e iωt dt and explained why δ(ω) = 0 for ω 0 but δ(0) =. A crucial property of the delta function, however, is that

Transcription

1 Phys 531 Fourier Transforms In this handout, I will go through the derivations of some of the results I gave in class (Lecture 14, 1/11). I won t reintroduce the concepts though, so you ll want to refer to the lecture notes and the text. I ll try to provide enough detail in the calculations to illustrate some of the techniques for working with Fourier transforms. I hope it is helpful, but you shouldn t need any of this specific material for the homework assignments or final exam. 1 Delta Functions I defined δ(ω) e iωt dt (1) and explained why δ(ω) = for ω but δ() =. A crucial property of the delta function, however, is that δ(ω) dω. I ll derive this property here. The infinite bounds on the integrals cause some difficulty, so let us interpret as Then define e iωt dt T lim e iωt dt. T T δ T (ω) T e iωt dt T so that δ(ω) = lim T δ T (ω). We can simply integrate to get δ T : δ T (ω) T T iω eiωt e iωt dt T T ( e iωt e ) iωt iω πω sin(ωt ) Now we need to evaluate δ T (ω) dω. This is a bit harder. The professional way to do it is using contour integration, but we can avoid that with some tricks. First, note that sin(ω)/ω is symmetric, so δ T (ω) dω = π 1 sin(ωt ) ω dω.

2 Change variables to u = ωt, giving π Now, make the unintuitive substitution to get δ T (ω) dω = π iπ iπ 1 u = The u integrals are elementary, giving sin(u) u δ T (ω) dω iπ iπ = π du. e uv dv e uv sin(u) dv du e uv ( e iu e iu) du dv [ e u( v+i) e u( v i) du dv ( 1 v + i + 1 i 1 + v dv v dv ) dv v i You can look the v integral up, or derive it by substituting v = tan θ, so that dv sec θ dθ and the limits become to π/. But also, 1 + v becomes 1 + tan θ = sec θ which cancels the sec θ term from the differential. The integral becomes π/ dθ = π/, so that δ T (ω) dω. Since this holds independently of T, it is reasonable to conclude that δ(ω) dω. () There are some formal mathematical issues regarding the limit T, but we don t need to worry about them here. Our result (), combined with the other properties of δ(ω) that we know, is enough to establish that for any function F (ω), F (ω)δ(ω ω ) dω = F (ω ) (3) which is the main result we ll need. It applies not only to frequency integrals; more generally, we have for any variable u. F (u)δ(u u ) du = F (u )

3 Calculating Transforms On slide 9 of the lecture, I gave a table with the Fourier transforms of several functions. We did the first in class, and I ll go through the rest here. At the back of this handout, I ll repeat the table for both space and time transforms. The second and third lines of the table are easy. If f(t) = e iωt then By definition (1), this is F (ω) = = e iωt e iω t dt e i(ω ω )t dt F (ω) = δ(ω ω ) as the table gives. Note that by choosing ω =, this gives us the transform of f(t). If f(t) = δ(t τ), then F (ω) = = e iωτ, e iωt δ(t τ) dt the third line shown. The fourth transform listed takes more work. Say f(t) = e t /τ. Then F (ω) = = e t /τ e iωt dt e t /τ +iωt dt. We can simplify the integrand with a trick called completing the square. Note that ( ) ( ) t t iωτ iωτ + iωt = τ τ + iωt +, since we re just adding and subtracting the same thing. But the first three terms can be factored: [ So we get t τ + iωt ( iωτ ) + ( iωτ F (ω) = e ω τ /4 ) = [ ( t τ iωτ ) ω τ 4. e (t/τ iωτ/) dt Change variables to u = t/τ iωτ/. Then dt τ du, and the limits of the integral don t change since they are infinite. Actually, that s a bit tricky: the limits really 3

4 become something like iωτ/, which isn t quite the same thing as plain. Sometimes the imaginary bit matters, but not here, so we ll just ignore it. (For people who know something about contour integrals, we can displace the integration path back to the real axis because the integrand has no poles.) In any case, we get F (ω) = τe ω τ /4 e u du. You could just look the final integral up, but I ll show you how to solve it for yourself. Define its value to be I: Then consider I = I = e u du. e u du e v dv where we introduce v to keep track of which integral is which. Combined, we have I = e (u +v ) du dv. Now do this double integral in polar coordinates: (u, v) (r, θ), where u = r cos θ and v = r sin θ. Then du dv r dr dθ and we have I = re r dr dθ The θ integral gives a factor of, and the r integral can be done with the substitution s = r. Then ds = r dr so I = π e s ds. The s integral is just unity, so I = π. But then I = π and as the table indicates. 3 Convolution Theorem F (ω) = τ π e ω τ /4 The convolution theorem states that if F (ω) = F 1 (ω)f (ω), then f(t) = f 1 (T )f (t T ) dt, where F, F 1 and F are respectively the transforms of f, f 1 and f. This is fairly easy to prove. 4

5 We have But in turn, and f(t) F (ω)e iωt dω F 1 (ω)f (ω)e iωt dω. F 1 (ω) = F (ω) = f 1 (t 1 )e iωt 1 dt 1 f (t )e iωt dt. Again, we can always relabel integration variables as we like. Combining all these, we have f(t) f 1 (t 1 )f (t )e iω(t 1+t t) dt 1 dt dω. Do the ω integral first, so that f(t) [ f 1 (t 1 )f (t ) e iω(t 1+t t) dω dt 1 dt. You should recognize the expression in brackets as δ(t 1 + t t), so f(t) = f 1 (t 1 )f (t )δ(t 1 + t t) dt 1 dt. Use the δ-function to do the t integral, so that we can replace t with t t 1 to get f(t) = f 1 (t 1 )f (t t 1 ) dt 1. Relabeling t 1 T gives us our convolution result. The correlation theorem and Parseval s theorem have derivations very similar to that of the convolution theorem. I ll go through Parseval s to demonstrate. The theorem states that f(t) dt F (ω) dω if f and F are a Fourier transform pair. Consider [ [ F (ω) dω = f(t 1 )e iωt 1 dt 1 f (t )e iωt dt dω. Again, reorder the integral to do the ω one first: [ F (ω) dω = f(t 1 )f (t ) e iω(t 1 t ) dω dt 1 dt. 5

6 The term in brackets is δ(t 1 t ) so F (ω) dω = and using the δ-function gives Parseval s theorem: F (ω) dω = 4 Transform pairs f(t 1 )f (t )δ(t 1 t ) dt 1 dt f(t 1 )f (t 1 ) dt 1 = f(t) dt. Finally, one thing I didn t really go through in class is how to get the inverse Fourier transform of a function F (ω) that has the same form as some g(t) whose transform you know already. For instance, in class (slide 4), I calculated the inverse transform of a square pulse function F (ω). We already knew the transform of a square pulse in time, and we could have used that information to obtain f(t) rather than doing the integral. The technique is nice to know, so I ll go through it here. Say we know that g(t) has transform G(ω). Then suppose we re given F (ω) = g(ω). That looks a little funny, so here s an example: Say g(t) = cos(at) for some constant a. Then G(ω) = π[δ(ω a) + δ(ω + a). The question is: if we re given a transform F (ω) = cos(aω), what is f(t)? Obviously, a must have different units in F, but the form is the same as g. We want f(t) F (ω)e iωt dω g(ω)e iωt dω. Now we know g(t) G(ω)e iωt dω so just relabeling variables t ω and ω t 1 gives g(ω) G(t 1 )e iωt 1 dt 1. Substitute this into the expression for f(t): f(t) G(t () 1 )e iω(t+t1) dt 1 dω. Once again, do the ω integral first f(t) () G( t) [ G(t 1 ) e iω(t+t1) dω dt 1 G(t 1 )δ(t + t 1 ) dt 1 6

7 which is the desired result. So in our example, if we have F (ω) = cos(aω), then f(t) [δ( t a) + δ( t + a). This can be simplified since δ( t) = δ(t), so f(t) [δ(t + a) + δ(t a). For another example, apply this to the calculation we did in class. We wanted the inverse transform of { 1 if ω < ω m F (ω) = else We know that the function g 1 g 1 (t) = { 1 τ if t < τ else has Fourier transform ( ωτ ) G 1 (ω) = sinc. Use the linearity properties to see that { 1 if t < a g(t) = else has transform G(ω) = a sinc(ωa). Then our F (ω) = g(ω) with a = ω m, so f(t) G( t) [ ωm sinc( tω m ) = ω m π sinc(ω mt) since sinc( ωt) = sinc(ωt). This agrees with the result we got in class by just doing the integral. 5 Summary I ve presented a pretty quick run through some of the important properties of Fourier transforms. Again, a lot of this is more complicated than what we ll be doing in class: I ll try to focus more on optics than on math. Nonetheless, you will be seeing many of these ideas again, and I hope that filling out some of the steps will help everyone feel more confident using the Fourier technique. 7

8 6 Transform Tables Time Transforms: f(t) F (ω) { } 1 t < τ τ sinc ( ) ωτ e iω t δ(ω ω ) δ(t τ) e iωτ e t /τ τ πe ω τ /4 f(t + τ) e iωτ F (ω) e iω t f(t) F (ω ω ) Space Transforms: f(x) F (k) { } 1 x < a a sinc ( ) ka e ik x δ(k k ) δ(x a) e ika e x /a a πe k a /4 f(x + a) e ika F (k) e ik x f(x) F (k k ) 8