Convergence of Fourier Series

Subjects: Fourier Analysis
Links: Hilbert Spaces, Good Kernels and Convergence in Fourier Analysis, Lp spaces

In L2([0,2π])

Let's consider the L2([0,2π]), then we know that the inner product is of the form $$\langle f, g\rangle = \frac{1}{2\pi} \int_0^{2\pi} f(\theta) \overline{g(\theta)} , d\theta$$and the norm $$|f|^2 = \frac{1}{2\pi} \int_0^{2\pi}|f(\theta)|^2, d\theta$$
For each integer n, let en(θ)=einθ, and observe that that the family {en}nZ is orthonormal; that is $$\langle e_n, e_m\rangle = \delta_{n,m}$$with δn,m being the Kronecker delta.

Let f be integrable function on the circle and let an denote its Fourier coefficients. We see that the Fourier coefficients are represented by inner products of f with the elements in the orthonormal set {en}nZ: $$\langle f, e_n\rangle = \frac{1}{2\pi} \int_0^{2\pi} f(\theta) e^{-in\theta}\ d\theta = a_n$$
We see that SN(f)=|n|Nanen. Then the orthonormal property of the family {en} and the fact that f,en imply that f|n|Nanen is orthogonal to en, for all |n|N. There for we have that $$(f- S_N(f)) \perp \sum_{|n| \le N} b_n e_n$$for any complex numbers bn. From this fact we get that, f=(fSN(f))+SN(f), and $$|f|^2 = |f-S_N(f)|^2+\left|\sum_{|n| \le N}a_n e_n\right|^2$$by the orthonormality we have that $$\left|\sum_{|n| \le N}a_n e_n\right|^2 = \sum_{|n| \le N}|a_n|^2$$, we deduce that $$|f|^2 = |f-S_N(f)|^2+\sum_{|n| \le N}|a_n|^2$$
Best Approximation: If f is integrable on the circle with Fourier coefficients an, then $$|f-S_N(f)| \le \left|f-\sum_{|n|\le N} c_n e_n\right|$$for any complex numbers cn. Moreover, equality holds precisely when cn=an for all |n|N.

Th: Let f be an integrable function on the circle with fnZaneinθ. Then we have that:

Riemann-Lebesgue Lemma If f is integrable on the circle, then f^(n)0 as |n|. An equivalent reformulation, if that if f is integrable in [0,2π], then $$\int_0^{2\pi} f(\theta) \sin(N\theta), d\theta \to 0 \qquad N \to \infty$$and $$\int_0^{2\pi} f(\theta) \cos(N\theta), d\theta \to 0 \qquad N \to \infty$$
Cor: Let f be a periodic and of class Ck, then $$\hat f(n) = o(1/|n|^k)$$

Lemma: suppose that F and G are integrable on the circle with $$F \sim \sum a_m e^{in\theta} \quad \text{and}\quad G \sim \sum b_n e^{in\theta}$$
Then $$\frac{1}{2\pi} \int_0^{2\pi} F(\theta) \overline{ G(\theta)}, d\theta = \sum_{n = -\infty}^\infty a_n \overline{b_n}$$
and if we a make F and G be L periodic, we get that $$\frac{1}{L} \int_0^L F(\theta) \overline{ G(\theta)}, d\theta = \sum_{n = -\infty}^\infty a_n \overline{b_n}$$

With this we have more things to prove the absolute convergence and hence uniform convergence.

We say that f satisfies a Hölder condition of order α, namely $$|f(x)-f(y)| \le C|x-y|^\alpha $$for some 0<α1, and some C>0 and all x,y. Then $$\hat f(n) = O(1/|n|^\alpha)$$We note that for α=1, then f is Lipschitz continuous.

Bernstein Theorem: If f satisfies a Hölder condition of order α>1/2, then the Fourier series of f converges absolutely.

Pointwise Convergence

Th: Let f be an integrable function on the circle which is differentiable at a point θ0. Then SN(f)(θ0)f(θ0) as N.

We can weaken further if we only assume that f satisfies a Lipschitz condition at θ0; that is $$|f(\theta) - f(\theta_0) \le |\theta - \theta_0|$$For some M0 and all θ.

Riemann Localisation Theorem: Suppose f and g are two integrable functions defined on the circle, and for some θ0 there exists an open interval containing θ0 such that $$f|_I = g|_I$$then SN(f)(θ0)SN(g)(θ0)0 as N.

We also have that near a discontinuity, we have that the Fourier series tend to overshoot/undershoot, by around 9% of the jump at this is called Gibb's phenomenon