2.1.

GENERAL SOLUTION TO WAVE EQUATION

1.138J/2.062J/18.376J, WAVE PROPAGATION Fall, 2004 MIT Notes by C. C. Mei CHAPTER TWO

ONE DIMENSIONAL WAVES

General solution to wave equation

It is easy to verify by direct substitution that the most general solution of the one dimensional wave equation: 2 2 = c2 2 t2 x can be solved by (x, t) = f (x ct) + g(x + ct) (1.2) (1.1)

where f () and g() are arbitrary functions of . In the x, t (space,time) plane f (x ct) is constant along the straight line x ct = constant. Thus to the observer (x, t) who moves at the steady speed c along the positivwe xaxis., the function f is stationary. Thus to an observer moving from left to right at the speed c, the signal described initially by f (x) at t = 0 remains unchanged in form as t increases, i.e., f propagates to the right at the speed c. Similarly g propagates to the left at the speed c. The lines x ct =constant and x + ct = contant are called the characteristic curves (lines) along which signals propagate. Note that another way of writing (1.2) is (x, t) = F (t x/c) + G(t + x/c) Let us illustrates an application. (1.3)

Branching of arteries

References: Y C Fung : Biomechanics, Circulation. Springer1997 M.J. Lighthill : Waves in Fluids, Cambridge 1978.

2.2 BRANCHING OF ARTERIES Recall the governing equations for pressure and velocity 2p 2p = c2 2 t2 x 2u 2u = c2 2 t2 x The two are related by the momentum equation The general solutions are : p = p+ (x ct) + p (x + ct) u = u+ (x ct) + u (x + ct) Since p = p+ + p , x and p u = t x

(2.1) (2.2)

(2.3)

(2.4) (2.5)

u = cu+ + cu t where primes indicated ordinary dierentiation with repect to the argument. Equation (2.3) can be satised if p+ = cu+ , p = cu Denote the discharge by Q = uA then ZQ = u A = p where p c (2.8) = A Q is the ratio of pressure to ux rate and is call the impedance. It is the property of the Z= tube. Now we examine the eects of branching; Referening to gure x, the parent tubes branches into two charaterized by wave speeds C1 and C2 and impedaces Z1 and Z2 . An incident wave approaching the junction will cause reection in the same tube p = pi (t x/c) + pr (t + x/c) (2.9) (2.7) (2.6)

2.2. WAVES DUE TO INITIAL DISTURBANCES

and transmitted waves in the branches are p1 (t x/c1 ) and p2 (t x/c2 ). At the junction x = 0 we expect the continuity of pressure and uxes, hence pi (t) + pr (t) = p1 (t) = p2 (t) p1 p2 pi pr = + Z Z1 Z2 wave, then pr (t) R= = pi (t) Similarly the tranmission coecients are T = p2 (t) p1 (t) = = pi (t) pi (t)

2 Z 1 Z1 1 Z 1 Z

(2.10) (2.11)

Dene the reection coecient R to be the amplitude ratio of reected wave to incident +
1 Z 1 1 Z1

+ +

1 Z2 1 Z2

(2.12)

1 Z

1 Z2

(2.13)

Note that both coecients are constants depending only on the impedances. Hence the transmitted waves are similar in form to the incident waves except smaller by the factor T . The total wave on the incidence side is however very dierent.

Waves in an innite domain due to initial distur bances

Recall the governing equation for onedimensional waves in a taut string

2 2u 2 u c 2 = 0, t2 x

< x < .

(3.1)

Let the initial transverse displacement and velocity be given along the entire string u(x, 0) = f (x) u (x, t) = g(x), t (3.2) (3.3)

where f (x) and g(x) are nonzero only in the nite domain of x. At innities x , u and u/t are zero for any nite t. These conditions are best displayed in the spacetime diagram as shown in Figure 1.

2.2. WAVES DUE TO INITIAL DISTURBANCES

u tt =c u xx

x u=f(x ) ut =g(x)

Figure 1: Summary of the initialb oundaryvalue problem In (3.1.1) the highest time derivative is of the second order and initial data are prescribed for u and u/t. Initial conditions that specify all derivatives of all orders less than the highest in the dierential equation are called the Cauchy initial conditions. Recall that the general solution is u = () + () = (x + ct) + (x ct), (3.4)

where and are so far arbitrary functions of the characteristic variables = x + ct and = x ct respectively. From the initial conditions we get u(x, 0) = (x) + (x) = f (x) u (x, 0) = c (x) c (x) = g(x). t The last equation may be integrated with respect to x = 1 c

x x0

(3.5)

g(x )dx + K,

(3.6)

where K is an arbitrary constant. Now and can be solved from (3.1.6) and (3.1.7) as functions of x, 1 1 x (x) = [f (x) + K] + g(x )dx 2 2c x0 1 x 1 (x) = [f (x) K] g(x )dx , 2 2c x0

2.2. WAVES DUE TO INITIAL DISTURBANCES

Range of influence
c c

(x,t)
1

x0

x-ct Domain of dependence

x+ct

Figure 2: Domain of dependence and range of inuence where K and x0 are some constants. Replacing the arguments of by x + ct and of by x ct and substituting the results in u, we get 1 1 xct f (x ct) u(x, t) = g dx 2 2c x0 1 x+ct 1 + f (x + ct) + g dx 2 2c x0 1 x+ct 1 [f (x ct) + f (x + ct)] + = g(x ) dx , 2 2c xct

(3.7)

which is dAlemberts solution to the homogeneous wave equation subject to general Cauchy initial conditions. To see the physical meaning, let us draw in the spacetime diagram a triangle formed by two characteristic lines passing through the observer at x, t, as shown in Figure 2. The base of the triangle along the initial axis t = 0 begins at x ct and ends at x + ct. The solution (3.1.9) depends on the initial displacement at just the two corners x ct and x + ct, and on the initial velocity only along the segment from x ct to x + ct. Nothing outside the triangle matters. Therefore, to the observer at x, t, the domain of dependence is the base of the characteristic triangle formed by two characteristics passing through x, t. On the other hand, the data at any point x on the initial line t = 0 must inuence all observers in the wedge formed by two characteristics drawn from x, 0 into the region of t > 0; this characteristic wedge is called the range of inuence. Let us illustrate the physical eects of initial displacement and velocity separately.

2.2. WAVES DUE TO INITIAL DISTURBANCES

u, t

Figure 3: Waves due to initial displacement Case (i): Initial displacement only: f (x) = 0 and g(x) = 0. The solution is 1 1 u(x, t) = f (x ct) + f (x + ct) 2 2 and is shown for a simple f (x) in Figure 3 at successive time steps. Clearly, the initial disturbance is split into two equal waves propagating in opposite directions at the speed c. The outgoing waves preserve the initial prole, although their amplitudes are reduced by half. Case (ii): Initial velocity only: f (x) = 0, and g(x) = 0. Consider the simple example where g(x) = g0 when |x| < b, and

= 0 when |x| > 0. Referring to Figure 4, we divide the x t diagram into six regions by the characteristics with B and C lying on the x axis at x = b and +b, respectively. The solution in various regions is: u=0 in the wedge ABE; u= in the strip EBIF ; 1 x+ct go g0 dx = (x + ct + b) 2c b 2c 1 x+ct u= go dx = go t 2c xct

3.2. REFLECTION FROM THE FIXED END

u, t E F

Figure 4: Waves due to initial velocity in the triangle BCI; u= in the wedge F IG;

1 b go b g0 dx = 2c b c

go 1 b u= g0 dx = (b x + ct) 2c xct 2c u=0

in the strip GICH; and

in the wedge HCD. The spatial variation of u is plotted for several instants in Figure 4. Note that the wave fronts in both directions advance at the speed c. In contrast to Case (i), disturbance persists for all time in the region between the two fronts.

Reection from the xed end of a string

Let us use the dAlembert solution to a problem in a half innite domain x > 0. Consider a long and taut string stretched from x = 0 to innity. How do disturbances generated near the left end propagate as the result of initial displacement and velocity? At the left boundary x = 0 must now add the condition u = 0, x = 0, t > 0. (4.1)

In the spacetime diagram let us draw two characteristics passing through x, t. For an observer in the region x > ct, the characteristic triangle does not intersect the time

3.4. FORCED WAVES IN AN INFINTE DOMAIN

axis because t is still too small. The observer does not feel the presence of the xed end at x = 0, hence the solution (3.1.9) for an innitely long string applies, u= 1 1
[f (x + ct) + f (x ct)] + 2 2c

x+ct

g( )d,
xct

x > ct.

(4.2)

But for x < ct, this result is no longer valid. To ensure that the boundary condition is satised we employ the idea of mirror reection. Consider a ctitious extension of the string to < x 0. If on the side x < 0 the initial data are imposed such that f (x) = f (x), g(x) = g(x), then u(0, t) = 0 is assured by symmetry. We now

have initial conditions stated over the entire x axis u(x, 0) = F (x) and ut (x, 0) = G(x) where F (x) =
f (x)

< x < ,

if x > 0 if x < 0

f (x) g(x)

G(x) =

if x > 0 if x < 0.

g(x)

These conditions are summarized in Figure 5. Hence the solution for 0 < x < ct is u =
x+ct 0 1 1 [F (x + ct) + F (x ct)] + + G(x )dx 2 2c xct 0 0 x+ct 1
1 [f (x + ct) f (ct x)] + =
+
g(x )dx 2 2c ctx 0 1 ct+x 1 [f (x + ct) f (ct x)] + = g(x )dx . 2 2c ctx

(4.3)

The domain of dependence is shown by the hatched segment on the x axis in Figure 6.

Forced waves in an innite domain

Consider the inhomogeneous wave equation 2u 2u = c2 2 + h(x, t) t2 x t > 0, |x| < ,

3.4. FORCED WAVES IN AN INFINTE DOMAIN

10

u=0

u tt =c 2u xx

x u=-f(-x) ut =-g(-x) u=f(x ) ut =g(x)

Figure 5: Initialb oundaryvalue problem and the mirror reection

x+ct=x 0+ct 0 t x-ct=x0-ct 0 ( x0,t 0 )

( x 0-ct 0 ,0)

( ct0-x 0 ,0) ( x 0+ct0 ,0)

Figure 6: Reection from a xed end

3.4. FORCED WAVES IN AN INFINTE DOMAIN

11

where h(x, t) represents forcing. Because of linearity, we can treat the eects of initial data separately. Let us therefore focus attention only to the eects of persistent forcing and let the initial data be zero, u(x, 0) = 0, The boundary conditions are u 0, |x| . (5.2)

t=0

u t

= 0,

(5.1)

Let the Fourier transform of any function f (x) and its inverse f () be dened by f () =

f (x) eix dx

(5.3)

1 f () eix d f (x) = (5.4) 2 The transformed wave equation is now an ordinary dierential equation for the transform of u(x, t), i.e., u(, t),
d2 u
2 2 +c u=h t>0 2 dt where h(, t) denotes the transform of the forcing function. The initial conditions for u are: u(, d 0) = g(). dt Let us hide the parametric dependence on for the time being. The general solution u(, 0) = f (), to the the inhomogeneous secondorder ordinary dierential equation is u = C1 u1 (t) + C2 u2 (t) +

0 t

h( ) [1 ( )2 (t) u2 ( )2 (t)] d, u u u W

(5.5)

where u1 and u2 are the homogeneous solutions u1 = eict and W is the Wronskian W = u1 u2 u2 u1 = 2ic = constant. The two initial conditions require that C1 = C2 = 0, hence the Fourier transform is u=

0 t

u2 = eict

h(, ) ic(t ) eic(t ) d. e 2ic

(5.6)

2.6. STRING EMBEDDED IN AN ELASTIC SUROUNDING The inverse transform is u(x, t) = 1 2

0 t

12

1 2c

h(, ) i(x+c(t )) e ei(xc(t )) d i

(5.7)

The integrand can be written as an integral: u=

xc(t ) 1 t 1 d d h(, ) d ei 2 0 2c xc(t )

By interchanging the order of integration u= 1 2c

0 t

xc(t )

d
xc(t )

1 2

d ei h(, )

and using the denition of Fourier inversion, we get 1 u(x, t) = 2c

0 t

xc(t )

d h(, )
xc(t )

(5.8)

The righthand side is the integration of h over the characteristic triangle dened by the two characteristics passing through (x, t). in the x t plane. Thus the observer is aected only by the forcing inside the characteristic triangle. For nonzero initial data u(x, 0) = f (x) and ut (x, 0) = g(x), we get by linear super position the full solution of DAlambert 1 x+ct 1 [f (x + ct) + f (x ct)] + u(x, t) = dg() 2 2c xct x+c(t ) 1 t + d d h(, ), 2c 0 xc(t ) The domain of dependence is entirely within the characteristic triangle.

(5.9)

String embeded in an elastic surounding

If the lateral motion of the string is restrained by elastic springs along the entire

Reference : Gra : Waves in Elastic Solids

length, the governing equation can be found from 1.1 by replacing the external pressure by the elastic restoring force KV per unit length, 2V 2V = T 2 KV t2 x (6.1)

2.6. STRING EMBEDDED IN AN ELASTIC SUROUNDING which can be written as 1 2V 2V K = V 2 t2 2 co x T where co =

13

(6.2)

(6.3)

6.1

Monochromatic waves

For any linearized wave problem, if the range of the spatial corrdinate x is (, ), and all coecients are independent of x, t, then the rst task is to examine the physics of sinusoidal wave train of the form: V (x, t) = |A| cos(kx it A ) = Aeikxit

(6.4)

where A = |A|eiA is a complex number with magnitude |A| and phase angle A . After examining the physical meaning of this special type of waves, it is possible to use the principle of superposition to construct more general solutions. It is customary to omit the symbol = the real part of for the sake of brevity, i.e., V (x, t) = Aeikxit First of all a few denitions about sinusoidal waves in general. We shall call (x, t) = kx t (6.6) (6.5)

the wave phase. Clearly the trigonometric function is periodic in phase with the period 2. In the x, t plane, V has a constant value along a line of contant phase. In particular, = 2n, (n = 0, 1, 2, ) correspond to the wave crests where V = |A| is the greatest. On the other hand, = (2n + 1), (n = 0, 1, 2, ) correspond to the wave troughs where V = |A| is the smallest. |A| is half of the separation between adjacent crests and troughs and is called the wave amplitude; we also call A the complex amplitude. Clearly
x

represents the number of phase lines per unit distance, i.e., the density of

phase lines, at a given instant; it is called the wavenumber, wavenumber = k = x (6.7)

2.6. STRING EMBEDDED IN AN ELASTIC SUROUNDING

14

On the other hand represensts the number of phase lines passing across a xed x t per unit time; it is called the wave frequency. wave frequency = = t (6.8)

To stay with a particular line of contant phase, say a crest, one must have d = kdx dt = 0 namely one must move at the phase velocity, dx c= = dt =constant k Now back to the string. Substituting (6.10) in (6.1) we get

(6.9)

K 2 + 2 V =0 k 2 T co

(6.10)

or

= co k 2 + The phase speed is c=

K T
1/2

(6.11)

K = co 1 + |k| T k2

> co

(6.12)

See gure ??. Note that, due the stiening by the lateral support, the phase speed is always greater than co , and decreases monotonically with the wave number. Longer waves (small |k|) are faster, while shorter waves (larger |k|) are slower. As |k| increases, c approaches the nite limit co for the shortest waves. In general a sinusoidal wave whose phase velocity depends on the wavelength, i.e., is a nonlinear function of k, is called a dispersive wave, and (6.11) or its equivalent (6.20) is called the dispersion relation. An interesting physical feature for dispersive waves in general can be found by su perposing two trains of sighltly dierent frequencies and wave numbers: V = A ei(k where k = k k , = (k ), k k. (6.14)

+ x + t)

+ Aei(k

x t)

(6.13)

2.6. STRING EMBEDDED IN AN ELASTIC SUROUNDING

15

Figure 7: Phase and group velocities of a string in elastic surrounding. Point of Sta tionary phase k0 . Let us approximate by = (k) k then V = Aeikxit eik (x(d/dk)t) + eik (x(d/dk)t) = A(x, t)eikxit where A = 2 cos k (x (d/dk)t) (6.16)

d + O(k 2 ) dk

(6.15)

The factor exp(ikx it) is called the carrier wave and A(x, t) the envelope. Thus the result is a sinusoidal wave train with a slowly varying envelope which has a very long wavelength 2/k 2/k and moves at the group velocity cg = d dk (6.17)

which is in general dierent from the phase velocity for dispersive waves. In our string problem, the group velocity is easily found from the dispersion relation (6.11) cg = Tk c2 o , = c hence cg c2 = o <1 c c2 (6.18)

2.6. STRING EMBEDDED IN AN ELASTIC SUROUNDING

16

Thus the group velocity is always smaller than the phase velocity, and increases with the wavenumber from 0 for the longest wave to the nite limit co (equal to the phase velocity) for the shortest waves. When > c where c = co is called the cuto frequency c , k is real k = |k|, with |k| =

K . T
1/2

(6.19)

2 K c2 T o

(6.20)

being the real and positive square root. V (x, t) is a propagating wave, with the plus (minus) sign corresponding to right (left) going wave. If < c, k = i is imaginary, with being the postive root: =
1/2

K 2 2 T co

(6.21)

Then eikx = ex . For boundedness one must choose the minus (plus) sign for x > 0 (x < 0). Oscillations are localized or evanecent; there is no wave radiation. As a simple application, (6.4) is the response in a semiinnite string forced to oscillate at the left end x = 0, y(0, t) = Aeit If > c, then

(6.22)

V (x, t) = Aei|k|xit

(6.23)

where |k| is dened by (6.20). The requirement that waves due to a local disturbance can only radiate outwards is called the radiation condition. More will be said on it later. If < c, we must require boundedness at innity so that V (x, t) = Aexit , x>0 (6.24)

At the cuto frequency, k = 0, V is constant in x; the innitely long string would oscillate in unison. This absurd result signes the breakdown of the the linearized theory.

2.7. DISPERSION FROM A LOCALIZED DISTURBANCE

17

6.2

Energy transport
2

Along a unit length of the string the densities of kinetic energy per unit length is, KE = 2 V t (6.25)

The potential energy in any segment dx of the string is the work needed to deform it from static equilibrium. The part due to lengthening of the string against tension is T (ds dx) = T
2 2

V 1+ x

T dx T dx 2
2

V x

dx

Adding the part against the springs, the total potential energy per unit length is T PE = 2 V x + K 2 V 2 (6.26)

We now calculate their time averages. If two timeharmonic functions are written in the complex form: a = Aeit , the time average of their product is given by ab 2

t t+ 2

b = Beit

(6.27)

1 1 ab dt = (AB ) = (A B) 2 2

(6.28)

Using this formula, the periodaveraged energy densities are, KE = (iAei )2 = 2 |A|2 2 4

T K T 2 K |A|2 k + P E = (ikAe )2 + (Ae )2 = 2 2 4 4 where |k |x t is the wave phase. Hence T E = KE + P E = 4 after using the dispersion relation. Now the averaged rate of energy inux across any station x is 2 2 V V T = T (ikV )(iV ) = k|A|2 = |A| T x t 2 2

K 2 |A|2 = 2 |A|2 + k2 + T T 2

(6.29)

Tk

= Ecg

Thus the speed of energy transport is the group velocty;. This result is quite general for many physical problems and is not limited to the springs.

2.7. DISPERSION FROM A LOCALIZED DISTURBANCE

18

Dispersion from a localized initial disturbance

The solution for monochromatic waves already shows that waves of dierent wavelengths move at dierent velocities. What then is the consequence of an initial disturbance? Since a general initial disturbance bounded in space can be represented by a Fourier integral which amounts to the sum of innitely many sinusoids with a wide spectrum, we shall employ the tools of Fourier transform. In addition to the governing equation: 1 2V KV 2V , = 2 t2 2 c x T we add the initial (Cauchy) conditions V (x, 0) = f (x), V (x, 0) =0 t (7.2) < x < , t > 0 (7.1)

Let us dene the Fourier transform and its inverse by f (k) =

ikx

f (x) dx,

1 f (x) = 2

eikx f (k) dk

(7.3)

The transforms of (7.1) is 1 d2 V KV , = k 2 V c2 dt2 T and of the intitial conditions are, dV (k, 0) =0 V (k, 0) = f (k), dt The solution for the tranform is V (k, t) = f (k) cos t where

t>0

(7.4)

(7.5)

(7.6)

= co k 2 + The Fourier inversion is

K T

(7.7)

1 V (x, t) = dkf (k)eikx cos t 2

(7.8)

2.7. DISPERSION FROM A LOCALIZED DISTURBANCE

19

Figure 8: Neighborhood of Stationary Phase Any real function f (x) can be expressed as the sum of an even and an odd function of x. For simplicity let us assume that f (x) is even in x so that f (k) is real and even in k, then 1 V (x, t) = which can be manipulated to V (x, t) =
1 ikxit dkf (k) e + eikx+it dk 2 0
0

dkf (k) cos kx cos t

(7.9)

The rst term in the integand represents the rightgoing wave while the second, left going. Each part correponds to a superposition of sinusoidal wave trains over the entire range of wave numbers, within the small range (k, k + dk) the amplitude is f (k)/2. The function f (k)/2 is called the Fourier amplitude spectrum. In general explicit evalua tion of the Fourier integrals is not feasible. We shall therefore only seek approximate information. The method of stationary phase is particularly useful here. It aims at the asymptotic approximation of the integral I(t) =

a b

F (k)eit(k) dk

(7.10)

2.7. DISPERSION FROM A LOCALIZED DISTURBANCE

20

for large t. Let us rst give a quick derivation of the mathematical result. Assume that F (k), (k) are ordinary functions of k. If t is large, then as k increases along the path of integration both the real and imaginary parts of the exponential function cos(t(k)) + i sin(t(k)) oscillates rapidly between 1 to +1, unless there is a point of stationary phase ko within (a,b) so that d(ko ) = (ko ) = 0, dk a < ko < b. (7.11)

Then important contribution to the Fourier integral comes only from the neighborhood of ko . Near the point of stationary phase, we approximate the phase by 1 (k) = (ko ) + (k ko )2 (ko ) + 2 and the integral by I(t) F (ko )eit(ko )

a b

exp

it (k ko )2 (ko ) dk 2

With an error of O(1/t), we also replace the limits of the last integral by ; the justication is omitted here. Now it is known that

itk 2

dk =

i/4 e t

It follows that I(t) =

a b

F (k)e

it(k)

dk F (ko )e

it(ko )i/4

2 (k )| t| o

1/2

+O

1 , t

if ko (a, b), (7.12)

where the sign is + (or ) if (ko ) is positive (or negative). It can be shown that if there is no stationary point in the range(a,b), then the integral I(t) is small I(t) = O

1 , t

if ko (a, b). /

(7.13)

Let us apply this result to the rightgoing wave 1 V+ (x, t) = 4

dkf (k)eit(kx/ti) dk

(7.14)

2.7. DISPERSION FROM A LOCALIZED DISTURBANCE For a xed x/t =constant, we have (k) = kx/t (k), with x/t = , = co k 2 +

21

K T

(7.15)

For an observer travelling at the speed x/t, there is a stationary point ko at the root of x/t = (ko ) = co ko T ko = 2 (ko ) ko +
K T

(7.16)

which increases from zero for long waves to the maximum co for very short waves. For any observer slower than co , there is a stationary point ko . If the observer is faster than co , there is no stationary point. Since (ko ) = (ko ) < 0, the nal result is that 1 2 eit(ko )i/4 f (ko ) V+ (x, t) 2 t (ko )

(7.17)

2 1 eiko x(ko )ti/4 f (ko ) = 2 t (ko )

(7.18)

The transform f (k) depends on the specic prole of the initial disturbance. For the special case where f (x) = Sb (x2 + b2 ) (7.19)

which has an area S and characteristic width b, the Fourier transform is f (k) = Se|k|b (7.20)

To a specic observer identied by the speed of travel x/t, the approximate result can be viewed as a simple harmonic wave train with wave number ko , frequency (ko ) and amplitude f (ko ) A= 2

2 t (ko )

(7.21)

Since (k) increases from 0 with increasing k to the nite maximum T /co = co , an observer faster than co sees no waves. However any observer slower than co is accom panied by a train of progressive sinusoidal waves. The local wavelength ko is such that the group velocty matches the observers speed. The faster the observer, the shorter

2.7. DISPERSION FROM A LOCALIZED DISTURBANCE

22

Figure 9: Snapshots of wave eld on the side x > 0. the waves. If a snapshot is taken, then the shortest wave, whose phase velocity is the lowest are seen at the front which moves as fast as the crests of the shortest waves. The crests of longer waves advance much faster than the local envelope. Because f (k) is the greatest at k = 0, the envelope of the longest waves which stays near the source, is the biggest. The envelope of the shortest waves is lower in amplitude and spreads out with the wave front. The entire disturbance attenuates in time as t1/2 . See gure 7 for an overview. Note also that very near the wave front, cg co ; the second derivative (ko ) = (ko ) vanishes. Hence the asymptotic formula breaks down. A better approximation is needed, but is omitted here. (See C. C. Mei, 1989, Applied Dynamics of Ocean Surface Waves). Finally we examine the propagation of wave energy in this transient problem. Using (7.21) the local energy density is: 1 2 f (ko )2 E = 2 |A|2 = 2 4t (ko ) At any given t, the waves between two observers moving at slightly dierent speeds,

2.8. SCATTERING OF SINUSOIDAL WAVES

23

cg (k1 ) and cg (k2 ), i.e., between two points x1 /t = cg (k1 ) and x2 /t = cg (k2 ) are essentially simple harmonic so that the total energy is

x2

dxE =
x1

x2

dx
x1

2 (f (ko ))2 4t (ko )

Since x = (ko )t for xed t, we have dx = (ko )dko t Now for x2 > x1 , k2 > k1 , it follows that

x2

dxE =
x1

k2 k1

2 (f (ko ))2 = constant dko 4

Therefore the total energy between two observers moving at the local group velocity remains the same for all time. In other words, waves are transported by the local group velocity even in transient dispersion.

Scattering of sinusoidal waves

If along a long rod there are some localized inhomogeneities, an incoming train of sinu soidal waves will be partly reected and partly transmitted. The scattered signals tell us something about the scatterer. To determine the scattering properties for a known scatterer is called the scattering problem. To determine the scatterer from the scattering data is called the inverse scattering problem. We shall only study the former. Various mathematical techniques are needed for dierent cases: (i) Weak scatterers characterized by small amplitude relative to the wavelength, or slow variation within a wavelength, (ii) Strong scatterers if their dimensions are comparable to the wave length.

8.1

Weak scattering

Let the long rod have a slightly nonuniform cross section, S(x) = So (1 + a(x)), 1. (8.1)

where a(x) diminishes to zero at x . The wave equation reads 2u u S(x) 2 = E S(x) t x x

(8.2)

2.8. SCATTERING OF SINUSOIDAL WAVES For sinusoidal waves

24

u(x, t) = U (x)eit the spatial factor is governed by d dU S dx dx

+ 2 SU = 0

(8.3)

Then the solution can be sought by the perturbation method U (x) = U0 (x) + U1 (x) + 2 U2 (x) + Substituting this into the governing equation
d d (1 + a(x)) U0 (x) + U1 (x) + 2 U2 (x) + ESo dx dx

(8.4)

+ 2 So (1 + a(x)) U0 (x) + U1 (x) + 2 U2 (x) + = 0

(8.5)

and equating the coecients of like powers of to zero, we get from the terms of order O(0 ): ESo d2 U0 + 2 S0 U0 = 0 dx2

(8.6)

The solution is simply the incident wave U0 = Aeikx , with k = /E At the order O() we get d2 U1 d2 U0 da dU0 ESo 2 + 2 S0 U1 + a(x) ESo 2 + 2 S0 U0 + ESo =0 x x dx dx or, da dU0 d2 U1 + k 2 U1 = 2 dx dx dx equation. Let us dene the fundamental solution (Greens function) by d2 G(x, x ) + k 2 G(x, x ) = (x x ) dx2 G outgoing waves at (8.9) (8.10) (8.8)

(8.7)

Thus the scattered wave is at most of the order and is governed by an inhomogeneous

2.8. SCATTERING OF SINUSOIDAL WAVES It will be shown in Appendix A that eik|xx | G(x, x ) = 2ik

25

(8.11)

By straightforward dierentiation, it can be veried that U1 (x) =

1 da(x ) dU0 (x ) G(x, x )dx = dx dx 2ik

da(x ) dU0 (x ) ik|xx | e dx (8.12) dx dx

Since U0 (x ) = eikx , we have U1 = = = = A da(x ) ikx ik|xx | e e dx 2 dx

A ikx x da da
2ikx ikx e dx + e e dx (8.13) 2 dx x dx A ikx da 2ikx ikx dx e a(x) + e e 2 x dx A d 2ikx ikx ikx 2ikx a(x)e + e a(x )e dx a2ike dx 2 dx x

= ikAeikx Far to the right we get

a(x )e2ikx dx

(8.14)

U1 = 0, x .

(8.15)

Thus on the transmission side modication of the incident waves can at most be of the order O(2 ). Far to the left x , U U1 = Aeikx ik =

a(x )e2ikx dx , x .

(8.16)

Let us dene the reection coecient by R = ik

a(x )e2ikx dx ,

(8.17)

For instance if a is a hump of unit total dimensionless area and length b, a(x) = then R = ikbe2kb (8.19) (x2 b + b2 ) (8.18)

Hence reection is small if kb is large, i.e., long and gentle obstacles are not eective in reecting short waves . For very long waves or a short obstacle, reection is also

2.8. SCATTERING OF SINUSOIDAL WAVES

26

Figure 10: Reection coecient given by ( 8.19). vanishingly small. The maximum reection coecient is R = iao /2 when kb = 1/2. Note that far to the right, U1 (x) 0 as x . The transmission vwave is U U0 + U1 = Aeik0 x (1 + O(2 )). = (8.20)

The tranmission coecient is T = 1 + O(2 ). This is consistent with the general law of energy conservation |R|2 + |T |2 = 1 to be proven later. Homework: Check if the governing equation at O(2 ) for U2 is: da dU0 da dU1 d2 U2 ESo 2 + 2 S0 U2 = ESo a x dx dx dx dx

(8.21)

Carry out the solution to nd the transmission coecient T from the amplitude of U0 + U1 + 2 U2 at x , and show that energy is conserved to the order O(2 ).

8.2

Strong scattering of long waterwaves by a shelf

Consider an ocean bottom with a stepwise variation of depth. h1 , h2 , h3 = h1 , x < a; a < x < a; x>a (8.22)

h=

2.8. SCATTERING OF SINUSOIDAL WAVES

27

How does the step interupt the propagation of an incident wave of unit amplitude arriving from x ? In each zone of contant depth (i = 1, 2, 3), the shallow water equations read: ui i + hi =0 t x i ui +g =0 t x For sinusoidal waves i = i eit , ui = Ui eit we get ii + hi Ui =0 x i =0 iUi + g x where ki = ghi (8.26) (8.27) (8.25) (8.23) (8.24)

or d2 i 2 + ki i = 0, dx2 (8.28)

At a junction, the pressure and the ux must be equal, hence 1 = 2 , 2 = 3 , and h1 d1 d2 = h2 , dx dx d3 d2 = h1 , dx dx x = a; x = a. (8.29) (8.30)

and h2

The forms of the solutions in each zone of constant depth are:

1 = eik1 (x+a) + Reik1 (x+a) , 2 = Aeik2 x + Beik2 x , 3 = T eik1 (xa) , x < a; (8.31)
(8.32)
(8.33)

a<x<a x>a

The reection and transmission coecients R and T as well as A and B are yet unknown. Applying the matching conditions at the left junction, we get two relations 1 + R = Aeik2 a + Beik2 a k1 h1 (1 R) = k2 h2 (Aeik2 a Beik2 a ). (8.34) (8.35)

2.8. SCATTERING OF SINUSOIDAL WAVES Similarly the matching conditions at x = a gives Aeik2 a + Beik2 a = T k2 h2 (Aeik2 a Beik2 a ) = k1 h1 T. These four equations can be solved to give T = R= (1 + s)2 e2ik2 a 4s (1 s)2 e2ik2 a

28

(8.36) (8.37)

(8.38) (8.39) (8.40) (8.41)

(1 s)2 (e2ik2 a e2ik2 a ) (1 + s)2 e2ik2 a (1 s)2 e2ik2 a T A = eik2 a (1 + s) 2 T B = eik2 a (1 s) 2

where

k1 h1 c1 h1 s= = = k2 h2 h1 c2 The energy associated with the tranmsitted and reected waves are : |T |2 = |R|2 = It is evident that |R|2 + |T |2 = 1. Over the shelf the free surface is given by 2 = 2s (1 + s)eik2 (xa) + (1 s)eik2 (xa) (1 + s)2 eik2 a (1 s)2 eik2 a

(8.42)

4s2 4s2 + (1 s2 )2 sin2 2k2 a (1 s2 ) sin2 2k2 a 4s2 + (1 s2 )2 sin2 2k2 a

(8.43) (8.44)

(8.45)

Recalling the time factor eit , we see that the free surface over the shelf consists of two wave trains advancing in oppopsite directions. Therefore along the shelf the two waves can interfere each other constructively, with the crests of one coinciding with the crests of the other at the same moment. At other places the interference is destructive, with the crests of one wave train coinciding with the troughs of the other. The envelope of energy on the shelf is given by ||2 = 4s2 cos2 k2 (x a) + s2 sin2 k2 (x a) 4s2 + (1 s)2 sin2 2k2 a

(8.46)

2.8. SCATTERING OF SINUSOIDAL WAVES

29

Figure 11: Transmission and reection coecients for a rectangular shelf. From Mei, C.C., Applied Dynamics of Ocean Surface Waves. At the downwave edge of the shelf, x = a, the envelope is 4s2 || = 2 4s + (1 s)2 sin2 2k2 a
2

(8.47)

Note that the reection and transmission coecients are oscillatory in k2 a. In par ticulatr for 2k2 a = n, n = 1, 2, 3..., that is 4a/ = n, |R| = 0 and |T | = 1. The shelf is transparent to the incident waves, correspoinidng to the most constructive inter ference and the strongest transmission Mininum transmision and maximum reection occur when 2k2 a = (n 1/2), or 4a/ = n 1/2, when the interference is the most destructive. The corresponding transimssion and reection coecients are min|T |2 = See Figure 11 The features of interference can be explained physically. The incident wave train consists of periodic crests and troughs. When one of the crests rst strikes the left edge at x = a, part of the it is transmitted onto the shelf and part is reected towards x . After reaching the right edge at x = a, the tranmitted crest has a part reected to the left and rereected by the edge x = a to the right again. When the 4s2 , (1 + s2 )2 max|R|2 = (1 s2 )2 . (1 + s2 )2 (8.48)

2.9. GENERAL IDENIITIES ON SCATTRERING

30

rereected crest arrives at the right edge the second time, its total travel distance is 4a. If 4a is an integral multple of the wave length 2 , the crest is in phase with all the other crests entering the shelf either before (the third, fourth, fth, ... time) or after. Thus all the crests reinforce one another at the right edge. This is constructive interference, leading to the strongest tranmission to the right x . On the other hand if 2k2 a = (n 1/2) or 4a/ = n 1/2, some crests will be in opposite phase to some other crests, leading to the most destructive interference at the right edge, and smallest transmission.

General identities in scattering by arbitrary inho mogeneities

Scattering due to inhomogeneities, caused by nonuniformities either in geometry or in material properties, requires the solution of ordinary dierential equations with variable coecients. In general one must resort to numerical means. For the sake of checking numerical accuracy and for gaining physical insight, identities which must be true are useful. They are often deduced by arguments typical in the derivation or the use of Greens theorem. We illustrate these identities by the example of an innitely long rod with variable cross section S(x), u S x x

S 2 u E t2

(9.1)

E and are taken to be constants. S(x) is nonuniform only in a nite neighborhood around x = 0. Elsewhere S = So =constant. Consider monochromatic waves u(x, t) = U (x)eit so that U (x) is governed by the ordinary dierential equation (SU ) + b2 SU = 0,

(9.2)

b2 =

2 E

(9.3)

The boundary conditions depend on the source of the incident waves.

2.9. GENERAL IDENIITIES ON SCATTRERING

31

If the incident waves arrive from x , the asymptotic form of the solution must be U1 (x) A eikx + B eikx = A eikx + R1 eikx , and U1 (x) A+ eikx = A T1 eikx kx (9.5)

kx

(9.4)

where T1 and R1 are the transmission and reection coecients which ar a part of the unknown solution of the leftincidence problem. Let us dene the Jost function by f1 (x) = then f1 (x) and f1 (x) eikx kx 1 (9.8) 1 ikx R1 ikx e + e T1 T1 kx 1 (9.7) U1 A T1 (9.6)

On the other hand, if the incident waves arrive from x , the asymptotic form of the solution must be U2 (x) b eikx = b+ T2 eikx, and

kx

(9.9)

U2 (x) a+ eikx + b+ eikx = b+ R2 eikx + eikx ,

kx

(9.10)

where T2 and R2 are the unknown transmission and reection coecients of the right incidence problem. Similarly we dene the Jost function by the following asymptotic solution f2 (x) eikx, and f2 (x) 1 ikx R2 ikx e + e , T2 T2 kx (9.12) kx (9.11)

Since both f1 and f2 satisfy (9.3), we have

(Sf1 ) + b2 Sf1 = 0,

and

(Sf2 ) + b2 Sf2 = 0,

(9.13)

2.9. GENERAL IDENIITIES ON SCATTRERING

32

Multiplying the rst by f2 and the second by f1 , then taking the dierence, we get after partial integration, d (S(f1 f2 f2 f1 )) = 0, dx or
W (f1 , f2 ) = (f1 f2 f2 f1 ) =

C S(x)

(9.14)

The constant C can be related to the asymptotic limits at k |x| 1 where S So . Far on the incident side, the Wronskian is
(f1 f2 f2 f1 )x 1 ikx R1 ikx ik ikx ikR1 ikx ikx ikx = e + e (ik)e e e e T1 T1 T1 T1 2ik = T1

Far on the transmission side, the Wronskian is

(f1 f2 f2 f1 )x ikR2 ikx 1 ikx R2 ikx ikx ik ikx =e e + e e + e (ik)eikx T2 T2 T2 T2 2ik = T2

Since the two Wronskians are equal, we conclude that T1 = T2 (9.15)

Thus, the complex transmission coecients are equal for incident waves coming from either side, no matter how asymmetric the geometry may be in the near eld! Next we consider the leftincidence problem. Multiplying the rst of (9.3) by the
complex conjugate of U1 , i.e., U1 , and taking the dierence of the result with its complex

conjugate, we get
U1 S(U1 ) U1 (SU1 ) =

d S(U1 U1 U1 U1 ) = 0 dx

(9.16)

hence
S(U1 U1 U1 U1 ) = constant

(9.17)

2.10. REFRACTION IN A SLOWLY VARYING MEDIUM By using the asymptotic expression (9.4) on the incidence side, we get

S(U1 U1 SU1 U1 )

33

x B eikx ikA eikx + ikB eikx

= So A e

ikx

So A eikx + B eikx

ikA eikx ikB eikx

= 2ik |A |2 + 2ik |B |2 Similarly by using the asymptotic expression (9.5) on the transmission side, we get

S(U1 U1 U1 U1 )

x (ik)A eikx So A+ eikx +

= So A+ e

ikx

(ik)A+ eikx

= 2ikSo |A+ |2 Equating the two limits we conclude that |A |2 = |B |2 + |A+ |2 or |R1 |2 + |T1 |2 = 1. (9.19) (9.18)

This is merely a statement of conservation ux : the ux rate of total scattered energy (reected and transmitted) is equal to the ux rate of the incident wave energy; the speed of energy tranport being the same on both sides of the scatterer. Clearly the same energy conservation must hold for the rightincidence problem. |R2 |2 + |T2 |2 = 1 (9.20)

If the scattering problem is solved numerically by, say, nite elements, it is necesssary that the computed scattering coecients R and T satisfy the identities (9.15) (9.19) and (9.20).

10

Refraction in a slowly varying medium

For timeharmonic cases, we again use shallow water waves for demonstration. The governing equation is d d 2 h + =0 dx dx g

(10.1)

2.10. REFRACTION IN A SLOWLY VARYING MEDIUM Consider a sea depth which varies slowly within a wavelength, i.e., 1 dh = O() 1 kh dx

34

(10.2)

Earlier analysis suggests that reection is negligibly small. Thus the solution is expected to be a locally progressive wave with both the wavenunmber and amplitude varying much more slowly than the wave phase in x . Hence we try the solution = A(x)ei(x) where (x) t is the phase function and k(x) = d dx (10.4) (10.3)

is the local wave number and A is the complex amplitude. Note the spatial rate of variation of the phase function, i.e., the wave number is in gneral not small, therefore itself is not a slowly varying function of x. Let us calculate the rst derivative: d dA i = ikA + e dx dx and assume = O(kL)1 1 kA In fact we shall assume each derivative of h, A or k is times smaller than kh, kA or k 2 . Futhurmore, d d 2 dA d dA d(khA) 2 A i + h + = ik ikh + h + h +i e =0 dx dx g dx dx dx dx g Note that the complex amplitude A can be written as A = |A(x)|eiA (x) (10.5)

dA dx

The phase of A: A , is a slowly varying function of x. It can be clonsidered as a part of the wave phase, although its spatial gradient is much smaller than k. Now lets expand A = A0 + A1 + A2 + (10.6)

with A1 /A0 = O(), A2 /A0 = O(2 ), . From O(0 ) the dispersion relation follows: 2 = ghk 2 , or k = gh (10.7)

2.10. REFRACTION IN A SLOWLY VARYING MEDIUM

35

Thus the local wave number and the local depth are related to frequency according to the well known dispersion relation for constant depth. As the depth decreases, the wavenumber increases. Hence the local phase velocity c= also decreases. From O() we get, ikh or, after multiplication by A , 0 d(khA0 ) dA0 +i =0 dx dx d (kh|A0 |2 ) = 0 dx kh|A0 |2 = C 2 = constant or,

= gh k

(10.8)

(10.9)

which means

gh|A0 |2 = constant =

gh |A |2

(10.10)

Since in shallow water the group velocity equals the phase velocity, the above result means that the rate of energy ux is the same for all x and is consistent with the original assumption of unidirectional propagation. Furthermore, the local amplitude increases with depth as A0 (x) = A This result is called Greens law. In summary, the leading order solution is = A
1/4 1/4

h h

1/4

(10.11)

h h

iit

= A

h h

exp i

k(x )dx i

(10.12)

Solution for the Green function

Let us divide the domain into two parts, corresponding to two sides of the concentrated forcing. On the right side x > x0 , we dene G(x, x0 ) G+ (x, x0 ) which must be governed by : d2 G 2 + k0 G+ = 0, 2 dx x > x0 (A.1)

2.10. REFRACTION IN A SLOWLY VARYING MEDIUM

36

To satisfy the radiation condition we take the sol,ution to beTo satisfy the radiation condition we take G+ = Ceik0 x On the left side we dene G(x, x0 ) = G (x, x0 ), which must be governed by d2 G 2 + k0 G = 0, dx2 To satisfy the radiation condition we take G+ = Deik0 x At x = x0 we require continuity: G+ (x0 , x0 ) = G (x) , x) ) Hence Ceiko x0 = Deik0 x0 (A.6) (A.5) (A.4) x < x0 (A.3) (A.2)

In addition we get another matching condition at x = x0 by integrating eq(8.8) across the concentrated forcing

x0 + x0

d2 G 2 + k0 G dx = dx2

x0

x0 + x0

(x x0 )dx = 1.

The lefthand side can be integrated to give the second matching codition, dG+ dx dG dx

x0

=1

(A.7)

implying that ik0 Ceik0 x0 + ik0 Deik0 xo = 1 Thus C0 = The nal solution is eik0 (xx0 ) G+ = , x x0 > 0; 2ik0 or more compactly, G(x, x0 ) = eik0 |xx0 | 2ik0 (A.10) eik0 (xx0 ) G = , x x0 < 0. 2ik0 eik0 x0 eik0 x0 , D0 = , 2ik0 2ik0 (A.9) (A.8)