Many of the computations involving Fourier series are simpler when we use the complex exponential instead of sine and cosine, so in this section we will consider complex valued solutions to the wave equation.
Euler’s Formulas
By definition
eiθ=cosθ+isinθfor all θ∈R.(1)
This implies
e−iθ=cosθ−isinθfor all θ∈R
and hence
cosθ=2eiθ+e−iθ,sinθ=2ieiθ−e−iθ(2)
Derivative of the complex exponential
Euler’s definition of eiθ implies
dθdeiθ=ieiθ.(3)
In the language of Linear Algebra, eiθ is an eigenvector of the linear transformation f↦dθdf with eigenvalue i.
Trigonometric polynomials
A function of the form
f(x)=A0+n=1∑N(Ancosnx+Bnsinnx)
where An,Bn∈C are constants is called a trigonometric polynomial.
Using Euler’s formulas we can rewrite a trigonometric polynomial in terms of complex exponentials:
The wave equation is linear. If u,v:R2→R are solutions to the wave equation,
then w(x,t)=au(x,t)+bv(x,t) is also a solution for any choice of a,b∈R. This is true both for classical solutions and for weak solutions; i.e. if u,v are classical solutions, then w=au+bv is a classical solution, and if u,v are weak solutions, then w=au+bv is a weak solution.
Solutions that are periodic in space. Instead of considering arbitrary solutions to the wave equation, we look for solutions for which
u(x+2π,t)=u(x,t)
holds for all x,t. By direct substitution in the equation we can verify that each of the functions
einteinx,e−inteinx
are classical solutions of the wave equation, and that they are 2π–periodic in the x variable. It follows that for any choice of a^n,b^n∈Cthe linear combination
u(x,t)=n=−N∑Na^neinteinx+b^ne−inteinx(4)
is again a solution of the wave equation. If we have infinitely many coefficients a^n,b^n then we could try to show that the series converges and that its limit is a weak or classical solution of the wave equation.
Sometimes it is convenient to rewrite (4) by applying Euler’s formula e±int=cosnt±isinnt, with result
u(x,t)=n=−N∑Nu^ncos(nt)einx+v^nsin(nt)einx(5)
where u^n=a^n+b^n and v^n=i(a^n−b^n).
The initial value problem. Suppose we want to find a solution to
utt=uxx, for all x,tu(x+2π,t)=u(x,t), for all x,tu(x,0)=f(x),ut(x,0)=g(x) for all x
by looking for a function u(x,t) that is given by (5). The function already satisfies the PDE, and it is 2π–periodic in the x variable, so we only have to choose the coefficients u^n,v^n so that u satisfies the initial conditions. It follows from (5) that
So if you can write the initial function and time derivative as a Fourier series (8) then we have a solution given in (9). This leaves us with a few questions
for which 2π–periodic functions f and g can we find a Fourier expansion (8)?
Fourier’s remarkable answer to the first question was: every 2π–periodic function f has a Fourier expansion and the coefficients can be found by computing the integrals in (7).
For example, the “sawtooth function” is given by
f(x)=2π−x for 0<x<2π
Fourier computed the coefficients in its expansion and claimed
f(x)=n=1∑∞nsinnx=sinx+21sin2x+31sin3x+⋯
Wikipedia offers a one paragraph history of Fourier’s claims.
Digression: convergence of Fourier series
There is not enough space in this aside on Fourier series to do the topic justice. If you want to read more, the book Fourier Analysis by Stein and Shakarchi (Princeton lectures in analysis) offers a good introduction. Alternatively, T.W.Körner’s book Fourier Analysis (yes, same title) from Cambridge University press is very readable.
Questions: was Fourier right?
Let f be a 2π–periodic function. Does the Fourier series for f converge, and does it converge to f? To be more precise, define the partial sums of the series
Proof. We show that ∫sin(λx)f(x)dx→0, the proof that ∫cos(λx)f(x)dx→0 being nearly identical.
Let ϵ>0 be given. Then, by definition of Riemann integrability of the function f, there exists a partition a=x0<x1<x2<⋯<xm−1<xm=b, and numbers mi<Mi such that
mk≤f(x)≤Mk for all x∈[xk−1,xk], and all k=1,…,m
∑k=1m(Mk−mk)(xk−xk−1)<ϵ/(2(b−a))
Define the step function
s(x)=mk for all x∈(xk−1,xk] and for all k=1,2,…,m
Then s(x)≤f(x) and 0≤∫ab(f(x)−s(x))dx<ϵ/(2(b−a)).
To show that ∫absin(λx)f(x)dx→0 we rewrite the integral as follows
∣g(s)∣≤2π∣sin(s/2)∣C∣s∣≤2C for all s∈(−π,π),s=0.
Here we have used that sinx≥2πx for all x∈[0,2π].
using the boundedness of g and the fact that f is Riemann integrable, it is now a somewhat lengthy exercise in real analysis to show that g is also Riemann integrable on [−π,π]. We can therefore invoke the Riemann–Lebesgue lemma and conclude that there is a λϵ such that for all λ>λϵ one has
∫−ππsin(λs)g(s)ds<ϵ.
and therefore ∣SNf(a)−f(a)∣<ϵ for all λ>λϵ, i.e. for all N>λϵ−21.
Q.E.D.
Theorem. If f is a 2π–periodic C2 function then its Fourier coefficients f^n satisfy
∣f^n∣≤n2∥f′′∥∞ for all n=0.
The Fourier series f(x)=∑−∞∞f^neinx converges uniformly, and absolutely.
Proof. Integration by parts in the definition of f^n gives us
The terms in the Fourier series for f are bounded by
f^neinx=∣f^n∣≤n2∥f′′∥∞(n=0).
Therefore, by the Weierstrass M-test, the series ∑−∞∞f^neinx converges uniformly, and absolutely.
Since f is C1 it is also Hölder continuous at all x∈R. Therefore Theorem 3.5 implies that ∑−∞∞f^neinx=f(x) for all x∈R.
Q.E.D.
Fourier series and the L2 Inner Product
An inner product
Let Rper be the set of Riemann integrable 2π–periodic functions, and let Cper be the subset of continuous 2π–periodic functions. For any two functions f,g∈Rper we define
⟨f,g⟩=∫02πf(x)g(x)dx.
Then ⟨…⟩ has the following properties:
Symmetry: ⟨f,g⟩=⟨g,f⟩
Bilinear: ⟨f+g,h⟩=⟨f,h⟩+⟨g,h⟩ for all f,g,h∈Rper
Bilinear: ⟨cf,g⟩=c⟨f,g⟩ and ⟨f,cg⟩=cˉ⟨f,g⟩ for all f,g∈Rper and c∈xC.
Nonnegative: ⟨f,f⟩≥0 for all f∈Rper
Theorem. If f∈Cper and ⟨f,f⟩=0 then f=0.
On the other hand there exist functions f∈Rper with f=0 and ⟨f,f⟩=0.
Proof. Arguing by contradiction we suppose ⟨f,f⟩=0 and f(x0)=0 for some x0∈[0,2π). Since f is continuous, there is an δ>0 such that ∣f(x)∣≥21∣f(x0)∣ for all x∈(x0−δ,x0+δ). This implies
This is a contradiction, so we conclude that f(x0)=0 for all x0.
To prove the second part of the theorem, consider the function f for which f(x)=0 at all x∈R except x=nπ (n∈Z) where f(nπ)=1. Then f is Riemann integrable and 2π–periodic, and ⟨f,f⟩=0 even though f=0.
Q.E.D.
Theorem.Cper with ⟨f,g⟩ defined as above is a complex vector space with an inner product.
We define the L2 norm of f∈Rper to be ∥f∥2 where
∥f∥22=⟨f,f⟩=∫02π∣f(x)∣2dx.
The L2 and the supremum norm. The L2 norm and the supremum norm ∥f∥∞=supx∣f(x)∣ are generally different. They always satisfy
∥f∥2≤2π∥f∥∞ for any f∈Rper.
This follows from the following short computation
∥f∥22=∫02π∣f(x)∣2dx≤∫02π∥f∥∞2dx=2π∥f∥∞2.
Fourier coefficients as inner products. Let ek(x)=eikx. Then for any k∈Z and any l=k one has
∥ek∥2=2π, and ⟨ek,el⟩=0, i.e. ek⊥el.
If f=f^−Ne−N+⋯+f^NeN then we have
⟨f,ek⟩=f^k∥ek∥2=2πf^k.
The kth Fourier coefficient of f∈Rper is given by
f^k=2π1⟨f,ek⟩
The Nth partial sum of the Fourier series of f∈Rper is given by
SNf=2π1k=−N∑N⟨f,ek⟩ek
Bessel’s inequality. For any f∈Rper we have
∥SNf∥22+∥f−SNf∥22=∥f∥22.
In particular ∥SNf∥2≤∥f∥2 always holds.
Proof. We begin by showing that SNf⊥(f−SNf). For any integer k with ∣k∣≤N we have
⟨ek,f−SNf⟩=⟨ek,f⟩−2π1l=−N∑n⟨ek,⟨f,el⟩el⟩=⟨ek,f⟩−2π1l=−N∑n⟨f,el⟩⟨ek,el⟩=⟨ek,f⟩−2π1l=−N∑n⟨el,f⟩⟨ek,el⟩=⟨ek,f⟩−2π1⟨ek,f⟩⟨ek,ek⟩=0.⟨ek,el⟩=0 if k=l⟨ek,ek⟩=2π
Thus ek⊥f−SNf for all k=−N,…,N. Since SNf is a linear combination of e−N,…,eN it follows that SNf⊥f−SNf.
Bessel’s inequality now follows from Pythagoras: since SNf⊥f−SNf we have
∥f∥22=∥f−SNf+SNf∥22=∥f−SNf∥22+∥SNf∥22.
Q.E.D.
Partial sums as best approximations. If g=∑−NNg^kek is a trigonometric polynomial of degree at most N then
∥f−SNf∥2≤∥f−g∥2,
with strict inequality if g=SNf.
Proof. First note that SNf−g is a linear combination of e−N,…,eN, and therefore that SNf−g⊥SN−f. Then apply Pythagoras to f−g=f−SNf+SNf−g, to get
∥f−g∥22=∥f−SNf∥22+∥SNf−g∥22.
This implies ∥f−SNf∥2≤∥f−g∥2, and if ∥f−SNf∥2=∥f−g∥2 then ∥SNf−g∥2=0.
Since SNf−g is a trigonometric polynomial it is a continuous function. Therefore ∥SNf−g∥2=0 implies that Sf−g=0, i.e. SNf=g.
Q.E.D.
Convergence in L2 of Fourier series. Let f∈Rper. Then
N→∞lim∥SNf−f∥2=0
and
∥f∥22=2π1n∈Z∑∣f^n∣2.
This statement is known as the Plancherel or Parseval identity.
Proof. Let ϵ>0 be given. Since f is Riemann integrable, a C2 function g exists that is 2π–periodic, and that satisfies
∫02π∣f(x)−g(x)∣2dx≤4ϵ2, i.e. ∥f−g∥2<2ϵ.
(details in lecture). Since g is C2, we know that the Fourier series of g converges uniformly to g, i.e.
∥SNg−g∥∞→0.
For any given ϵ2>0 we can therefore find an Nϵ such that ∣∣SNg−g∥∞<22πϵ for all N≥Nϵ2. This implies
The Parseval—Plancherel identity now follows by letting N→∞ in ∥f∥22=∥SNf∥22+∥f−SNf∥22.
Q.E.D.
Problems
Consider the solutions u:R2→R to the wave equation described in (4).
Find functions F,G:R→R such that u(x,t)=F(x+t)+G(x−t) for all x,t. (Hint: don’t use the derivation of d’Alembert’s solution but instead use eiθeiϕ=ei(θ+ϕ) and take a good look at (4).
Various questions about the Dirichlet kernel and Fourier coefficients
(a) Show that ∫02πDN(s)ds=1. (b) The kth Fourier coefficient f^k of a 2π–periodic function f:R→C is defined as an integral from x=0 to x=2π. Show that f^k=2π1∫−ππe−ikxf(x)dx.
Let f(x)=∣sin2x∣.
(a) Compute the Fourier series of f (the integrals simplify if you write sin2x in terms of complex exponentials.) (b) For which x∈R does the Fourier series converge to f? (c) What do you get if you set x=0 in the Fourier expansion of f?
The “plucked string” function
Let ℘ be the 2π–periodic function that satisfies ℘(x)=x for x∈[0,π], and ℘(x)=2π−x for x∈[π,2π]
(a) Compute the Fourier series of ℘ (b) For which x∈R does the Fourier series converge to ℘? (c) What do you get if you set x=0 in the Fourier expansion of ℘?
Solving ordinary differential equations with Fourier series
(a) Let f:R→C be a 2π–periodic function that is C1. Show that if g=f′ then g^k=ikf^k. (b) Let f:R→C be a 2π–periodic function that is C2. Show that if g=f′′ then g^k=−k2f^k. (c) Show that there exist numbers mk∈C such that the following holds for any C22π–periodic function f:R→C: if f′′−4f=g then f^k=mkg^k. (d) Find the Fourier series of the 2π–periodic solution f:R→C of f′′(x)−4f(x)=sin(x) (0≤x≤2π). (e) Find the Fourier series of the 2π–periodic solution f:R→C of f′′(x)−4f(x)=℘(x) (0≤x≤2π) where ℘ is the “plucked string” function from the previous problem
Let f:[0,2π]→R be Riemann integrable, and let f^k be the kth Fourier coefficient of f.
(a) Show that ∣f^k∣≤∥f∥∞ for all k∈Z. (b) Show that limk→±∞f^k=0. (c) Suppose that f is m times continuously differentiable. Show that ∣f^k∣≤∣k∣m∥f(m)∥∞ for all k=0.
Let z:R→R be the 2π–periodic “saw tooth function” given by z(x)=2π−x for 0<x<2π.
Compute ∥z∥2 and the Fourier coefficients of z. Which identity do you find by applying the Plancherel—Parseval’s identity?