Bernstein s inequality and Nikolsky s inequality for R d

Similar documents
Zygmund s Fourier restriction theorem and Bernstein s inequality

TOOLS FROM HARMONIC ANALYSIS

1 Fourier Integrals on L 2 (R) and L 1 (R).

FOURIER TRANSFORMS. 1. Fourier series 1.1. The trigonometric system. The sequence of functions

MATH MEASURE THEORY AND FOURIER ANALYSIS. Contents

Topics in Harmonic Analysis Lecture 1: The Fourier transform

THE RESTRICTION PROBLEM AND THE TOMAS-STEIN THEOREM

1.5 Approximate Identities

TOPICS IN FOURIER ANALYSIS-III. Contents

SOLUTIONS TO HOMEWORK ASSIGNMENT 4

CHAPTER VI APPLICATIONS TO ANALYSIS

RESULTS ON FOURIER MULTIPLIERS

FRAMES AND TIME-FREQUENCY ANALYSIS

MATH 220 solution to homework 5

Math 172 Problem Set 5 Solutions

Average theorem, Restriction theorem and Strichartz estimates

Bochner s Theorem on the Fourier Transform on R

Positive definite functions, completely monotone functions, the Bernstein-Widder theorem, and Schoenberg s theorem

II. FOURIER TRANSFORM ON L 1 (R)

The Hilbert transform

u t = u p (t)q(x) = p(t) q(x) p (t) p(t) for some λ. = λ = q(x) q(x)

The Fourier Transform and applications

n 1 f = f m, m=0 n 1 k=0 Note that if n = 2, we essentially get example (i) (but for complex functions).

Harmonic Analysis: from Fourier to Haar. María Cristina Pereyra Lesley A. Ward

Real Analysis Chapter 8 Solutions Jonathan Conder. = lim n. = lim. f(z) dz dy. m(b r (y)) B r(y) f(z + x y) dz. B r(y) τ y x f(z) f(z) dz

7: FOURIER SERIES STEVEN HEILMAN

Notes. 1 Fourier transform and L p spaces. March 9, For a function in f L 1 (R n ) define the Fourier transform. ˆf(ξ) = f(x)e 2πi x,ξ dx.

1.3.1 Definition and Basic Properties of Convolution

Math 172 Problem Set 8 Solutions

Reminder Notes for the Course on Distribution Theory

Hermite functions. Jordan Bell Department of Mathematics, University of Toronto. September 9, 2015

3 (Due ). Let A X consist of points (x, y) such that either x or y is a rational number. Is A measurable? What is its Lebesgue measure?

Tools from Lebesgue integration

LECTURE Fourier Transform theory

Fourier Series. ,..., e ixn ). Conversely, each 2π-periodic function φ : R n C induces a unique φ : T n C for which φ(e ix 1

The Wiener algebra and Wiener s lemma

THE MATTILA INTEGRAL ASSOCIATED WITH SIGN INDEFINITE MEASURES. Alex Iosevich and Misha Rudnev. August 3, Introduction

Fourier Transform & Sobolev Spaces

Some basic elements of Probability Theory

2 (Bonus). Let A X consist of points (x, y) such that either x or y is a rational number. Is A measurable? What is its Lebesgue measure?

INTRODUCTION TO NONLINEAR WAVE EQUATIONS

Probability and Measure

13. Fourier transforms

Differentiation and function spaces

Outline of Fourier Series: Math 201B

THE LAPLACIAN AND THE DIRICHLET PROBLEM

STRUCTURE OF (w 1, w 2 )-TEMPERED ULTRADISTRIBUTION USING SHORT-TIME FOURIER TRANSFORM

Folland: Real Analysis, Chapter 8 Sébastien Picard

Finite-dimensional spaces. C n is the space of n-tuples x = (x 1,..., x n ) of complex numbers. It is a Hilbert space with the inner product

Fourier Series. 1. Review of Linear Algebra

Topics in Fourier analysis - Lecture 2.

REAL AND COMPLEX ANALYSIS

Complex Analysis, Stein and Shakarchi The Fourier Transform

Advanced Real Analysis

Best constants in the local Hausdorff Young inequality on compact Lie groups

Fourier transforms, I

be the set of complex valued 2π-periodic functions f on R such that

The Bernstein and Nikolsky inequalities for trigonometric polynomials

Real Analysis Problems

Notions such as convergent sequence and Cauchy sequence make sense for any metric space. Convergent Sequences are Cauchy

Chapter 4. The dominated convergence theorem and applications

UNCERTAINTY PRINCIPLES WITH FOURIER ANALYSIS

THEOREMS, ETC., FOR MATH 515

A Bowl of Kernels. By Nuriye Atasever, Cesar Alvarado, and Patrick Doherty. December 03, 2013

Solutions to Tutorial 11 (Week 12)

Part II Probability and Measure

The uniform uncertainty principle and compressed sensing Harmonic analysis and related topics, Seville December 5, 2008

3: THE SHANNON SAMPLING THEOREM

Chapter 9. Parabolic Equations. 9.1 Introduction. 9.2 Some Results from Functional Analysis

g(x) = P (y) Proof. This is true for n = 0. Assume by the inductive hypothesis that g (n) (0) = 0 for some n. Compute g (n) (h) g (n) (0)

Bernoulli polynomials

8 Singular Integral Operators and L p -Regularity Theory

FINITE FOURIER ANALYSIS. 1. The group Z N

A. Iosevich and I. Laba January 9, Introduction

Linear Independence of Finite Gabor Systems

Wavelets and applications

6.2 Fubini s Theorem. (µ ν)(c) = f C (x) dµ(x). (6.2) Proof. Note that (X Y, A B, µ ν) must be σ-finite as well, so that.

For example, the real line is σ-finite with respect to Lebesgue measure, since

Summer Jump-Start Program for Analysis, 2012 Song-Ying Li

(1) Consider the space S consisting of all continuous real-valued functions on the closed interval [0, 1]. For f, g S, define

Heisenberg s inequality

Lebesgue s Differentiation Theorem via Maximal Functions

The Lindeberg central limit theorem

THE DIRICHLET-JORDAN TEST AND MULTIDIMENSIONAL EXTENSIONS

NAME: MATH 172: Lebesgue Integration and Fourier Analysis (winter 2012) Final exam. Wednesday, March 21, time: 2.5h

ANALYSIS QUALIFYING EXAM FALL 2017: SOLUTIONS. 1 cos(nx) lim. n 2 x 2. g n (x) = 1 cos(nx) n 2 x 2. x 2.

MATH 5640: Fourier Series

Marcinkiewicz Interpolation Theorem by Daniel Baczkowski

2. Function spaces and approximation

44 CHAPTER 3. CAVITY SCATTERING

Sobolev spaces. May 18

Chapter 7: Bounded Operators in Hilbert Spaces

Rudin Real and Complex Analysis - Harmonic Functions

Gaussian Hilbert spaces

HARMONIC ANALYSIS TERENCE TAO

MATH6081A Homework 8. In addition, when 1 < p 2 the above inequality can be refined using Lorentz spaces: f

Let R be the line parameterized by x. Let f be a complex function on R that is integrable. The Fourier transform ˆf = F f is. e ikx f(x) dx. (1.

Rademacher functions

Compact operators on Banach spaces

Lecture Notes on Metric Spaces

Transcription:

Bernstein s inequality and Nikolsky s inequality for d Jordan Bell jordan.bell@gmail.com Department of athematics University of Toronto February 6 25 Complex Borel measures and the Fourier transform Let ( d ) = rca( d ) be the set of complex Borel measures on d. This is a Banach algebra with the total variation norm with convolution as multiplication; for µ ( d ) we denote by µ the total variation of µ which itself belongs to ( d ) and the total variation norm of µ is µ = µ ( d ). For µ ( d ) it is a fact that the union O of all open sets U d such that µ (U) = itself satisfies µ (O) =. We define supp µ = d \ O called the support of µ. For µ ( d ) we define ˆµ : d C by ˆµ(ξ) = e 2πiξ x dµ(x) ξ d. d It is a fact that ˆµ belongs to C u () the collection of bounded uniformly continuous functions d C. For ξ d ˆµ(ξ) e 2πiξ x d µ (x) = µ ( d ) = µ. () d Let m d be Lebesgue measure on d. For f L ( d ) let Λ f = fm d which belongs to ( d ). We define ˆf : d C by ˆf(ξ) = Λ f (ξ) = e 2πiξ x dλ f (x) = d f(x)e 2πiξ x dm d (x) d ξ d. The following theorem establishes properties of the Fourier transform of a complex Borel measure with compact support. Thomas H. Wolff Lectures on Harmonic Analysis p. 3 Proposition.3.

Theorem. If µ ( d ) and supp µ is compact then ˆµ C ( d ) and for any multi-index α D α ˆµ = F (( 2πix) α µ). For > if supp µ B( ) then D α ˆµ (2π) α µ. Proof. For j =... d let e j be the jth coordinate vector in d with length. Let ξ d and define (h) = ˆµ(ξ + he j) ˆµ(ξ) h. h We can write this as e 2πihxj (h) = e 2πiξ x dµ(x). h d For any x d e 2πihxj h = e 2πihxj h 2πihx j h = 2π x j. Because µ has compact support 2π x j L (µ). Furthermore for each x d we have e 2πihxj 2πix j h. h Therefore the dominated convergence theorem tells us that lim (h) = 2πix j e 2πiξ x dµ(x). h d On the other hand for α k = for k = j and α k = otherwise so (D α ˆµ)(ξ) = lim h (h) (D α ˆµ)(ξ) = ( 2πix) α e 2πiξ x dµ(x) = F (( 2πix) α µ)(ξ) d and in particular ˆµ C ( d ). (The Fourier transform of a regular complex Borel measure on a locally compact abelian group is bounded and uniformly continuous. 2 ) Because µ has compact support so does ( 2πix) α µ hence we can play the above game with ( 2πix) α µ and by induction it follows that for any α D α ˆµ = F (( 2πix) α µ) 2 Walter udin Fourier Analysis on Groups p. 5 Theorem.3.3. 2

and in particular ˆµ C ( d ). Suppose that supp µ B( ). The total variation of the complex measure ( 2πix) α µ is the positive measure hence Then using () (2π) α x α x d α d µ ( 2πix) α µ = (2π) α x α x d α d d µ (x) d = (2π) α x α x d α d d µ (x) (2π) α = (2π) α B() B() B() = (2π) α d d µ (x) = (2π) α µ. α α d d µ (x) d µ (x) F (( 2πix) α µ) ( 2πix) α µ (2π) α µ. But we have already established that D α ˆµ = F (( 2πix) α µ) which with the above inequality completes the proof. 2 Test functions For an open subset Ω of d we denote by D(Ω) the set of those φ C (Ω) such that supp φ is a compact set. Elements of D(Ω) are called test functions. It is a fact that there is a test function φ satisfying: (i) φ(x) = for x (ii) φ(x) = for x 2 (iii) φ and (iv) φ is radial. We write for k = 2... φ k (x) = φ(k x) x d. For any multi-index α hence (D α φ k )(x) = k α (D α φ)(k x) x d D α φ k = k α D α φ. (2) We use the following lemma to prove the theorem that comes after it. 3 3 Thomas H. Wolff Lectures on Harmonic Analysis p. 4 Lemma.5. 3

Lemma 2. Suppose that f C N ( d ) and D α f L ( d ) for each α N. Then for each α N D α (φ k f) D α f in L ( d ) as k. Proof. Let α N. In the case α = φ k f f = φ k (x)f(x) f(x) dx d = φ k (x)f(x) f(x) dx x k x k f(x) dx. Because f L ( d ) this tends to as k. Suppose that α >. The Leibniz rule tells us that with c β = ( α β) we have for each k D α (φ k f) = φ k D α f + c β D α β fd β φ k. For C = max β c β D α (φ k f) φ k D α f <β α C <β α cβ D α β fd β φ k <β α D β φ k D α β f. Let C 2 = max <β α D β φ. By (2) for < β α we have Thus D β φ k = k β D β φ C 2 k β C 2 k. D α (φ k f) φ k D α f C C 2 k <β α D α β f which tends to as k. For any k φ k D α f D α f = φ k (x)(d α f)(x) (D α f)(x) dx d = φ k (x)(d α f)(x) (D α f)(x) dx x k x k (D α f)(x) dx and because D α f L ( d ) this tends to as k. But D α (φ k f) D α f D α (φ k f) φ k D α f + φ k D α f D α f which completes the proof. 4

Now we calculate the Fourier transform of the derivative of a function and show that the smoother a function is the faster its Fourier transform decays. 4 Theorem 3. If f C N ( d ) and D α f L ( d ) for each α N then for each α N D α f(ξ) = (2πiξ) α ˆf(ξ) ξ d. (3) There is a constant C = C(f N) such that ˆf(ξ) C( + ξ ) N ξ d. Proof. If g Cc ( d ) then for any j d integrating by parts ( j g)(x)e 2πiξ x dx = 2πiξ j g(x)e 2πiξ x dx. d d It follows by induction that if g C N c ( d ) then for each α N D α g(ξ) = (2πiξ) α ĝ(ξ) ξ d. Let α N. For k = 2... let f k = φ k f. For each k we have f k C N ( d ) hence D α f k (ξ) = (2πiξ) α fk (ξ) ξ d. On the one hand D α f k D α f = F (D α f k D α f) D α f k D α f and Lemma 2 tells us that this tends to as k. On the other hand for ξ d D α f k (ξ) (2πiξ) α ˆf(ξ) = (2πiξ) α fk (ξ) (2πiξ) α ˆf(ξ) = (2πiξ) α F (f k f)(ξ) (2πiξ) α f k f which by Lemma 2 tends to as k. Therefore for ξ d D α f(ξ) (2πiξ) α ˆf(ξ) D α f k D α f + D α f k (ξ) (2πiξ) α ˆf(ξ) and because the right-hand side tends to as k we get D α f(ξ) = (2πiξ) α ˆf(ξ). If y S d then there is at least one j d with y j from which we get y β >. β =N 4 Thomas H. Wolff Lectures on Harmonic Analysis p. 4 Proposition.4. 5

The function y β =N yβ is continuous S d so there is some C N > such that y β y S d. C N β =N For nonzero x d write x = x y with which β =N xβ = x N β =N yβ. Therefore x N C N x β x d. β =N For α N because the Fourier transform of an element of L belongs to C we have by (3) that ξ ξ α ˆf(ξ) belongs to C ( d ) and in particular is bounded. Then for ξ d ξ N ˆf(ξ) C N ξ β ˆf(ξ) β =N = C N C N = C. On the one hand for ξ we have hence giving β =N β =N + ξ 2 ξ ξ β ˆf(ξ) ξ β ˆf(ξ) ( ) N + ξ ξ N = 2 N ( + ξ ) N 2 On the other hand for ξ we have ˆf(ξ) C ξ N C 2 N ( + ξ ) N. + ξ 2 and so Thus for ˆf(ξ) ˆf 2 N 2 N ˆf 2 N ( + ξ ) N. { } C = max 2 N C 2 N ˆf we have completing the proof. ˆf(ξ) C( + ξ ) N ξ d 6

3 Bernstein s inequality for L 2 For a Borel measurable function f : d C let O be the union of those open subsets U of d such that f(x) = for almost all x U. In other words O is the largest open set on which f = almost everywhere. The essential support of f is the set ess supp f = d \ O. The following is Bernstein s inequality for L 2 ( d ). 5 Theorem 4. If f L 2 ( d ) > and ess supp ˆf B( ) (4) then there is some f C ( d ) such that f(x) = f (x) for almost all x d and for any multi-index α D α f 2 (2π) α f 2. Proof. Let χ be the indicator function for B( ). By (4) the Cauchy-Schwarz inequality and the Parseval identity ˆf = χ ˆf χ 2 ˆf = m d (B( )) /2 f 2 < 2 so ˆf L ( d ). The Plancherel theorem 6 tells us that if g L 2 ( d ) and ĝ L ( d ) then g(x) = ĝ(ξ)e 2πix ξ dξ d for almost all x d. Thus for f : d C defined by f (x) = ˆf(ξ)e 2πix ξ dξ = F ( ˆf)( x) x d d we have f(x) = f (x) for almost all x d. Because f = f almost everywhere f = ˆf. Applying Theorem to dµ(ξ) = f ( ξ)dξ we have f C ( d ) and for any multi-index α D α f = F (( 2πiξ) α ˆf( ξ)). 5 Thomas H. Wolff Lectures on Harmonic Analysis p. 3 Proposition 5.. 6 Walter udin eal and Complex Analysis third ed. p. 87 Theorem 9.4. 7

By Parseval s identity D α f 2 = ( 2πiξ) α 2 ˆf( ξ) = (2πiξ) α χ (ξ) ˆf(ξ) 2 (2πiξ) α χ (ξ) ˆf (2π) α ˆf 2 = (2π) α f 2 proving the claim. 4 Nikolsky s inequality Nikolsky s inequality tells us that if the Fourier transform of a function is supported on a ball centered at the origin then for p q the L q norm of the function is bounded above in terms of its L p norm. 7 Theorem 5. There is a constant C d such that if f S ( d ) > and p q then Proof. Let g = f i.e. supp ˆf B( ) f q C d d( p q ) f p. g(x) = d f( x) x d. Then for ξ d ĝ(ξ) = g(x)e 2πiξ x dx = d d f( x)e 2πiξ x dx = ˆf(ξ) d showing that supp ĝ = supp ˆf B( ). Let χ D( d ) with χ(ξ) = for ξ with which ĝ = χĝ. Then g = (F χ) g and using Young s inequality with + q = r + p g q F χ r g q = ˆχ r g q. (5) 7 Camil uscalu and Wilhelm Schlag Classical and ultilinear Harmonic Analysis volume I p. 83 Lemma 4.3. 2 8

oreover ( g a = d f( x) a dx ( d = da+d f(y) a dy d = d( a ) f a ) /a ) /a so (5) tells us d( ) q f q ˆχ r d( ) p f p i.e. f q ˆχ r d( p q ) f p. Now r = + q p so r because p q namely r. By the log-convexity of L r norms for r = θ we have Thus with we have proved the claim. ˆχ r ˆχ θ ˆχ θ. C d = max{ ˆχ ˆχ } 5 The Dirichlet kernel and Fejér kernel for The function D C () defined by D (x) = sin 2πx x πx and D () = 2 is called the Dirichlet kernel. Let χ function for the set [ ]. We have for x χ (x) = χ (ξ)e 2πixξ dξ = = e 2πixξ 2πix e 2πixξ dξ = e 2πix 2πix + e2πix 2πix = e 2πix e 2πix πx 2i sin 2πx =. πx be the indicator 9

For x = χ () = 2 = D (). Thus D = χ. For f L () and > we define (S f)(x) = ˆf(ξ)e 2πiξx dξ x. It is straightforward to check that sin 2πt (S f)(x) = f(x t)dt = (D f)(x) x. πt For f L () > and x (S m f)(x)dm = = = = = = = ( m m ( m ( f(y) m ( We define the Fejér kernel K C () by K (x) = ) ˆf(ξ)e 2πiξx dξ dm f(y)e 2πiξy dy ( m ) e 2πiξ(y x) dξ m ) ( f(y) D m (y x)dm ) e 2πiξx dξ dm ) ) dy dm ( ) sin 2πm(y x) f(y) dm dy π(y x) ( ) cos 2πm(y x) f(y) 2π 2 (y x) 2 dy ( f(y) cos 2π(y x) 2π 2 (y x) 2 2π 2 (y x) 2 cos 2πx 2π 2 x 2 x and K () =. Thus because K is an even function (S m f)(x)dm = (K f)(x). One proves that K is an approximate identity: K K (x)dx = dy ) dy.

and for any δ > lim K (x)dx =. x >δ The fact that K is an approximate identity implies that for any f L () K f f in L () as. We shall use the Fejér kernel to prove Bernstein s inequality for. 8 Theorem 6. If µ () > and supp µ [ ] then ˆµ 4π ˆµ. Proof. For x let dµ x (t) = e 2πixt dµ(t). µ x has the same support has µ and µ x (x) = e 2πixt dµ x (t) = e 2πixt e 2πixt dµ(t) = ˆµ(x + x ). It follows that to prove the claim it suffices to prove that ˆµ () 4π ˆµ. Write f = ˆµ C u (). Define C c () by { t t < (t) = t. t We calculate for x so (t)e 2πixt dt = e 2πix ( + e 2πix ) 2 4π 2 x 2 (sin πx)2 = π 2 x 2 cos 2πx = 2π 2 x 2. (x) = K (x). Then for t [ ] (e 2πiξ e 2πiξ )K (ξ)e 2πiξt dξ = K (t ) K (t + ) = ( t + ) ( t ) = t. 8 ark A. Pinsky Introduction to Fourier Analysis and Wavelets p. 22 Theorem 2.3.7.

On the one hand the integral of the left-hand side with respect to µ is (e 2πiξ e 2πiξ )K (ξ)e 2πiξt dξdµ(t) = (e 2πiξ e 2πiξ )K (ξ)f(ξ)dξ. On the other hand the integral of the right-hand side with respect to µ is t dµ(t) = 2πitdµ(t) 2πi = 2πi F (( 2πit)µ)() = 2πi f (). Hence giving proving the claim. 2πi f () = (e 2πiξ e 2πiξ )K (ξ)f(ξ)dξ f () 4π f K = 4π f 2