﬁnition A. Then f ∗ g Lp (T) ≤ f f ∗ g Lp (Rd ) ≤ f g L1 (T) . Deﬁne the convolution of functions f and g on T by 1 f (t − τ )g(τ ) dτ.3 (Young’s theorem).1. x ∈ Rd .162 g ∈ Lq (X). t ∈ T. . Argue similarly on T. Fix 1 ≤ p ≤ ∞. Proof. MINKOWSKI’S INTEGRAL INEQUALITY f (x. Theorem A.e. Lp (Rd ) g L1 (Rd ) . g L1 (Rd ) . Now the theorem follows from the dual characterization of the norm on Lp (X) (see [Folland. In particular. y)|p dµ(x) Y X 1/p ≤ = Y g Lq (X) dν(y) by H¨lder o f (x. y) Lp (X) dν(y) · g Lq (X) . y)||g(x)| dµ(x)dν(y) |f (x. Theorem A.

Appendix B Lp norms and the distribution function Goal Express Lp -norms in terms of the distribution function Given a σ-ﬁnite measure space (X.2) In particular. The distribution function of f is µ(E(λ)). when r = 0 < p < ∞ and α = 1.1. (B.1) If −∞ < p < r < ∞ then ∞ 0 λp−r−1 E(λ/α)c |f (x)|r dµ(x)dλ = αp−r r−p |f (x)|p dµ(x).1) expresses the Lp -norm in terms of the distribution function: ∞ 0 pλp−1 µ(E(λ)) dλ = X |f (x)|p dµ(x) = f 163 p Lp (X) . write E(λ) = {x ∈ X : |f (x)| > λ} for the level set of f above level λ. Let α > 0. Lemma B. µ) and a measurable function f on X.3) . formula (B. If −∞ < r < p < ∞ then ∞ 0 λp−r−1 E(λ/α) |f (x)|r dµ(x)dλ = αp−r p−r |f (x)|p dµ(x). X (B. X (B.

λ) ∈ X × (0. λ) ∈ E ⇔ x ∈ E(λ). and (B. LP NORMS AND THE DISTRIBUTION FUNCTION Proof. by changing variable with λ → αλ.164 APPENDIX B. λ) dλdµ(x) λp−r−1 dλdµ(x) by Fubini since λ < |f (x)| on E |f (x)| = = X |f (x)|r 1 |f (x)|p−r dµ(x) p−r since p − r > 0. so that (x. Write E = {(x.1) equals ∞ 0 λp−r−1 X 1E (x. ∞) : |f (x)| > λ}.1). Then the left hand side of (B. . Thus we have proved (B. We can assume α = 1 without loss of generality. λ)|f (x)|r dµ(x)dλ ∞ 0 = X |f (x)|r |f (x)| X r 0 λp−r−1 1E (x.2) is similar.

p) whenever p0 < p < p1 . µ) and (Y. all c ∈ C.1. Assume T : Lp0 + Lp1 (X) → {measurable functions on Y } is sublinear. Let 1 ≤ p0 < p1 ≤ ∞ and suppose (X.3 Deﬁnition C. ν) are measure spaces. 165 . An operator is sublinear if |T (f + g)(y)| ≤ |(T f )(y)| + |(T g)(y)| |T (cf )(y)| = |c||(T f )(y)| for all f. g in the domain of T . and all y in the underlying set. If T is weak (p0 . p1 ). Theorem C.Appendix C Interpolation Goal Interpolation of operators on Lp spaces. then T is strong (p.2 (Marcinkiewicz Interpolation). assuming either weak endpoint bounds (Marcinkiewicz) or strong endpoint bounds (Riesz–Thorin) References [Folland] Chapter 6 [Grafakos] Section 1. p0 ) and weak (p1 .

1). λ > 0. INTERPOLATION Proof. p /(p −p ) p /(p −p ) Choosing α = 2A11 1 0 /A00 1 0 gives simple constants: Tf Lp (Y ) ≤ 2p1/2 1 1 + p − p0 p1 − p 1/p A1−θ Aθ f 0 1 Lp (X) (C. Consider f ∈ Lp (X). By sublinearity. A1 for the constants in the weak (p0 .1) = 0 pλp−1 ν ({y ∈ Y : |T f (y)| > λ}) dλ αp−p0 f p − p0 p Lp (X) ≤ p(2A0 )p0 + p(2A1 )p1 αp−p1 f p1 − p p Lp (X) by (C. {x:|f (x)|≤λ/α} ∞ + (2A1 )p1 λ−p1 Therefore Tf p Lp (Y ) (C. (B.166 APPENDIX C. Split f into “large” and “small” parts: g = f 1{x:|f (x)|>λ/α} Notice that g ∈ Lp0 (X) since |g|p0 ≤ |f |p (λ/α)p0 −p . Then ν ({y ∈ Y : |T f (y)| > λ}) ≤ ν ({y ∈ Y : |T g(y)| > λ/2}) + ν ({y ∈ Y : |T h(y)| > λ/2}) by sublinearity p0 p1 A0 A1 g Lp0 (X) h Lp1 (X) ≤ + by the weak estimates on T λ/2 λ/2 = (2A0 )p0 λ−p0 {x:|f (x)|>λ/α} and h = f 1{x:|f (x)|≤λ/α} .2) 1 where 0 < θ < 1 is determined by expressing p as a convex combination of 1 and p11 : p0 1−θ θ 1 = + . p p0 p1 Note the estimate in (C. p0 ) and (p1 . Case 1. |T f | ≤ |T g| + |T h|. p). Write A0 . p1 ) estimates.1) and formulas (B. Hence f = g + h ∈ Lp0 + Lp1 (X).2) blows up as p approaches p0 or p1 .2). . We have proved strong (p. |f (x)|p0 dµ(x) |f (x)|p1 dµ(x). Let α > 0. h ∈ Lp1 (X) since |h|p1 ≤ |f |p (λ/α)p1 −p . Assume p1 < ∞.

That is. Given a measure space (X. f ∈ Σ(X) provided f = αj 1Fj where the sum is ﬁnite. Let α = 2A1 . Now argue like in Case 1 to get strong (p. Proof. p) on Lp (X) whenever p0 < p < p1 . Suppose T : Σ(X) → {measurable functions on Y } is linear. p0 ) and weak (p1 . If f is simple with support of ﬁnite measure. and the sets Fj have ﬁnite measure and are disjoint.) Our next interpolation result needs: Lemma C. ∞) is deﬁned to mean strong (∞. Hence ν ({y ∈ Y : |T f (y)| > λ}) ≤ ν ({y ∈ Y : |T g(y)| > λ/2}) because |T f | ≤ |T g| + |T h|. 1]. p1 ) on the simple functions in Σ(X). then so are g = f 1{|f |>λ/α} and h = f 1{|f |≤λ/α} . Deﬁnition C. . Remark C. write Σ(X) = {simple functions on X with support of ﬁnite measure}. Assume H(z) is holomorphic on U = {z ∈ C : 0 < Re(z) < 1} and continuous and bounded on U = {z ∈ C : 0 ≤ Re(z) ≤ 1}. Hence the proof of Marcinkiewicz Interpolation gives a strong (p. then T is strong (p. p). αj ∈ C \ {0}. and so Th L∞ (Y ) ≤ A1 λ λ = α 2 by deﬁnitions of h and α. Next we weaken the hypotheses of Marcinkiewicz Interpolation.5 (Hadamard’s Three Lines).4 (Linear Operators). By density of Σ(X) in Lp (X) (using here that p < p1 implies p < ∞). 1−θ θ Then |H(z)| ≤ B0 B1 whenever Re(z) = θ ∈ [0. Assume p1 = ∞. ∞). Then Marcinkiewicz Interpolation still holds: if T is weak (p0 . (This extension step uses linearity of T . p) bound for T on Σ(X). µ). because weak (∞. Let B0 = supRe(z)=0 |H(z)| and B1 = supRe(z)=1 |H(z)|. And T f = T g + T h by linearity. We have T h L∞ (Y ) ≤ A1 h L∞ (X) .3. we deduce T has a unique extension to a bounded linear operator on Lp (X).167 Case 2.

Theorem C. T Lp (X)→Lq (Y ) ≤ T 1−θ Lp0 (X)→Lq0 (Y ) T θ Lp1 (X)→Lq1 (Y ) . If B0 = 0 or B1 = 0. m > 0. Let Gm = G(z)e(z .) Suppose T : Lp0 + Lp1 (X) → Lq0 + Lq1 (Y ) is linear. Therefore the Maximum Principle applied to Gm in U says sup |Gm (z)| ≤ sup |Gm | z∈U ∂U ∪{∞} = sup |Gm | ∂U ≤ sup |G| ∂U ≤ 1. . Let 1 ≤ p0 . and (X. which proves the lemma. Let Bθ = supRe(z)=θ |H(z)|. q0 ) and (p1 . since |H| ≤ B0 on {Re(z) = 0} and |H| ≤ B1 on {Re(z) = 1}. q1 ≤ ∞. p q p0 q0 p1 q1 for some 0 < θ < 1. q1 ). −(y 2 +1)/m Hence Gm → 0 as |z| → ∞ in U . +θ . ν) be measure spaces. then further assume ν is semi-ﬁnite. Assume B0 > 0 and B1 > 0. then T is strong (p. p1 . Then G(z) = H(z) 1−z z B0 B1 1−z z is holomorphic on U and bounded on U . q0 . then add ε to H and argue as above. B1 ) > 0. If T is strong (p0 . since H is bounded and |B0 B1 | = 2 −1)/m 1−Re(z) Re(z) B0 B1 ≥ min(B0 . INTERPOLATION 1−θ θ (Exercise.6 (Riesz–Thorin Interpolation).168 APPENDIX C.) Proof. Show that θ → log Bθ is convex. Then Gm is holomorphic on U with |Gm (z)| = |G(z)|e−(y ≤ |G(z)|e 2 +1)/m e(x 2 −1)/m where z = x + iy since x2 ≤ 1 on U . Letting m → ∞ gives |G(z)| ≤ 1. = (1 − θ) . Speciﬁcally. (If q0 = q1 = ∞. µ) and (Y. q) whenever 1 1 1 1 1 1 . Let ε → 0. so that Bθ ≤ B0 B1 by the Lemma.

In applications of the theorem. and the two deﬁnitions agree on the intersection Lp0 ∩ Lp1 (X). Then one deﬁnes T on f = f0 + f1 ∈ Lp0 + Lp1 by T f = T f0 + T f1 . p1 q1 1 1 . p0 q0 0 1. Recall from measure theory that Lp ⊂ Lp0 + Lp1 . The space Lp0 + Lp1 (X) = {f0 + f1 : f0 ∈ Lp0 (X). if θ = 0 then (p. 2. For suppose f = f0 + f1 . p q 1 1 . This deﬁnition is independent of the decomposition. f1 ∈ Lp1 (X)} consists of all sums of functions in Lp0 and Lp1 . The relationship between the p and q parameters is shown in Figure C. q) = (p0 .169 1 q 1 1 . as follows.1: Parameters in the Riesz–Thorin theorem. usually one has T deﬁned and linear on p0 L (X) and Lp1 (X).1 1 p Figure C. q0 ). q1 ). by splitting f ∈ Lp into large and small parts. 1. In particular. A subtle aspect of the theorem is that when we assume T maps Lp0 + Lp1 (X) into Lq0 + Lq1 (Y ). Remark C. we need the value of T f to be independent of the choice of decomposition f = f0 + f1 . Then f0 − f0 = f1 − f1 ∈ Lp0 ∩ Lp1 (X) . and if θ = 1 then (p.1.7. q) = (p1 .

We will prove an Lp → Lq bound on T f for f ∈ Σ(X). q1 ) bounds can be applied directly to give the (p. Fix θ ∈ (0. p0 p1 q q Q (z) = (1 − z) + z q0 q1 P (z) = .170 APPENDIX C. When T = identity. deﬁne p p (1 − z) + z. where on the left side we use T deﬁned on Lp0 (X) and on the right side we use T on Lp1 (X). p1 Here is a direct proof: p Lp (X) |f |p dµ X = X |f |p(1−θ) |f |pθ dµ |f |p(1−θ)·p0 /p(1−θ) dµ X p(1−θ) Lp0 (X) p(1−θ)/p0 X ≤ = f |f |pθ·p1 /pθ dµ pθ/p1 by H¨lder o f pθ Lp1 (X) Proof of Riesz–Thorin Interpolation. so that p0 = p1 = p. q) bound. Now the (p0 . so that p < ∞. 1). q0 ) and (p1 . 3. Next suppose p0 = p1 .3) applied to T f on Y . say f = αj 1Fj and g = βj 1Gj . Then at the end we will prove the bound for f ∈ Lp (X). which ﬁxes p and q. Then T f Lq (Y ) ≤ T f 1−θ(Y ) T f θ q1 (Y ) L Lq0 by (C. For z ∈ C. First suppose p0 = p1 . Let f ∈ Σ(X) and g ∈ Σ(Y ).3) = = 1−θ p0 + θ . Linearity of T now yields T f0 + T f1 = T f0 + T f1 so that the deﬁnition of T f is independent of the decomposition of f . Riesz–Thorin says that Lp0 ∩ Lp1 ⊂ Lp with f where f 1 p Lp (X) ≤ f 1−θ Lp0 (X) f θ Lp1 (X) (C. INTERPOLATION and so T f0 − f0 = T f1 − f1 .

z ∈ U . T fz = by linearity.6) is well-deﬁned and bounded for z ∈ U . Hence the function H(z) = Y (T fz )(y)gz (y) dν(y) (C. o as one sees by substituting (C.) Let fz (x) = |f (x)|P (z) ei arg f (x) .6) and taking the sums outside the integral. (C. e . by H¨lder. gz (y) = |g(y)| Q (z) i arg g(y) x ∈ X. Next.5) into (C. Clearly gz = |βj |Q (z) ei arg βj 1Gj . and it has support (independent of z) with ﬁnite measure.4) and (C. since P (θ) = 1 and Q (θ) = 1. since 1Fj ∈ Lp0 ∩ Lp1 (X). Note fθ = f and gθ = g. so that |T fz | ≤ |αj |Re P (z) |T 1Fj | |T 1Fj | for z ∈ U . p0 Re Q (z) = q q0 ⇒ |fz |p0 = |f |p0 Re P (z) = |f |p |gz |q0 = |g|q0 Re Q (z) = |g|q ⇒ fz gz Lp0 (X) Lq0 (Y ) = f = g p/p0 Lp (X) q /q0 Lq (Y ) (valid even when p0 = ∞ or q0 = ∞) ⇒ |H(z)| ≤ T fz Lq0 (Y ) gz Lq0 (Y ) by H¨lder o |H(z)| ≤ T Lp0 (X)→Lq0 (Y ) f p/p0 Lp (X) g q /q1 .5) ≤ (const.4) Therefore gz (y) is bounded for y ∈ Y. |αj |P (z) ei arg αj (T 1Fj ) (C. Similarly. (The in Q does not denote a derivative.) The right side belongs to Lq0 ∩ Lq1 (Y ) by the strong (p0 .171 where 1 q + 1 q = 1. And H is holomorphic. p1 ) bounds. here. y ∈ Y. Lq (Y ) . Re(z) = 0 ⇒ Re P (z) = p . p0 ) and (p1 .

Suppose p0 < p1 . 1−θ Lp0 (X)→Lq0 (Y ) θ Lp1 (X)→Lq1 (Y ) f Lp (X) .14] for the dual characterization of the norm. Now the dual characterization of the norm on Lq (Y ) implies Tf Lq (Y ) ≤ T 1−θ Lp0 (X)→Lq0 (Y ) T θ Lp1 (X)→Lq1 (Y ) f Lp (X) .10]. Then gn → g in Lp0 (X) by dominated convergence. and so Tf Lq (Y ) ≤ lim inf T fn n Lq (Y ) by Fatou’s lemma T T θ Lp1 (X)→Lq1 (Y ) ≤ T = T 1−θ Lp0 (X)→Lq0 (Y ) lim inf fn n Lp (X) by (C. By passing to a subsequence we can suppose T hn → T h a. Hence T hn → T h in Lq1 (Y ). by swapping p0 and p1 if necessary.e. Theorem 6.7) from f ∈ Σ(X) to f ∈ Lp (X). (See [Folland.172 Similarly. so that f = g + h. Re(z) = 1 ⇒ |H(z)| ≤ T APPENDIX C. (C. Choose a sequence of simple functions fn ∈ Σ(X) with |fn | ≤ |f | and fn → f at every point.e. by the uniform convergence fn → f on X \ E). fn = gn + hn .) We must extend this bound (C. gn = fn 1E .7). and so T gn → T g in Lq0 (Y ). p) bound. Lq (Y ) Hence by the Hadamard Three Lines Lemma C. hn = fn 1X\E . |hn | ≤ |h|. So ﬁx f ∈ Lp (X) and let E = {x : |f (x)| > 1}.7) which is the desired strong (p. g | = |H(θ)| ≤ T 1−θ Lp0 (X)→Lq0 (Y ) T θ Lp1 (X)→Lq1 (Y ) f Lp (X) g Lq (Y ) . and h = f 1X\E . Deﬁne g = f 1E . INTERPOLATION Lp1 (X)→Lq1 (Y ) f p/p1 Lp (X) g q /q1 . Theorem 2. such a sequence exists by [Folland.e. we have T fn → T f pointwise a. if z = θ then | T f. the strong (p. and with fn → f uniformly on X\E.5 and a short calculation. Also hn → h in Lp1 (X) by dominated convergence (or. which uses semi-ﬁniteness of ν when q = ∞. By passing to a subsequence we can further suppose T gn → T g pointwise a. Therefore by linearity of T . p) bound on the simple functions. and |gn | ≤ |g|. if p1 = ∞.

We have proved the desired strong (p. p) bound for all f ∈ Lp (X). and so the proof is complete.173 since fn → f in Lp (X) by dominated convergence. .

174 APPENDIX C. INTERPOLATION .

Littlewood-Paley and Multiplier Theory.Bibliography [Benedetto] J. CruzUribe. 2001. Prentice Hall. Fourier Anal. 1972. John Wiley and Sons. S. Goh and C. American Mathematical Society. Fourier series and integrals. o [Gr¨chenig] K. Pure and Applied Mathematics (New York). Academic Press. Band 90. 8:335–373. Springer–Verlag. Appl. [Dym and McKean] H. J. 14. Foundations of Time–Frequency Analysis. J. Approx. J. Boston. CRCPress. 2001. 2001. Micchelli. Uncertainty principles in Hilbert spaces. Dym and H. [Edwards and Gaudry] R. McKean. 2002. [Folland] G. New York. Edwards and G. [Goh and Micchelli] S. [Duoandikoetxea] J. Inc. New Jersey. Jaming. Modern Techniques and their Applications. 1977. Grafakos. Folland. Classical and Modern Fourier Analysis. A. Harmonic Analysis and Applications. Gaudry. Benedetto. Graduate Studies in Mathematics. Probability and Mathematical Statistics. Nazarov’s uncertainty principles in higher dimension. 175 . New York-London. Providence. P. 2007. 2003.. E. o Birkh¨user. a [Jaming] P. Upper Saddle River. Berlin–New York. Fourier Analysis. [Grafakos] L. Gr¨chenig. A Wiley-Interscience Publication. I. 29. Theory 149:30–41. RI. No. Real Analysis. B. Ergebnisse der Mathematik und ihrer Grenzgebiete. Duoandikoetxea. 1999. D. Transl. Second edition.

Fourier analysis. [K¨rner] T. 30.176 BIBLIOGRAPHY [Katznelson] Y. Cambridge.J. Inc. S. Stein and R. Cambridge University Press.. Cambridge. [Zygmund] A. [Stein and Shakarchi] E. Katznelson. Princeton University Press. [Rudin] W. Introduction to Fourier Analysis on Euclidean Spaces. 2003. [Stein and Weiss] E. A Guide to Distribution Theory and Fourier Transforms. Princeton. 1990. Second corrected edition. M. 2003. 1. New York. Feﬀerman. Princeton Mathematical Series. NJ. Princeton.. 1971.. No. M. Stein. An introduction. Strichartz. 1970. 1989. Vol. Fourier Analysis on Groups. [Stein] E. M. Third edition. Weiss. Princeton Mathematical Series. New York. I. World Scientiﬁc Publishing Company. 2002. Princeton University Press. K¨rner. Princeton University Press. Second edition. Princeton. . An Introduction to Harmonic Analysis. 32. Stein and G.J. Zygmund. Wiley-Interscience. Foreword by R. Cambridge Unio o versity Press. W. Fourier Analysis. N. Cambridge Mathematical Library. Rudin. No. N. Trigonometric Series. Princeton Lectures in Analysis. Dover Publications. A. 1976. Shakarchi. Singular Integrals and Diﬀerentiability Properties of Functions. II. [Strichartz] R.