S. D. V. Stephens

notes / Studies in Real and Complex Analysis / chapter 6: Analysis in Several Variables

23 Fourier Series; Differentiation in Several Variables

· Denis Auroux

Original Professor Auroux’s handwritten notes (PDF), behind a password for now:

Fourier series

We consider continuous -periodic functions with complex values, or equivalently functions on , with inner product

The complex exponentials , , satisfy (orthonormality).

Definition 23.1. The Fourier coefficients of are

The Fourier series of is then .

Question 23.2. (Fourier, Dirichlet, Fejér, …) Does the Fourier series accurately represent ? (E.g. does it converge? To ?)

Definition 23.3. Trigonometric polynomials = the vector space of finite linear combinations of the (including the real and imaginary parts , ).

Clearly this is an algebra, contains and , and separates points of , which is compact. Hence by Stone–Weierstrass, real/complex trigonometric polynomials are dense in (real/-valued)… hence also in the norm, since

The th Fourier sum

(if is -valued then so is ) is the orthogonal projection of onto for . Indeed:

Thus: , , i.e. is the point of closest to for . (This follows from :

Theorem 23.4 (Parseval). Let , its Fourier coefficients, the partial sums.

  1. in , i.e. as .
  2. (in particular converges, so as ).

Proof. (1) Since the trigonometric polynomials are dense in , such that with . Now for , and is the closest point to , so . This shows in .

(2) Since and , , where

by orthonormality, and by the first part. ∎

Corollary 23.5. If have the same Fourier series then , hence .

The fact that in is the best approximation (in norm) of by trigonometric polynomials, and that trigonometric polynomials are dense in (so trigonometric polynomials uniformly), makes one hope that uniformly, or at least pointwise… alas not!

Remark (Fact). such that the Fourier series of does not converge ( unbounded!) (but the example is hard to construct).

Theorem 23.6 (Dirichlet). If is then uniformly.

The proof uses convolution—redefine, for periodic functions,

and note

So:

where

is the Dirichlet kernel. Dirichlet’s proof studies this convolution for to prove uniform convergence. (See HW8 for another proof.)

The fact that convergence can sometimes fail makes it remarkable that , can be recovered from the partial sums …

Theorem 23.7 (Fejér). If then

converges uniformly to .

The reason is that this process amounts to convolution with the Fejér kernel , which actually approximates the identity (in the sense seen last time), unlike . (See HW8.)

Differentiation in several variables

Definition 23.8. open; is differentiable at if a linear map such that

(also write: , where means , i.e. ).

The differential of at is then , aka or .

  • Conceptually, the input of is a tangent vector to at , and the output is a tangent vector at .
  • Natural norm on : the operator norm .
  • Say if is differentiable and is continuous.
  • As a matrix, the entries of are the partial derivatives = the derivative of as a function of (keeping the other constant); then . (Proof: take in the definition of the differential.)

Theorem 23.9. iff , exists and is continuous.

is clear, but is more subtle: the existence of the does not imply the differentiability, or even the continuity, of !

Example 23.10. , . Then , so , and , so . So if exists, it maps . However !

Proof of . Enough to consider , one component at a time. Applying the mean value theorem successively, for and such that :

with . All these points are within distance of , so using continuity of we get that for this is well approximated (within ) by . Hence is differentiable and , which depends continuously on . ∎

Notes not reached in this lecture, covered at the start of Lecture 24

Usual rules of differentiation hold; in particular:

Theorem 23.11 (Chain rule). If is differentiable at and is differentiable at , then is differentiable at and

Proof. where (i.e. ), so

∎

The mean value theorem doesn’t hold: e.g. , has . However we have the mean value inequality:

Theorem 23.12 (Mean value inequality). differentiable at every point of the line segment

Proof. Let = unit vector in the direction of , = unit vector in the direction of , and let . Then , so . The result then follows from the single-variable mean value inequality for on . ∎

Higher-order derivatives

is if is , etc. The main important fact about higher partial derivatives is:

Proposition 23.13. If the exist and are continuous, then .

Proof. Enough to consider the case of . For and small, , consider

Writing this in terms of , this is , so by the mean value theorem for , such that this equals

In turn, by the mean value theorem for , such that this equals .

Doing the same calculation in the opposite order shows it also equals for some , . Since these 2nd derivatives are continuous by assumption, taking limits as gives the result. ∎

Hence: the Hessian matrix is symmetric, and should be interpreted as a symmetric bilinear form on tangent vectors. If then

(and so on, Taylor!).

Local behavior and diffeomorphisms

Because of the local approximation , the behavior of governs that of near . In particular:

  • if is injective then is injective on a (sufficiently small) neighborhood of ;
  • if is surjective then maps a neighborhood of surjectively onto a neighborhood of .

When both hold, is a local diffeomorphism, by the inverse function theorem.

Definition 23.14. A map between open subsets of is a diffeomorphism if it is a homeomorphism and both and are .