CHAPTER | ] Fourier Series, Integrals, and Transforms Fourier series (Sce. 11.1) are infinite series designed to represent general periodic functions in terms of simple ones, namely, cosines and sines, They constitute a very important tool, in particular in solving problems that involve ODEs and PDEs. In this chapter we discuss Fourier series and their engineering use from a practical point of view, in connection with ODES and with the approximation of periodic functions Application to PDEs follows in Chap. 12. ‘The theory of Fourier series is complicated, but we shall see thatthe application of these series is rather simple. Fourier series are in a certain sense more universal than the familiar ‘Taylor series in calculus because many discontinuous periodic functions of practical interest ccan be developed in Fourier series but, of course, do not have Taylor series representations In the last sections (11.7-11.9) we consider Fourier integrals and Fourier transforms, which extend the ideas and techniques of Fourier series to nonperiodic functions and have basic applications to PDEs (to be shown in the next chapter). Prerequisite: Elementary integral calculus (needed for Fourier coefficients) Sections that may be omitted in a shorter course: 114-119 References and Answers to Problems: App. 1 Part C, App. 2. 11.1 Fourier Series 478 Fourier series are the basic tool for representing periodic functions, which play an important role in applications. A function f(x) is called a periodic function if f(x) is defined for all real x (perhaps except at some points, such as x 2, #3m22, © + for tan x) and if there is some positive number p, called a period of f(x), such that a fix + p) = fa) for all x ‘The graph of such a function is obtained by periodic repetition of its graph in any interval of length p (Fig. 255). Familiar periodic functions are the cosine and sine functions. Examples of functions that are not periodie are x, x%, 13, e*, cosh x, and In x, to mention just a few. IE f(a) has period p, it also has the period 2p because (1) implies fx + 2p) = file + p] + p) = fox + p) = fla), ete. thus for any integer n = 1,2, 3, ++ oy fix + np) = f0) for all x.SEC.111. Fourier Series 479 Fig. 255. Periodic funetion Furthermore if f(x) and g(x) have period p, then af(x) + bg(x) with any constants @ and Jr also has the period p. (Ous problems in the fist few sections ofthis chapter will be the representation of various functions §(x) of period 2r in terms ofthe simple functions @) 1, cosx, sinx, — cos2x, sin2x,---, cosa, sin nx, All these functions have the period 2. They form the so-called trigonometric system. Figure 256 shows the first few of them (except forthe constant I, which is periodic with any period). The series to be obtained will be a trigonometric series, that is, a series of the form day + a4 cO8.x + By Sin x + ag COs 2x + by Sin 2x + + @ = a9 + D (a, cos nx + by sinnx). diy, Das da, Da,“ are constants, called the coefficients of the series. We see that each term has the period 2. Hence ifthe coeficients are such that the series converges, its sum willbe a function of period 2. It can be shown that if the series on the left side of (4) converges, then inserting parentheses on the right gives a series that converges and has the same sim as the series 6 the left. This justifies the equality in (4. Now suppose that f(x) is @ given function of period 2a and is such that it ean be represented by a scrcs (4), tht is, (4) converges and, moreover, has the sum f(x). Then, using the equality sign, we write 6) $0) = a9 + D (ay cos nx + by sin nx) Te) NM Fig. 256. Cosine and sine functions having the period 2ar430 EXAMPLE 1 CHAP. 11 Fourier Series, Integrals, and Transforms and call (5) the Fourier series of f(s). We shall prove that in this case the coefficients of (5) ate the so-called Fourier coefficients of f(x), given by the Euler formulas =f sera 6) ) J 100) cos me dr n=l, © J seosinnede n= 1, ‘The name “Fourier series” is sometimes also used in the exceptional case that (5) with coefficients (6) does not converge or does not have the sum f(x)—this may happen but is merely of theoretical interest. (For Euler see footnote 4 in Sec. 2.5.) A Basic Example Before we derive the Euler formulas (6), let us become familiar with the application of (5) and (6) in the case of an important example. Since your work for other functions will be quite similar, try to fully understand every detail ofthe integrations, which because of the n involved differ somewhat from what you have practiced in calculus. Do not just routinely use your software, but make observations: How are continuous functions (cosines and sines) able to represent a given discontinuous function? How does the quality of the approximation increase if you take more and more terms of the series? Why are the approximating functions, called the partial sums of the series, always zevo at O and =? Why is the factor 1/n (obtained in the integration) important? Periodic Rectangular Wave (Fig. 2578) ‘Find the Fourier coeticons ofthe period function fx) in Fig. 2572, The formula is o so-{ et forte = denen Functions ofthis kind occur as external fores acting on mechanical systems, electromotive forces in electric ‘lnc, ce. (The value of f() a a single point doesnot affect the integra; Hence we can eave J) defines stx=Oandx= =z) Solution. From (62) we obtain ay = 0. This ean also be Seen without integration, since the area under the ture off Between 7 and is zero. From (6b), [== ‘]-e 2,15. Similarly, from (62) we obtain 1 because sin = Oat ~ 0, and forall af]SEC. TU Fourier Series 481 fe) 4 A cr aceaatl Li, Lo (a) The ven function fi) Periodic rectangular wave) 8, 4 (b) The fist thee pot sums ofthe eesponding Fourier sres Fig. 257. Eample 1 Since 0s —a) = cosa and cos 0 = 1 this yields 2 seo tondmon tan antes) = 2a Noronha ont eigen ae-{ atts ence the Fourie coeticions by of our function ane482 CHAP. 11 Fourier Series, Integrals, and Transforms Since the ay are zero, the Fourier series of J) is : 2 (inst 1 snare 1 anset ® a Fine + F sine The paral sums ane so +d sings : = Hains, inx + 5 sin3r) oe, “Thee graphs in Fig 257 seom wo indicate that the series is convergent and has the sum fhe given fenton ‘We notice that a s = D and x = the pins of discontinuity offs), all paral sums have te vale ze, the ssithmetc mean of the limi —K and of or function, a these points urtermere, assuming that f(s he sum of he series and sting x = m2, we have This is famous result obtained by Lebniz in 1673 from geometric considerations. Illustrates tha he values of various series with constant terms can he obained by evaluating Fourie seis at specific points. Derivation of the Euler Formulas (6) ‘The key to the Euler formulas (6) is the orthogonality of (3), concept of basic importance, as follows. THEOREM 1 Orthogonality of the Trigonometric System (3) ‘The trigonometric system (3) is orthogonal on the interval ~77 5 x = (hence also fon 0 = x S 2m or any other interval of length 277 because of periodicity); tha is, the integral ofthe product of any two functions in (3) over that interval is 0, so that Jor any integers n and m, () Preosmcomar=0 atm o J sinmx sin ms ax nem (© J sinnxcosmeds =O (n# morn =m). PROOF This follows simply by transforming the integrands trigonometrically from products into sums, In (9a) and (9b), by (11) in App. A3.1, f [costn meee ep cxensommeas= + [ont emede+ } io ie imme sin d= > J cos (n—myxde~ > [cos + md,‘SEC.1L1 Fourier Series 483 Since m # n (integer!), the integrals on the right are a O. Similarly, in (9e), for all integer ‘mand n (without exception; do you see why?) fcc 1 sina t meds + 2 [sna myrar=0-+0 m Application of Theorem 1 to the Fourier Series (5) We prove (6a), Integrating on both sides of (5) from —a to 7, we get J so a0= J [ + D (a, cosine + by son | de. We now assume that termwise integration is allowed. (We shall say in the proof of ‘Theorem 2 when this is true,) Then we obtain (on cornea +o ‘The first term on the right equals 27rdo, Integration shows that all the other integrals are 0. Hence division by 27 gives (6a). i) co a ‘We prove (6b). Multiplying (5) on both sides by cos mx with any fixed positive integer ‘m and integrating from ~ to 7, we have (10) ff cosme ds = J [» +3 (on cosne + by sin | cos mx de. We now integrate term by term, Then on the right we obtain an integral of ay cos mx, which is 0; an integral of a, cos mx cos mux, which is dy.7 for n = m and 0 for # m by (Ga); and an integral of b,, sin nx cos mx, which is O for all n and m by (9c). Hence the right side of (10) equals 4,7. Division by a gives (6b) (with m instead of n). We finally prove (6c). Multiplying (5) on both sides by sin mor with any fixed positive integer m and integrating from ~7 to 7, we get ay Ff seosinmeae= J [« £3 (corms + fan | in ere Integrating term by term, we obtain on the right an integral of a sin mx, which is 0: an integral ofa, cos nx sin mx, which is O by (8c) and an integral of sin nx sin mx, which is bz ifn ~ m and O if m # m, by (Ob). This implies (6c) (with denoted by m). This completes the proof of the Euler formulas (6) forthe Fourier coeficents. 7484 THEOREM 2 PROOF Fig 288 Lee and right-hand limits f-0)=1, £0 0) of the function ?ofx we ny ban ot [_ swat mer a (+ en _E [aon snes), ote st “On Ten 1m sod, this s qual 1 710, and for even x we have = ee ee 2 Diet De ae {Ina similar fashion we find from (6) that By = £2 and by = 0 for x = 2, 3, +=, Consequently.490 CHAP. Fourier Series, Integrals, and Transforms T-Ti] FOURIER SERIES FOR PERIOD p = 2L Find the Fourier series ofthe function f(s), of period p = 21, and sketch or graph the first three partial sums, Show the details of your work.) 1. fle) = -1(-2 a, cos nx (f even) with coefficients = 2 ee Fras ge? [00 cos mas n= 1,2, “ and f(x) = >) by sin nx (f odd) with coefficients ae ee be = & | 1e9 sinned, * For instance, x) in Example 1, Sec. 11.1, is odd and is represented by a Foutier sine Further simplifications result from the following property, whose very simple proof is left to the student ‘Sum and Scalar Multiple ‘The Fourier coefficients ofa sum fy + fz are the sums of the corresponding Fourier coefficients of fy and fo. The Fourier coefficients of cf are ¢ times the corresponding Fourier coefficients off Rectangular Pulse ‘The function f(x) in Fig. 264i the sum ofthe faction f(x) in Example of See 11.1 and the constant k Heac, from tat example and Theorem 2 we conclude that mavens (uss bunses bases) . sale ave Rete ‘We conelude frm this and Theorem 2 thi (f)~ U0) must be a even function. Verify this graphically. (See Fig 265) Snes Fig. 264. Example 1 Fig. 265. u(t) ~ off) with E=1, 0SEC. 113 Even and Odd Functions. Half-Range Expansions 493 EXAMPLE 3 Sawtooth Wave Find the Fourier series of the function (Fi, 2665), foyrxtm if —mexcm and fle} 2m) = for re) (a) Me fanetion 0) iL a ay aN Pt (0) Partial arm Sy, Soy Sap Fig. 266. Example 3 Solution. We have f= fy + fo, where fy = rand fa = 1. The Fourier coeticients of fare zero, excep or the fst one the constant term), which is Hence, by Theorem 2, the Fourier coeticients dy by ate tose Off, except for dg, which is Since J 1 odd, ay ~ O or and no [hasan negating by pas, we obtain hone Peei(ueleseteu se) Half-Range Expansions Half-range expansions are Fourier series. The idea is simple and useful. Figure 267 explains it, We want to represent f(x) in Fig. 267a by a Fourier series, where f(x) may be the shape of a distorted violin string or the temperature in a metal bar of length L, for example. (Corresponding problems will be discussed in Chap. 12.) Now comes the idea.494 EXAMPLE 4 iP TE Fig. 268. The given function in Example 4 CCHAP.11 Fourier Series, Integrals, and Transforms (a) The gen urcton fs) 1) | 7 t = (0) fis extended 9629 a8 pecod tunetion of peed 2 Fig. 267. (a] Function f(x given on an interval 0. x SL (b) Even extension to the fll “range” (interval) ~L = x = L (heavy curve) and the periodic extension of period 2L to the a-axis, (c) Odd extension to ~L = x 5 (heavy curve) and the periodic extension of period 21 to the xaxis We could extend fx) asa function of period Land develop the extended function into a Fourier series. But this series would in general contain borh cosine and sine terms. We can do beter and got simpler series. Indecd, for our given f we can calculate Fourier coefficients from (2) or from (4) in Theorem 1. And we have a choice and ean take what seems more practical If we use (2), we get (1). This isthe even periodic extension of f in Fig. 267. If we choose (4) instead, we get (3), the odd periadic extension fof Fig. 2676 Both extensions have period 2. This motivates the name half-range expansions: f is siven (and of physical interest) only on half the range, half the interval of pericity of length 2 et us illustrate these ideas with an example that we shal also need in Chap. 12. “Triangle” and Its Half-Range Expansions Fo the two blag expansions ofthe fant ig. 268) He woceet wo={e : Solution. (a) Bren periodic extension. From (2) we obain anf [BL ae [eve] 4.SEC. 113 Even and Odd Functions. Half-Range Expansions 495 We consider dy. For the fist integral we obtain by Intgration by pats L Eg BE paar Similarly, forthe second integral we obtain ee re ee eee “6-5 (S04) ow) for dy The sine ems cancel and so dows a factor 12. This gives Seba fom) eset (aon © care Thos, dy = ~164(2 0), ig = —16K6? a), ayq = — 16410? x”), aa. og = Oita #26 10 ++» Heh a lesen ff) 268) kW oe oe jose (Hoe ee ‘This Fourier cosine seis represents the even periodic extension ofthe given funtion J) of period 2 (b) O44 periodic extension, Siilry, from (4) we obtain Bh o aaa sn ence the other hal-range expansion of f(s) s (Fig. 2686) HL ye sin ~ a This ros een theol prod eteson ff of pr Basic appiaton ofthese ess wl be shown i Sec. 123 and 125. . ° % = (a) Even extanion ©) 048 extension Fig. 269. Periodic extensions of f(x) in Example 4496 CHAP. TT Fourier Series, Integrals, and Transforms ois S02-5 So ees eee [E3]_EvEN AND opp FUNCTIONS ‘Are the following fanctions even, od or neldhe even nor oad? 1. |x|, x? sin mx, x + Hl, In x, x cosh x 2. sin (x), sin® x, x sinh x, x9], e%, xe", tan 2x, xf(1 + 24) A the fllowing fancios, which ae assumed 10 be pio of ped 2m, even, od or tele even oro? 3B fixy= x (<4 0), then the steady-state solution isa harmonic oscillation with frequency equal 0 that of r(). However, if r() is nota pure sine or cosine function but is any other periodic function, then the steady-state solution will be a superposition of harmonic oscillations with frequencies equal to that of r() and integer multiples of the later. And if one of these frequencies is close 0 the (practical) resonant frequency of the vibrating system (see Sec. 2.8), then the corresponding oscillation may be the dominant part of the response of te system to the external force. This is what the use of Fourier series will show us. Of cours, this is quite surprising to an observer unfamiliar with Fourier series, which are highly important in the study of vibrating systems and resonance. Let us discuss the entire situation in terms ofa typical example Forced Oscillations under a Nonsinusoidal Periodic Dr In (0), let m = 1 (gm, ¢ = 0085 (gsc), and ing Force 25 (asec, so tat (1) booomes o yh + 0059" + 255 = 1) wheter) fs measured in gm m/sec, Let (Fig. 273) eZ it wero, r= no + 2m) = rt, akin ocr Find the steady-state solution 3) Fig. 273. Force in Example 1 Solution. We represent») by a Fourier ste, fiaing (ake the answer to Pro 11 in Problem Set 11.3 minus Jrand write # for), Then we consider the ODE. 4 o y" $0053" + 25y= SE coemt w= 13-9) whose right side i a single term of the series (3). From Sec. 2.8 we know thatthe steady-state solution ¥0) of (4) is of the form ° 605 + By sn aSEC.IS Forced Oscillations sor By sung his ft 0) Hi 4es— 8) 02 Boose © 4- 4B Ee vice Dg = 0s + (00508 Since the ODE (2) is ner, we may expect the steady-state solution to be a sotto whet s 6siven by (5) and (6), In fac, his follows readily by substitng (7) into (2) and sing the Fourier series ofr), provided tha terme differentiation of 7) is prmissibe, (Readers already familiar ith the nation oF uniam convergence (See. 15.5] may prove that (7) may e diferenated term by tem.) From (6) we ind thatthe ample of (5) i (afactar VB, cancels out) 4 cua Varo [Numeri values are C1 ~ 0.0831 x = 0.0088 co = 0.2087 = 00011 y= 0.008, Figure 274 shows the input (muiipied by 0.1) andthe ouput. For 1 = 5 the quanity Dy. is very smal the denominator of Cis smal, and Ca i 0 lrg that yi the dominating term in (7). Hence the oupot is almost ssharmonic oscillation of five times the frequency of the driving fre, ile dstoned due tothe ter y, whose Smplitud is about 25% ofthat of yp. Yow could make the station sil more extreme by decreasing the damping constant. Try it 7 Fig. 274. Input and steady-state output in Example 1 Pach tes eee 1, (Cooficents) Derive the formula for C, froma, andB,. 3. (Damping) In Example 1 change ¢ 0 0.02 and discuss 2. (Spring constant) What would happen to the amplitudes ‘how this changes the output. Cy in Example I (and thas tothe form of the vibration) 4 Input) What would happen in Example 1 if we if we changed the spring constant tothe value 9? If we replaced r( with its derivative (the rectangular wave)? took a sifferspeing with & = 81? First guess. ‘What isthe ratio of the new C, to the old ones?502 cHaP.11 Exit] GENERAL SOLUTION Find a general solution of the ODE y" + oy ‘(tas given. (Show the details of your work.) 5. 1(0) = e08 ot, w = 05, 08, ll 15,50, 100 6. rl) = cos ant + 608 wat (a® # a2, a3?) ith 7. re) =D ay cost, fol #1,2,000N int + fsin3e + bsinSr + 4 sin 7¢ & rl) ten it ano [ and r(t + 2m) = r(0, [a] # 0,13, { 1 if —eR 0 and r@) as given, (Show the details of your work.) 14. (0) = ay cos nt 15, r(Q) = sin3 { of a(t) if and r(r+ 2m) = 7) -midere a2 16. 109) wid<1< 30/2 17. ris) = EJ by sin ne 18. CAS EXPERIMENT, Maximum Output ‘Term. Graph and discuss outputs of »" + es" + ky = rit) with r(9 as in Example 1 for various © and & with nphasis on the maximum Cy and its ratio t0 the second largest |C;|. 19-20] RLC-CIRCUIT Find the steady-state current 1(f) in the RUC-citeuit in Fig. 272, where R = 100.0, L = 10 H, C = 10°? F and E() V as follows and periodie with period 2. Sketch of graph the first four partial sums. Nowe thatthe coefficients ofthe solution decrease rapidly 19, EQ) = 20017? = 1) ( 100(m + 2) if 100(mt — 1%) 11.6 Approximation by Trigonometric Polynomials Fourier series play a prominent role in differential equations. Another field in which they hhaye major appli tions is approximation theory, which concems the approximation of functions by other (usually simpler) functions. In connection with Fourier series the idea is as follows. Let f(x) be a function on the interval ~a x a that can be represented on this interval by a Fourier series. Then the Nth partial sum of the series wo Hk) ~ a5 +S (ay cos wx + By sin nx) is an approximation of the given f(x). It is natural to ask whether (1) is the “best” approximation of f by a ti of the form @ Fla) = Ay +X (Ay cos nt + By six) mnometric polynomial of degree N, that is, by a function (W fixed) where “best” means that the “error” of the approximation is as small as possible.SEC. 116 Approximation by Trigonometric Polynomials 503 Of course, we must first define what we mean by the error £ of such an approximation We could choose the maximum of |f ~ F]. But in connection with Fourier series itis better to choose a definition that measures the goodness of agreement between f and F on the whole interval 7 = x= 7. This seems preferable, in particular iff has jumps: F in Fig. 275 is a good overall approximation of f, but the maximum of |f — F (more precisely, the supremum) is large (it equals at least half the jump of f at xo). We choose ° e-[o-ma This is called the square error of F relative to the function f on the interval ~ +3 w+ | ‘We finally subtract (6) from (5). Then the integrals drop out and we get terms Ay? = 2Aydy + ay? = (Ay — a) and similar terms (By ~ By)?! E-E #26 ~ eo* +3 Ue 2? Bn ~ byl} Since the sum of squares of real numbers on the right cannot be negative, E-E*20, tus BEBE, and E = £* if and only if Ag = do, * » By = by This proves the following fundamental ‘minimum property of the partial sums of Fourier series. Minimum Square Error The square error of F in 2) (with fxed N) relative tof on the interval — w= x= we is minimum if and only ifthe coefficients of Fin (2) are the Fourier coefficients of | §. This mininum value E° is given by (6). From (6) we see that &* cannot increase as N increases, but may decrease. Hence with increasing N the partial sums of the Fourier series of f yield better and better ‘approximations 10 f, considered from the viewpoint of the square error. Since E* = 0 and (6) holds for every N, we obtain from (6) the important Bessel’s inequality zi ws nat + Sat + 0.9 5 2 fpr ae for the Fourier coefficients of any function f for which integral on the right exists. (For F. W. Bessel see Sec. 5.5.) It can be shown (sce [C12] in App. 1) that for such a function f, Parseval’s theorem holds; that is, formula (7) holds with the equality sign, so that it becomes Parseval’s identity’ 8) Dag? + (a,? + b,2) = 2 [soot MARC ANTOINE PARSEVAL (1755-1836), French mmitcian, A physical interpretation ofthe idemty Follows ithe next section,SEC.1L6 Approximation by Trigonometric Polynomials 505 EXAMPLE 1 Minimum Square Error for the Sawtooth Wave (Compute the minimum square exo E* of F(x) with N= 1,2, «++, 10,20, ++, 100 and 1000 relative to fiyerte Crexem on the interval Sx = Solution. Fis) = = + 2(6inx ~ ‘See. 11.3. From this and (6), sin.) by Example 3 in w= far nta aoe oad 4) eo [-w N ge N gE N BE ] 700.782 8001361 800.1389 100 0.1250 1000 0.0136 F = 5, Sy, Sy ave shown in Fig. 266 in See. 11.3, and F = Sa is shown in Fi. 276, Although [f) ~ FU is large at (how large, where fis dscomtnaous,F approximates fguite well onthe whole interval except oF 0 _& ear 7, where “waves” remain owing tothe Gibbs phenomenon (see CAS Experiment 20 in Problem Set Fig. 276. Fwith 113) N= 20 in Example 1 Can you think of fanetions f for which B* decreases more quickly with increasing N? . This is the end of our discussion of Fourier series, which has emphasized the practical aspects of these series, as needed in applications. In the last three sections of this chapter ‘we show how ideas and techniques in Fourier series can be extended to nonperiodie funetions.. -PROBLEM-SET—H-6 1-9] MINIMUM SQUARE ERROR f it -jrcx 2 piven by tf -Lexe hwayt i text lo tee ‘The left pr of Fig. 277 shows this function for 2L = 4,8, 16 as well asthe nonpeiodie Function fe), which ‘we obtain fom f, if we let b> 2, 1 if-texet f(s) = hi fut) = va lo otherwise ‘We now explore what happens to the Fourier coeticients of f, a Lcreases. Since fy, i ven, b, all Foe ay the Euler formulas (6), See. 112, give of Lf nme ona fart on Tf oe EE ae “Ths sequence of Four coffe cll he ampttde spectrum of fy because fi the maximum ample ofthe wave a, eos naa/). Figure 277 shows ths spectrum forthe periods 21 = 4 8,16, We see that for increasing bese amplitudes become more and more dense onthe pstve nauk Wht = nal, Ind for 20 = 48 16 e ag 3.7 apts pe “halve” f the fnton in 9) lah in the figure). Hence for 2/. = 2 we have 2*~? — | amplitudes per half-wave. so that these amplitudes will veal be everywiere dene nthe postive w,-ais (and wil estes o 260), “The tome ofthis example ives an nutve resson of what about to expec if we tu rom our special function wo an abivary ones We sal doeSEC.TU7 Fourier Integral 507 aretorm 2) spa, Aipltde spectrum) ane pened 1) oOo, Dy or ROH wee ee oa babes he ome fa) “ror ee Fig. 277. Waveforms and amplitude spectra in Example 1 From Fourier Series to Fourier Integral We now consider any periodic function f(x) of period 2 that can be represented by a Fourier series Fi02) = do + DY (@q C08 Wx + By iM WX), W_ = and find out what happens if we let L—> 2°, Together with Example 1 the present calculation will suggest that we should expect an integral (instead of a series) involving cos we and, sin wx with w no longer restricted to integer multiples w = w,, = nail. of il. but taking all values. We shall also see what form such an integral might have. If we insert a, and b,, from the Euler formulas (6), Sec. 11.2, and denote the variable of integration by v, the Fourier series of f,(x) becomes. Hb) = 3 ff fut) do + > [ow wax J filo) €05 mye do + singe f Fe(v) sin wo «| We now set Q@+De nt ir: EL Aw = Wye = Wa =508 THEOREM 1 CHAP. 11 Fourier Series, Integrals, and Transforms ‘Then I/L = Aw/rr, and we may write the Fourier series in the form 1 1 i: f= 5p J ful do + 3 [loosen aw Ff) cos wv do + (sin wx) Aw J fe(0) sin we «| ‘This representation is valid for any fixed L, arbitrarily large, but finite. ‘We now let L—» 2 and assume that the resulting nonperiodie function fla) = Jim, f.t0) is absolutely integrable on the x-axis; that is, the following (finite!) limits exist: 2 lim, fi {peo ae + fim J * p00) ax (writen ie "Weal «) ‘Then VL — 0, and the value of the first term on the right side of (I) approaches zero, Also Aw = wil. 0 and it seems plausible that the infinite series in (1) becomes an integral from 0 to %, which represents f(x), namely, @ fo=% f [eo wr fs) cos wo do + sin we fs) sin we «| dw Ir we introduce the notations a= Lf soreormod, Bo = + [0 sin wo a we can write this inthe form © fo) = fam cos wx + Bow) sin ws) dv This is called a representation of f(x) by a Fourier integral. Its clear that our naive approach merely suggests the representation (5), but by no ‘means establishes it; in fact, the limit of the series in (1) as Aw approaches zero is not the definition of the integral (3). Sufficient conditions for the validity of (5) are as follows. itera es en fcr b eed bre es beg) ak and hen by). Arapoin whee fl) ediconinious he veo he Four ese marnes aneSEC. 117 Fourier Integral 509 Applications of Fourier Integrals ‘The main application of Fourier integrals is in solving ODEs and PDEs, as we shall see for PDEs in Sec. 12.6. However, we can also use Fourier integrals in integration and in discussing functions defined by integrals, as the next examples (2 and 3) illustrate. EXAMPLE 2 Single Pulse, Sine Integral Find the Fourier integral representation ofthe function het so ie 27 lo it Waa * fe rc a4 3 Fig. 278 Example 2 Solution, From (4) we obain if” 1 Peep a poeta nd (5) gives the answer © foo 2 [7 coswx sinw ims ete age ‘Th average of he es igh hand iis of fs) atx = 1 elt (1+ 0/2 ht, 72 Furthermore, frm (6) and Theorem I we obtain (tiply by /2) anf Ose), o ait rou. o i eo ‘We mention that this integral i called Direhlet’s discontinous factor. (For PL Dirichlet see Sec. 108) ‘The cake x = Os of partial interest. If += 0, then () gives on [ant arias We eh is ine i he init of he s-aldsn Integead ® su) = [5% ae 4s w+ 2, The graphs of Sie) and ofthe integrand are shown ia Fig. 279. In te ease of « Fourier serie the graphs ofthe partial sums are approximation curves ofthe curve of the psi fonction represented by the erie. Silay inthe eae ofthe Fourie integral (5), approximations are ‘obtained hy replacing by numbers, Hence the integral % i fower,. approximates the right sie in (6) and therefore J)510 CHAP. 11 Fourier Series, Integrals, and Transforms Fig. 279. Sine integral Siu) and integrand Figure 280 shows oscillations nesr the points of discontinuity of f(x). We might expect tat these oscillations isappear aso approaches infinity. Bu this isnot tre; with increasing, they are shied closer othe points “x= =I. This expected bevior, which also occurs in connection with Fourier sere, is known a the Gibbs ‘Phenomenon. (Soo also Problem Set 11.2) We can explain it by represeting (9) in terms of sine integrals as follows. Using (11) in App. A3.1, we have Inthe fint integra on the right we set w+ OS 130+ Da. nthe list inegral we set w — wx 0213 (x= Nha. Sincesin(-1) Then die = aft and 0 = w = a comesponds to 1. Then dw = dit, and 0 = w = eoesponds (0 sin, we thus obtain From this and (8) we se that ou integral (9) equals siete + 1) = 4 ste = sd the onillations in Fig. 280 result fom those ia Fig. 279. Te increase of « amounts to a transformation ofthe scale onthe axis and causes the shit ofthe osilltion (the waves) toward the points of discontinuity “Land 210 tee 2d fa The integral (9) for a = 8,16, and 32SEC.T17 Fourier integral sn EXAMPLE 3 Fig. 261. fla) in Example 3 Fourier Cosine Integral and Fourier Sine Integral For an even or odd function the Fourier integral becomes simpler. Just as in the case of Fourier series (See. 11.3), this is of practical interest in saving work and avoiding errors. The simplifications follow immediately from the formulas just obtained, Indeed, if f(x) is an even function, then BOs) = 0 in (4) and (10) G0) = = [fie os we de. ‘The Fourier integral (5) then reduces to the Fourier cosine integral an $08) = [AQ cos wx de (F even). Similarly, if (6) is odd, then in (4) we have A(w 0 and si (2) 0) = = ff fle) sin wo do. ‘The Fourier integral (5) then reduces to the Fourier sine integral a3) $08) = [Bt sn x dw (F oda) Evaluation of Integrals Earlier inthis section we pointed out thatthe main application of the Fourier integral is in differential equations but that Fourier integral representations also help in evaluating certain integrals To see this, we show the method for an important case, the Laplace imtegras Laplace Integrals ‘We shall derive the Fourier cosine and Fourier sine integrals of 3) ‘The result willbe used to evaluat the so-called Laplace itera where x> O and > O(Fg, 281. Fe nn yg ys Solution. (a) From (10) we have As k Jrvonved | Iv = 0,the expression on the right equals “L(G? + w®,IFv approaches infinity, that expression approaches ‘ero because ofthe exponential ator. Tus 14 A) = ad 0 Ft ‘By substituting this imo (11) we thus obi the Fourier cosine integral representationsm CHAP.11_ Fourier Series, Integrals, and Transforms ‘rom this representation wesc that a5 i te 0, 40) ae © sina, om we emo) = 2 ™ ste 87 gan by a Vo =(: ) e-* sinned eo (E sine 4 con ‘This equals —wi(A® + w) if w = 0, and approaches 0 as v — =. Thus ave 6 aw = From (3) wt i he ont sie gr proetion te 2 [Tw sine Fone S| aw From is we at an aren 0. 80) The inepal (5) and (7 ae aed he Laplace interes 1 Ee ssa Sees 1-6] EVALUATION OF INTEGRALS wi if OR X<1 Show that the given inegral repreens the indicted iO” cowdv alot it ce incon Hie Ue Oh AnverGaroemepatiayes hoped yee = whieh one andits valu tells you what funtion wo consider, 0 red (Show te dtl of your work) 7 ‘ gee £ fee meee eee Saint 5 TW Joosx if 0< [el < wiz wet it x>0 o bean 2 [RSE sw te [ss gga ffiee ORE g 0 law 0 if a> m2 if Oe sf Oe sn if O yo. ay fen = 1 ['a(%) cosa aw (a> 0) (Seale change 13, CAS EXPERIMENT. Approximate Fourier Cosine Integrals Gap the integrals nb. 7,9.and 11382) agen) = [BMH sin de, functions of Graph approximations obexned by h replacing with fine upper limits of your choice a Compare the quay of the approximations. Wate a w=- 4. awinan short report on your empirical results and observations. pl roumen are arsed ey [farce cone REPRESENTATIONS 2 Represent fs) 8 an integra (13, a 24 ws a[L TOES? (b) Solve Prob. § by applying (4) 0 here oF 0 if x>a@ Prob. 7. (©) Verily (2) fr fo) = 10 < ¢ a sins if Oa Solution. Fron he defitions (2) and (5) we obi by iteration 0s wr de o- fi tw fE fum= 2 framerar= ‘This agrees with formulas 1 inthe fst 0 tables in Sec. 11.10 (where & = 1). [Note that for fs} = k= conus (0 = x-< 2, these transforms donot exis. (Why) . Fourier Cosine Transform of the Exponential Function Find 56) Solution. By integration by parts and eeusion, sie fe [etemmedce [2 Ay (comme + wom This ares with formula 3 in Tale 1, Ses 11.10, witha ~ 1, Se aso the next example . What did we do t0 introduce the two integral transforms under consideration? Actually rot much: We changed the notations A and B to get a “symmetric” distribution of the constant 2/7 in the original formulas (10)-(13), Sec. 11.7. This redistribution isa standard convenience, but it is not essential. One could do without it ‘What have we gained We show next that these transforms have operational properties that permit them t0 convert differentiations into algebraic operations Gust asthe Laplace transform does). This isthe key to their application in solving differential equations. Linearity, Transforms of Derivatives If f(a) is absolutely imtegrable (see Sec. 11.7) on the positive x-axis and piecewise continuous (see Sec. 6.1) on every finite interval, then the Fourier cosine and sine transforms of f exis Furthermore, if f and g have Fourier cosine and sine transforms, so does af + bg for any constants a and b, and by (2), Flas + bg) = af 2 [tate + vets 608 wx dx ~af2 [rercommede +» iE fs cose a ‘The right side is aF,(f) + b3,(g). Similarly for &,, by (5). This shows that the Fourier cosine and sine transforms are linear operations, (a) Flaf + bg) = aF (f) + Fg), tb) Flaf + bg) = aF.f) + BFA). o516 THEOREM 1 PROOF ‘CHAP. 11 Fourier Series, Integrals, and Transforms Cosine and Sine Transforms of Derivatives Let f(x) be continuous and absolutely integrable on the x-axis, let f'(x) be piecewise Continuous on every fnite interval, and let let f(x) + 0.as x %. Then 2 (@)— FeAs"@)) = wF, LF} ~ 2 40). | (b) FASO) = —wF LF). This follows from the definitions by integration by parts, namely, FAS) = tz fre cos wx de = /2 Va [rw cos wee | + w [900 sin wear} = £0) + wR: and similarly, FAO) {109 sins ax [1 sin we w [re cos wx «] — wFLF)). . Formula (8a) with f" instead of f gives (when f’, f” satisfy the respective assumptions for f, f" in Theorem 1) 2 FAL) = wH AFD) — fz FO hence by (8b) (oa) Fels") = -w°F LF} ~ 2 FO. Similarly, (9b) FAL) = —WL + I WFO. ‘A basic application of (9) to PDEs will be given in Sec. 12.6. For th show how (9) can be used for deriving transforms.SEC.1L8 Fourier Cosine and Sine Transforms EXAMPLE 3 317 ‘An Application of the Operational Formula (9) Find the Fourier cosine transform &(e7*) of fx) =e, where a > 0. Solution. By citereataion, @~®)" = ae; tas 70) = F0. From this, (9a), and dhe linoaiy (73), “The anewer i (see Table I, See 1.10) Fae*) (8 + W549 = VIR. -Elete) oo . ‘Tables of Fourier cosine and sine transforms are included in Sec. 11.10. FOURIER COSINE TRANSFORM 1 Let f) = -180< 4 <1, fo) = 1ifl 2, Find jaw). 2. Let fo) = vif 0 k Find $00. 3, Derive formula3in Table lof See. 11.10by integration, 4. Find the inverse Fourier cosine transform 2) from the answer to Prob. 1. Hint. Use Prob. 4 in See. 11.7 8. Obiain F"(1(1 + w2) from Prob. 3 in See. 11.7 6, Obtain F2%(e"*) by integration 7. Find FAC) ~ x 00s (x12). Hint. Use Prob. 5 in See. 11.7. 8 Let fix) = 27if0 L Find Fup) 9. Does the Fourier cosine transform of x" sin.x exist? Of x°* cos.x? Give reasons. 10, J) = 1 (0 m Find 4). Compare with Prob. 6 in Sec, 11.7. Comment. 18, In Table I of Sec. 11.10 obtain formula 2 from formula 4, osing MQ) = Va (0) in App. 3.11 16. Show that F(x") y-¥? by setting wa = 1? and using S(@) = Vm in G8) of App. 3.1 17, Obtain ¥,(e-* from (8) and formula 3 in Table To See. 11.10. 18, Show that Fs-%) = 20! Hin, Set we = 2, integrate by parts, and use C=) = Vai in (38) of App. 3.1, 419, (Seale change) Using the notation of (S), show that flax) bas the Fourier sine transform (1/a)jw/a, 20. WRITING PROJECT. Obtaining Fourier Cosine and Sine Transforms. Write a short report on ways ‘of obaining these transforms, giving illstations with examples of your own,518 CHAP. 11 Fourier Series, Integrals, and Transforms 11.9 Fourier Transform. Discrete and Fast Fourier Transforms “The two transforms in the lst section are real. We now considera third one, called the Fourier transform, which is complex. We shall obtain this transform from the complex. Fourier integral, which we explain first. Complex Form of the Fourier Integral ‘The (real) Fourier integral is (ee (4), (5), See. 11.7] 505) = [1A cos er + BO) sin ws de where JF 500) cos we do, 0) = 2 fe sin wo Substituting A and B into the integral for f, we have ie f= = [J 10) loose cos wx + sin ww sin ws] do di By the addition formula for the cosine [(6) in App. A3.1] the expression in the brackets [++] equals cos (wo — wx) of, since the cosine is even, cos (wx ~ wo). We thus obtain as) fey = = [UL cos wx — wo} aw ‘The imegral in brackets san een function of ll i Fe), beat cs (ns — wo) an even fon fw, the fenton j does ot depend on, and we integrate with expect from "10 Ths (nt the change of the ization nit) fear 10)= 2 J [fs eo toe wey do] ‘We claim that the integral of the form (1) with sin instead of eos is zero: @ FL [se sin our we) ao] aw = 0 brackets ano anton of allt Gn Hence the inal of Gi) from == 0 seo, asclaned We now take the integrand of (I) plus # (= V=T) times the integrand of (2) and use the Euler formula [(11) in Sec. 2.2] @ Me osx + i sinx,SEC. 119. Fourier Transform. Discrete and Fast Fourier Transforms 519 THEOREM 1 ‘Taking wx ~ we instead of x in (3) and multiplying by f(v) gives Flv) cos (wx — we) + if(0) sin (wx — we) = Floyer*—™, Hence the result of adding (1) plus times (2), called the complex Fourier integral, is, ses ® sa= 52 Jf sere* do dw G=V It is now only a very short step to our present goal, the Fourier transform, Fourier Transform and Its Inverse Writing the exponential function in (4) as a product of exponential functions, we have ° o= Tf [+— foe «| aw ‘The expression in brackets is a function of w, is denoted by f(w), and is called the Fourier transform of f: writing » = x, we have 6 fon = See [poner ae With this, (5) becomes 1 ons o $0) = Fee | fone aw and is called the inverse Fourier transform of f(w). ‘Another notation for the Fourier transform is F= FN, so that f= FMS). ‘The process of obtaining the Fourier transform (4) = f from a given f is also called the Fourier transform or the Fourier transform method. Conditions sufficient for the existence of the Fourier transform (involving concepts defined in Secs. 6.1 and 11.7) are as follows, as we state without proof a Existence of the Fourier Transform | YF) is absolutely integrable on the x-axis and piecewise continuous on every finite interval, then the Fourier transform f(w) of f(x) given by (6) exists. i: a —__|520 EXAMPLE 1 EXAMPLE 2 CHAP. Il Fourier Series, Integrals, and Transforms Fourier Transform Find the Fourier transform of f(s) = 1 if|a|< 1 and fs) = 0 eterwise, Solution, Using (6) and integrating. we obtain pre oe Vaz’ =v |. > Seva “ va J, {Asin @) we have & = cos w + Fin e° = cos w ~ isin wy ad by subercton © = 2isinw Substituting tis in the previous formula onthe right, we se that drops out and we obtain the answer : Vin ~@+ iw) |g V2mr(a + iw) Physical Interpretation: Spectrum ‘The nature ofthe representation (7) of (2) becomes clear if we think oft sa superposition of sinusoidal oscillations of all possible frequencies, called spectral representation. This name is suggested by optics, where light is such a superposition of colors (Grequencies). In (1), the “spectral density” fw) measures the intensity off) in the frequency interval between w and w + Aw: ‘(Aw small, fixed). We claim that in connection with vibration, the integral jroo? aw can be interpreted asthe total energy ofthe physical system. Hence an integral of Lf)? from ato gives the contribution of the frequencies w between a and bo te total ners To make this plausible, we begin with a mechanical system givin a single frequency, namely, the harmonie osilator (mass on a spring, Se. 24) my" + ky = 0. Here we denote time ¢ by x. Multiplication by y’ gives my’y” + ky'y = 0. By integration, dmv? + By? = Eo = const where v = "is the velocity. The first term is the kinetic energy, the second the potential energy, and Ey the total energy of the system. Now a general solution is (use (3) in Sec. 11.4 with t = x)SEC. 119 Fourier Transform. Discrete and Fast Fourier Transforms 521 THEOREM 2 PROOF = ay COS wox + By sin wox = cye™* + ce, ang? = hm where cy = (ay — iby)/2, cy = 2 = (ay + iby)/2. We write simply A = cye™>*, B = c.ye~*", Then y = A + B. By differentiation, v = y' = A’ + B' = iwg(A ~ B). Substitution of v and y om the left side of the equation for Ep gives Eq = fv? + dy? = dmn(iwe)*(A — BY? + $R(A + BY. Here wo” = k/m, as just stated; hence mw? = k. Also i? = —1, so that Ey = 3e[-(A ~ BY + (A + BY] = 2kAB = heyee%c_pe“ "= Deyo = le Hence the energy is proportional to the square of the amplitude |ey| [As the next step, if a mare complicated system leads to a periodic solution y = f(x) that can be represented by a Fourier series, then instead of the single energy term |cy)? we get a series of squares |cy|® of Fourier coefficients c,, given by (6), Sec. 11.4. In this case wwe have a “discrete spectrum” (or “point spectrum”) consisting of countably many isolated frequencies (infinitely many, in general), the corresponding |c,|® being the contributions to the total energy. Finally, a system whose solution can be represented by an integral (7) leads tothe above integral for the energy, as is plausible from the cases just discussed. Linearity. Fourier Transform of Derivatives ‘New transforms can be obtained from given ones by ir ity of the Fourier Transform The Fourier transform isa linear operation; that is, for any functions f(x) and g(x) whose Fourier transforms exist and any constants a and b, the Fourier transform | of af + bg exists, and | @) Flaf + be) = aF(f) + bHg). ‘This is true because integration is a linear operation, so that (6) gives Flaflx) + be) Tae [late + vate ax aoe Pipe macs 1 Pert a aF(f(x)} + bF{g@)}. = In applying the Fourier transform to differential equations, the key property is that differentiation of functions corresponds to multiplication of transforms by iw:sa CHAP. Fourier Series, Integrals, and Transforms THEOREM 3. Fourier Transform of the Derivative of f(x) Let f(x) be continuous on the x-axis and f(x) > 0 as |x| —» >. Furthermore, let §'(8) be absolutely integrable on the x-axis, Then ) FF") = wFFO)}. Da a em a PROOF From the definition of the Fourier transform we have so) = ee [seve ax Integrating by parts, we obtain FIFO) = iw fF faye «| Since fx) — 0 as |x| > ~, the desired result follows, namely, FU") = 0+ iw FFE) . Two successive applications of (9) give FS") = HS) = GwP FN. —w?, we have for the transform of the second derivative of f a0) FSO) = —w?F (FO) Similarly for higher derivatives. ‘An application of (10) to differential equations will be given in See. 12.6. For the time being we show how (9) can be used to derive transforms, EXAMPLE 3 Application of the Operational Formula (9) Find the Fourier transform of e“* from Table I, See 11.10, Solution, We wse (9) By formals 9 in Table ISEC.119. Fourier Transform. Discrete and Fast Fourier Transforms 523 THEOREM 4 PROOF Convolution ‘The convolution f » g of functions f and g is defined by ay nay = eo = f_Flrrece=prdo = J fee ~ Prato) ap ‘The purpose is the same as in the convolution of two functions and then ‘as multiplying the transforms of these functions (and multiplying them by V se of Laplace transforms (Sec. 6.5): taking the the transform of the convolution is the same /2a): Convolution Theorem Suppose that fix) and g(x) are piecewise continuous, bounded, and absolutely integrable on the x-axis. Then «2 FE « g) = Via F( NF). By the definition, FE a) JF seoses— nap ea Vie An interchange of the order of integration gives xf . i “fo (x — pe“ dx dp FE 9 = ae I Instead of we now take + ~ p = q as a new variable of integration, Then x = p + q and 1 a lik -twcp@ a0e9= Te [J sore aq ap. ‘This double integral can be written as a product of two integrals and gives the desired result oe a SUF #8) = ee [sie dp J ataert dg = L [viz anliviz 5) = Vie! = Sag [V2 FNllV27 Hw) = V27 INF. . By taking the inverse Fourier transform on both sides of (12), writing f = F(f) and Fg) as before, and noting that V2" and U/V/Ze in (12) and (7) cancel each othe, we obtain a3) = 00 = J fonsone aw 4 formula that will help us in solving pastil differential equations (See. 12.6).524 CHAP. Tl Fourier Series, Integrals, and Transforms Discrete Fourier Transform (DFT), Fast Fourier Transform (FFT) In using Fourier series, Fourier transforms, and trigonometric approximations (See. 11.6) ‘we have to assume that a function f(x), to be developed or transformed, is given on some interval, over which we integrate in the Euler formulas, etc. Now very often a function ‘G2 is given only in terms of values at finitely many points, and one is interested in extending Fourier analysis to this case. The main application of such a “discrete Fourier analysis” concerns large amounts of equally spaced data, as they occur in telecommunication, time series analysis, and various simulation problems. In these situations, dealing with sampled values rather than with functions, we can replace the Fourier transform by the so-called discrete Fourier transform (DFT) as follows. Let f(x) be periodic, for simplicity of period 27, We assume that N measurements of (2) ate taken over the interval 0 2mrat regularly spaced points ant ua na i k=O1 N= 1 We also say that f(2) is being sampled at these points, We now want to determine a complex trigonometric polynomial as) a) =D ce that interpolates f(x) at the nodes (14), that is, g(x.) = Flx.), written out, with f,, denoting. Sods 06) f= fle) = gad = Zien K=O LN. Hence we must determine the coefficients ¢q *** cy such that (16) holds. We do this by an idea similar to that in Sec. 11.1 for deriving the Fourier coefficients by using the orthogonality ofthe trigonometric system. Instead of integrals we now take sums, Namely, we multiply (16) by e~*"** (note the minus!) and sum over k from 0 to N ~ 1. Then we interchange the order of the two summations and insert x, from (14). This gives i) Spe a ST eg aS cy Felt rtm Now [erommanarp We donote [-- “] by r. For n = m we have r = e° = 1. The sum of these terms over k equals WV, the number of these terms. For n # m we have r # I and by the formula for a geometric sum {(6) in Sec. 15.1 with q = rand n = N — 1]SEC. 119. Fourier Transform. Discrete and Fast Fourier Transforms 325 EXAMPLE 4 because r™ + indeed, since k, m, and m are integers, PX = eimmiok = cos Dakin — m) + i sin 2mk(n — m) = 1+ O= 1 ‘This shows that the right side of (17) equals ¢,.V. Writing m for m and dividing by N, we thus obtain the desired coefficient formula (18*) on i D fee f= flows ms N ‘Since computation of the c,, (by the fast Fourier transform, below) involves successive halfing of the problem size Nit is practical to drop the factor LAV from, and define the discrete Fourier transform of the given signal f = [fy **~ fy-s]” to be the vector f= [Fo --- fai} with components as) n= Neu =D fre fx = flrs This is the frequency spectrum of the signal. In vector notation, f = Fyf, where the NV x N Fourier matrix Fy = [éq,) has the entries {given in (18)) et Lr Pe ayy ene = eH = where n, k -1 Discrete Fourier Transform (DFT). Sample of N= 4 Values HAN 2 gant thas w™* LeLW’= 4 measurements (sample values) be given Then Lette sample values be, say f= [0 1 4 9]¥. Then by (18) and (19, we Pw hor. afer 4 wt “! tone -t alfa] [see 20) f= R= f t = ¥ " ieeeie eal 4 - “ wl los o) Lae From the fist matrix in (20) is easy to fer what Fy 1oks like for ablvary W, which in practice may be 1000 or more, for reasons given below: . From the DFT (the frequency spectrum) f {we can recreate the given signal [a] satisty = Fyvf, as we shall now prove, Here Fry and its complex conjugate By = (21a) 1b)526 PROOF CCHAP.T1_ Fourier Series, Integrals, and Transforms We prove (21). By the multiplication rule (row times column) the product matrix Gy = FyFy = [gy in 2la) has the entries gy, = Row j of Fy, times Column k of Fy. ‘That is, writing W= iw, we prove that = GW) + WE! + + GW 0 if jek = WOR Wa Wht Nit 7=k Indeed, when j = k, then whw* = (wy = (2ANe~2™)F = 1 = 1, 50 that the sum, of these N terms equals N; these are the diagonal entries of Gy. Also, when j # k, then W # | and we have a geometric sum (whose value is given by (6) in Sec. 15.1 with q= Wandn=N-1) _ we Wawa gwete t 1-W because WN = (wiw) = (ple Prk = VIE = 1 . We have seen that f is the frequency spectrum of the signal f(x). Thus the components £,, of give a resolution of the 2-periodie function f(x) into simple (complex) harmonics. Here one should use only n’s that are much smaller than N/2, o avoid aliasing. By this we ‘mean the effect caused by sampling at too few (equally spaced) points, so that, for instance, {na motion picture rotating wheels appear as rotating too slowly or even in the wrong Sense Hence in applications, is usually large. But this poses a problem. Eq, (18) requires O(N) ‘operations for any particular n, hence O(N*) operations for, say, all n < N/2. Thus, already {or 1000 sample points the straightforward calculation would involve millions of operations. However, tis difficulty ean be overcome by the So called fast Fourier transform (FFT), for which coxes are readily available (e.g. in Maple). The FFT is a computational methoxt for the DFT that needs only O(\) loge.N operations instead of O(N?) It makes the DFT a practical tool for large N. Here one chooses N = 2° (p integer) and uses the special form Of the Fourier matrix to break down the given problem into smaller problems. For instance, ‘when N= 1000, those operations are reduced by a factor 1000Vlog 1000 ~ 100. ‘The breakdown produces two problems of size M = N/2. This breakdown is possible because for N = 2M we have in (19) = (eran? ete ‘The given vector f = [fy *** fy-1l" is split into two vectors with Mf components each, namely, fey = [fo fa. *** foal” containing the even components of f, and fos = Uf fx ‘++ yal" comtaining the odd components of f. For fyy and foq We determine the DFTs fee = [Foro fous Fex.v-a]" = Fufer and fea = oars fous *** Foan-i)” = Fufoa involving the same M X M matrix Fy. From these vectors we obtain the components of the DFT of the given vector f by the formulas @ fe (0) Frvar Fenn + Wr" Foam .M— 2 eon — Wr" FoanSEC.1L9. Fourier Transform. Discrete and Fast Fourier Transforms 327 EXAMPLE 5 For N = 2? this breakdown can be repeated p — 1 times in order to finally artive at N/2 problems of size 2 each, so that the number of multiplications is reduced as indicated above. ‘We show the reduction from NV = 4 10 M = 72 = 2 and then prove (22), Fast Fourier Transform (FFT). Sample of N= When N= 4, then w = wy Consequently, Values at vo [P] m= [) DES £e7) toffee TE From this and (223) we obtain hence w= Wap fo Sexo * wn*foan = Un + Sa) + Us f= fot fat Sat Sa Fo Ja) ~ Sy + fa) = fo thy ~ fa * Hh fi = fexa + wn Peas Sinilary, by 226) 42> fevo ~ wr°foao ~ (fo + fa) ~ + fad fo~ fit fa~ fa a= feva ~ wx'foas = So ~ fa) ~ (iS ~ fa) = fo + if ~ fa ~ ie ‘This agrees with Example 4, as canbe seen by replacing O 1, 4,9 with fo Sas Fa Soe . We prove (22). From (18) and (19) we have for the components of the DFT Dwr Splitting into two sums of M = N/2 terms each gives n= wae + Dw ana 0 io We now use wy? = wy, and pull out wy" from under the second sum, obtaining 23) Fn = Z wit foru + wn" D wal Foe io ra ‘The two sums are fy, and fogs the components of the “hal F fog Formula (22a) is the same as (23). In (22) we have n + M instead of n. This causes a sign change in (23), namely —w" before the second sum because 1" wransforms F fy and yl = enn = pte ‘This gives the minus in (22) and completes the proof. .528 CHAP. 11 Fourier Series, Integrals, and Transforms (acc ese eee 1, (Review) Show that 1/7 oF =e = ising, ie +e = 2 cons, FOURIER TRANSFORMS BY INTEGRATION. Find the Fourier transform of f(x) (without using Table HI in See. 11.10). Show the details. fe ite<0 K=O 2. fa) = 0 ifs>0 ik oce O and Oifx <0 fom formula 5 in Table HL and (9) in the text, Hin: Consider se and e*, 11. Obtain Fe") from formula 9 in Table HL, 12, Oban formula 7 in Table I from forma 8. 13, Obian formula Lin Table I from forma 2. 1M, TEAM PROJECT. Shifting. (a) Show thst ff) tas n Fourier transform, so does flr ~ 2) and Ff — ay} = eo F( FO). (b) Using, oban formula Lin Tale Se 1.10, from formula 2. (©) Shitng onthe w- Anis, Stow that ff isthe Fourier transfor of then fw ~ a) she Fourier transform of e*f(x). (@ Using (¢), obtain formula 7 in Table TT from tnd formula § from 2.SEC.TMI0. Tables of Transforms 529 11.10 Tables of Transforms Table |. Fourier Cosine Transforms See (2) in See. 11.8. | fy fn) = FD) | { ifoo a) ere own Cet 1 wn sles @>y |e Via fees «oo | [Papeete RO 6 > | be are wip Ree + wy Real part feosx if0o| + (é - 2) costar?) @ zo (T-F * 1 wom 9) sincas®) (a> 0) | em oos (= + 2) | sin ax [= 10) as (a>0) Vz (1 = ulw — a) (See See. 6.3.) " | | iz | 2 12) Jofax) (a > 0) ly ema! (1 = uw = a) (See Sees. 5. Va Ve- Lt __ IfCHAP. 11 Fourier Series, Integrals, and Transforms Table ||, Fourier Sine Transforms See (5) in See. 11.8. 0) 1 if0o | @>0 7 | ate @>o) | Im (a + wy" Va @ + wy Imaginary part 8 wen (a> oe ene c oe | Ga sins f00 (See Sec. 6.3.) 2a 12 actin (a> 0)SEC. 1110 Tables of Transforms 31 Table Ill. Fourier Transforms See (6) in See. 119. x if0 | set 0 otherwise Vina + iw) ; | [Rewer chon — gacton oO otherwise WV2e(a — iw) aoe [Z sate =o 0 otherwise Ve ws. (ie ib o532 CHAP. 11 Fourier Series, Integrals, and Transforms Care es eeee eta nes Be eo el 1, What is Fourier series? A Fourier sine series? A balfsange expansion? 2. Can a discontinuous function have a Fourier series? A “Taylor series? Explain, 3. Why did we start with period 2” How did we proceed to funetions of any period p? 4. Whatis the trigonometric system? Its main property by which we obtained the Euler formulas? '5. What do you know about the convergence of a Fourier {6 What isthe Gibbs phenomenon? 7. What is approximation by trigonometric polynomials? ‘The minimum square error? ‘8, What is remarkable about the response of a vibrating system to an arbitrary periodic force? 9. What do you know about the Fourier integra? Its applications? 10, What is the Fourier sine wansform? Give examples. 11-20] FOURIER SERIES Find the Fourier series of f(x) as given over one period. Sketch f(a). (Show the details of your work.) wesw-[t ise os & @ O 0, fl) = 0ix <0. 40. Find F_(e"**) and F,(e°**) by Formulas involving second derivatives. Biviteaiae tis ol el gmt at OTK Fourier series concern periodic functions f(x) of period p = 2L, that is, by definition f(x + p) = f(x) for all x and some fixed p > 0; thus, f(x + mp) = f(x) for any integer n, These series are of the form | wo $0) = 9 +B (0 AE x + bgsin x) (See. 11.2) with coefficients, called the Fourier coefficients of f(x), given by the Euler formulas (Sec. 11.2) Lf nm n= 7 [_f09 008 Fae | 1 f nm by = 7 J foo sin TE ae ay $69) = ay + Z (ay cos nx + by sin nx) || with the Fourier coefficients of f(x) (See. 11.1) =a I “fla tf Fed eosme ds, by =~ se sine th Foner sist oe findansnsl in conection with peat pheomeaa, | particularly in models involving differential equations (Sec. 11.5, Chap. 12). If f(x) is even [f(—x) = f(x)] or odd [f(—x) = ~f(0)], they reduce to Fourier cosine or Fourier sine series, respectively (Sec. 11.3). If f(a) is given for 0 5 x 5 L only, | it has two half-range expansions of period 2Z, namely, a cosine and a sine series (Sec. 11,3),534 CHAP. Il Fourier Series, Integrals, and Transforms ‘The set of cosine and sine functions in (1) is called the trigonometric system. Its most basic property is its orthogonality on an interval of length 2L; that is, for all integers m and n # m we have and for lingers adm 0s ig MA ae = eras re This orthogonality was crucial in deriving the Euler formulas (2) Partial sums of Fourier series minimize the square error (Sec. 11.6). [Ideas and techniques of Fourier series extend to nonperiodic functions f(x) defined ‘on the entire real line; this leads to the Fourier integral @ $65) = [1400 cos ee + BOW) sin wx] dw (See. 11.7 where ae a @ Ao =— fi fey coswe de, — Bow = — f_fle) sin we ao or, in complex form (See. 11.9), 1 * ay) glo ;o $9) = Tee J fone ae where A eT Abe ae © fon = Tee [seers Formula (6) transforms f(x) into its Fourier transform f(w), and (5) is the inverse transform, Related to this are the Fourier cosine transform (Sec. 11.8) | Fi eae | o Faoy= JZ |, se cosmeds and the Fourier sine transform (Sec. 11.8) ®) Fo = J |) 100 sinwrde. \a The discrete Fourier transform (DFT) and a practical method of computing it, called the fast Fourier transform (FFT), are discussed in See. 11.9.