Convergence of Fourier Series Basic Results From Orthogonality Outline The following three results are valid in any finitedimensional inner product space, and the infinitedimensional spaces `2 (Z), and R. I. Key Concepts • The Pythagorean Theorem: If X and Y are orthogonal, then II. Summary III. Basic Results From Orthogonality kX + Y k2 = kXk2 + kY k2 IV. Special Infinite-Dimensional Vector Spaces • The Cauchy-Schwartz Inequality: V. Approximation Theorems |hX, Y i| = kXkkY k VI. Mean-Square Convergence • The Triangle Inequality VII. Proof of Mean-Square Convergence VIII. Other Results kX + Y k ≤ kXk + kY k Key Concepts Special Infinite Dimensional Vector Spaces • Orthogonality • The space of all (double-sided) sequences • Completeness (. . . , a−n , . . . , a−1 , a0 , a1 , . . . , an , . . . ) • Hilbert Space such that Summary ∞ X |an |2 < ∞ n=−∞ The mean-square of a Riemann integrable function defined on [a, b] is the integral Z b 1 |f (t)|2 dt b−a a – Addition & Scalar Multiplication: {an } + c{bn } = {an + cbn } – Inner Product & Norm: This short chapter is devoted to answering the question of whether the Fourier series of an integrable function converges in the mean-square sense: Z π 1 |f (t) − SN (f )(t)|2 dt → 0 2π −π han , bn i = ∞ X an b n n=−∞ kan k = ∞ X n=−∞ – Denoted by `2 (Z) as N → ∞. 1 !1/2 2 |an | Mean-Square Convergence – A Hilbert space • The space of all complex-valued Riemann integrable functions on the circle Let f be a Riemann integrable function on P∞ the circle with f (θ) ∼ n=−∞ an e−inθ . Then we have: – Addition & Scalar Multiplication: (f + cg)(x) = f (x) + cg(x) (a) Mean-square convergence of the Fourier series: Z π 1 |f (t) − SN (f )(t)|2 dt → 0 2π −π – Inner Product & Norm: Z 2π 1 f (θ)g(θ) dθ hf, gi = 2π 0 Z 2π 1/2 1 2 kf k = kf (θk 2π 0 as N → ∞. (b) Parseval’s identity: – Denoted by R 1 2π – A pre-Hilbert space Z π 2 |f (θ)| dθ = −π ∞ X |an |2 n=−∞ Parseval’s identity is a consequence of the meansquare convergence of the Fourier series. Approximation Theorems • Best Approximation: Proof of Mean-Square Convergence For any sequence of complex numbers {cn }|n|≤N , we have Fix > 0. Since f is integrable on the circle, there exists a continuous function g such that kf − SN (f )k ≤ f − X cn einθ sup |g(θ| ≤ sup |f (θ| = B |n|≤N θ∈[0,2π] and – Proved from the observation that for all n ∈ Z [f − SN (f )] ⊥ en . (Here en (x) = e inx θ∈[0,2π] π 2 π |f (θ) − g(θ)| dθ < 9B −π Z This continuous function will be the key to the proof. We want to prove that there exists N ∈ N such that n ≥ N implies kf − SN k < . The answer is the degree of the trigonometric polynomial P such that kg − P k < /3. Since g is continuous ) 2 on the circle, such a polynomial exists: Z 2π 1 2 kf − gk = |f (θ) − g(θ)|2 dθ 2π 0 Z B 2π |f (θ) − g(θ)| dθ ≤ π 0 2 < 9 So, if M is the degree of P , then N > M implies eiN θ + P kf − SN k ≤ f − 3 ≤ kf − P k + 3 ≤ kf − gk + kg − P k + 3 ≤ + + = 3 3 3 Notice how I played with the epsilons! The best approximation lemma tells us that the first equality holds if P and SN have the same degree, which isn’t the case. So I had to add that extra (/3)eiN θ to make the degrees match and be able to use the lemma. But I couldn’t have done this without g approximating f and P approximating g. It feels amazing being able to approximate a function. I feel like a god. This concludes the proof. Other Results • Riemann-Lebesgue Lemma • Generalized Parseval’s Identity There’s actually two more sections in the chapter. It’s about pointwise convergence, and an example of a continuous function with a diverging Fourier series. I haven’t read them yet tho. 3