Math 412-501 Theory of Partial Differential Equations Lecture 3-9: Convolution theorem. Applications of Fourier transforms. Fourier transform Given a function h : R → C, the function Z ∞ 1 ĥ(ω) = F[h](ω) = h(x)e −iωx dx, 2π −∞ ω∈R is called the Fourier transform of h. Given a function H : R → C, the function Z ∞ −1 H(ω)e iωx dω, Ȟ(x) = F [H](x) = −∞ x ∈R is called the inverse Fourier transform of H. Theorem Suppose h is an absolutely integrable function on (−∞, ∞) and let H = F[h] be its Fourier transform. (i) If h is smooth then h = F −1 [H]. (ii) If h is piecewise smooth then the inverse Fourier transform F −1 [H] is equal to h at points of continuity. Otherwise h(x+) + h(x−) F −1 [H](x) = . 2 In particular, any smooth, absolutely integrable function h : R → C is represented as a Fourier Z ∞ integral H(ω)e iωx dω. h(x) = −∞ Proposition 1 (i) F[af + bg ] = aF[f ] + bF[g ] for all a, b ∈ C. (ii) If g (x) = f (x + α) then ĝ (ω) = e iαω fˆ(ω). (iii) If h(x) = e iβx f (x) then ĥ(ω) = fˆ(ω − β). R∞ Proposition 2 Suppose that −∞ |f (x)| dx < ∞. Then (i) fˆ is well defined and bounded; (ii) fˆ is continuous; (iii) fˆ(ω) → 0 as ω → ∞. Theorem 1 Let f be a smooth function such that f , f ′ , f ′′ , . . . , f (k) are all absolutely integrable on R. (k) (ω) = (iω)k fˆ(ω); Then (i) fd (ii) fˆ(ω) = α(ω)/ω k , where lim α(ω) = 0. ω→∞ Theorem 2 Let f be a function on R such that R k R (1 + |x| )|f (x)| dx < ∞ for some integer k ≥ 1. Then (i) fˆ is k times differentiable; (ii) fˆ(k) (ω) = (−i)k F[x k f (x)](ω). Convolution Suppose f , g : R → C are bounded, absolutely integrable functions. The function Z f (y )g (x − y ) dy (f ∗ g )(x) = R is called the convolution of f and g . Lemma f ∗ g = g ∗ f . Proof: Let z = x − y . Then Z ∞ f (y )g (x − y ) dy (f ∗ g )(x) = −∞ Z ∞ = f (x − z)g (z) dz = (g ∗ f )(x). −∞ Convolution Theorem (i) F[f · g ] = F[f ] ∗ F[g ]; (ii) F[f ∗ g ] = 2π F[f ] · F[g ]. 1 F[f ∗ g ](ω) = 2π Proof of (ii): 1 2π Z Z 1 = 2π Z Z = R R R Z R (f ∗ g )(x)e −iωx dx f (y )g (x − y )e −iωx dx dy R (x = y + z) f (y )g (z)e −iω(y +z) dz dy = 2π fˆ(ω)ĝ (ω). Plancherel’s Theorem (a.k.a. Parseval’s Theorem) (i) If a function f is both absolutely integrable and square-integrable on R, then F[f ] is also square-integrable. Moreover, Z Z 2 |f (x)| dx = 2π |fˆ(ω)|2 dω. R R (ii) If functions f , g are absolutely integrable and square-integrable on R, then Z Z f (x)g (x) dx = 2π fˆ(ω)ĝ (ω) dω. R That is, hf , g i = 2π hfˆ, ĝ i. R 2 Gaussian g (x) = e −αx , α > 0 (density of the normal probability distribution) Z ∞ 1 2 g (x) = e , ĝ (ω) = e −αx e −iωx dx. 2π −∞ Z 1 ∞ −αx 2 cos ωx dx. e ĝ (ω) = π 0 Z d ĝ 1 ∞ −αx 2 ∂ e = (cos ωx) dx dω π 0 ∂ω Z Z ∞ 1 1 ∞ −αx 2 2 sin ωx dx = xe sin ωx d e −αx =− π 0 2απ 0 ∞ −αx 2 Z ∞ e sin ωx 1 2 x=0 e −αx d(sin ωx). = − 2απ 2απ 0 −αx 2 ω d ĝ =− dω 2απ Z 0 ∞ 2 e −αx cos ωx dx = − ω ĝ (ω). 2α ω ĝ ′ ω ω =− =⇒ (log ĝ )′ = − ĝ = − ĝ =⇒ 2α ĝ 2α 2α ′ ω2 2 =⇒ log ĝ = − + C =⇒ ĝ (ω) = ce −ω /(4α) , 4α Z ∞ 1 2 where c = ĝ (0) = e −αx dx. 2π −∞ 2 2πĝ (0) = = ZZ Z ∞ e −αx 2 2 e −αx e −αy dx dy = = π −π 2 e −αy dy ZZ e −α(x 2 +y 2 ) dx dy R2 R2 Z ∞ −∞ −∞ 2 dx Z Z ∞ e −αr 2 r dr dθ = 2π ∞ 2 e −αr r dr 0 0 2 e −αr ∞ π = 2π = −2α r =0 α g (x) = e −αx Z 2 =⇒ ĝ (0) = √ ĝ (ω) = √ 1 4πα 1 2 e −ω /(4α) 4πα Heat equation on an infinite interval Initial value problem: ∂u ∂ 2u =k 2 (−∞ < x < ∞), ∂t ∂x u(x, 0) = f (x). We assume that f is smooth and rapidly decaying as x → ∞. We search for a solution with the same properties. Apply the Fourier transform (relative to x) to both sides of the equation: 2 ∂ u ∂u . =kF F ∂t ∂x 2 Let U = F[u]. That is, Z ∞ 1 U(ω, t) = F[u(·, t)](ω) = u(x, t)e −iωx dx. 2π −∞ 2 ∂ u ∂U ∂u = , F = (iω)2 U(ω, t). Then F 2 ∂t ∂t ∂x ∂U = k(iω)2 U(ω, t) = −kω 2 U(ω, t). ∂t 2 General solution: U(ω, t) = ce −ω kt , where c = c(ω). Hence Initial condition u(x, 0) = f (x) implies that 2 U(ω, 0) = fˆ(ω). Therefore U(ω, t) = fˆ(ω)e −ω kt . We know that 2 F[e −αx ] = √ It follows that e −ω 2 ω2 1 e − 4α , 4πα α > 0. kt = F[g (x, t)], where p x2 g (x, t) = ktπ e − 4kt , t > 0. Hence U(ω, t) = fˆ(ω)ĝ (ω, t). By the convolution theorem, u = (2π)−1 f ∗ g , that is, Z ∞ 1 u(x, t) = f (x̃)g (x − x̃) d x̃ 2π −∞ Z ∞ (x−x̃)2 1 f (x̃)e − 4kt d x̃. =√ 4πkt −∞ Initial value problem: ∂ 2u ∂u (−∞ < x < ∞), =k 2 ∂t ∂x u(x, 0) = f (x). Z ∞ G (x, x̃, t) f (x̃) d x̃, Solution: u(x, t) = −∞ where G (x, x̃, t) = √ 2 1 − (x−x̃) 4kt e . 4πkt The solution is in the integral operator form. The function G is called the kernel of the operator. Also, G (x, x̃, t) is called Green’s function of the problem.