Jim Lambers MAT 773 Fall Semester 2018-19 Lecture 5 and 6 Notes These notes correspond to Section 1.3 in the text. Convergence Theorems for Fourier Series The Riemann-Lebesgue Lemma Theorem 21 (Riemann-Lebesgue Lemma) Let f be piecewise continuous on [a, b]. Then Z b Z b f (x) sin kx dx = 0. f (x) cos kx dx = lim lim k→∞ a k→∞ a Proof: In the case where f is continuously differentiable on [a, b], we have Z b Z f (b) sin kb − f (a) sin ka 1 b 0 − f (x) cos kx dx = f (x) sin kx dx k k a a which approaches zero as k → ∞. The case of integrating f (x) sin kx is similar. 2 Convergence at a Point of Continuity Let f have the Fourier series F (x) = a0 + of f converges to f at x if lim N →∞ a0 + P∞ N X k=1 ak cos kx + bk sin kx. We say that the Fourier series ! ak cos kx + bk sin kx = f (x). k=1 Theorem 22 Let f be continuous and 2π-periodic. If f 0 (x) exists, then the Fourier series of f at x converges to f (x). P To prove this theorem, we let SN (x) = a0 + N k=1 ak cos kx + bk sin kx be the truncated Fourier series of f (x) on [−π, π]. Using the identity for the cosine of a difference, and the formulas for the Fourier coefficients, this series can be rewritten as SN (x) = a0 + N X ak cos kx + bk sin kx k=1 Z π N Z π π X 1 1 1 f (t) dt + f (t) cos kt dt cos kx + f (t) sin kt dt sin kx 2π −π π −π π −π k=1 ! Z N 1 π 1 X f (t) + cos kt cos kx + sin kt sin kx dt π −π 2 k=1 ! Z N 1 π 1 X + cos(k(t − x)) dt. f (t) π −π 2 Z = = = k=1 1 The following lemma will help us to evaluate the summation. Lemma 23 For u ∈ [−π, π], sin((N + 1/2)u) 1 2 sin(u/2) + cos u + cos 2u + · · · + cos N u = 2 N + 1/2 u 6= 0, u = 0. Proof: The case where u = 0 is straightforward, so we assume u 6= 0. Using the formula for the partial sum of a geometric series, N X rN +1 − 1 rk = , r−1 k=0 we obtain N N k=1 k=0 1 X 1 X + cos ku = − + cos ku 2 2 N X 1 eiku = − + Re 2 1 = − + Re 2 k=0 N X (eiu )k k=0 ei(N +1)u 1 −1 = − + Re 2 eiu − 1 ei(N +1)u/2 ei(N +1)u/2 − e−i(N +1)u/2 1 = − + Re 2 eiu/2 eiu/2 − e−iu/2 sin((N + 1)u/2) 1 = − + Re eiN u/2 2 sin(u/2) 1 sin((N + 1)u/2) = − + cos(N u/2) . 2 sin(u/2) Using a product-to-sum identity, we obtain N 1 X 1 1 (N + 1)u N u (N + 1)u N u + cos ku = − + sin + + sin − 2 2 2 sin(u/2) 2 2 2 2 k=1 1 1 [sin((N + 1/2)u) + sin (u/2)] = − + 2 2 sin(u/2) sin((N + 1/2)u) = . 2 sin(u/2) 2 Substituting the result of Lemma 23 into SN (x) yields Z π Z π 1 sin((N + 1/2)(t − x)) SN (x) = f (t) dt = f (t)PN (t − x) dt 2π −π sin((t − x)/2) −π 2 where PN (u) = 1 sin((N + 1/2)u) . 2π sin(u/2) The following result pertaining to PN (u) will be helpful. Lemma 24 Z π PN (u) du = 1. −π Proof: Because PN (u) is an even function, we have, by Lemma 23, Z π Z π PN (u) du = 2 PN (u) du −π 0 Z 1 π sin((N + 1/2)u) = du π 0 sin(u/2) Z 2 π1 = + cos u + cos 2u + · · · + cos N u du π 0 2 1 1 2 u + sin u + sin 2u + · · · + sin N u = π 2 2 N = 1. π 0 2 To show that SN (x) converges to f (x), we must show Z π [f (x) − f (t)]PN (t − x) dt → 0 −π as N → ∞, from Lemma 23. We have Z π 1 f (x) − f (t) f (x) − SN (x) = sin((N + 1/2)(t − x)) dt. 2π −π sin((t − x)/2) Consider the function g(t) = f (t) − f (x) , sin((t − x)/2) t ∈ [−π, π]. This function is continuous, except possibly at t = x. However, lim g(t) = lim 2 t→x t→x f (t) − f (x) (t − x)/2 = 2f 0 (x). t−x sin((t − x)/2) Therefore, g(t) can be made continuous for all t by defining g(x) = 2f 0 (x). We conclude that Z π 1 f (x) − SN (x) = − g(t) sin((N + 1/2)(t − x)) dt → 0 2π −π as N → ∞ by the Riemann-Lebesgue Lemma. This completes the proof of Theorem 22. 3 Convergence at a Point of Discontinuity Definition 25 The right and left limits of f at x are f (x + 0) = lim f (x + h), h→0+ f (x − 0) = lim f (x − h). h→0+ We say that f is right differentiable if the limit f 0 (x + 0) = lim h→0+ f (x + h) − f (x + 0) h exists, and that f is left differentiable if the limit f 0 (x − 0) = lim h→0+ f (x − 0) − f (x − h) h exists. Example 26 Let f (x) be the periodic extension of y = x on [−π, π]. Then f (x) is discontinuous at x = kπ where k is an integer. We also have f (π + 0) = −π, f (π − 0) = π, and f 0 (π + 0) = 1, f 0 (π − 0) = 1. ( x f (x) = π−x if 0 ≤ x ≤ π/2, if π/2 ≤ x ≤ π. 2 Example 27 Let Then f is continuous on [−π, π] but not differentiable at x = π/2. We have f 0 (π/2 − 0) = 1 and f 0 (π/2 + 0) = −1. 2 Theorem 28 Let f be periodic and piecewise continuous, and let f be left and right differentiable at x. Then the Fourier series of f at x converges to f (x + 0) + f (x − 0) . 2 The proof of this theorem is very similar to that of Theorem 22. Example 29 Let f be the function from Example 26. By Theorem 28, at x = π the Fourier series of f (x) converges to f (π + 0) + f (π − 0) −π + π = = 0. 2 2 This is consistent with the Fourier series of f computed earlier, ∞ X 2(−1)k+1 F (x) = sin kx k k=1 which is zero at x = π. 2 4 Uniform Convergence Theorem 30 Let f be continuous and piecewise smooth on [−π, π]. Then the Fourier series of f on [−π, π] converges uniformly to f on [−π, π]. Proof: We prove this theorem under the weaker assumption that f is twice continuously differentiable. Assume f has the Fourier series ∞ X f (x) = a0 + ak cos kx + bk sin kx. k=1 As in the proof of Theorem 22, we let SN (x) = a0 + N X ak cos kx + bk sin kx k=1 be the truncated Fourier series of f . We then have ∞ X f (x) − SN (x) = ak cos kx + bk sin kx. k=N +1 We also have 00 f (x) = ∞ X (−k 2 )(ak cos kx + bk sin kx). k=1 f 00 Because is continuous, we must have k 2 ak and k 2 bk converge to zero by the Riemann-Lebesgue Lemma. As a convergent sequence is bounded, it follows that |k 2 ak | and |k 2 bk | are both bounded by some number M . We then have ∞ X |ak | + |bk | ≤ k=1 ∞ X 2M k=1 k2 < ∞. The result follows from Lemma 33 below. 2 Example 31 Consider the sawtooth wave from Example 27. Its periodic extension is piecewise smooth. Therefore, by Theorem 30, its Fourier series on [0, π] converges uniformly. 2 Example 32 Consider the Fourier series for f (x) = x on [−π, π]. Its periodic extension is piecewise smooth but discontinuous, so Theorem 30 does not apply. Its Fourier series does not converge uniformly. 2 Lemma 33 Let f (x) = a0 + ∞ X ak cos kx + bk sin kx k=1 with ∞ X |ak | + |bk | < ∞. k=1 Then the Fourier series converges uniformly and absolutely to f (x). 5 P∞ k=N +1 |ak | Proof: Let > 0. Then, there exists N0 > 0 such that for N ≥ N0 , SN (x) = a0 + N X + |bk | < . Let ak cos kx + bk sin kx. k=1 Then, we also have, for N ≥ N0 , |f (x) − SN (x)| = ∞ X ak cos kx + bk sin kx ≤ k=N +1 ∞ X |ak | + |bk | < . k=1 That is, SN (x) converges to f (x). Because the choice of N0 does not depend on x, convergence is uniform. 2 Convergence in the Mean Lemma 34 Let V = L2 ([−π, π]) and let f ∈ V . Let VN = span{1, cos kx, sin kx, . . . , cos N x, sin N x} and fN (x) = a0 + N X ak cos kx + bk sin kx k=1 where ak and bk are the Fourier coefficients of f . Then fN is the orthogonal projection of f onto VN . Theorem 35 Let f ∈ L2 ([−π, π]), and let fN (x) = a0 + N X ak cos kx + bk sin kx k=1 where ak and bk are the Fourier coefficients of f . Then fN converges to f in L2 ([−π, π]). Proof: Let > 0 and let g be a smooth 2π-periodic function that approximates f to within /2; that is, kf − gkL2 < /2. Such a g exists by Lemma 38 below. Let fN and gN be truncated Fourier series for f and g, respectively. Because g is smooth, gN converges uniformly to g on [−π, √ π] by Theorem 30, so there exists an N0 > 0 such that for N ≥ N0 , |g(x) − gN (x)| < /(2 2π) for x ∈ [−π, π], and therefore kg − gN kL2 ≤ /2. Because fN is the closest function from VN to f in the L2 norm, where VN = span{1, cos x, sin x, . . .}, we must have kf − fN kL2 ≤ kf − gN kL2 . Putting all of these results together, we have kf − fN kL2 ≤ kf − gN kL2 ≤ kf − gkL2 + kg − gN kL2 ≤ . 2 6 Theorem 36 Let f ∈ L2 ([−π, π]), and let N X fN (x) = αk eikx k=−N where 1 αk = 2π Z π f (t)e−ikt dt, k∈Z −π are the complex Fourier coefficients of f . Then fN converges to f in L2 ([−π, π]). Example 37 All Fourier series presented in the examples in these notes converge in the mean, as they arise from functions in L2 over an appropriate interval. 2 Lemma 38 Let f ∈ L2 ([−π, π]). Then f can be approximated arbitrarily closely by a smooth 2π-periodic function. Theorem 39 (Parseval’s Identity, Real Version) Let f ∈ L2 ([−π, π]) have the Fourier series ∞ X f (x) = a0 + ak cos kx + bk sin kx. k=1 Then ∞ X 1 kf k2 = 2|a0 |2 + |ak |2 + |bk |2 . π k=1 Theorem 40 (Parseval’s Identity, Complex Version) Let f ∈ L2 ([−π, π]) have the Fourier series ∞ X f (x) = αk eikx . k=−∞ Then ∞ X 1 kf k2 = |αk |2 . 2π k=−∞ Furthermore, ∞ X 1 hf, gi = αk β k , 2π k=−∞ where the βk , k ∈ Z, are the complex Fourier coefficients of g. Proof: Let fN (x) = N X αk eikx , gN (x) = k=−N N X k=−N 7 βk eikx . Then, by the orthogonality of the functions {eikx }k∈Z , we have Z π fN (x)gN (x) dx hfN , gN i = −π Z N X π = −π = αk eikx k=−N N N X X N X ! ! β` ei`x dx `=−N αk βk heikx , ei`x k=−N `=−N N X = 2π αk βk . k=−N We then have |hf, gi − hfN , gN i| = |hf, gi − hf, gN i + hf, gN i − hfN , gN i ≤ |hf, g − gN i| + |hf − fN , gN i| ≤ kf kkg − gN k + kf − fN kkgN k. By Theorem 36, hfN , gN i converges to hf, gi, from which the Theorem follows. 2 One consequence of Parseval’s Identity is Bessel’s Inequality N X |αk |2 ≤ k=−N 1 kf k2 . 2π Example 41 Let f (x) = x on [−π, π]. From Example 29, f has the Fourier series ∞ X 2(−1)k+1 F (x) = k k=1 sin kx. By the real version of Parseval’s identity, ∞ 2 X 2 k=1 k We conclude that 1 = π Z π x2 dx = −π ∞ X 1 π2 = . k2 6 k=1 2 Exercises Chapter 1: Exercises 32, 33, 36, 37, 38, 39 8 2π 2 . 3