8.5 FOURIER SERIES 8.3 101 THE TRIGONOMETRIC FUNCTIONS • skipped 8.4 THE ALGEBRAIC COMPLETENESS OF THE COMPLEX FIELD • skipped 8.5 FOURIER SERIES 8.9 Orthogonal Functions, Orthonormal: Let { of complex functions on [a, b], such that Z Then { n} for all n, { n }, n = 1, 2, 3, . . . , be a sequence b n (x) m (x) dx n 6= m. = 0, a is said to be an orthogonal system of functions on [a, b]. In addition, if Z n} b a | n (x)| 2 dx = 1 is said to be orthonormal. Example (a) (2⇡) 1/2 e inx , n = 0, ±1, ±2, . . . , is an orthonormal system on [ ⇡, ⇡]. (b) 1, cos x, sin x, cos 2x, sin 2x, . . . form an orthogonal system on [ ⇡, ⇡]. (c) The sequence 1 cos x sin x cos 2x sin 2x p , p , p , p , p ,..., ⇡ ⇡ ⇡ ⇡ 2⇡ form an orthonormal system on [ ⇡, ⇡]. 8.10 Fourier Series: Let { n} cn = be orthonormal on [a, b]. Put Z b f (t) n (t) dt, n = 1, 2, 3, . . . . a We call cn the nth Fourier coefficient of f relative to { f (x) ⇠ 1 X cn n }. We write n (x) n=1 and call this series the Fourier series of f (relative to { n }). 102 8 SOME SPECIAL FUNCTIONS 8.11 Theorem Let { n} be orthonormal on [a, b]. Let sn = n X cm m (x) m=1 be the nth partial sum of the Fourier series of f , and suppose tn (x) = n X m m (x). m=1 Then Z Z b a sn |2 dx |f b a tn |2 dx, |f and equality holds if and only if m Proof Since { n} = cm , m = 1, . . . , n. is orthonormal, elementary calculations give ! Z b Z b n X f tn dx = f· dx m m a a = n X m=1 b Z m m=1 f m a dx = n X cm m, m=1 and Z b a |tn |2 dx = = Z Z b tn tn dx a b a ! n X m m (f tn ) · (f m=1 n X · m m m=1 ! dx = n X m=1 | m| 2 . Hence Z b a |f tn |2 dx = = = = Z Z Z Z b a b 2 |f | dx a b |f |2 dx a b a |f |2 dx Z tn ) dx b f tn dx a n X m=1 n X m=1 cm m Z b f tn dx + a n X cm m=1 |cm |2 + n X m=1 | m + Z b |tn |2 dx a n X m=1 m cm |2 . | m| 2 8.5 FOURIER SERIES In particular, if m = cm , we have Z These give Z Z b sn |2 dx = |f a b a 103 |f tn |2 dx = Z b |f |2 dx a a sn |2 dx + f (x) ⇠ n X | m=1 The results follow this inequality. 8.12 Theorem (Bessel Inequality) If { |cm |2 . m=1 b |f n X m cm |2 . n} is orthonormal on [a, b], and if 1 X cn n (x), n=1 then the Bessel inequality holds: 1 X n=1 |cn |2 Z b |f (x)|2 dx. a In particular, lim cn = 0. n!1 Proof In the proof of Theorem 8.11, we have 0 Z b a |f sn |2 dx = Z n X b a |f |2 dx m=1 |cm |2 . Let n ! 1, we have the Bessel inequality. 8.13 Trigonometric Series: We shall limit our attention to the orthonormal system {(2⇡) 1/2 einx } on [ ⇡, ⇡]. Consider functions f that have period 2⇡ and that are Riemann-integrable on [ ⇡, ⇡]. The Fourier series of f is given by f (x) ⇠ where cn = 1 2⇡ Z 1 X cn einx , n= 1 ⇡ f (x)e inx dx. ⇡ Let sN (x) = sN (f ; x) = N X n= N be the N th partial sum of the Fourier series of f . cn einx 104 8 SOME SPECIAL FUNCTIONS For the convenience of argument, we introduce the Dirichlet kernel N X DN (x) = einx = e iN x =e iN x n= N [1 + eix + · · · + e2iN x ] · sin(N + 12 )x e2iN x+ix = . 1 eix sin( x2 ) 1 By the definition, N X sN (f ; x) = 1 2⇡ n= N 1 = 2⇡ 1 2⇡ = Z Z Z ⇡ ⇡ f (t) ⇡ f (t)e int dteinx ⇡ N X ein(x t) dt n= N ⇡ f (t)DN (x t) dt. ⇡ For f being 2⇡-periodic, the last integral is the same as Z ⇡ 1 sN (f ; x) = f (x t)DN (t) dt. 2⇡ ⇡ 8.14 Theorem Suppose f is Riemann-integrable on [ ⇡, ⇡] and 2⇡-periodic. If, for some x, there are constants > 0 and M < 1 such that |f (x + t) for all t 2 ( f (x)| M |t| , ), then lim sN (f ; x) = f (x). N !1 Proof Since 1 2⇡ Z ⇡ DN (x) dx = ⇡ we have sN (f ; x) 1 f (x) = 2⇡ = 1 2⇡ If we put g(t) = Z Z 1 2⇡ Z ⇡ N X ⇡ f (x t)DN (t) dt ⇡ ⇡ [f (x ⇡ 8 > < f (x > :0, t) einx dx = 1, ⇡ n= N f (x)] · t) f (x) , sin(t/2) 1 2⇡ Z ⇡ f (x)DN (t) dt ⇡ sin(N + 12 )t dt. sin(t/2) 0 < |t| ⇡, t = 0, 8.5 FOURIER SERIES 105 by the hypothesis, it is clear that g is continuous at t = 0, so that g is Riemannintegrable. From Z ⇡ Z ⇡ 1 1 sN (f ; x) f (x) = [g(t) cos(t/2)] sin N t dt + [g(t) sin(t/2)] cos N t dt, 2⇡ 2⇡ ⇡ ⇡ by Theorem 8.12, we know that both integrals tends to 0 as N ! 1. 8.14’ We can similarly prove a weaker result: Theorem Suppose f is Riemann-integrable on [ ⇡, ⇡] and 2⇡-periodic. If, for some x, both f (x ) and f (x+) exist, and there exists a constants > 0 and M < 1 such that |f (x + t) f (x+)| M t, |f (x t) f (x )| M t, for all 0 < t < , then lim sN (f ; x) = N !1 f (x+) + f (x ) . 2 • Corollary: If f (x) = 0 for all x in some segment J, then lim sN (f ; x) = 0 for every x 2 J. This result show that the behavior of {sN (f ; x)} depends only on the values of f in the neighborhood of x. This is very di↵erent from power series. 8.15 Theorem Suppose f is continuous and 2⇡-periodic. For any ✏ > 0, there is a trigonometric polynomial P such that |P (x) f (x)| < ✏ for all real x, where a trigonometric polynomial is a finite sum of the form a0 + N X (an cos nx + bn sin nx). n=1 Proof Consider ✓ ◆n 1 1 + cos x un (x) = , n = 1, 2, 3, . . . , ↵n 2 ◆n Z ⇡ ✓ 1 + cos x = dx. Then un (x) 0, un (x + 2⇡) = un (x), and 2 ⇡ Z where ↵n Z ⇡ un (x) dx = 1. We claim that for any fixed ⇡ 2 (0, ⇡), |x|⇡ un (x) dx ! 0 as n ! 1. In fact, since 1 + cos x is decreasing on [0, ⇡], for any x 2 [ , ⇡] and y 2 [0, /2], we have 1 + cos x 1 + cos , 1 + cos y 1 + cos( /2) 106 8 SOME SPECIAL FUNCTIONS so that 1 + cos x < r(1 + cos y), 1 + cos < 1. Taking powers and dividing by ↵n , we have un (x) 1 + cos( /2) n r un (y). This implies Z /2 Z /2 un (x) = un (x) dy rn un (y) dy rn ! 0, 2 0 0 where r = as Z n ! 1. Thus un (x) ! 0 uniformly on [ , ⇡] as n ! 1. By Theorem 7.16, un (x) dx ! 0. |x|⇡ By Euler’s formula eix = cos x + i sin x, we see that un (x) = n X me imx for some m= n constants m. For f continuous and 2⇡-periodic, if we put Z ⇡ Pn (x) = un (y)f (x y) dy, ⇡ then Pn is a trigonometric polynomial for each n, since Z x ⇡ Pn (x) = un (x z)f (z)( 1) dz = = = Z x+⇡ x+⇡ n X me x ⇡ m= n Z ⇡ X n me ⇡ m= n n X me m= n imx im(x z) im(x z) Z f (z) dz f (z) dz ⇡ f (z)e inz dz. ⇡ Let ✏ > 0 be given. We shall prove that there is N such that n |Pn (x) f (x)| < ✏, for all x 2 [ ⇡, ⇡]. Then the assertion of the theorem follows. Since f is uniformly continuous on [ 2⇡, 2⇡], there exists x 2 [ ⇡, ⇡] and y with |y| < , |f (x Put M = that n sup 1<y<1 N implies y) N implies > 0 such that for all f (x)| < ✏/2. |f (y)| < 1. Then, from the discussion above, there is N such Z un (y) dy < ✏/(4M ). |y|⇡ 8.5 FOURIER SERIES Thus, for n |Pn (x) N, f (x)| = = Z Z Z Z ⇡ un (y)f (x y) dy ⇡ ⇡ ⇡ un (y) |f (x |y|< Z un (y) |f (x y) ⇡ un (y)f (x) dy ⇡ f (x)| dy Z f (x)| dy + un (y) |f (x |y|⇡ Z un (y) dy + 2M un (y) dy y) ✏ · 2 |y|< ✏ ✏ < · 1 + 2M · = ✏. 2 4M < 107 y) f (x)| dy |y|⇡ • piecewise continuous, piecewise di↵erentiable, continuously di↵erentiable: A function f is said piecewise continuous on [a, b], if there is a partition P = {x0 , . . . , xn } of [a, b] such that f is continuous on each (xi 1 , xi ). Similarly, a function f is said piecewise di↵erentiable on [a, b] if f 0 exists on each (xi 1 , xi ). A function f is said continuously di↵erentiable on [a, b] if f 0 is continuous on [a, b]. 8.15U Theorem Suppose f is continuous, 2⇡-periodic, and piecewise continuously di↵erentiable. Then sN converges to f uniformly. Proof Without loss of generality, we assume that f 0 has only one discontinuity at a 2 [ ⇡, ⇡]. In the case of more discontinuities, the proof is similar. We expand f as a Fourier series: f (x) ⇠ with sN (f ; x) = N X n= N cn e inx 1 X cn einx , n= 1 1 and cn = 2⇡ Z ⇡ f (x)e inx dx. ⇡ The proof is divided into two steps: first we prove that the series 1 X cn einx n= 1 converges uniformly to a function, say S(x); next we prove f (x) = S(x) for all x. To prove the P Fourier series converges uniformly, by Theorem 7.10, we only need to prove that |cn | < 1. Since f is continuously di↵erentiable on ( ⇡, a) and (a, ⇡), we integrate by parts on each of these intervals: Z a Z ⇡ 2⇡cn = f (x)e inx dx + f (x)e inx dx ⇡ a " # Z a a e inx e inx 0 = f (x) f (x) dx in x= ⇡ in ⇡ Z ⇡ ⇡ e inx e inx + f (x) f 0 (x) dx . in x=a in a 108 8 SOME SPECIAL FUNCTIONS Put 1 2⇡ = n Z ⇡ f 0 (x)e inx dx. ⇡ Since f ( ⇡) = f (⇡), the above calculation gives n cn = in n = ±1, ±2, ±3, . . . . , By Bessel’s inequality for f 0 , we have N X n= N 2 n| | 1 2⇡ Z ⇡ ⇡ |f 0 (x)|2 dx. By the Schwarz inequality (Theorem 1.35), X 1|n|N which implies say S(x). |cn | = P X 1|n|N 1 | n n| 0 @ 0 @ |cn | converges. Thus, X 1|n|N X 1|n|N P 11/2 0 1A @ X n2 1|n|N 11/2 Z 1A 1 · n2 2⇡ ⇡ ⇡ | n| 11/2 2A |f 0 (x)|2 dx, cn einx converges uniformly to a function, Next we prove that f (x) = S(x). In fact, since 1 X cn einx converges to S(x) uni- 1 formly, by Theorem 7.12, S is continuous. Notice that, by Theorem 8.11, the value of the integral Z ⇡ |f (x) sN (f ; x)|2 dx ⇡ decreases as N increases. Hence Z lim N !1 ⇡ ⇡ |f (x) sN (f ; x)|2 dx exists. By the first step, we know that sN converges uniformly to S, thus, for any N, Z ⇡ Z ⇡ Z ⇡ 2 2 |f (x) S(x)| dx = lim |f (x) sN (f ; x)| dx |f (x) sN (f ; x)|2 dx. N !1 ⇡ ⇡ ⇡ This implies that, by Theorem 8.11, for any trigonometric polynomial P (x), Z ⇡ Z ⇡ |f (x) S(x)|2 dx |f (x) P (x)|2 dx. ⇡ ⇡ 8.5 FOURIER SERIES 109 Let ✏ > 0 be given. By Theorem 8.15, there is a trigonometric polynomial P such that |P (x) f (x)| < ✏ for all x. Thus, Z ⇡ |f (x) S(x)|2 dx < 2⇡✏2 , ⇡ which implies the value of the integral is zero. We conclude that f (x) = S(x) for all x, since f (x) and S(x) are continuous. 8.16 Theorem (Parseval’s Theorem) Suppose f and g are 2⇡-periodic Riemannintegrable function, and f (x) ⇠ Then 1 X cn einx , g(x) ⇠ n= 1 1 N !1 2⇡ lim Z 1 X ne inx . n= 1 ⇡ ⇡ |f (x) sN (f ; x)|2 dx = 0. Moreover, the following equalities hold: 1 2⇡ Z and 1 2⇡ ⇡ Z f (x)g(x) dx = ⇡ 1 X cn n, n= 1 ⇡ ⇡ |f (x)|2 dx = 1 X n= 1 |cn |2 . Proof Let ✏ > 0 be given. We show that for 2⇡-periodic Riemann-integrable function f , there is a function h that is continuous, 2⇡-periodic and piecewise di↵erentiable such that Z ⇡ 1 |f (x) h(x)|2 dx < ✏. 2⇡ ⇡ In fact, we only prove this assertion for real f . If f is complex, we just need to apply the conclusion for real and imaginary parts separately. Since f is Riemann-integrable, by Theorem 6.6, there is a partition P = {x0 , . . . , xn } of [ ⇡, ⇡], such that U (P, f ) L(P, f ) = n X (Mi mi ) X i < n=1 where Mi = sup xi 1 xxi f (x), mi = xi inf 1 xxi f (x), and M = xi x f (xi xi xi 1 1) sup ⇡x⇡ define a piecewise linear (continuous) function on [xi h(x) = ⇡✏ , M + x xi 1 , xi ] xi xi 1 1 by f (xi ). |f (x)|. Now we 110 8 SOME SPECIAL FUNCTIONS The facts that h(xi 1 ) = f (xi 1 ), h(xi ) = f (xi ), and f is 2⇡-periodic imply that h can be continuously extended to the whole real line. We still denote the extended function as h. Hence we have a continuous, 2⇡-periodic, and piecewise di↵erentiable function h. To see the estimate, we notice that on [xi 1 , xi ], mi h(x) Mi , 1 i n. Hence, Z ⇡ X X |f (x) h(x)|2 dx (Mi mi )2 xi 2M (Mi mi ) xi < 2⇡✏. ⇡ To prove the limit in the theorem, for the function h, we know that, by Theorem 8.15U, there exists N such that n N implies p |h(x) sN (h; x)| < ✏ for all x 2 [ ⇡, ⇡]. By the Bessel inequality, we know that Z ⇡ Z ⇡ Z 1 1 1 2 2 |sN (f ; x) sN (h; x)| dx = |sN (f h; x)| dx 2⇡ 2⇡ 2⇡ ⇡ ⇡ ⇡ ⇡ |f (x) h(x)|2 dx < ✏. Thus, if n N , Z ⇡ 1 |f (x) sN (f ; x)|2 dx 2⇡ ⇡ Z ⇡ 1 = |f (x) h(x) + h(x) sN (h; x) + sN (h; x) sN (f ; x)|2 dx 2⇡ ⇡ Z ⇡ Z ⇡ Z ⇡ 3 |f (x) h(x)|2 dx + |h(x) sN (h; x)|2 dx + |sN (h; x) 2⇡ ⇡ ⇡ ⇡ sN (f ; x)|2 dx < 3(✏ + ✏ + ✏) = 9✏. That is, 1 N !1 2⇡ lim Z ⇡ ⇡ |f (x) sN (f ; x)|2 dx = 0. Re-writing the integral in the limit gives Z ⇡ 1 |f (x) sN (f ; x)|2 dx 2⇡ ⇡ Z ⇡ 1 = [f (x) sN (f ; x)] · [f (x) sN (f ; x)] dx 2⇡ ⇡ Z ⇡ Z ⇡ N X 1 1 = |f (x)|2 dx f (x) · cn e 2⇡ 2⇡ ⇡ ⇡ n= N ! Z ⇡ Z ⇡ N X 1 1 inx cn e · f (x) dx + 2⇡ 2⇡ ⇡ ⇡ n= N = 1 2⇡ Z ⇡ ⇡ |f (x)|2 dx N X n= N |cn |2 , inx ! N X dx n= N cn e inx ! · N X n= N cn e inx ! dx 8.5 FOURIER SERIES so that 1 2⇡ Z ⇡ ⇡ |f (x)|2 dx = In general, we can similarly prove Z ⇡ 1 lim [f (x) N !1 2⇡ ⇡ 1 X n= 1 |cn |2 . sN (f ; x)]g(x) dx = 0, which leads to the general Parseval’s identity. 111