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)

Similar documents
Sobolev spaces. May 18

Review of Multi-Calculus (Study Guide for Spivak s CHAPTER ONE TO THREE)

Sobolev Spaces. Chapter 10

L p Spaces and Convexity

MATH MEASURE THEORY AND FOURIER ANALYSIS. Contents

Notes on Distributions

REAL ANALYSIS I HOMEWORK 4

2. Function spaces and approximation

1/12/05: sec 3.1 and my article: How good is the Lebesgue measure?, Math. Intelligencer 11(2) (1989),

Exercise 1. Let f be a nonnegative measurable function. Show that. where ϕ is taken over all simple functions with ϕ f. k 1.

L p Functions. Given a measure space (X, µ) and a real number p [1, ), recall that the L p -norm of a measurable function f : X R is defined by

Integral Jensen inequality

Continuous Functions on Metric Spaces

Lebesgue Integration on R n

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?

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

l(y j ) = 0 for all y j (1)

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.

THEOREMS, ETC., FOR MATH 515

g 2 (x) (1/3)M 1 = (1/3)(2/3)M.

Homework 11. Solutions

Real Analysis Problems

TOPICS IN FOURIER ANALYSIS-IV. Contents

Real Analysis Notes. Thomas Goller

Reminder Notes for the Course on Distribution Theory

Problem Set. Problem Set #1. Math 5322, Fall March 4, 2002 ANSWERS

Most Continuous Functions are Nowhere Differentiable

Chapter 6. Integration. 1. Integrals of Nonnegative Functions. a j µ(e j ) (ca j )µ(e j ) = c X. and ψ =

3. (a) What is a simple function? What is an integrable function? How is f dµ defined? Define it first

Outline of Fourier Series: Math 201B

Math 328 Course Notes

MATH 202B - Problem Set 5

APPLICATIONS OF DIFFERENTIABILITY IN R n.

Riesz Representation Theorems

consists of two disjoint copies of X n, each scaled down by 1,

APPROXIMATE IDENTITIES AND YOUNG TYPE INEQUALITIES IN VARIABLE LEBESGUE ORLICZ SPACES L p( ) (log L) q( )

Folland: Real Analysis, Chapter 8 Sébastien Picard

CHAPTER 6. Differentiation

Solutions: Problem Set 4 Math 201B, Winter 2007

THE INVERSE FUNCTION THEOREM

Tools from Lebesgue integration

MATH 6337: Homework 8 Solutions

Defining the Integral

converges as well if x < 1. 1 x n x n 1 1 = 2 a nx n

Laplace s Equation. Chapter Mean Value Formulas

Lebesgue s Differentiation Theorem via Maximal Functions

The optimal partial transport problem

Austin Mohr Math 704 Homework 6

1.3.1 Definition and Basic Properties of Convolution

Measure and Integration: Solutions of CW2

Course 212: Academic Year Section 1: Metric Spaces

Lecture 4 Lebesgue spaces and inequalities

Problem set 4, Real Analysis I, Spring, 2015.

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

Notes on uniform convergence

MATH 426, TOPOLOGY. p 1.

Distributions. Chapter INTRODUCTION

Unbounded operators on Hilbert spaces

Generalized pointwise Hölder spaces

Thus f is continuous at x 0. Matthew Straughn Math 402 Homework 6

Solutions to Tutorial 11 (Week 12)

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

TOOLS FROM HARMONIC ANALYSIS

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

Harmonic Functions and Brownian motion

Definition A.1. We say a set of functions A C(X) separates points if for every x, y X, there is a function f A so f(x) f(y).

Geometric intuition: from Hölder spaces to the Calderón-Zygmund estimate

Math 5051 Measure Theory and Functional Analysis I Homework Assignment 3

Distribution Theory and Fundamental Solutions of Differential Operators

Math 209B Homework 2

Fourier Transform & Sobolev Spaces

A note on some approximation theorems in measure theory

Commutative Banach algebras 79

Green s Functions and Distributions

Problem set 1, Real Analysis I, Spring, 2015.

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?

Measurable functions are approximately nice, even if look terrible.

2) Let X be a compact space. Prove that the space C(X) of continuous real-valued functions is a complete metric space.

Math 172 Problem Set 5 Solutions

Analysis III Theorems, Propositions & Lemmas... Oh My!

MATH 217A HOMEWORK. P (A i A j ). First, the basis case. We make a union disjoint as follows: P (A B) = P (A) + P (A c B)

Problem Set 5. 2 n k. Then a nk (x) = 1+( 1)k

MASTERS EXAMINATION IN MATHEMATICS SOLUTIONS

Functional Analysis Exercise Class

Bounded point derivations on R p (X) and approximate derivatives

Chapter 5. Measurable Functions

HARMONIC ANALYSIS. Date:

CHAPTER 1. Metric Spaces. 1. Definition and examples

McGill University Math 354: Honors Analysis 3

Fall f(x)g(x) dx. The starting place for the theory of Fourier series is that the family of functions {e inx } n= is orthonormal, that is

1.5 Approximate Identities

On some shift invariant integral operators, univariate case

Continuous functions that are nowhere differentiable

Continuity of convex functions in normed spaces

Metric Spaces and Topology

Functional Analysis, Stein-Shakarchi Chapter 1

Analysis Comprehensive Exam Questions Fall F(x) = 1 x. f(t)dt. t 1 2. tf 2 (t)dt. and g(t, x) = 2 t. 2 t

(z 0 ) = lim. = lim. = f. Similarly along a vertical line, we fix x = x 0 and vary y. Setting z = x 0 + iy, we get. = lim. = i f

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

SOLUTIONS TO HOMEWORK ASSIGNMENT 4

Transcription:

Mollifiers and Smooth Functions We say a function f from C is C (or simply smooth) if all its derivatives to every order exist at every point of. For f : C, we say f is C if all partial derivatives to every order exist and are continuous. Proposition. The function g(x) = { 0 if x 0 e x if x > 0 is C. This will follow from several lemmas. Note the only thing we need to prove is that g (n) (0) exists for all n. We will show g (n) (0) = 0 for all n. Lemma 2. For x > 0, n 0 the n th derivative g (n) (x) = P ( x )e x for P a polynomial. Proof. This is true for n = 0. Now inductively, if g (n) (x) = P ( x )e x, compute for x > 0 g (n+) (x) = P ( x ) ( x 2 ) e x + P ( x ) e x ( x 2 ). Now note that P ( x ) ( x 2 ) + P ( x ) ( x 2 ) is also a polynomial in x. Lemma 3. If P is a polynomial, then lim P ( x 0 + x )e x = 0. Proof. Mae the substitution y =, and note that x Lemma 4. For all n, g (n) (0) = 0. P (y) lim = 0. y e y Proof. This is true for n = 0. Assume by the inductive hypothesis that g (n) (0) = 0 for some n. Compute g (n+) (0) = lim g (n) (h) g (n) (0) h = lim g (n) (h) h by the inductive hypothesis. For h < 0, it is clear that g (n) (h) = 0, and thus that g (n) (h) lim = 0. h

For h > 0, use Lemma 2 to compute g (n) (h) lim + h = lim P ( + h h )e h = 0 by applying Lemma 3 for the polynomial P (y) = yp (y). Since the left and right limits are equal, we find g (n+) (0) = 0 and by induction this holds for all n. The Proposition is proved. Theorem. There is a function φ on which satisfies () φ(x) 0 for all x. (2) φ C ( ). (3) supp φ = B (0). (4) φ dm =. Proof. For g from Proposition above, let { φ(x) = c g( x 2 0 if x ) = c exp( ) if x < x 2 where c is defined so that φ dm =. Then φ is C since φ = c g h for h = x 2 = x 2 x 2. h is C since it s a polynomial. Then we can verify that all the various partial derivatives of φ = c g h are continuous by using the usual rules of differentiation (the chain rule in multiple variables, the product rule, etc.) to compute them. A mollifer is a family of functions φ δ based on φ given by φ δ (x) = δ φ( x ) for all δ > 0. Then it is easy to chec that δ Proposition 5. For all δ > 0, () φ δ (x) 0 for all x. (2) φ δ C ( ). (3) supp φ δ = B δ (0). (4) φ δ dm =. Proof. All of these properties are obvious except the last one. For the last one, we use the change of variables formula in multiple integrals to prove φ δ (x) dm(x) = δ φ( x) dm(x) = δ φ(y) dm(y) =, by using using the substitution y = x, which implies dm(y) = δ δ dm(x) for δ the Jacobian determinant. 2

A function f is mollified by convolution with φ δ. Define f δ (x) = (f φ δ )(x) = f(x y)φ δ (y) dm(y) = f(y)φ δ (x y) dm(y), as long as the integrals converge. In this case, note that the integrals in the line above are equal by maing the substitution z = x y, which implies dm(z) = dm(y) (the multiplicative factor is the absolute value of the Jacobian determinant, which is ). The idea is that f δ (x) is a weighted, smoothed average all the values of f in the ball B δ (x) of radius δ and center x. To see this, we mae an analogous construction using the function α(x) = m(b (0)) χ B (0)(x). Then α satisfies the same properties as φ in Theorem except it is not smooth. Then if we define α δ (x) = δ α( x ), the convolution δ (f α δ )(x) = f(y)α δ (x y) dm(y) = m(b δ (x)) f(y) dm(y) B δ (x) is the average value of f over the ball B δ (x) of radius δ and center x. Since φ δ 0 and has integral over its support B δ (0), we can consider (f φ δ )(x) to be a weighted average of f over B δ (x). We say a complex function f on is locally L if χ K f L for every compact subset K of. It is an easy consequence of Hölder s inequality that every f L p ( ) is locally L for p. Compute for q the conjugate exponent of p: fχ K dm = f dm f L p (K) L q (K) f L p ( ) m(k) q < K Theorem 2. If f : C is locally L and ψ : C is C with compact support, then f ψ is C. In order to prove this theorem, we need a few lemmas: Lemma 6. If f L loc ( ) and ψ C c ( ), then f ψ is continuous. Proof. Let x n x be a convergent sequence. The lemma is proved if we can show (f ψ)(x n ) = f(y)ψ(x n y) dm(y) f(y)ψ(x y) dm(y) = (f ψ)(x). Note that the definition of f ψ is unchanged if f is redefined on a set of measure 0. Thus, without loss of generality, assume f is defined everywhere. 3

We may assume x n B (x). Choose r > 0 so that supp ψ B r (0) (since ψ has compact support). Since ψ is continuous, we have for all y, f(y)ψ(x n y) f(y)ψ(x y). Moreover, if C = sup ψ <, then for all x n, we have f(y)ψ(x n y) C f(y) χ Br+ (x) (y). (To show the last term is justified, note that if y > r +, then x n y y x n > (r + ) = r and so ψ(x n y) = 0.) By assumption, the function on the right-hand side is integrable. So the Dominated Convergence Theorem applies to show that lim (f ψ)(x n) = lim f(y)ψ(x n y) dm(y) n = lim f(y)ψ(x n y) dm(y) n = f(y)ψ(x y) dm(y) n = (f ψ)(x). Lemma 7. If f L loc ( ) and ψ C c ( ), then for all i =,...,, ψ (f ψ) = f xi x. i Proof. We may assume ψ is real, since we may otherwise consider the real and imaginary parts. Compute at x, for e i the i th coordinate vector (f ψ)(x) = lim x [(f ψ)(x + he i h i) (f ψ)(x)] [ = lim f(y)ψ(x + he i y) dm(y) h ] f(y)ψ(x y) dm(y) = lim f(y) h = lim f(y) ψ x (x y + c(h)e i) dm(y), i ( ψ(x + hei y) ψ(x y) 4 ) dm(y)

where we use the Mean Value Theorem to find a value c(h) so that c(h) h (here we use that ψ is real). Continue to compute at x x i (f ψ) = lim = ( f ψ x i ( f ψ x i ) (x) ) (x + c(h)e i ) 5 where we apply Lemma 6 to tae the last limit. Proof of Theorem 2. Apply Lemma 7 to compute the first partial derivative ψ (f ψ) = f xi x. i This function is continuous by Lemma 6 and since ψ C x i c ( ). The higher-order partial derivatives can be handled by induction. Lemma 8. If f is locally L and f δ = f φ δ is the standard mollifier, then Proof. If suppf δ suppf + B δ(0) = {x + y x suppf, y Bδ(0) }. 0 f δ (x) = f(x y)φ δ (y) dm(y) = B δ (0) f(x y)φ δ (y) dm(y), then f(x y) cannot be identically zero for all y B δ (0). So for some such y, x y supp f, and thus x = (x y) + y supp f + B δ (0). Theorem 3. If f : C is continuous, and φ δ is the standard mollifier, then f δ = f φ δ f uniformly on compact subsets of. Proof. Let K be compact. Then there is an r > 0 so that K B r (0). On the compact set B r+ (0), f is uniformly continuous. Choose ɛ > 0. There is an η > 0 so that if z, w B r+ (0), then z w < η implies f(z) f(w) < ɛ. We may additionally assume η <. Let δ (0, η).

For x K B r (0), compute f δ (x) f(x) = f(x y)φ δ (y) dm(y) f(x) = f(x y)φ δ (y) dm(y) f(x)φ δ (y) dm(y) f(x y) f(x) φ δ (y) dm(y) = f(x y) f(x) φ δ (y) dm(y) < B δ (0) B δ (0) ɛ φ δ (y) dm(y) = ɛ The last line is justified since x B r (0) B r+ (0) and x y B r (0)+ B δ (0) B r+ (0). Since this estimate applies to all x in the compact set K, we have that f δ f uniformly on K. Corollary 9. If f C c ( ), then f δ f uniformly on. Theorem 4. For p <, C c ( ) is dense in L p ( ). Proof. Let f L p ( ) and ɛ > 0. Then Theorem 3.4 of udin implies there is a g C c ( ) so that f g p < ɛ. 2 Now let g δ = g φ δ for φ the standard mollifier. Then Corollary 9 implies g δ g uniformly as δ 0. Assume supp g B r (0) for some r > 0. Then supp g δ B r (0) + B δ (0) = B r+δ (0). Compute g δ g p p = g δ g p dm = g δ g p dm sup g δ g p m(b r+δ (0)). B r+δ (0) Now sup g δ g 0 since g δ g uniformly. Thus g δ g p 0 and there is a δ > 0 so that g δ g p < ɛ. 2 Therefore, f g δ p f g p + g g δ p < ɛ. g δ Cc ( ) since supp g δ supp g + B δ (0) is compact. g δ is C by Theorem 2. A similar but more precise result is Theorem 5. For any p < and f L p ( ), then f φ δ f p 0 as δ 0, where φ is any nonnegative measurable function on with total integral one. The proof is essentially contained in Theorems 9.5, 9.9 and 9.0 in udin. Proposition 0. Let f L p ( ) for p <, and let y. Let f y (x) = f(x y). Then the map y f y is a continuous map from to L p ( ). 6

Proof. Let ɛ > 0. By udin, Theorem 3.4, choose a continuous function g with compact support so that f g p < ɛ. Let > 0 be so that supp g B (0). Since g is continuous with compact support, it is uniformly continuous. Thus there is a δ (0, ) so that s t < δ implies g(s) g(t) < m(b 0 (2)) p ɛ, for m Lebesgue measure on. Then compute for s t < δ g s g t p p = g(x s) g(x t) p dx < m(b 2 (0)) ɛ p m(b +δ (s)) < ɛ p. Whenever s t < δ, f s f t p f s g s p + g s g t p + g t f t p = (f g) s p + g s g t p + (g f) t p = f g p + g s g t p + g f p < 3ɛ. Here we have used the change of variables z = x s for α = f g to compute α s p p = α(x s) dx = α(z) dz = α p p. Proof of Theorem 5. We need to prove that lim f φ δ f p = 0. δ 0 As in the proof of Theorem 3 above, we have (f φ δ )(x) f(x) f(x y) f(x) φ δ (y) dm(y). Since φ δ is a positive function with integral one, we may apply Jensen s Inequality for the convex function t t p to find ( ) p (f φ δ )(x) f(x) p f(x y) f(x) φ δ (y) dm(y) f(x y) f(x) p φ δ (y) dm(y). 7

Now we may integrate this inequality over in x and use Fubini s Theorem to find f φ δ f p p f(x y) f(x) p φ δ (y) dm(y) dm(x) = f(x y) f(x) p dm(x) φ δ (y) dm(y) = f y f p p φ δ (y) dm(y). Let g(y) = f y f p p. Proposition 0 above shows g is a continuous function, and it is clear that g(0) = 0. Moreover, g is bounded since g(y) = f y f p p ( f y p + f p ) p = (2 f p ) p. We continue computing to find f φ δ f p p g(y)φ δ (y) dm(y) = g(y)δ φ(yδ ) dm(y), = g(δs)φ(s) dm(s) for the change of variables s = yδ. As δ 0, g(δs)φ(s) g(0)φ(s) = 0 pointwise on. Moreover, the integrand g(δs)φ(s) g φ(s) for all δ. Since g is bounded and φ is integrable, the Dominated Convergence Theorem applies to show that lim g(δs)φ(s) dm(s) = 0. δ 0 This implies f φ δ f p 0 as δ 0. 8