Kernels arising from the Fourier series

Here, we present a different way of looking at the Fourier series using kernels. We see that the partial Fourier sum of a function f(x) can be expressed as the convolution between f and the Dirichlet kernel. We have already seen that the partial Fourier sum converges uniformly to f if f(x) is continuously differentiable. Similarly, we show that the convolution of the function f and the Fejér kernel converges uniformly to f, but this time, if f(x) is continuous. These kernels are used to prove the L2 convergence of the Fourier series.



Definition (Kernel function)

Let X be a set, and K:X×XR be a two-variable function. K(x,y) is said to be a kernel function (or integral kernel or nucleus) if it satisfies the following conditions:

  1. (symmetry) For any x,yX, K(x,y)=K(y,x).
  2.  (positive semi-definiteness) For any nN, {x1,x2,,xn}X, {c1,c2,,cn}R, i=1nj=1ncicjK(xi,xj)0.

We now consider kernel representations of the Fourier series.

The Dirichlet kernel

Consider the partial sum Sn(x) of S[f]:
Sn(x)=|k|nckeikx=|k|n(12πππf(y)eikydy)eikx=ππ12πk=nneik(xy)f(y)dy.
Let us define Dn(θ) (θR) by
(eq:Ddef)Dn(θ)=12πk=nneikθ.
Noting
k=nneikθ=einθei(n+1)θ1eiθ=ei(n+1/2)θei(n+1/2)θeiθ/2eiθ/2=sin(n+12)θsinθ2,
we have
(eq:Dfconv)Sn[f](x)=Sn(x)=ππDn(xy)f(y)dy
where
Dn(xy)=12πsin(n+12)(xy)sinxy2 (n=0,1,)
is called the Dirichlet kernel. As the name suggests, it is a kernel function. Dn(θ) is an even function of θ, and has a period of 2π. Thus, Dn(xy) is symmetric with respect to x and y. Using (eq:Ddef), we can show that it is also positive semi-definite (exercise!).
For the constant function f(x)=1cn=0 for n0 and c0=1 so that Sn[f](x)=1 for all n. Therefore
ππDn(xy)dy=1   (n=0,1,).
The following figure shows the Dirichlet kernels with n=0,1,5,20.

The integral of the form of (Eq:Dfconv) is so common that it has a name.

Definition (Convolution)

Let f,gP2π . The convolution of f and g, often denoted as fg, is defined as
(fg)(x)=ππf(xy)g(y)dy.

See also: Convolution (Wikipedia)

Thus, we have
Sn(x)=(Dnf)(x).

The Fejér kernel

Given the sequence {an}n=0, define another sequence {bn} by
bn=1n+1k=0nak=a0+a1++ann+1.
It is known that, if {an} converges, then {bn} converges to the same value and the convergence is generally ``faster'' than {an}. Based on this observation, we introduce another kernel representation of the Fourier series. For the partial sum Sn of the Fourier series of f, we define Fejér's partial sum σn[f](x) by
(eq:fejer)σn[f](x)=σn(x)=1n+1k=0nSk(x)   (n=0,1,).
The integral representation of σn(x) can be obtained as follows.
σn(x)=1n+1k=0nππDk(xy)f(y)dy=ππ(1n+1k=0nDk(xy))f(y)dy
where Dn is the Dirichlet kernel. Noting
k=0nsin(k+12)θ=1sinθ2k=0nsin(k+12)θsinθ2=12sinθ2k=0n(coskθcos(k+1)θ)=12sinθ2(1cos(n+1)θ)=sin2n+12θsinθ2,
we have
σn[f](x)=σn(x)=ππFn(xy)f(y)dy=(Fnf)(x)
where
Fn(xy)=12(n+1)πsin2(n+12(xy))sin2xy2   (n=0,1,).
Fn(xy) is called the Fejér kernel. Fn(θ) is an even function and has a period of 2π, so Fn(xy) is symmetric. The positive semi-definiteness is inherited from that of the Dirichlet kernel.
For the constant function f(x)=1, Sn1 (n=0,1,) so that σn(x)1, and hence
(eq:Fnormal)ππFn(xy)dy=1.
Unlike Dn, Fn is non-negative:
Fn(xy)0   (x,y).
The following figure shows the Fejér kernels for n=0,1,5,20.


Lemma (Uniform convergence of σn[f])

If fC2π0, then σn[f] converges uniformly to f

Proof. Using (eq:Fnormal), we have
f(x)=ππFn(xy)f(x)dy.
Therefore,
(eq:fms)f(x)σn(x)=ππFn(xy){f(x)f(y)}dy
where the range of integration could be any one period.

For any ε>0, there exists a δ>0 such that 
|xy|<δ|f(x)f(y)|<ε
as f is uniformly continuous. Fixing this δ, we split the right-hand side of (eq:fms) into two parts:
f(x)σn(x)=I1+I2
where
I1=|yx|<δFn(xy){f(x)f(y)}dy,I2=δ|yx|πFn(xy){f(x)f(y)}dy.
Since Fn is non-negative and normalized (eq:Fnormal),
|I1|ε|yx|<δFn(xy)dyε|yx|<πFn(xy)dy=ε.

Next, if |xy|δ, noting that
Fn(xy)12(n+1)π1sin2δ2
and letting M=max|y|π|f(y)|,
|I2|δ|yx|π12(n+1)π1sin2δ22Mdy2M(n+1)sin2δ2.
Thus, for a sufficiently large N1, we have
nN1|I2|<ε.
Putting these together, we have
|f(x)σn(x)|<2ε   (nN1).
Here, N1 does not depend on x, which proves that σn uniformly converges to f with respect to x. ■


Comments

Popular posts from this blog

Birth process

Branching processes: Mean and variance

Informal introduction to formal logic