M ATH 140 B - HW 8 S OLUTIONS Problem 1 (WR Ch 8 #13). Put f (x) = x if 0 ≤ x < 2π, and apply Parseval’s Theorem to conclude that ∞ 1 X π2 = . 2 6 n=1 n Solution. First we calculate the Fourier coefficients of f (x) = x using integration by parts. For n 6= 0, Z 1 π f (x)e −i nx d x 2π −π Z 1 π = x e −i nx d x 2π −π · −i nx ¸π Z 1 π e −i nx e 1 − x dx = 2π −i n −π 2π −π −i n · ¸ · ¸π 1 (−1)n 1 e −i nx = 2π − 2 2π −i n 2π−n −π n (−1) =i . n Z π Z 1 1 π c0 = f (x) d x = x d x = 0. 2π −π 2π −π cn = Parseval’s identity says that 1 2π Z π −π | f (x)|2 d x = ∞ X |c n |2 . −∞ Computing the right side we get ∞ X −∞ And the left side is 1 2π Z π −π |c n |2 = ¯ ¯ ∞ 1 X ¯ (−1)n ¯2 X ¯i ¯ =2 . ¯ n ¯ 2 n=1 n n6=0 |x|2 d x = · ¸π 1 x3 1 2π3 π2 = · = . 2π 3 −π 2π 3 3 Putting this together, ∞ 1 X π2 =2 2 3 n=1 n =⇒ 1 ∞ 1 X π2 = . 2 6 n=1 n Problem 2 (WR Ch 8 #19). Suppose f is a continuous function on R1 , f (x + 2π) = f (x), and α/π is irrational. Prove that Z N 1 π 1 X f (x + nα) = f (t ) d t N →∞ N n=1 2π −π lim for every x. Solution. First we prove it for f (x) = e i kx with k ∈ Z and k 6= 0. 1 2π Z π e −π i kt " #π 1 e i kx = 0. dt = 2π i k −π Also, ¯ ¯ ¯ ¯ ¯1 X ¯ 1 ¯X ¯ N N ¯ ¯ i k(x+nα) ¯ i kx i knα ¯ e e e ¯ ¯= ¯ ¯ ¯ N n=1 ¯ N ¯n=1 ¯ ¯ ¯ ¯ N 1 ¯¯ i kx X i knα ¯ e = ¯e ¯ ¯ N¯ n=1 ¯ ¯ ¯ N 1 ¯¯ i kx ¯¯ ¯¯ X ¯ = ¯e ¯ ¯ e i knα ¯ ¯n=1 ¯ N ¯ ¯ ¯ N 1 ¯¯ X ¯ e i knα ¯ = ¯ ¯ N ¯n=1 ¯ ¯ ´n ¯ N ³ 1 ¯¯ X i kα ¯ e = ¯ ¯, ¯ N ¯n=1 and since α/π is irrational, that means kα is not a multiple of π, so e i kα 6= 1, and we can use the partial sum formula from page 61 of Rudin to get ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯X ¯ ¯ 1 − ¡e i kα ¢N +1 ´n ¯ ¯ X ´n N ³ N ³ 2 ¯ ¯ ¯ ¯ i kα ¯ ¯ i kα ¡ ¢ − 1¯ < +1 = M, e − 1¯ = ¯ e ¯ ¯=¯ ¯ |1 − e i kα | ¯n=1 ¯ ¯n=0 ¯ ¯ 1 − e i kα so the sum is bounded by M . Therefore, ¯ ¯ ¯ ¯ ¯ ¯ ¯1 X ¯ N N M 1 X ¯ ¯ i k(x+nα) ¯ i k(x+nα) ¯ e e = 0. ¯ lim ¯ = lim ¯ ¯ ≤ lim ¯N →∞ N n=1 ¯ N →∞ ¯ N n=1 ¯ N →∞ N This establishes the equality for f (x) = e i kx with k 6= 0. If k = 0, we have f (x) = 1, so Z N N 1 X 1 X 1 1 π f (x + nα) = lim 1 = lim N =1= f (t ) d t . N →∞ N n=1 N →∞ N n=1 N →∞ N 2π −π lim Now that we’ve shown it for all functions of the form e i kx , we set some ² > 0 and we use TheoP rem 8.15 to get a trigonometric polynomial P (x) = Kk=−K c k e i kx such that |P (x) − f (x)| < ²/2 2 for all real x. Then N N ¡£ ¤ ¢ 1 X 1 X f (x + nα) = lim f (x + nα) − P (x + nα) + P (x + nα) N →∞ N n=1 N →∞ N n=1 lim N £ N ¤ 1 X 1 X f (x + nα) − P (x + nα) + lim P (x + nα) N →∞ N n=1 N →∞ N n=1 = lim 1 2π Z π Z 1 π f (t ) d t = [ f (t ) − P (t )] + P (t ) d t 2π −π −π Z π Z 1 1 π = [ f (t ) − P (t )] + P (t ) d t . 2π −π 2π −π Z N 1 π 1 X P (x + nα) = P (t ) d t N →∞ N n=1 2π −π lim ¯ ¯ Z ¯ N ¯ 1 π 1 X ¯ f (t ) d t ¯¯ f (x + nα) − ¯ lim ¯N →∞ N n=1 2π −π ¯ ¯ Z ¯ ¯ N 1 X 1 π ¯ ¯ ≤ ¯ lim [ f (x + nα) − P (x + nα)] − [ f (t ) − P (t )] d t ¯ ¯N →∞ N n=1 ¯ 2π −π ¯ ¯ ¯ ¯ Z ¯ ¯ 1 π ¯ N ¯ 1 X ¯ ¯ [ f (x + nα) − P (x + nα)]¯ + ¯¯ [ f (t ) − P (t )] d t ¯¯ ≤ ¯ lim ¯ 2π −π ¯N →∞ N n=1 Z N 1 X 1 π ≤ lim | f (x + nα) − P (x + nα)| + | f (t ) − P (t )| d t N →∞ N n=1 2π −π Z N 1 X 1 π < lim ²/2 + ²/2 d t N →∞ N n=1 2π −π ² ² = + = ². 2 2 Since ² was arbitrary, this proves the equality. Problem 3 (Supp-3 #1). Suppose f : [0, π] → R is continuous and Rπ 0 f (x) sin nx = 0, n = 1, 2, . . .. Does it follow that f ≡ 0? Proof or counterexample. Solution. Let c = R 1 π π 0 f (x) d x and define g : [−π, π] → R by ( f (x) x ∈ [0, π], g (x) = − f (−x) x ∈ [−π, 0). Then g is an odd function (i.e. g (−x) = −g (x)), and since cos nx is an even function, that means that g (x) · cos nx is an odd function, and therefore Z π g (x) cos nx d x = 0, −π 3 because integrating an odd function symmetrically around 0 gives 0. Also note that, (with a change of variables in the first integral using u = −x) Z π −π Z 0 Z π g (x) sin nx d x = − f (−x) sin nx d x + f (x) sin nx d x Z−π Z π 0 π = f (u) sin nu d u + f (x) sin nx d x 0Z 0 π nx = 2 f (x)sin dx 0 = 0. Calculating the Fourier coefficients for g for n ≥ 0: Z 1 π cn = g (x) e −i nx d x 2π −π Z 1 π = g (x) (cos nx − i sin nx) d x 2π −π µZ π ¶ Z π 1 g (x) cos nx d x − i g (x) sin nx d x = −π 2π −π = 0. Z 1 π g (x) d x 2π −π Z π 1 2 ( f (x) − c) d x = 2π 0 Z π Z 1 1 π = f (x) d x − c dx π 0 π 0 Z π 1 = f (x) d x − c π 0 c0 = = 0. Now by Parseval’s identity, 1 2π Z π −π |g (x)|2 d x = ∞ X |c n |2 = 0, −∞ and since g 2 is continuous and g 2 ≥ 0, then we have g 2 ≡ 0 by Problem 6.2 proved on a previous homework. Therefore, g ≡ 0, so f ≡ 0. Problem 4 (Supp-3 #3). If f is real analytic in a neighborhood of x 0 and f (x 0 ) = 0, show that f (x)/(x − x 0 ) is real analytic in the same neighborhood. Solution. If f is real analytic in a neighborhood of x 0 and f (x 0 ) = 0 then there is some ² > 0 such that for x ∈ B (x 0 ; ²), f (x) = ∞ X c n (x − x 0 )n and 0 = f (x 0 ) = c 0 n=0 =⇒ f (x) = ∞ X n=1 4 c n (x − x 0 )n . Letting f k = k P n=1 c n (x − x 0 )n , we have f k â f on B (x 0 ; ²) because f is real analytic on B (x 0 ; ²). Also, by Theorem 8.1, f k0 = k P n=1 nc n (x − x 0 )n−1 is such that { f k0 } converges uniformly to f 0 on B (x 0 ; ²) (this means the radius of convergence of f k0 is some number R 0 ≥ ²). We’re going to prove that f k /(x − x 0 ) has the same radius of convergence as f k0 , and that will finish the proof. Therefore, k k X X f k (x) = (x − x 0 )−1 c n (x − x 0 )n = c n (x − x 0 )n−1 , x − x0 n=1 n=1 and checking the radius of convergence R of this last power series, we have p p p p 1 1 = lim sup n−1 |c n | = lim sup n−1 n n−1 |c n | = lim sup n−1 n|c n | = 0 , R R n→∞ n→∞ n→∞ p since limn→∞ n−1 n = 1. Then R = R 0 ≥ ². That means f k /(x − x 0 ) converges uniformly on B (x 0 ; ²), and thus f (x)/(x − x 0 ) is real analytic on B (x 0 ; ²). Problem 5 (Supp #4). Prove that if f (x) is real analytic on (a, b) and c ∈ (a, b), then F (x) = Rx c f (t ) d t is also real analytic on (a, b). b+a 2 (the point in the middle of a and b). If f is real analytic on (a, b) and k P c ∈ (a, b), then letting f k (x) = c n (x − x 0 )n we have f k â f on [a, b]. Let R 1 be the radius of n=0 ¯ ¯ ¯ ¯ convergence of f k . What we have shown so far is that R 1 ≥ ¯ b−a 2 ¯. Now we set Solution. Let x 0 = Z G k (x) = x x0 k x X Z f k (t ) d t = x 0 n=0 c n (t − x 0 )n d t = k Z X x n=0 x 0 c n (t − x 0 )n d t = k X cn (x − x 0 )n+1 , n + 1 n=0 then the radius of convergence R 2 of this last power series is given by 1 = lim sup R2 n→∞ s n+1 |c n | = lim sup n +1 n→∞ 1 p n +1 p p 1 |c n | = lim sup n+1 |c n | = , R1 n→∞ n+1 n+1 ¯ ¯ p ¯ ¯ n + 1 = 1, and thus we have R 2 = R 1 ≥ ¯ b−a 2 ¯, so G k converges uniformly on n→∞ Rx [a, b]. If we let G(x) = x0 f (t ) d t , then by Theorem 7.16 we have G k â G on [a, b]. Lastly, if Rx we let C = c 0 f (t ) d t , we have since lim n+1 x Z F (x) = c x0 Z f (t ) d t = c Z f (t ) d t + x x0 f (t ) d t = G(x) +C , and similarly we define F k (x) = G k (x)+C , so that F k â F , and thus F is real analytic on [a, b]. 5