Fourier transforms. R. C. Daileda. Partial Differential Equations April 17, Trinity University

Transcription

1 The Fourier Transform R. C. Trinity University Partial Differential Equations April 17, 214

2 The Fourier series representation For periodic functions Recall: If f is a 2p-periodic (piecewise smooth) function, then f can be expressed as a sum of sinusoids with frequencies,1/p,2/p,3/p,...: f(x) = a + n=1 ( a n cos ( nπx p ) +b n sin ( nπx p )), where p a n = 1 p }{{} 2p when n= p sine for b n ( ) {}}{ nπt f(t) cos dt. p We can obtain a similar representation for arbitrary (non-periodic) functions, provided we replace the (discrete) sum with a (continuous) integral.

3 The Fourier integral representation For L 1 (R) functions Theorem If f is piecewise smooth and f(x) dx <, then where A(ω) = 1 π f(x) = (A(ω) cos(ωx) + B(ω) sin(ωx)) dω, f(t)cos(ωt)dt, B(ω) = 1 π f(t)sin(ωt)dt Remarks: When f(x) dx <, we say that f L1 (R). The integral actually equals f(x+)+f(x ) 2, which is almost f.

4 Example If a >, find the Fourier integral representation for { 1 if x < a, f(x) = otherwise. Note that f is piecewise smooth and that f(x) dx = a so f has an integral representation. We find that A(ω) = 1 π = 1 π a dx = 2a <, f(t)cos(ωt)dt = 1 π sin(ωt) t=a ω = 2sin(ωa). πω t= a a a cos(ωt) dt

5 Likewise B(ω) = 1 π f(t)sin(ωt)dt = 1 π a It follows that the integral representation of f is 2 sin(ωa) cos(ωx) dω = f(x+)+f(x ) = π ω 2 Remarks: a sin(ωt) dt =. }{{} odd 1 if x < a, 1 2 if x = a, otherwise. It is difficult to evaluate the integral on the left directly; we have indirectly produced a formula for its value. Expressing a difficult integral as the Fourier representation of a familiar function gives a powerful technique for evaluating definite integrals.

6 Complexifying the Fourier integral representation f(x) = From Euler s formula e iθ = cosθ+i sinθ one can deduce cosθ = eiθ +e iθ 2 and sinθ = eiθ e iθ. 2i If we apply these in the Fourier integral representation we obtain = 1 2 = 1 2 = 1 2 (A(ω) cos(ωx) + B(ω) sin(ωx)) dω (A(ω) ib(ω))e iωx dω + 1 (A(ω)+iB(ω))e iωx dω 2 }{{} (A(ω) ib(ω))e iωx dω (A(ω) ib(ω))e iωx dω sub. ω for ω (A( ω) +i B( ω) )e iωx dω }{{}}{{} even odd

7 The Fourier transform According to their definitions, we have A(ω) ib(ω) = 1 π = 1 π f(x)(cos(ωx) i sin(ωx)) dx f(x)e iωx dx. Definition: The Fourier transform of f L 1 (R) is π ˆf(ω) = F(f)(ω) = 2 (A(ω) ib(ω)) = 1 f(x)e iωx dx. Note that the Fourier integral representation now becomes f(x) = F 1 (ˆf)(x) = 1 which gives the inverse Fourier transform. ˆf(ω)e iωx dω,

8 Example For a >, find the Fourier transform of { 1 if x < a, f(x) = otherwise. We already found that A(ω) = 2sin(ωa) πω and B(ω) =, so ˆf(ω) = π 2 2 (A(ω) ib(ω)) = π sin(ωa). ω Remark: As this example demonstrates, if one already knows the Fourier integral representation of f, finding ˆf is easy.

9 Example For a >, find the Fourier transform of 1 if a < x <, f(x) = 1 if < x < a, otherwise. We have ˆf(ω) = 1 ( = 1 1 = f(x)e iωx dx = 1 ( a ) iω e iωx a 1 iω e iωx 1 iω (2 2cos(ωa)) = i 2 π a = 1 e iωx dx + (cos(ωa) 1). ω a ) e iωx dx 1 ( 2 e iωa e iωa) iω

10 Example Find the Fourier transform of f(x) = e x. We have ˆf(ω) = 1 = 1 ( f(x)e iωx dx e x e iωx dx + = 1 ( e x(1 iω) dx + ( = 1 e x(1 iω) 1 iω 1+iω = 1 ( 1 1 iω + 1 ) = 1+iω e x(1+iω) ) e x e iωx dx ) e x(1+iω) dx ) 2 π 1 1+ω 2.

11 Properties of the Fourier transform Linearity: If f,g L 1 (R) and a,b R, then af +bg = aˆf +bĝ. Transforms of Derivatives: If f,f,f,...,f (n) L 1 (R) and f (k) (x) as x for k = 1,2,...,n 1, then f (n) (ω) = (iω) nˆf(ω), i.e. differentiation becomes multiplication by i ω. Derivatives of Transforms: If f,x n f L 1 (R), then ˆf (n) (ω) = 1 i n x n f(ω), i.e. multiplication by x (almost) becomes differentiation.

12 Remarks Linearity follows immediately from the definition and linearity of integration. For the second fact, if f,f L 1 (R) then f (ω) = 1 f (x)e iωx dx }{{} u=e iωx,dv=f (x)dx = 1 ( f(x)e iωx +iω ) f(x)e iωx dx = iωˆf(ω), provided f(± ) =. The general result follows by iteration, e.g. f (ω) = (f ) (ω) = iω f (ω) = (iω) 2ˆf(ω). The final fact follows by differentiating under the integral.

13 Example Find the Fourier transform of f(x) = (1 x 2 )e x. Using properties of the Fourier transform, we have f = ê x x 2 e x = ê x i 2ê x ( ( ) ) = π 1+ω ω 2 ( ) 2 1 = π 1+ω 2 + 6ω2 2 2 ω 4 +8ω 2 1 (1+ω 2 ) 3 = π (1+ω 2 ) 3. Remark: The Fourier inversion formula tells us that ω 4 +8ω 2 1 (1+ω 2 ) 3 e iωx dω = π(1 x 2 )e x. Imagine trying to show this directly!

14 Example Find the Fourier transform of the Gaussian function f(x) = e x2. Start by noticing that y = f(x) solves y +2xy =. Taking of both sides gives (iω)ŷ +2iŷ = ŷ + ω 2ŷ =. The solutions of this (separable) differential equation are We find that C = ŷ() = 1 ŷ = Ce ω2 /4. and hence ŷ = ˆf(ω) = 1 2 e ω2 /4. e x2 e ix dx = 1 π = 1 2,

15 Example (Another useful property) Show that for any a, f(ax) = 1 a ˆf ( ω a). We simply compute f(ax) = 1 f(ax)e iωx dx }{{} sub. u=ax = 1 a f(u)e iωu/a du = a ˆf 1 ( ω a) Example Find the Fourier transform of f(x) = e x2 /2. Since f(x) = e (x/ 2) 2, taking a = 1/ 2 above and using the Gaussian example gives ˆf(ω) = e ( 2ω) 2 /4 = e ω2 /2.

16 Example Find the general solution to the ODE y y = e x2 /2. The corresponding homogeneous equation is y h y h = y h = c 1 e x +c 2 e x. It remains to find a particular solution y p. We assume that an L 1 (R) solution exists, and take the Fourier transform of the original ODE: (iω) 2 ŷ ŷ = e ω2 /2 ŷ = e ω2 /2 ω Applying the inverse Fourier transform we obtain y p = 1 e ω2 /2 ω 2 +1 eiωx dω.

17 We can eliminate i by a judicious use of symmetry: y p = 1 = e ω2 /2 ω 2 +1 cos(ωx) } {{ } even 2 e ω2 /2 π ω 2 +1 cos(ωx)dω. Hence, the complete solution is y = y h +y p = c 1 e x +c 2 e x + dω + i e ω2 /2 ω 2 +1 sin(ωx) } {{ } odd 2 e ω2 /2 π ω 2 +1 cos(ωx)dω. dω