8 Fourier Series Our aim is to show that under reasonable assumptions a given 2π-periodic function f can be represented as convergent series ∞ a0 X (an cos nx + bn sin nx) . + 2 n=1 f (x) = (8.1) By definition, the convergence of the series means that the sequence (sn (x)) of partial sums, defined by n sn (x) = a0 X + (ak cos kx + bk sin kx) , 2 k=1 converges at a given point x to f (x), sn (x) → f (x). We start with some definitions. A real valued function defined on R is said to be periodic with period L, or L-periodic, is f (x + L) = f (x) for all x. In this case f is completely determined by its values on any interval [a, a+L). A function f defined on [a, a+L) can be extended in a unique way to a periodic function on R. Indeed, given x, there exists a unique integer n so that x belongs to [a + nL, a + (n + 1)L) so that x − nL ∈ [a, a + P ) and we set f (x) = f (x − nL). A function f on [a, b] is piecewise continuous if f is continuous except a finite numbers of points and at each such point the one-sided limits f (x+ ) = limε→0− f (x + ε) and f (x− ) = limε→0− f (x + ε) exist and are finite. We say that f is piecewise smooth (or piecewise differentiable) on [a, b] if f and f ′ are piecewise continuous on [a, b]. If f is piecewise continuous, then it is Riemann integrable over any bounded interval contained in its domain. In addition, if f is L-periodic, then Z L Z a+L f x) dx f (x) dx = 0 a for every a. In what follows we focus on functions which are 2π-periodic. Assume ∞ a0 X (an cos nx + bn sin nx) . + f (x) = 2 n=1 We also assume that the convergence is well-behaved so that we can integrate term-by-term. We want to compute the coefficients ak and bk . We use 66 integral identities 0, cos mx cos nx dx = 2π, −π π, ( Z π 0, sin mx sin nx dx = π, −π Z π cos mx sin nx dx = 0. Z π m 6= n m=n=0 m = n 6= 0 m 6= n m = n 6= 0 −π To compute a0 , we multiply both sides of (8.1) by the constant function 1 and integrate over [−π, π] and get Z Z π Z π Z ∞ X a0 π f (x) dx = sin nx dx cos nx dx + bn 1 dx + an 2 −π −π −π −π n=1 a0 · (2π) = πa0 = 2 π To compute am with m ≥ 1 we we multiply both sides of (8.1) by cos mx and integrate over [−π, π] to get Z π Z a0 π f (x) cos mx dx = cos mx dx 2 −π −π Z π Z π ∞ X sin nx cos mx dx + cos nx cos mx dx + bn an = am −π −π n=1 Z π cos mx cos mx dx = πam . −π Similarly, multiplying both sides of (8.1) by sin mx and integrating over [−π, π], we find that Z π f (x) sin mx dx = πbm . −π Summing up, Z 1 π an = f (x) cos nx dx, π −π Z 1 π f (x) sin nx dx, bn = π −π 67 n≥0 (8.2) n ≥ 1. The coefficients an and bn are called the Fourier coefficients of f and the series ∞ a0 X [an cos nx + bn sin nx] + 2 n=1 the Fourier series of f . We denote this fact by ∞ f (x) ∼ a0 X [an cos nx + bn sin nx] . + 2 n=1 The symbol ∼ should be read as f “has Fourier serier”. Example 8.1. Let f be a 2π-periodic function given by f (x) = x for x ∈ (−π, π]. Then integrating by parts π Z Z π 1 π 1 x sin nx 1 an = − x cos nx dx = sin nx dx = 0 π −π π n nπ −π −π and 1 bn = π Z π 1 1 h x cos nx iπ − + x sin nx dx = cos nx dx π n nπ −π −π −π Z π =− 2(−1)n+1 2 cos nπ = . n n Consequently, f (x) ∼ 2 X (−1)n+1 sin nx. n n≥1 The series converges for every x in view of the following Dirichlet’s test. If (an ) is a decreasing sequence converging to 0 and P P the sequence (sn ) of partial sums of 1the series bn is bounded, then the series an bn converges. In our case, an = n is decreasing and converges to 0, (−1)n sin nx = sin n(x + π), and n (n + 1)ϑ nϑ ϑ X sin kϑ = sin · sin sin · 2 2 2 k=1 which implies that if ϑ 6= 2πk for all k ∈ Z. n X 1 sin kϑ ≤ sin ϑ2 k=1 Example 8.2. Consider f be a 2π-periodic function defined by f (x) = |x| for x ∈ [−π, π]. Note that f is even. Since the product of even function with the odd function is odd, it follows that Z π f (x) sin mxdx = 0. −π 68 Hence bn = 0 for all n ≥ 1. To compute bn note that the product of an even function with an even function is even so that Z Z Z 1 π 2 π 2 π an = f (x) cos nx dx = f (x) cos nxdx = x cos nx dx. π −π π 0 π 0 If n = 0, then a0 = 2 π Z π x dx = π, 0 and if n ≥ 1, then integrating by parts, one finds that π Z Z π 2 2 x sin nx 2 π − x cos nx dx = sin nx dx π 0 π n nπ 0 0 π 2 2 = 2 cos nx = 2 [(−1)n − 1] . n π n π 0 Hence an = 0 if n is even and an = − n24π when n is odd, and hence f (x) ∼ 4 X cos(2n − 1)x π − . 2 π (2n − 1)2 n≥1 Since the series series converges. P 1 n2 converges, the M-Weierstarss test implies that the above In the above examples we have used the following identities, ( RL Z L 2 0 g(x) dx if g is even g(x) dx = 0 if g is odd. −L Using that the sum of two even functions is even and the sum of two odd functions is odd, that the product of two even functions or two odd functions is even, and the product of an even function and an odd function is odd, we have the following proposition. Proposition 8.3. (i) If f is even, then its Fourier sine coefficients bn are equal to 0 and f is represented by Fourier cosine series ∞ f (x) ∼ where 2 an = π Z a0 X an cos nx + 2 n=1 π f (x) cos nx dx −π 69 for n ≥ 0. (ii) If f is odd, then its Fourier cosine coefficients an are equal to 0 and f is represented by Fourier sine series f (x) ∼ ∞ X an sin nx n=1 where 8.0.1 2 bn = π Z π f (x) sin nx dx for n ≥ 1. −π Convergence Theorem Theorem 8.4. Assume that f is a 2π-periodic function which is piecewise smooth. Then n a0 X 1 − + [ak cos kx + bk sin kx] + f (x ) + f (x ) = 2 2 k=1 where f (x± ) = limy→x± f (y). In particular, if f is continuous at x, then n a0 X f (x) = + [ak cos kx + bk sin kx] . 2 k=1 Example 8.5. Consider the function f from Example 8.1. Then f is smooth at all points except points mπ where m is odd. The one sided limits at these points are f (mπ − ) = lim f (x) = π x→mπ − and f (mπ + ) = lim f (x) = −π x→mπ + provided that m is odd. Consequently, the Fourier series converges of f converges to f at every point except points mπ with m odd. At these points the Fourier series converges to 21 [f (mπ − ) + f (mπ + )] = 0. In particular, ∞ x X (−1)n+1 = sin nx, 2 n=1 n for x ∈ (−π, π). Taking x = π/2, we find that ∞ X (−1)n−1 π = . 4 2n − 1 n=1 70 Example 8.6. Consider the function 2π-periodic function f given by f (x) = |x| for x ∈ (−π, π). (See Example 8.2.) Then f is continuous at every point and it is smooth except points mπ with m odd. Hence the Fourier series of f converges to f at every point. In particular, |x| = ∞ π 4 X cos(2n − 1)x − 2 π n=1 (2n − 1)2 for x ∈ [−π, π]. Substituting x = 0, we find that ∞ X 1 π2 . = 8 (2n + 1)2 n=0 Set S = P∞ 1 n=1 n2 . Then we note that ∞ ∞ X S 1 1X 1 = = 2 4 4 n=1 n2 (2n) n=1 and hence ∞ ∞ ∞ X X X 1 1 1 π2 S 3 − = = S=S− = . 2 2 2 4 4 n (2n) (2n + 1) 8 n=1 n=0 n=1 From this we conclude that S= ∞ X 1 π2 = . 2 n 6 n=1 Integration of Fourier series We start with the following observation. R x Assume that f is 2π-periodic, piecewise continuous, and let F (x) = 0 f (y) dy. Then F is 2π periodic if Rπ and only if −π f (y)dy = 0. Indeed, we have F (x + 2π) − F (x) = Z x+2π f (y)dy = x Z π f (y)dy. −π This means that R π the constant term in the Fourier series of f is equal to 0. 1 Since a20 = 2π −π f (y) dy, the number a0 /2 is the mean of the function f over the interval [−π.π]. Theorem 8.7. Assume that f is 2π-periodic and piecewise continuous and its mean is equal to 0. Then its Fourier series X f (x) ∼ [an cos nx + bn sin nx] n≥1 71 can be integrated term by term and produce the Fourier series Z x X bn an sin nx f (y) dy ∼ C0 + − cos nx + F (x) = n n 0 n≥1 Rπ 1 where the constant C0 = 2π −π F (y) dy. Example 8.8. Consider a 2π-periodic function f given by f (x) = x on (−π, π]. (See Example 8.1.) Since f is odd, its mean value over [−π, π] is equal to 0. Its Fourier series is given by X (−1)n−1 sin nx, for x ∈ (−π, π]. x=2 n n≥1 Integrating term by term we find that X (−1)n−1 x2 π2 cos nx = −2 2 6 n2 for x ∈ (−π, π] n≥1 where π 2 /6 = 1 2π Rπ −π (x2 /2) dx. Differentiation of Fourier series Proposition 8.9. Assume that f is 2π-periodic, continuous, and piecewise smooth. Abbreviate by an , bn the Fourier coefficients of f and by a′n , b′n the Fourier coefficients of f ′ . Then a′n = nbn and b′n = −nan . Proof. Integrating by parts, a′n = 1 π Z π f ′ (x) cos nx dx = −π Similarly, b′n = −nan . Z iπ 1h n π f (x) cos nx + f (x) sin nx dx = nbn . π π −π −π As a consequence, we get: Theorem 8.10. Let f be 2π-periodic, continuous, and piecewise smooth. In addition, assume that f ′ is piecewise smooth. If ∞ then a0 X f (x) = [an cos nx + bn sin nx] , + 2 n=1 ∞ i X 1h ′ − ′ + [nbn cos nx − nan sin nx] . f (x ) + f (x ) = 2 n=1 In particular, at points x at which f ′ (x) exists, the series converges to f ′ (x). 72 Proof. By Proposition 8.9, f ′ (x) ∼ ∞ X [nbn cos nx − nan sin nx] , n=1 and since f ′ is piecewise smooth, the theorem follows from Theorem 8.4. Example 8.11. Recall the function 2π-periodic function f given by f (x) = |x| for x ∈ (−π, π). (See Example 8.2 and Example 8.6.) The function f is continuous and piecewise smooth, and |x| = ∞ 4 X cos(2n − 1)x π , − 2 π n=1 (2n − 1)2 x ∈ [−π, π]. The derivative of f is given by d |x| = dx ( 1, −1, x > 0, x < 0. In view of Theorem 8.10, ∞ 4 X sin(2n − 1)x = π n=1 (2n − 1) ( 1, −1, 0 < x < π, −π < x < 0. At x = 0, the series converges to 12 [f ′ (x− ) + f ′ (x+ )] = 12 [−1 + 1] = 0. 8.0.2 Absolute and Uniform Convergence Theorem 8.12. Let f be 2π-periodic, continuous, and piecewise smooth. Then its Fourier series converges absolutely and uniformly. Proof. We prove the result under additional assumption that f ′ is piecewise smooth. This means that f ′′ is piecewise continuous. Abbreviate by an , bn the Fourier coefficients of f and by a′n , b′n the Fourier coefficients of f ′ . Since f is continuous, ∞ f (x) = a0 X [an cos nx + bn sin nx] . + 2 n=1 To prove the absolute convergence it suffices to show that |an | , |bnP | ≤ M/n2 for some constant M independent of n. Then, by the comparison test, n≥1 |an | and P n≥1 |bn | converge which imply the absolute convergence of the Fourier series. Also, the Weierstrass M-test implies the uniform convergence. By Proposition 8.9, an = −b′n /n. That is, Z Z π 1 1 π f (x) cos nx dx = − f ′ (x) sin nx dx. an = π −π nπ −π 73 By assumption, f ′ is piecewise continuous and so f ′′ is continuous except a finite number of point at which it has a jump. If (a, b) is an interval on which f ′′ is continuous, then 1 nπ b Z b 1 1 ′ f (x) cos nx − f ′′ (x) cos nx dx 2π n2 π n −π a a Z b 1 1 − + f ′′ (x) cos nx dx. = 2 f (b ) cos nx − f (a ) cos nx + 2 n π n π a Z π f ′ (x) sin nx dx = Since |f ′ (x)| , |f ′′ (x)| ≤ K for some constant K and all x (at points x where f ′ and f ′′ has a jump discontinuity, f ′ (x) one has to take left- and hand-side derivatives). So, Z K(2 + (b − a) 1 π ′ 2K(1 + π 4K ≤ ≤ 2. f (x) sin nx dx ≤ 2 2 nπ −π n π n π n Rπ ′ ′′ Since f has a finite number of discontinuities, the integral −π f (x) sin nx dx can Rb be written as the finite number, say m, of integrals a f ′ (x) sin nx dx over intervals on which f ′′ is continuous. Then Z 4Km 1 π ′ |an | = . f (x) sin nx dx ≤ nπ −π n2 Similarly, |bn | ≤ 4Km n2 . Theorem 8.13. Let f be 2π-periodic. Assume that f is of class C k−1 and that f (k−1) is piecewise smooth. If the Fourier coefficients satisfy |an | ≤ C nk+α and |bn | ≤ C nk+α for some α > 1 and C > 0, then f is of class C k . (k) Proof. Let a(j) series of f (j) . Then using n and bn are coefficients of the Fourier (j) (j) Proposition 8.9 an = nj |an | if j is even and an = nk |bn | if j is odd. Similarly, (k) for bn . Observe, using the assumption, that if j ≤ k, then X X n j an ≤ C n≥1 n≥1 1 nk−j+α ≤C X 1 nα n≥1 P for j ≤ k. By the Weierstrass M-test, the series n≥1 nj an converges absolutely P and uniformly for j ≤ k. Similarly, the series n≥1 nj bn converges absolutely P j) j) and uniformly for j ≤ k Hence, f (j) = n≥1 [an cos nx + bn sin nx] for j ≤ k and f (k) is continuous. 74 8.1 Changing a scale So far we have considered 2π-periodic functions and their Fourier series over the interval [−π, π]. Now, given a function f defined on [−L, L], we define L F (y) = f π y . Then F is defined on [−π, π] and its Fourier series is given by a0 X an cos ny + bn sin ny F (y) ∼ + 2 n≥1 where an = 1 π Z π F (y) cos ny dy and bn = −π 1 π Z π F (y) sin ny dy. (8.3) −π Since f (x) = F ( Lπ x) for x ∈ [−L, L], we deduce that f (x) ∼ a0 X nπ nπ an cos + x + bn sin x 2 L L n≥1 The coefficients an and bn can be computed by integrating (8.3) by substitution x = Lπ y. We get Z Z L 1 π 1 π y cos ny dy F (y) cos ny dy = f an = π −π π −π π (8.4) Z 1 L nπx = dx f (x) cos L −L L and similarly bn = 1 π Z π F (y) sin ny dy = −π 75 1 L Z L −L f (x) cos nπx dx. L (8.5)