Chapter 11.

Time-Dependent Green's Functions

524
11.3.5. Consider
au
at

a 2u
k dx'

au
ox (O,t)

A(t)

+ Q(:r, t)

u(x,O)

f(x)

au (L ,t;i
"'1
C.T

B(t),

(a) Solve for the appropriate Green's function using the method of eigen
function expansion.
(b) Approximate the Green's function of part (a), Under what conditions is
your approximation valid?
(c) Solve for the appropriate Green's function using the infinite space Green's
function.
(d) Approximate the Green's function of part (c). Under what cOlluitiulls is
your approximation valid?
(e) Solve for u(x, t) in terms of the Green's function.

11.3.6. Determine the Green's function for the. heat equation subject to zero bound
ary conditions at x - 0 and x = L by applying the lllp.thod of eigenfunction
expansions directly to the defining differential equation. [Hint: The answer is
given by (11.3.35)1

Chapter 12

The Method of
Characteristics
for Linear and Quasi-Linear
Wave Equations
12.1

Introduction

In previous chapters we Obtained cert.ain results concerning the one _dimensional

wave equation,

8 2u

') D2,U

Ot' =c- dx2 '

(12.1.1)

,,(x.O) = f(:r)

(12.1.2)

subject t.o the initial condit,ions

au

at (:r.O)

= g(:r).

(12.1.3)

For a vibrating string with zero displacernent at :r = 0 and x = L we obtained a

~omewhat complicated Fourier .sine .serie8 solution by the method of separation of
variables in Chapter 4:
.
~ . mrx ( CLrL cos --y;mret
u(:r, t) = L sm L

.mrct)
+ bn S111
L .

(12.1.4)

n=l

Further analysis of this solution [see (4.4.14) and Exercises 4.4.7 and 4.4.8j shows
that the solution can be represented as the Stun of a forward and backward moving
wave. In particular,
u(x, t)

= fix -

et)

+2 fix + ct + 2.
2e
525

x+d

x-c'

g(xo) dxo,

(12.1.5)

Chapter 12. Method of Characteristics

12.2. First-Order Wave Equations

527

526
where f(x) and g(x) are the odd periodic extensions of the functions given in (12.1.2)
aV_Cav=o.
(12.2.5)
and (12.1.3). We also obtained (12.1.5) in Chapter 11 for the one- dimensional wave
iJt
ax
equation without boundaries, using the infinite space Green's function.
In this chapter we introduce the more powerful method of characteristics to solve
the one- dimensional wave equation. We will show in general that u(x, t) = F(x
12.2.2 Method of Characteristics for First-Order
ct) + G(x + ct), where F and G are arbitrary functions. We will show that (12.1.5)
Partial Differential Equations
follows for infinite space problems. Then we will discuss modifications needed to
We begin by discussing either one of these simple first-order partial differential
solve semi-infinite and finite domain problems. In Section 12.6, the method of
equations:
characteristics will be applied to quasi-linear partial differential equations. There
shock waves will be introduced when characteristics intersect.

aw

aw

tit +ca;: =

12.2 Characteristics For First-Order

Wave Equations

12.2.1

The methods we will develop \vill be helpful in analyzing the one- dimensional wave
equation (12.2.1). We consider the rate of change of w(.r(t), t) as measured by a
moving observer, x = x(t). The chain rule l implies that

Introduction

The one-dimensional wave equation can be rewritten as

[)2

U
-a
2 t

iJ2 u
c2 _ [ 2) =
x

o.

(12.2.6)

O.

aw

-d w(x(t), t) = -iJ
t
t

(12.2.1)

The firot term

aw / iJ t

dxiJw

+ -dt -:;-.
U.r

(12.2.7)

represents the change in w at the fixed position, while

(dx/dt)(oW/al') represents the change due to the fact that the observer moves
A short calculation shows that it can be "factored" in two ways:

+ c!!....)
(!!....
iJt
ax

into a region of possibly different w. Compare (12.2.7) with the partial differential
equation for w, equation (12.2.6). It is apparent that if the observer moves with
velocity C, that is , if

(iJU _ au) = 0
at
ax

iJ
a) (au
1ft + c au)
iJx = 0,
( at - c ax

since the mixed

second~derivative terms vanish in both. If we let

au
iJ':\
w=lft-c
axJ

au
au
v=lft+ciJx'

then
(12.2.2)

(12.2.8)

[B

(12.2.9)

dt

(12.2.3)

we see that the one _dimensional wave equation (involving second derivatives) yieldS

O.

Thus , w is constant. An observer moving with this special speed

no changes in IV.

would measure

Characteristics. In this way, the partial differential equation (12.2.6) has

been replaced by two ordinary differential equations, (12.2.8) and (12.2.9). Inte
grating (12.2.8) yields

two first-order wave equations:

ow
aw
tit+cax =0

x = ct + Xn,

(12.2.10)

(12.2.4)
1 Here

d/ dt as measured by a moving observer is sometimes called the substantial derivative.

Chapter 12. Method of Char""teristics

528

the equation for the family of parallel characteristics 2 of (12.2.6), sketched in

Fig. 12.2.1. Note that at t = 0, x = .TO. w(x, t) is constant along this line (not
necessarily constant ever,ywhere). w propagates a.<; a wave with wave speed c
[see (12.28)J

a.:i. First-Urder Wave Equations

\iVe have shown that w is constant along' the characteristics x - 2t = constant,

keeping its same shape moving at velocity 2 (t.o the right.). The important charac

teri'5tics, x ~ 2t

+0

and x = 2t + 1, as well as a sketch of the solution at various

times, appear in Fig. 12.2.2. w(x, t)

shift.ing
w(x,t)

~~~

~~

L /~-

Figure 12.2.1: Characteristics for the

w(x, t)

w(x,O)
US

determine

'W

at the point (x,

teristic)

if 2t

(12.2.11)

= P(x),

t). Since w is constant along the charac

2t. Otherwise, by

< x < 2t+ 1.

= xo:

= 2t + xu.

= w(xo, 0) = 4xo = 4(x -

~~"'-t=2
~
t=l
t=O

thus
(12.2.12)

w(x, t) = P(x - ct),

which we call the general solution of (12.2.6).

We can thiuk of P(x) as being an arbitrary function. 1b verify this, we substitut.e
(12.2.12) back int.o t.he partial differential equation (12.2.6). Using the chain rule
dP
a(x - ct)
OJ' = d(x - el)
Ox

dP

Ow

2t),

~t=3

w=o

Giveu x and t, the parameter is known from the characteristic. Xo = x - ct, and

and

<

as before. This is valid if 0 < Xo < 1 or equivalently 0 < x _ 2t < 1.

0,

= w(xo, 0) = P(xo).

w(x, t)

2t)

or if x

Along this characteristic, w(x, t) is constant. If 0 < Xo <.: 1, then

General solution. If w(x, t) is given initially at t

then let

first-order wave equation.

"

""

= 4(x -

= 0 if x > 2t + I

To derive this analytic solution, we Use the characteristic which starts at x

52

Figure 12.2.2: Propagation for the first-order wave equation.

Same shape. In general w(x, t) = P(.T - et). At fixed t, thc solution of the
first-Order wave equation is the same shape shifted a distance ct (distance = velocity
times t.ime). We illustrate this in Fig. 12.2.3.
t=o

t = 11

d(J' - ct)

Ow
dP
O(x-d)
dP
_= .
.
=-r
at
d\x - et)
aI
. d(x - cf:)'

Thus, it is verified that (12.2.6) is "'t.isfied by (12.2.12). The general solution of

a first-order partial differential equation contains an arbitrary function, while the
general solution to ordinary differential equations contains arbitrary constants.

Example. Consider

ow
ow
_+2-=0.
at

ax

x=o

= ell

Fignre 12.2.3: Shape invariance for the first-order wave equation.

subject to the initial condition

w(x,O) = {

--0--------

~x

x<O
O<x<l

x>!.

:.: A characteristic is a curve along which a PDE reduces to an ODE.

Summary. The method of characteristics solves the first-order wave eqnation

(12.2.6). In Sections 12.3-12.5, this method is applied to solve the wave equation
(12.1.1). The reader may proceed directly to Section 12.6 where the method of
characteristics is described for quasi-linear partial differential equations.

L'lIapter Lt. lVlctnoC1 or G'tlMacrensncs

530

EXERCISES 12.2
12.2.1. Show that the wave equation can be considered as the following system of
two coupled first-order partial differential equations:

au

au

--cat
ax

*12.2.2. Solve

Ow

aw

at

ax

ow

ow

--3-=0
12.2.3. Solve

ow
ow
-+c- at
01'

-+4-=0
f) t
a.r.

Vllt:-1JJ11H:;1J~lUllUl

.lL,.u.

vrctvt:

~LJua"lUl1

vve

12.2.9. As motivated by the analysiH of a llloving observeL llHtke a change of in

dependent variables from (x, t) to a coordinate system moving with velocity
c, t.~, I'), whf're ~ = 2' - ct and t' = ( in order to solve (12.2.6).
12.2.10. For the first-order (;quasi-linear" partial differential equation

O.
ij"
au
a-a
+ be>
=
x
uy

C,

where a, o. and c are fllnction~ of x, y and 11" show that the method of
characterititics (if necessary, see Section 12.6) )rield.;;

with w(:r,O) = cos I.

dx

with w(O. t) = sin3t.

dy

du

-;;=b"=-;;

12.2.4. Solve

ow
aw
-+c-=O
at
ax
w(x. 0)

f(x)

(c > 0)

x>p

for x > 0 and t > 0 if

h(t)

w(O. t)

+ c~w
= e'x with W(I, 0) = f(x)
o:r:
~"( + x~:: = 1 with 11I(x, 0) = f(x}
~"( + t%'i- = 1 with w(x, 0) = f(x)
~~ + 3t~:: = w with w(:r, 0) ~ f(x)

(a) D.w
dt
(c)
*(d)

{)u

-+2u-=0
at
ax

12.3.1

Introduction

From the one - dimensional wave equation,

a2u
a'u- 0
ih,2 - ,
at 2 c.2 _

*12.2.6. Consider (if necessary, see Section 12.6):

au

Method of Characteristics
for the One-Dimensional Wave Equation

t > O.

12.2.5. Solve using the method of characteristics (if necessary, see Section 12.6):

*(b)

12.3

we derived two first-order partial differential el{\mtions, () w! D t +

u(x,O)

with

f(x).

have

ShO\\'1l

x < 0

.E > O.

Obtain the solution u(x, t) by considering the lirrlit as L ~ 0 of the charac

teristics obtained in Exercise 12.2.7. Sketch characteristics and u(x, t) for t
fixed.

et).

au + C-.~au = Q(.r + et).

v = -

at

(12.3.2)

ax

(12.3.:J)

By combining (12.3.2) and (12.3.3) we Obtain, for example,

*12.2.8. Consider ExefClse 12.2.6 with

all
au
= -at - cax- = P(x -

The problem for V is identical (replace c by -c). Thus, we could have shown that
u is trl'lJlslRted unchanged at velocity e

(a) Determine equations for the characteristics. Sketch the characteristics.

(b) Determine the solution u(x,t). Sketch u(x, t) for t fixed.

0 and

that w remains the same shapE' moving at ve10rity c:

12.2.7. Consioer Exercise 12.2,.6 with

1
x<O
u(x,O) = f(.r.) = { 1 + xlL O<x<L
x> L.
2

co 11.' / ax =

iJ v/D t - ca,,/ax = 0, where u = iJ u/a t - co u/ax and v = iJ ul iJ t + co u/ax. We

Show that the characteristics are straight lines.

u(x,O) = f(I) = { 2

(123.1)

au

'"
ut = -[P(x
2

et) + Q(x + et)],

and thus

u(x, t) = F(x - et)

+ G(x + ct),

(123.4)

Chapter 1:2. Method at

532

cnaractenSlJGj

1L.J.

where F and G are arbitrary functions (-cF' = tP and cG' = tQ) This result
was obtained by d'Alembert in 1747.
The general sulutiull is the sum of F(x ct)l a wave of fixed shapP moving to
the right with velocity c, and G(x + eI), a wave of fixed shape moving to the left
with velocity -c. The solution may be sketched if F(x) and G(x) are known. We
shift F(x) to the right a distance et and shift G(x) to the left a distance ct and add
the two. Although each shape is unchanged, the sum will in general be a shape that
is changing in time. In Sec. 12.3.2 we will show how to detcrmine F(x) and G(x)
from initial conditions.

une-1JlInensional Wave Equation

g(x)

= _ dF + dG
dx

(12.3.9)

dJ'

We solve. for G(X) by eliminating F(x); for example, adding the derivative of (12.3.8)
to (12.3.9) yields

dG
= ~ (df + g(X)).
dx
2dx
c
By integrating this, we obtain

Characteristics. Part of the solution is constant along the family of char

acteristics x - ct = constant, while a different part of the solution is constant along

;l' + ci = constant,. For the one-dimensional wa,ve equation, (12.3.1), there are two

farnilies of characteristic curves, as sketched in Fig. 12,3.1.

G(x) = -1 fix)
2

+ -2c1

1
1
F(x) = -fix)
- ---:
2

2,

1,'
1,"
()

g(7) d7 +k
,

(12.3.10)

g(E) d7 -k,

(123.11)

x-ct=a

",+cl=;3 ~

where the latter equation was obtained from (12.3.8). k can be nel(lectcd since
u("',I) b obtained from (12.3.5) by adding (12.3.10) and (12.3.11) (with appropriate
shifts).

/XX><IX><x.>
13

Sketching technique. The solution u(.1', t) can be graphed based on

(12.3.5) in the following straightforward manner:

- -x

'"

1. Given f(x) and g(x). Obtain the graphs of

Figure 12.:3.1: Characteristics for the one - dimen::;ional wave equation.

2f (x)
12.3.2

533

and

11

2c

g(7) dE

'

the latter by integrating first.

Initial Value ProbleIll (Infinite Domain)

[n Sec. 12.3.1 we showed that the general solution of the one - dimellBional wave

2. By addition and subtraction, from F(x) and G(x); see (12.3.10) and (12.3.11).

equation is

3. lraJlslate (shift) F(x) to the right a dbtance ct and G(x) to the left ct.
n(x, t)

= FCT -

(12.3.5)

r1;) + G(x + eI).

4. Add the two shifted fUllctions, thus satisfying (12.3.5).

Here we -will determine the arbitrary functions in order to satisfy the initial condi

Initially at rest. If a vibrating string is initially at rest [8n/D I(x. 0) =

g(x) = 0], then from (12.3.10) and (12.3.11) F(x) = G(x) = tf(x). Thus.

tions:

U(x.O)

= fix)

ou( x,OJ.
)
at
= g(x

-oo<x<oo

00

<x <

(12.3.6)

,,(x, t)

F(x)

eI) + f(x

+ ct)].

(12.3.12)

The initial condition nix, 0) ~ f(x) sphts into twu parts; half moves to the lett and
half to the right.
00.

Example. Suppose that an infinite vibrating string is initially stretched into the
shape of a single rectangular pnlse and is let go from rest. The corresponding initial
conditions are
1 Ixl < h
u(x, 0) = fix) = { 0 Ixl > h.

These initial conditions imply that

fix)

= 2[J(x -

+ G(x)
L.

___

Chapter 12. Method of Characteristics

534
and

au
i)t(X, 0) =

g(x)

O.

!
f
l
j
t
=
f
,
!
L
~
I
J
~
I
L
.
J ~L ~]l
't-~ ,~_
1-

02

r\

o~___
1

-~ ~
~

~'----

~~_
-

~-<~.
~
~-,-4
8
6 ~__
--'
~ __
4

-~~~_~
0
0

6--------

t=

Ix] <h

Ixl > h.

This information also appears in Fig. 12.3.3.

t

x+d=h

10

x - ct = -h

F~ol

x+cf= -h

x-ct=h

G~OI

x
F=O
G~O

'__

,.
I

F~O

t/2

t> hie
F~

1/2 (x t)

g~o

,_o~,O

F_~ 1/2

",

x
F_G-~O

0_
0.4

F~ 0

F=G=o----~--G~11/2'
_~---.--.'----G=O
-h F=G=lI2
. _
'..
t<hlc

08

= C(x) = {

distance
2h
h
=_=_
veioel ty
2c
c

Important characteristics are sketched in Fig. 12.3.3. F stays constant moving to

the right at velocity c, while G stays constant moving to the left. From (12.3.10)
and (12.3.11)

F(x)

::1

535

apart at velocity 2c. The ends are initially a distance 2h apart, and hence the time
at which the two pulses separate is

The solution io given by (12.3.12). By adding toget.he.r these two rectangular pulses,
we obtain Fig. 12.3.2. The pulses overlap until the left end of the right-moving one
passes the right end of the othf';f. Since each is traveling at :::;peed c, they are moving

12.:3. Une-lJimensionaJ Wave f!;quation

1_

0.2

Figure 12.3.3: Method of characteristics for the one- dimensional wave equation.

o
10

U(x,O)

to

___..J
-h-et
~

.c:l

-h-et

-h+ct: h-et

1/2

III
h-ct ',-h+et

= f(.r)

au
ot (x,O) = g(x)

------h

Example not at rest. Suppose that a.n infinite otring is initially horizontally
stretched with prescribed initial velocity 1'1."\ follows:

Odd/e

o
{

I Ixl < h
o Ixl > h.

In Exercise 12.3.2 it is Shown that this corresponds to instantaneously applying

a constant impulsive force to the entire region Ix) < h, fiR though the string is
being struck by a broad ([xl < h) hammer. The calculation of the solution of the
wave equation with these initial conditions is more involved than in the preceding
example. From (12.3.10) and (12.3.11), we need J~' g(x) dx, representing the area
under g(x) from 0 to x:

h+et

t>hle

h+et

Figure 12.3.2: Initial value problem for the one - dimensional wave equation.

2cG(x) = -2cF(x) = io g(x)dx =

o

-h x <-h
-h < x < It
x> h.

x
h

Thc solut.ion u(x, t) is the SUIn of F(x) shifted to the right (at velocity c) and G(x)
shifted to the left (at velocity c). F(x) and G(x) are sketched in Fig. 12.3.4, as is

Chapter 12. Method of Characteristics

536

12.3. One-Dimensional ~Vavc Equation

537

(x,t)

x = Xu - ct

02

~'

-\i\ ~,~

\~

'--LiN,

.--------1]<..1=1 4

hic<t=ts

- -I---:-Ll T:-N--":--~--t----i-I--!--:!
~
-~-------:r~::P~H7"-t~
'

____

w
o

,.hj,
-~_ 11<1 = "2<h/c

O<f=i1<h!c
t =0

x-ct

------

The general solution of thE' OIle - dimensional wave equation can be simplified some
what. Substituting (12.3.10) and (12.3.11) into the general solution (12.3.5) yields

'u(x,t)

- ct)

"'2c :'lc1 x+dg(x) dx -l,-d g(x) dXJl

(J

(h)

12.3.1. Suppose that ulx, t) = F(x - ct) + G(x + et), where F and G are sketched
in Fig. 12.3.6. Sketch tIle solution for various times.

~~~L

(J

L __
x=2

Figure 12.3.6

--~~~
x=O

x=l

12.3.2. Suppose that a stretched string is unperturbed (horizontal, u = 0) and at

rest (au/at = 0). If all impulsive force is applied at t = 0, the initial value
problem is

a2 u

or

_ fIx - ct)

u(x, t) -

+ fix + ct ) +;-:
2

2(.
1

f'+d::J
g(x) dx.

Xo

EXERCISES 12.3

x=o

d'Alembert's Solution

+ et) + fIx

Figure 12.3.5: (a) Domain of dependence; (b) range of influence.

their ::;hifted sum. The striking of the broad hammer causes the displacement of the
string to gradually increase near where the hammer hit and to have this disturbance
spread out to the left and right as time increase8. Eventually, thp, string reaches an
elevated rest position. Alternatively, the solution can be obtained in an algebraic
way (see Exercise 12.3.5). The characteristics sketched in Fig. 12.3.3 are helpfnl.

fix

x+ct
(a)

Figure 12.3.4: Time evolution for a struck string.

12.3.3

+ ct

tr

tf
I ,--',

x = xo

(12.3.13)

at"
u(x, t)

a'u + a(x)5(t)
.

c-.-

ox'

t < O.

x-d

known as d' Alembert's solution (previously obtained by Fourier transform meth

ods). It is a very elegant result. However l for sketching solutions often it is easier
to work directly with (12.3.10) and (12.3.11), where these are shifted according to
(12.3.5).

Domain of dependence and range of influence. The importance

of the characteristics x - ct = constant and x + ct = constant is clear. At position x
at time t the initial position data are needed at x 1....1;, while all the initial velocity
data between x - ct and x + ct is needed. The region between x - et and x + ct
is called the domain of dependence of the solntion at (x, t) as sketched in Fig.
12.3.5. In addition, we sketch the range of influence, the region affected by the
initial data at one point.

(a) Without using explicit solutions, shm\i that this is equivalent to

a'u

a'u

at'

dx'

- - =c2 _ _

>0

subject to u(x, 0) = 0 and ~;' (x, 0) = a(x).

Thus, the initial velocity a(x) is equiva.lent to an impubive force.
(b) Do part (a) using the explicit solntion of both problems.
12.3.3. An alternative way to solve the one-dimensional wave equation (12.3.1)
is based on (12.3.2) and (12.3.3). Solve the wave equation by introdncing a
change of variables from (x, t) to two moving coordinates (e, I}) one moving
to the left (with velocity -c) and one moving to the right (with velocity c);

~ = x - ct

and

'] = x

+ ct.

Chapter 12. 1\lethod of Chm'aetcristio,

12.4. Serni-Infinite Strings and Reflections

539

538
A condition is lH'ct'ssary at the bounclary
at .r = 0:

*12.3.4. Suppose that u(x, t) = F(.r - el). Evaluate

~~ (0, t)

(ll)

(a) ~~ (x, 0)

{~

Du

iJt(x,O) = g(.r)

1.1'1> h.

different forms: depending on x ,I

u(:r, t) = F(x - ct)

= c \I

u.

Jl

= (1/ p)w(p. t) and verify that

a2 w

oj

8 2 1lJ

Dt' = c- Dp' .

Shuw that the mo~t genpral spherically symmetric solutlon of the wave

(b) equation COIl::iists of the ~um of two spherically synunetric waves, one
moving outward at speed c and the other inward at speed c. Note the

1 D2 u
PDE: I -:C-2
31

a2 u

c'-a
2
x

II

(12.4.5)

2c

l'

g(x) dx

.r

>0

(12.4.6)

1"

g(x) dx

x> O.

(12.4.7)

(I

2c

_ f(.r - ct) + f(T

u(.r, t) 2

+ et) + ~

.dd

g(x) dx,

2c .[ .1 '-('t

x> ct,

d'Alembert's solution. However, here thb is not valid if x < ct. Since x

Semi-Infinite Strings and Reflections

We wiH solve the one- dimensional wave e4.uation on a semi-infinite interval, x

+ G(T + ct).

Howevpr. it is very important to note that (unlike the case of the infinite string)
(12.4.6) awl (12.4.7) art' v"lid only for T > 0; the arbitrary functions an' only de
tennined from tIll' initial conditions for po:-:;itive argument.... In the general solution,
G(L' + d) requires only positive argument.1:; of G (since x > 0 and t > 0). On the
other h'1nd, F(.T - ct) requires po~itive arguments if .T > ct but rp<1uires negative
arguments if.r < ct. As indicat.ed by a space-time diagram, Fig. 12.4.1. the in
formation that. there is a fixed end at x = 0 travpls at a finite velocity c. Thus, if
:r > ct, the string does not know that. there is any boundary. In this case (.1' > ctL
the ,olntion is oblained as before [using (12.4.6) and (12.4.7)],

decay of the amplitude.

12.4

+ -1

-1 f(x) - -1

G(.r )

\I'll = (l/rl)(D/Dp)(p'Du/3p).

Make the transformation

-1 fl.r)
2

G(T)

Assume that the solution is spherically symmetric, so that

(a)

(12.4.4)

As in Sec, 12.3, the illiti<11 conditioll::; are satisfied if

12.3.6. Consider the three ~ dimensional wave equation

iJt'

that til{' string is fixen

ITI < "

(Hint: Using charaeteristies a~ sketched in Fi.g. 12,3,3, show there are two

distinct regions t < hie and t > hie, In each, show that the solution h8.-':I five

a2 u

SUppOSf'

Alt.hongh a Fourier sine transform can be Ilsed, we prefer to indicate how to use the
general solution and the method of characteristics:

U(X,O) = f(.r)

O. \Ve

I u(O, I) = 01

DC:

1
12.3.5. Dptennine analytic fonnula::; fora(x t) if

J.."

G(l:

> 0:

+ et)

1
= ;;f(.r
~

+ el) + -1

2c

1'+'"

(12.4.8)

+ ct > 0:

g(x) dx,

IJ

as determined earlipl'. To obtain F for negative argument.s, we cannot use the initial
conditions. Inst.ead: the boundary condition must be utiliz.ed. n(O, t) = 0 implies
t

lCl

lC2:

u(X,O) = f(x)

3u
at(x,O) = g(x).

I~x~ct
- x

Figure 12,4.1: Characteri1:>tic emanat

ing from the boundary.

Chapter 12. Method of Characteristics

12.4. Semi-Infinite Strings and Reflectiolls

541

540

that [from (12.4.5)]

o = F(~ct) + G(c!)

for t

> O.

(12.4.9)

Thus, F for negative arguments is -0 of the corresponding positive argument:

F(z) = -G(-z)
Thus, the solution for x - ct

n(r, t)

F(x - et)

for z

< O.

(12.4.10)

< 0 is

+ G(x + ctl

G(x + ct) - G(c! - x)

1
.
1
-If(x + et) - flct - x)] + 2
2c
1
.
1
-[f(.r + c!) - f(et - X)I + 2
2c

[1 x

+",

g(x) dx -

1,,,,-r g(x) dx]

0

1)

jX+c t g(x) dx.

Jo'

As further explanation, suppose that u(x, t) is any solution of the wave eqnation.
Since the wave equation is unchanged when;c is replaced by -;c, a( -;c, t) (and any
multiple of it) is also a solution of the wave equ<ttiuH. If the initial conditions
satisfied by 1L(.~, t) are odd functions of ;c, then hoth lI(;C, t) and -1L( -;c, t) solve
these initial conditions and r,he wave equation. Since the initial value problem has
a unique solution l u(;:r, t) = -11( -;T~ t); that is) u(;r t), which is odd initially, will
remain odd for all time. Thus, odd initial conditiom; yield a solution that will satisf:.r
a zero boundary condition at ;(: = O.
j

tt-x

To interpret this solution, the method of characteristic~ i~ helpful. Recall that

for infinite problem, n(x. t) is the sum of F (moving to the right) and G (moving to
the left). For se-mi-infinite probleIIls with x > ctl the boundary does not affect the
characteristics (see Fig. 12.4.2). If;c < ct, then Fig. 12.4.3 shows the left-moving
characteristic iG constant) not affected by the boundary, but the right-moving
characteristic emanates from the boundary. F is constant moving to the right. Due
to the boundary condition, F + G = 0 at x = 0, tIl{;' right-lllovillg wave is minus the
left-moving wave. The wave inverts as it "boullces off" the boundary. The resulting
right-moving wave -G(ct - x) is called the reflected wave. For x < ct, the total
solution is the reftee.ted wave plu.~ the as yet unrefleeted left-moving wave:

,"(;c,t)

One wi\Y to obtain this is to extend the initial position f(x) for ;C > 0 as an odd
function [such that f( -.r) = - f(x)1 and also extend t.he initial velocity g(.r) for
;c > 0 as au odd function [then its integral,
g(x) dx, will be an even function].
in summary, the solution of the semi-infinite problem with u = 0 at x = 0
is the same as an infinite problem with the initial positions and velocities
extended as odd functions,

:E > 0 with a fixed end u(O, t) = 0,

which is initially at rest, au/at(x, 0) = O. with an initial unit rectangular pulse,

Example. Consider a semi-infinite string

f(;c)
Since g(.x)

= 0,

4<x<5
otherwise.

it follows that

F(x)

G(;c + ct) -G(-(;c-ct)).

= {

= G(;c)

1
= 2f(x)

={

1
6

4<;c<5
ot.herwise (with x

>

0).

F moves to the right; G mOves to the left, negatively reflecting off ;c = O. This can
also be interpreted as an initial condition (on an infinite domain) with f(x) and
g(;c) extended as odd functions. The solution is sketched in Fig. 12.4.4. Note the
negative reflection.

I
I

Figure 12.4.2: CharacteristicS.

x=ct

Problems with nonhomogelleuu~ boundary conditions at x = 0 can be analyzed

in a similar way.

I
I
I

_~

The negatively reflected wave -G(-(x - ct)) moves to the right. It behaves as
at t = 0 it were -G( -;c). If there were no boundary, the right-moving
wave F(;c _ et) would be initially F(;c). Thus, the reflected wave is exactly the wave

if initially

;c =ct Figure 12.4.3:

characteristics.

that would have occurred if

F(;c) = -G(-;c)

for;c

< 0,

Fconstant

or eqUivalently
1
21 f (X) - 2c

1"
0

1
1
g(x) dx= -2J(-;C)
- 2c

1-'
0

"- G constant
g(7) dx.

G constant
;c

Reflected

Chapter 12. Method of Characteristics

542

12.5. Method of Characteristics

for a Vibrating String of Fixed Len/{th

u(x,O) =
8u
8t (x,O) =
8u

",,(O,t)
uT

Figure 12.4.4: Reflected pulse.

J(X)}
g(x)

543

x> 0

t> O.

(Assume that u is continuous at x = 0, t = 0.)

(b) Show that the solution of part (a) may be obtained by extending the
initial position and velocity as even functions (around x = 0),

o+

--<----~-~~~--c-2

(c) Sketch the solution if y(x) = 0 and

10

J(x) = {

4<x<5
othcnvjse.

EXERCISES 12.4

12.4.5. (a) Show that if u(x,t) and 8u/at are initially even around x = Xn,
u(x,t) will remain even for all time.

*12.4.1. Solve by the method of characteristics:

8'u
8'u
_=(;2_
8/'
8x'
subject to u(x,O)

= 0,

~~ (x, 0)

= 0,

(b) Show that this type. of e.ven initial condition yields a solution that will
satisfy a zero derivative boundary condition at x = XQ.

x>O

and u(O, t)

= h(t).

*12.4.6. Solve. (x > 0, t > 0)

8'u

*12.4.2. Determine u(x, ti if

a2 u

at2

282U

--=c
-
8 t2
8,,'

2
C

subject to the conditions u(x,O) = 0, ~~(x,O) ~ 0, and %;,(0, i) = iL(t).

0? U .
&x 2
for x <.. 0 only,
t

*12.4.7. Solve

t> O.
where u(x, 0) = COS1' X < 0, %7(x,O) = 0 x < 0, u(O,t) = eDo not sketch the solution. However 1 draw a. space - time diagnull l including
all important characteristics.

12.4.3. Consider the wave equation on a semi-infinite interval

8'u
a'u
at' =c' 8x'

forO<x<oc

00<X<2
1 2<x <~

{ Ox> 3

a and the initial conditions

~~(x,O) = O.

Determine the solution. Sketch the solution for various times. (Assume that

u is continuous at x = 0, t

==::

subject to u(:r, 0) = j(x), ~~(x,O) = 0, and %;-(O,t) = h(t). (Assume that

u is continllous at x = 0, t = 0.)

12.4.8. Solve

with the free bounda.ry condition ~~ (0, t) =

u(x,O) =

8'u
28'U x> 0

8t' ~c 8Il t>O

0.)

[)2 U
[) 2

8'u

c',...,.
with u(x,O) =
uX

t
subject to u(x, t) = y(t) along x = ~i(c > 0).

ui

12.5 Method of Characteristics

for a Vibrating String of Fixed Length
In Chapter 2 we solved for the vibration of a finite string satisfying

12.4.4. (aJ Solve for x > 0, t > 0 (using the method of characteristics)
8'u
8'u
8 t' = c? 8Il

8u

0 and ""(,,,0) = 0,

PDE:

a'u
_ = c2 _8'u
8 t'
8x'

(12.5.1)

544

Chapter 12. Method of Characteristics

12.5. Vibrating String of Fixed Length

5451
t

u(O.t) =0
ulL,t)=U

(12.5.2)

11(X,0) = f(x)
au
a;-(x,O) = g(x),

(12.5.3)

~.

B v.

Ie:

Figure 12.5.2:
characteristic-so

MUltiply reflected

using Fourier series methods. \Ve can obtain an equivatent 1 but in some ways more
useful, result by using the general solution of the one- dimensional wave equation:

u(x, t)

= F(x ~ ct) + G(x + et).

(12.5.4)

The initial conditions are prescribed only for 0 < x < L, <;tnd hence the formulas
for F(x) and G(x) previously obtained are valid only for 0 < x < L:

F(x).
G(x)
If 0 < x et < L and 0
solution is valid:

u(x, i) =

-1 f(x) -1

1
x

g(x) dx

(12.5.5)

r g(x) dx.

(12.5.6)

2c

+ 2e Jo

2f(x)

< x + ct < L

Alternatively, a solution on an infinite domain 1vithout boundaries can be Con

sidered which is odd around x = 0 and odd around x = L. as sketched in Fig. 12.5.3.
In t.hi,., way, the zero condition at both x = 0 and x = L will be satisfied. We note
that u(x, t) is periodic with period 2L. In fact, we ignore the oddness around x = L,
since periodic functionR that are odd around x = 0 are auLumatically odd around
x = L. Thus, the simplest way to obtain the solution is to extend the initial
conditions as odd functions (around x = 0) which are periodic (with pe
riod 2L). With these odd periodic initial conditions, the method of characteristics
can be utilized a~ \vel] as d'Alembert's solution (12.5.7).

a, shown in Fig. 12.5.1, then d'Alembert's

f(x - et) + f(x

2

+ ct) + -1

1'+0'

2c

g(x) dx.

(12.5.7)

x-ct

In this region the string does not know that either boundary exists; the information
that there is a boundary propa~ates at velocity c from x = 0 and x = L.
t

l~~~~1

."=".,,C'_,"'''..

If one's position and time is such that signals from the boundary have already
arrived, then modifications in (12.5.7) must be made. The hOllndary condition at
x = 0 implies that
0= Fl ~et) + G(ct) for i > 0,
(12.5.8)

Pigure 12.5.3: Odd periodic extension,

Example. Suppose that a string is initial1y at rest with prescribed initial con
ditions u(x,O) = f(x). The string is fixed at x = 0 and.T = L. Instead of using
Fourier series methods, we extend the initial conditions as odrl fllllctions around
x = 0 and J: = L. Equivalently, we introduce the odd periodic extension. (The odd
periodic extension is also used in the Fourier series solution.) Since the string is
initialfy at rest, g(x) = 0, the odd periodic extension is glx) = 0 for al1 x. Thus,
the solution of the One- dimensional Wave equation is the sum of two simple waves:

while at x = L we have

0= F(L ci)

+ GlL + et)

for t >

o.

These in turn imply reflections and multiple reflections, as illustrated in Fig. 12.5.2.

u(x, t)

= 2[f(x -

el)

+ f(x + Ci)],

Where f(x) is the odd periodic extension of tire given initiai position. This solution
is much simpler than the summation of the first 100 terms of its Fourier sine series.

546

Chapter 12. Method of Characteristics

12.6. Quasi-linear PDEs

547

EXERCISES 12.5

!nethod of characteristics, used to solve the wave equation, can be applied to partiaJ
differential equations of the form

12.5.1. Consider

a2 u

2 02U

D:(T, 0)

Dx'

at(x,O)

--=c--

Df'

u(O, t)

f(x) }
gCT)

u(L, f)

*(b) If g(]') ~ 0, show that part. (a) is equivalent. to the results of Chapter 12.
(c) If f(x) = 0, show that part (a) is equivalent to the results of Chapter 12.
characteri~tics:

iJ t2 =
u(x,O) = 0

n(O, t) = h(l)

12.5.3. Consider

dp
df = Q(p,x,!).

ax2

Bu(].
fi t ,

2u
{)2_U
_
_ _ (,2 _ _

iJ f' - .

ax'

'

0) =

au
8i(;r,O)

u(L, I) = O.

n(O,

di
t)

= [)

au

ax (L, f) = O.

(a) Sketch the solution using the method of characteristics.

(b) Obtain the solution using Fourier- series- type tcchniquL"'S.

(c) Obtain the solution by converting to an equivalent problem on an infinite

domain.
12.5.4. How should initial conditions be extended if au/D:r(O, t)
07

= a and u(L, I) =

12.6 The Method of Characteristics

for Quasi-linear Partial Differential Equations

(12.6.2)

if

d.T

otherwise

= g(.T) = 0

o < x < 10

u(x 0) = f(x) = { 1 4 <.r < 5

,

where c and Q may be functions of;T, t 1 and p. 'Vhen the coefficient. c depends on the
unknown solution p, (12.6.1) is not linear. Superposition i8 not valid. Nonetheless
(12.6.1) i~ called a quasi-linear partial differential equation, since it is linear in the
first part.ial derivatives, lip/a I and ap/a". To solve (12.6.1), we again consider an
observer moving in Some prescribed way .r(t). By comparing (12.2.7) and (12.6.1).
\VP obtain

2 82u

iJ2 n

(12.6.1 )

O.

(a) Obtain the solution by Fourier series techniques.

12.5.2. Solve using the method of

a
l
aapt + c-!!.
ax Q,

O<x<L

c(p, .7:, f),

(12.6.3 )

The partiaJ differential equation (12.6.1) reduC'c.-; to two coupled ordinary differen
tial equat.ions along the special traject.ory or direction defined by (12.6.,3), known
as a characteristic curve, or simply a characteristic for short. The velocity
defined by (12.6.3) is called the characteristic velocit.y, or local wave velocit.y.
A characteristic starting from x = ;ro, as illustrated in Fig. 12.6.1, is determined
from the coupled different.ial equations (12.6.2) and (12.6.3) using t.he init.ial condi
tions p(x,O) = f(x), Along the characterist.ic, the solution I' changes according to
(12.6.~). Other initial positions yield other characteristics. generating a family of
characteristics.
t
dx

(It = c

<ip~ Q

dt

Figure 12.6.1: Characteristic starting

from ;t = Xo at time t = O.
x=xo

p~p(xo,O) ~ f(.xo)

12.6.1 Method of Characteristics

fvlost of this text describes methods for solving linear partial differential equations
(separation of variables, eigenfunction ex-pansions, Fourier and Laplace transforms,
Green's functions) that cannot be extended to nonlinear problems. However 1 the

Exatnple. If t.he local wave velocit.y c is a const.ant Co and Q = 0, then the

quasi-linear partial different.ial equat.ion (12.6.1) becomes t.he linear one, (12,2.6),
which arises in the analysis of the wave equation. In this example, the characteris
t.ics may be obtained by direct.ly int.egrat.ing (12.6.3) wit.hout. using (12.6.2). Each

Cbapter 12. Metbod of Cbaracteristics

12.6. Quasi-linear PDEs

549

548
characteristic has the same constant velocity,

The falnily of characteristics are

C{).

parallel straight lines, as sketched in Fig. 12.2.l.

Thus, l,y taking the time-derivative inside the integral (making it a partial deriva
tive) and using (12.6.9) it follows that

Quasi-linear in two _ dimensional space. If the independent vari

ables are x and y instead of x and t, then a quasi-linear first-order partial differential

op

at

equation is usually written in the form

op

op

a - + b - =c

ox

where a, b, c may be functions of

X,

if

oy

oq =

01

~l

(12.6.10)

(12.6.4)

since a and b are arbitrary (see Section 1.2). We call (12.6.10) conservation of
cars.

y, p. The method of characteristics is

c
dp
dx =~'

(12.65)

dy
dx

(12.6.6)

=-;;,

Car velocity. The number of cars per hour passing a place equals the density
of cars times the velocity of cars. By introducing u(x, t) as the car velocity. we
have

= pu. ,

(12.6.11)

This is written in tb.e following (easy to memorize) equivalent form

dx
dy
dp
--;; = b = 7

(12.6.7)

12.6.2 Traffic Flow

Traffic density and flow. As an approximatiou it is possible to model a

In the mid-1950s, Lighthill and Whitham [1955J and, independently, Richards [1956]
made a simplifying assumption, namely, that the car velocity depends only on the
density, u = 1L(p), with cars slowing down as the traffic density increases; i.e.,
dujdp 0:: o. For further discussion, the interested reader is referred to Wb.itham
[1974] and Haberman [1977]. Under this assmnption, the traffic flow is only a
function of the traffic density, q = q(p). In this case, conservation of cars (12.6.10)

congested one _directional highway by a quasi-linear partial differential equation.

We introduce the traffic density p(x, t), the number of cars per mile at time t
located at position x. An easily observed and measured quantity is the traffic flow
q(x, t), the number of cars per hour passing a fixed place x (at time t).

becomes

Conservation of cars. We consider an arbitrary section of roadway, be

tween x = a and x = b. If there are neither entrances nor exits on this segment of
the road, then the number of cars between x = a and x = b (N =
p(x, t) dx, the
definite integral of the density) might still change in time. The rate of change of
the number of cars, dN j dt, equals the number per unit tlIne entering at x = a lthe
traffic flow q(a, t) there] minus the number of cars per unit time leaving at x = b

where c(p) = q'(p), a quasi-linear partial differential equation with Q = 0 (see

(12.6.1)). Here c(p) is considered to be a known function of the unknown solution
p. In any physical problem in whicb. a density p is conserved and tb.e flow q is a
function of density, p satisfies (12.6.12).

f:

lthe traffic flow q(b, t) there]:

op

op

at + c(p)&; =

12.6.3

0,

(12.6.12)

Method of Characteristics (Q = 0)

The equations for the characteristics for (12.6.12) are

1
b

dt
d a p(x, t) dx = q(a, t) - q(b, t).

(12.6.8)

dp = 0
dt

(12.6.13)

along
Equation (12.6.8) is called tb.e integral form of conservation of cars. As with heat
flow, a partial differential equation may be derived from (12.6.8) in several equivalent
ways. One way is to note that the boundary contribution may be expressed as an
illtegral over the region:
b

q(a, t) - q(b, t)

=-

1
a

ox q(x, t) dx.

dx
dt

= c(p).

(12.6.14)

The characteristic velocity c is not constant but depends on the density p. It is

known as the density wave velocity. From (12.6.13), it follows that the density
p remains constant along each as yet undetermined characteristic. The velocity of
each chara.cteristic, c(p), will be constant, since p is constant. Each characteristic

Chapter 12. Method of Characteristics

550

is thus a straight line (as in the case in which c(p) is a constant CO). However,
different characteristics will move at different constant velocities because they may
start with different densities. The characteristics, though each is straight, are not

parallel to one another. Consider the characteristic that is initially at the position
x = XO, as shown in Fig. 12.6.2. Along the curve dx/dt = c(p), dp/dt = 0 or pis
constant. Initially p equals the value at x = Xo (i.e., at t = 0). Thus, along this one
characteristic,

p(x, t) = p(xo, 0) = f(xo),

12.6. Quasi-linear PDEs

551

proce~s must be carried out for a large number of points (as is elementary to do on
any computer). In this way, we could obtain the density at time t.

Fan-like characteristics. As an example of the method of characteristics,

we consider the following initial value problem:

(12.6.15)

line,

since x

XQ

c(f(xo))t + Xo.

at t = O. Different values of

XQ

yield different straight-line charac

XQ

dx
dt

2p.

Thus, the characteristics are given by

that goes through that space - time point must

be obtained from (12.6.16).

x <0
x> O.

3
4

={

The density p(x, t) is constant moving with the characteristic velocity 2p:

(12.6.16)

teristics, perhaps as illustrated in Fig. 12.6.2. Along each characteristic, the traffic
density p is a constant; see (12.6.15). To determine the density at some later time,
the characteristic with parameter

op

ox

p(x,O)

which is a known constant. The local wave velocity that determines the character

istic is a constant, dx/dt = c(f(xo)). COl"equently, this characteristic is a straight

op

ot

-+2p-, =0

= 2p(xo, O)t + xo.

If Xo > 0, then p(xo, 0) = 4, while if Xo < 0 then p(xo, 0)

sketched in Fig. 12.6.4, show that

p(x.t)

~ f(xo) Figure

~x

12.6.2:

p(x. t)

Possibly nonparallel

straight-line characteristics.

Xo

Graphical solution. In practice, it is often difficult and not particularly

interesting actually to determine Xo from (12.6.16) as an explicit function of x and
t. Instead, a graphical procedure may be used to determine p(x, t). Suppose the
initial density is as sketched in Fig. 12.6.3. We know that each density Po stays the
same, moving at its own constant density wave velocity c(po). At time t, the density
Po will have moved a distance c(Po)t as illustrated by the arrow in Fig. 12.6.3. This

{4
x><
3

(12.6.17)

= 3. The characteristics,

8t
6t,

as illustrated in Fig. 12.6.5. The distance between p = 3 and p = 4 is increasing: we

refer to the solution as an expansion wave. But, what happens for 6t < x < St?
The difficnlty is caused by the initial density having a discontinuity at x = O. We
imagine that all values of p between 3 and 4 are present initially at x = O. There
will be a straight line characteristic along which p equals each value between 3 and
4. Since these characteristics start from x = 0 at t = O. it follows from (12.6.17)
that the equation for these characteristics is
x

= 2pt,

for 3

< p < 4,

also sketched in Fig. 12.6.4. In this way, we obtain the density in the wedge-shaped
region

"
,,

,
~

Xo

--_

for 6t

< x < 8t,

which is linear in x (for fixed t). We note that the characteristics fan out from x = 6t
to x = 8t and hence are called fan-like characteristics. The resulting density is
sketched in Fig. 12.6.5. It could also be obtained by the graphical procedure.

,,

p(x,O)

/1

p = 2t

p(x,t)

.... /

xo + c(Po)t

Figure 12.6.3: Graphical solution.

12.6.4

Shock Waves

Intersecting characteristics. The method of characteristics will not always

Work as we have previously described. For quasi-linear partial differential equations,
it is quite usual for characteristics to intersect.. The resolution will require the

Chapter 12. Method of Characteristics

552

Figure 12.6.4: Characteristics (in

cluding the fan-like ones).

12.6. Quasi-linear PDEs

553

the method of characteristics~ the fHster-moving characterititic PH.S~PS the slower.

Then we obtain Fig. 12.6.7(d). The method of characteristics predicts that the
density becomes a "rnultivalued" fnnction of position; that iR, at some later time our
mathematics predicts there will be three densitief) at sOme positions [as illustrated
in Fig. 12.6.7(d)]. We say the density wave breaks. However, in many physical
problems (such as traffic flow) it makes no sense to have three values of density at
3
One place. The density must be a siuRle-valued function of position.

p ~ xj2t
p

~"~_X

p=4
p~3

(.J
x = 6t

Figure 12.6.5:
wave.

Expansion

4,

35

Figure 12.6.7: Density wave r;teepens

(density becomes triple - valued).

x = 8t

.. 8

2
H

0.5

"0

introduction of moving discontinuities called shock waves. In order to make the

mathematic,al presentation relatively simple. we feHtrict our attention to quasi-linear
p"rti"l differential equations with Q = 0, in which case
[)p

ap

at + c(p) ax

(12.6.18)

Discontinuous solutions. On the bllliis of the qUlllii-linear partial differ

ential equation (12.6.18), we predicted the physically impossible phenomenon that
the density becomes multivalued. Since the method of characteristics is mathemati
cally justified, it is the partial differential equation itself which must not be entirely
valid. Some apprOXimation or assumption that we used must at times be invalid.
We will assume that the density (as illustrated in Fig. 12.6.8) and velocity have a
jllrup- dbcontinuity, which we call a shock wave, or simply a shock. 4 The shock
occurs at SOme unknown position X 8 and propagates in time. so that x,~ (1). We
illtruduce the notation ;r,~_ and x s -'- for the position of the shock on the two sides
of the discontinuity. The shock velocity, dx,/dt, is a5 yet unknown.
p(x,t)

p(xM,t)

~,_,t)
.

In Fig. 12.6.6 t.wo characteristics arE' sketched, one starting at x = Xt, with p ::=
!(XI,O) '" PI e.nd the other starting at x = X2 with P = !(X2'O) '" P2' These
characteristics intersect if c(pJl > C(P2), the faster catching up to the slower. The
density is constant along characteristics. As time increases, the distance between the
densities PI and P2 decrea<;es. Thus, this is called a compression wave. We sketch
the initial condition in Fig. 12.6.7(a). The density distribution hecomes steeper as
time i.ncreases [Fig. 12.6.7(b) and (e)]. Eventually charaderistics intersect; the
theory predicts the denslty if> simultaneously PI and pz If we continue. to apply

Figure 12.6.6: Intersecting character

istics.

C(PI) > e(p,)

.~
P = PI

I
P~P2

--

X,(t)

Figure 12.6.8: Density di::::continuity

at x = x,(t).

-x

3The partial differential equation~ describing the height of water wave'" near the shore i.e., in
shallow waleI') are similar to the equation:; for traffic density waves. In this situation the prediction
of breaking is then quite significant!

4The terminology .~h()ck wave is introduced because uf the analogous behavior that occurs in
gas dYnamics. There, changes in pressure and density of air, for example, propagate and are heard
(due to the sensitivity of the human ear). They are called sound wave.';. \\Then fluctuations of
prC.5sure and density are small, the equations describing sound waves can be linearized. Then
sound is propagated at a Constant speed known as the sound speed. However, if the amplitudes
of the fluctuations of pressure and density arc not ::;Illall. t}U,!H the partial differential equations
are quasi-linear. Characteristics may intersect. In this case, the pressure and density can be
mathematically modeled as being discontinuous, the result being called a shock 1l1avp. Exa.mples
are t.be 50und emitted from an explosion or the thunder resulting from lightning. If a shack wave
results from exceeding the sound barrier, it is known as a sonic boom.

Chapter 12. Method of Characteristics

554

Shock velocity.

On either side of t.he shock, t.lle quasi-linear partial dif

ferent.ial equation applies, ap/at + c(p)ap/a.r = O. where c(p) ~ dq(p)/dp. We
need to determine how the discontinuity propagates. If p is conserved even at, a
discontinuity: then the flow relative tu tbe moving ~hock on Olle sid", of the ~h()rk
mnst equal the flow [dative to the moving shock on the other ~ide. This statement
of relative inflow equaling relrttive outflow beeomes

dX.]

p(x.,_.t) u(x,-,t)-di' =p(x .+,t) U(XH,t)

rh.,l.
ell

(l26.1Q)

Cl.G. Quasi-linear PDEs

Fig. 12.o.9(a). The lllC't.hod of characteristics yields a multi-valued solution of the

partial differentia.l eqwltion. Tlli.'i diffiClllty is n>medied by introducing a shuck wave
[Fig. 12.6.9Ib)], a propagating wave indicating the path at which densities and ve
locities abrnptly change (i.e.. are discontinuous). On one side of the shock, the
method of charactf'risti(~s suggests t he densit~r is constant p = 4) and on the other
side I' = 3. We do !lot know as yet t.he pat.h of the shock. The theory for SUell a
discontiullouS solution implies that the path for any shock Illllst satisfy the ~hock
condition, (12.G.20). Suhstitutillg the jumps in flow and den",ity yields the following
equation for the shock \'E'locjt.y:
d:.c$

since flO'\' equals den..,it;r times velocity

(here relative velocUy). Solving for the
shock velocity from (12.6.19) yields

d:f.~

elt

[q]
q(:r . +, t) - qCr,_, i)
=
[pi'
p(x.,+. t) - p,:r,_, t)

(12.6.20)

where we recall that q = pu and where we introduce the notation [q] and [pi for the
jumps in q and p~ respectively. In ga." dynamics, (12.6,U) is calkd the Hankiue
Hug;oniot condition. In summary, for the conservation law 8pI8 t+8q/8x = 0
(if the quantity p dx is actually conserved), the shuck velocity equals
the jump in the flow divided by the jump in the density of the conserved
quantity. At points of discontinuit,v, thb ~1rock condition replaces the use of the
partial differential equation, which i~ valid elsewhere. However. ViE' have not yet
explained where shock~ uccur and how to determine p(X1j+' t) and fI(X,,-, t).

.r

505

dt

q(4)
q(:3)
4_ 3
-

4' - 3'
, ,,= 7,
"'i-a

sinc> in thi....: case q = p2. Thus, the shock moves at a constant velocity. The
initial po1:iition of the shock is Known, giving a condition for this first-order ordinal'.r
diH'erential equation. In this case. the shock must initiate at x., = 0 at t = o.
COllsequently', applying che initial condition results in the position of the shock
::1:'8

= 7t.

The n>sulting spacE'- time diagram is sketch(-,(l ill Fig. 12.6.9(c). For any time t > 0,
the traffic density is rliscnntinuOlIS, lis shown in Fig. 12.G.I0.

Example. We conr--icier the initial yallle problplll

81'

at

(a)

2 ap_

I' ax - 0

4
p(x,O) = { 3

(bJ

(c)

Figure 12.6.9: Shock cau.-;ed by inter1'lecting characteristics.

< (J
x> O.

,1;

We a.<;;sume that p is a conserved densit.y. Putting the partial differential equation

in conservat.iun form (ap/at + aq/ax = 01 shows that t.he flow q = 1'2 Thus, if
t.here is a discontinuity, the ,hock velocit.y sat.isfies elx1elt = [ql/[p] = [(J"l/Ip] The
densit;v p(.T, t) is constant. moving at the characteristic velocity ?p:
p~J

elx = 2(J
elt

"

t,

,,,

Figure 12.6.10: Density shock wave.

"A'

Therefore, the equation for the characterbtics is

x = 2p("'u, OJt

0.(

".

+ xu

If XQ < 0, then p(:l'o, 0) = 4, This parallel group of characteristict; intersects

those st.art.ing from Xu > a (with p(Xn,O) = 3) in the cross - hatched region in

Initiation of a shock.

We have described the propagation of shuck waves.

In the example considered, the density was initially discontinuous; thus, the shock

Chapter 12. Met1lOd of Characteristics

556

wave formed immediately. However 1 'we will now show that shock waves take a finiLe
time to form if the initial density is continuous. Supf.lose that the first shock 0('('Uf5
at t = r due to the intersection of two characteristics initially a distance D..x (not
necessarily small) itpart. However, any characteristic starting between the two at
t = 0 will almost always intersect one of the other two characteristic:; before t = r.
Thus shocks cannot first occur due to characteristics that are a finite distance ~x
apart. In,';tead~ the first shock actually occurs dup to the intersection of neighboring
charaderistics (the limit as t.x __ 0). We will show that even though fu -- 0, the
first intersection occurs at a finite positive time, the time of the earliest shock. The
density p is constant along <..:haracteri~ticsl satisfyjug d:rjdt = c(p). 'Ve will analyze
neighboring characteristics. Consider the characterbtic emanating from x = ;:ro at
t = 0, where p(x,O) = f(x),
(12.6.21)
x = c[J(xo)]t + Xu
and the characteristic :;tarting from .r -'
X

= elJ(xo

X(J

+ .6..x at t --::; 0,

+ t.";)]t + Xo

I t.x.

Ouly if clJ(xo)J > c[J(xo + i'lx); win the,c chmacteristicsintuscct (in" positive
time). Solving fUf the intersection point by pliminating X ~yield:"

c'J(XD)]t

+ Xv

~ CIJ(XO

12.6. Quasi-linear PDEs

5.S7
TIlis
7"

has also u,ed the result of partwl differentiation of (12.6.21) with re'pcet to

..

ax"

1-_

- 8T

[1 +-1
de]

d:eu'

The slope is inrinite at tho,e place' satisfying (12.6.22). Thi, shows that the turning
points of the triplc- valued solutiu!1 correspond to the intersection of neighboring
characteristics (the envelope of the eh"Taeteristies). Within the envelope> of charac
teristics, the 'olution is triple- valued. It is known that the envelope of the char
acteristics lS cU'P-shaped; aB indicated in Fig. 12.6.11. How{>ver, the triple-valued
,olution (within the cusp region) makes no seI"e. Iu,tcad, as d"cussed earlier. a
shock W"ve exists satis(yillg (12.0.20), initiating at the (;llSp point. The shock is
located within the envelope. In fact, the triple - valued solution, obtained by the
method of characteristics, may be used to determine the location of the shock.
Whitham [1974] ha' shown that the correct location of the shock may be deter
mined by cutting the lobes off to form equal arms (Fig. 12.6.12). The rpa."'11 for
thb is that the method of chamcteristics conserves cars and that, when a shock is
introoucpd. the number of cars (represented Ly the '"'ea Ip dx) must al,o be the
same aN it is initially.

+ L'>x)lt + Xo + D.x.

Therefore, the time at which nearly neighboring curves intersect is

t=

i'll;
elJ(xoIJ- c[J(xo

+ t.x)]

1
{clJ(xolJ - c[J(xu

--.

~~

+ t.x)]}/t.x

The characteristics aI'/? paths of obseryers following constant density. Then this
equation states that the time of intersection of the two characteristics is the initial
distance between the characteristics dividt:!d by the relative velocity of the two
characteristicti. Although the distance In hetween is small. the relative velocity is
also small. To consider neighboring chal'actel'istics 1 the limit as ~x --+ 0 must be
calculated:

(12.6.22)

Charaeteri'tics will intersect (t> 0) only if (d/d.ro)c[J(xo)] < O. Thus, we conclude

that all neighboring characteristics that emanate from regions where the charac
teristic velocit.y is locally decreasing will always intersect. To ddermine the first
time at which an intersection (shock) occurs, we must minimize the intersection
time over all possible neighboring ('haracteristics~ Le., find the absol'ute minimum
of t given by equation (12.6.22). This can be calculated by determining where
d2/dx6C[f(xo)] = o.

Shock dynamics, We will show that the slope of the solution is infinite
where neighboring characteristics intersect. Since p(x, t) = p(xo, 0), we have
8p
8x

dp DXD
dx o 8x

= .:!:!!.- /
da o

[1 + dxo
de t] .

Figure 12.611: Euvelupe of char

acteristics. locus of inter~ections of
neighborillg charact~ristics.

Figure 12.6.12:
area principle.

\Vhitham's equal-

~
xAt)

12,6,5

QuaSi-Linear Example

Consider the quasi-linear example

{}p

- - pP
8t
7h

= -2p,

(12.6.n)

Chapter 12. Method of Characteristics

l~Uj.

<..!uasi-linear }-'l)l'.,'s

559

558
Thus, an explicit solution can be obtained
subject to the initial conditions

p(x, 0)

(12.6.24)

= f(x).

Xo =

This could model a traffic flow problem (with p a scaled version of the density)
where cars are not conserved but irl,;tead leave the roadway (at exits) at a rate
proportional to the density (as though cars e"it the highway to avoid congestion).
The method of characteristics yields
dp _ -2p,
dt

along the characteristic

(12.6.26)

dx

I+c

2t'

Note that for each :r and t there is only one characteristic Xo (because the initial
condition was chosen such that the family of characteristics ooes not intersect itself.)
From (12,6.28), the solution of the initial value problem for the partial differential
equation is

(12.6.25)

-=-p.
dt

2x

.
2xe- 2t
2x
Plx,t)
= 1
2, = - 1
'
+c
+e 2t"

General solution. Quasi-linear partial differential equations have general

solutions (as with the linear wave equation). "Tithout specifying initial c-onditions,
integrating (12.6.25) and (12.6.26) yields

These equations are sometimes written in the equivalent form

X

Sometimes the coupled system of ordinary differential equations can be directly

solved. In this ewe the ordinary differential equation for p may be solved first, and
its solution used to determine the characteristic x. Because of the initi111 condi
tion (12.6.24), we introduce the parameter Xo representing the characteristic that
emanates from x = Xo (at t = 0). From (12.6.25), along the characteristic, we
ohtain

p(x,t) = p(xo,O)e- 2t ~ fl:ro)c-

2
'.

(12.6.28)

The pal'ameter Xu is constant along each characteristic. The solution (density) expo
nentially decays along the characteristk as time increases. Thus, the characteristic
velocity becomes

d:r
__

=-

p(x,t)

(12.6.27)

= dt.

dp _ dx
-2p - -p

dt

cle- 2t

cle-

2t

+c2'

In general) One constant can be an arbitrary function of the other contant. Thus
1
we obtain the general solution of (12.6.23),

p + f(pe").

EXERCISES 12.6
12.6.1 Determine the solution p(x, t) satisfying the initial condition p(x, 0) = f(x)
if

* (a)
* (c)

f()
xoe -2< .

~=O
~ = -3:rp

~=_~+~n
at

~ ~=X2~

(b1.

By integrating the velocity, we obtain the posltion of the characteristic:

1
-2t
x = '5/(xo)e

-"21 f( 1'0 ) + :ro,

(12.6.29)

since x must equal Xo at t = O. Here, the characteristics are not straight lines
The parametric representation of the solution is obtained fWIll (12.6,28), where "'0
should be considered as a function of :r and t from (12.6.29). Usually an explicit
solution is impractical.

Explicit solution of an initial value problem. For the quasi-linear

partial differential eqnation (12.6.23). suppose the initial conditions are

p(x,O)

f(x)

x = '2xoe

-2<

1 xo

+ "2

12.6.3. Suppose

1 + sinx

f!i + co~ = 0 with Co constant.

*(a) Determine p(x,t.) if p(:r, 0) =sinx.

*(b) If co> 0, determine p(x, t) for:r > 0 and t > 0, where p(x, 0) = f(x) for
x> and p(O, t) = get) for t > 0.
(c) Show that part (b) cannot be solved if Co < O.
*12.6.4. If u(p) = a+ r3p, determine (} ancl,q such that u(O) =
0.

x.

In this casc, froul (12,6.29), the characteristics satisfy

1

*12.6.2. Determine the solution of Dp/Dt = p, which satisfies p(x, t)

along :r = - 2t.

max and "(Pm.x)

(a) What is the flow as a function of density? Graph the flow as a function
of the density.

Chapter 12. Method of Characteristics

l~.U,

l..tW:LN-lIIH:::dl- rUjb

LlO.1

560
(b) At what density i, the flow ffiiL'Cimum? What i' the corresponding ve
locity? What is the maximum flow (called the capacity)?

12.6.5. RecIo Exercise 12.6.4 ifu(p) = umax(1-

p'j); = 31'

(b)

* (c) e.
at + t'pe.
a,T = _I'

p3/p~,ax)'

12.6.10. Solve
p(x,O) = O.

12.6.6. Consider the trathc flow problelll

31'
31'
3t + "(1') ox

* (a) ~ -

+ t'~

(d) '!e
Dt + pe.
Ox = -xl'

= 41' for x > 0 and t >

12.6.12. Consider (12.6.8) if there is a moving shock x, such that a < x,(I) < b.
By differentiating the integral [with a discontinuous integrand at x,(I)], derive
(12.6.20).

x<O

0 < x < a

O x > a

PIT,ax

' ",ox
p

ma

p,,~,

(c) p(x, 0) =

(bl

(c)

+ p2~ = U

Wi +41'* =0
~ + 3pj); =
8

(a) What sign should v have for this expression to be physically reasonable?
(b) What equation now describes conservation of cars?

p(x, 0) = {
p(x,O)

p(x, 0)

={
,

(c) Assume that U(p) = u rr,ax(1- P/Pmax)' Derive Burgers' equation:

x<o
X>O

(d) 7!t+6pax =Oforx>Oon.y

31' +u

x < 1

{) t

= -p
ax

* (e) at
* (g) %i + xj); = t

p(x, 0) = 5
plO,I)=2

x>
1>0

fl x )

(d) E.
at

+ 5te.
ax = 31'

(f) E.
at

+ tze.
ax =

12.6.9. Detennine a parametric representation of the solution satisfying p(x, 0)

f(x):

[1-~] 01'
Pmax

AX

v32p
Ox2 .

(12.6.30)

12.6.14. Consider Burgers' equation as derived in Exercise 12.6.13. Suppose that

a solution exists as a density wave moving without r;hange of shape at velocity
V, p(x, I) = f(x - VI).

x<o
O<x<1
x>1

3x
(b) ~ + 3xj); =
* (a) !ift + c~ ~ e

=5

max

x> 1

12.6.8. Solve subject to the initial condition p(x, 0) =

J2. _ t"2!!''

v 31'

--,
I' 3x

where v is a constant.

<0
I> 0

E.

+ tE.
* (e) E.
() i
ax

= U(p) -

12.6.7. Solve the following problems:

(a) ~

with 1'(0, I) = h(l) and

12.6.13. Suppose that, instead of u = U(p), a car's velocityu is

started by a red light turning green.

, {3

(a) p(x,O) = pmax for x < 0 and p(x,O) = for x > O. This corresponds to
the traffic density that results aft.er an infinite line of stopped traffic is

12.6.11. Solve ~ + (I + t)~ = 31' for I > 0 and x > -1/2 with p(x, 0) = f(l) for
x > and p(x, t) = 9(1) along x = -t/2.

= O.

Assume u(p) = u mnx (l- P/Pmax). Solve for p(x, t) if the initial conditions are

lb) p(x,O) =

if, + pj); =

'(a) What ordinary differential equation is 'etisfied by f?

(b) Integrate this differential equation once. By graphical techniques show
that a solution exists such that f -> 1'2 as x -> +00 and f -> PI Il.S
X -4 -00 only if P2 > Pl. Roughly sketch this solution. Give a physical
interpretation of this result.
*(c) Show that the velocity of wave propagation, Ii, is the same as the shock
velocity separating I' = Pl from I' = 1'2 (occurring if v = 0).
12.6.15. Consider Burgers' equation as derived in Exercise 12.6.13. Show that the
change of dependent variahles

VPmax cPx
1'=----,
U max
introduced independently by E. Hopf and J. D. Cole, transfuflns Burgers'
equation into a diffusion equation, ~ + U max ~ = v~. Use this to solve

Chapter 12. NIethod 01 CharactenstIci:>

562

1:L. /. rlTst-Urder Nonlmear JJUl!h'

;'OJ

the initial value problem ('Cr,O) = i(x) for -00 < x < 00. [In Whitham
[1974] it i~ shown that thi8 exact solnti01l can be asymptotically analy'Zed as
!J _ 0 using Laplace'::; method for exponential integrals to show that p(x, t)
approaches the solntion obta~nE'd for v = 0 using the method of characteristics
with shock dynamics.]
12.6.16. Suppose that the initIal traffic den"ity is p(x,O) = PD for x < 0 ann

p(x,O) = PI lor x > O. Consider the two case". Po < Pl and Pl < po For

which of the preceding ca...,es is a den~ity shock necessary? Briefly eX}lli:Lin.

12.6.17. Consider a traffic problem, with u(p) =

'1L11lUX(1 -

pi PmB,x)' Determine

then oyer a few wave lengths thp. wave sees nearly constant c. However, over long
distances (relative to short wave lengths) we may be interE'sted in the effects of
variahle c. Another situl.1tion in which nearly plane waves arise is the reflection of a
plane wave by a curved boundary (or reflection and refraction by' a curved interface
between two rnedias with different indices of refraction). We assume the radius of
curvature of the boundar,V if:' much longer thaD typical wave lengths. III mallY of
these situations the temporal frequency w b fixed (by an incoming plane wave).
Thus.

'(
) -iwt ,
E =fixlye

p(x,t) if

* (a)

{)2A

{ ,,/,,,,
~

p(x.O) =

P',"'

'C'O

~: 0

(b) p(x,O) =

2/?;",

x<O
~>0

12.6.18. Assume that u(p) -:- 'Uma x(1 ~ (>'21 P~w,J Determine the traffic density p

(for t > 0) if p(x. 0) = Pl for :r < 0 and p(x,O) = p, for x > O.

(a)

Assume that P,

* (b)

> Pl

Assume that P,

< Pl

(b)

(c)

(d)

Yi + 4p!i: = 0
.

+ 3p* = 0

pix
0)
. '

(12.7.4)

exist if

={

x < 0

43 x>O

p(x, IJ}

={

(12.7.6)
The wave numbers k1 and k2 for uniform media are usually called p and q respec

tively and are defined by

x <1

{ 3
2

p(x,O)

A(:e,y) = R(x,y)e'"lx. y ).

> I
x <0

Ozx<l

x>l

6p*-Oforx>Ocmly

p(0,t)~5

t>O

f),1./.

{)x

(12.7.7)

Ou
Oy'

(12.7.8)

As an approximation (which can be derived using perturbation methods), it can

be shown that the (slowly varying) wave numbers satisfy (12.7.5), corresponding to
the given temporal frequency associated with plane waves,

12.1
First-Order Nonlinear
Partial Differential Equations

12.7.1

+ ki)
(12.7 ..5)
For nearly plane \-"aves, we introduce the phase v(x, y) of the redueed wave
equation:

PIx, IJ) = 2 x>O

1

(12.7.3)

Again the temporal frequency W is fixed (and given) but c --.:. . c(J~, y) for inhomoge
neous media Or c = con:-:;tant for uniform media.
In 1lniform media (c = constant); plaue waves of the form .6' = Aoei(kp+k2y-wt)
or
A = Aoei(k1X-/-kZY)
2

= IJ

{)2A

-w'A = C2(ax2 + DyZ)'

w = c (kf

12.6.19. Solve the following problems:

( a.1 Qe
at + p2Qe
[-Ix

(12.7.2)

where A(.,y) satisfies the Helmholtz or reduced wave equation:

w 2 = c'(p'

+ q').

(12.7.9)

Derive Eikonal Equation from Wave Equation

For simplicity we cUllsider the two -dimensional wave equation

{)2

2 iJ2E

{) t' = c (ax2

This is a first order nonlinear partial difff'l'ntial equation (not quasi-linear) for the
phase u(x, V), known as the eikonal equation

iJ2E

+ iJ y2 ).

Plane waves and their reflections were analyzed in Section 4.6. Nearly p~ane waV'1

exist under many circumstances. If the t:oefficient c is not constant but varies slowl

w2 =

(~)2
+ (iJU)'
{)x
iJy

(12.7.10)

12.7. First-Urder ."IIonIinear PlJb:,

Chapter 12. Method of Character;stics

!j(i5

564

A first-order quasi-linear partial differential equation for q can be obtained by again

using 29= E2...
ax
By'

where 1.4-' is a fixed reference temporal frequency and c = C\T, y) for inhomogeneous
media or c = constant for uniform media. Sometimes the index of refraction
n(x, y) is introduced proportional to~. The amplitude R(x, y) solves equations

!
Ii

~
~
P-a
x + q"
uy =

(which we do not discuss) known as the transport equations, which describe the
propagation of energy of these nearly plane waves.

~;,

~I
I,

Thus,
result

II

III

dx
p

varies along this characteristic: du

(0 = constant),
!I!

(p'

au)' (au)'
w'
( ax + ay = c.,'

(12.7.11)

+ q')~

';? ~T,

p2 +q' =

-=

~,

dp

;=-q=O

1r - p' (from (12.7.12). Since (12.7.14) shows that p is constant along e

not cortlpletely straightforward.

We have differentiated the eikonal equation with respect to x.

differentiate with respect to y we obtain:

{)p

uq

p_+q_=O.
()y
ay
I:'
1111

= dq =~.2

_ dp
0

w' /c

(12.7.17)

+ q2

= _

c'

IkII

incoming wave and the phase of the retlected wave must be the same:

If there is a boundary condition for p, then (12.7.14) can be solved for p since:

characteristic, it abo fonows from (12.7.14) that each characteristic is a strai

line. In this way P can be determined. However, given p = ~, integrating for 1J,

+ 2JAdy
+ q2!!:JL
=
a y = pdx + qdy = p2cix
p
p

slowly varying reflected wave will always have the same wave length as the incident
wave). We assume the boundary condition on the curved boundary is that the t.otal
field is zero (other boundary conditions yield the same equations for the phase):
o = eiik,.X-wt) + R(x, t)e;~(x'Y)e-;w'. Thus, on the boundary the phase of the

Equation (12.7.13) may be solved by the method of characteristics [,ee (12.6.7)]

q=V

:r.

Can be interpreted as saying the slowly varying reflected wave number vector (P, q)
has the same length as the c.onstant incoming wave number vector (physically the

iJp
up

p-ur + q-uY = o.

dy

= dy

p2

p satisfies a first-order quasi-linear partial differential equation

dx

dB" dx

wish to determine the phase u(x, y) of the reflected wave. The eikonal equation

iJq

ax

p_+q- =0.
Since

Reflected waves. We consider an elementary incoming plane wave ei(kl.X~wt)

where k[ represents the given constant incoming wave numher vector and \vhere
w = clkII We assume the plane wave reflects off a curved boundary (as illus
trated in Figure 12.7.1) which we represent with a parameter T as x = XO(T) and
Y = YO(T). We intruuuce the unknown reflected wave, R(x , t)eiu(x'Y)e~iwt, and we

(12.7.12)

Differentiating (12.7.11) or (12.7.12) with respect to x yields

&p
ax

(12.7.16)

The Characteristics are straight lines since p and q are constants along the charac
teristics.

where wand c are constants. Rather than solve for u(x, y) directly, we will show
that it is easier to solve first for p = ~~ and q = %; . Thus, we consider

W'
?i.

where we have used (12.7.16) and (12.7.12). Thus, for the

dx

I
~I

eikonal equation

11

= dy = dp = d q

However 1 usually we want to determine u so that we wish to determine how u

The simplest example of the eikonal equation (12.7.10) occurs in uniform media

Ii

(12.7.15)

d; = 7 = ~ , which when combined with (12.7.14) yields the more general

12.7.2 Solving the Eikonal Equation in Uniform Media

and Reflected Waves

1\

O.

.'

u(xo, Yo) ~ k, "'0.

R

~
k

x = XO(T)
y=yo(r)

Figure 12.7.1: Reflected wave from curved boundary.

(127.18)

Chapter 12. Method of Characteristics

12.7. First-Order NonlineaJ' PDEs

567

566

Taking the partial derivative of (12.7..22) with respect to x, we obtain

Taking the derivative of (12.7.18) with respect to the parameter r shows that

~ dxo + au dyo = p dxo + qdyo = k R . dxo = k I

ax dr

0 y dr

dr

dT

dr

dxo

dr '

(12.7.19)

where we have noted that the vector (P, q) is the unknown reflected wave number
vector kR (because p and q are constant along the characteristic). Since dxo/ dr is

a vector tangent to the boundary, (12.7.19) shows that the tangential component of

the incoming and reflecting wave numbers must be the same. Since the magnitude

of the incident and reflecting wave number vectors are the same, it follows that the

normal component of the reflected wave must be minus the normal component of

the incident wave. Thus, the angle of reflection off a curved boundary is the same

as the angle of incidence. Thus at any point along the boundary the constant value

Fx

op

oq

+ Fup + Fp ax + Fq ox

= 0,

where we use the subscript notation for partial derivatives. For example F u =:
~~ keeping x,y,p,q con~tant. Since
= ~, we obtain a quasi-linear partial

differential equation fo1' p:

op
Fp ax

op
= -p, - Fup.
uy

+ Fq" ,

Thus, the method of characteristics for p yields

of p and q is known for the reflected wave. Because q = J;?- p2 there are two
solutions of the eikonal equation: oue represents the incoming wave and the other

dx = dy =
dp
Fp
Fq
-F, - Fup

(of interest to us) the reflected wave.

To obtain the phase of the reflected wave, we DlUst solve the characteristic
equations (12.7.17) for the eikonal equation with the boundary condition specified
by (12.7.18). Since for uniform media ~ = Ik I I,2 is a constant, the differential

Similarly, taking the partial derivative of (12.7.22) with respect to y, yields

~; = ;?~ =

equation for u along the characteristic,

'lk:

I' , can be integrated (since

ap
uy

oq

Fy + Puq + Fpc> + FqHere

f};

oy

(12.723)

O.

; yields a quasi-linear partial differential equation for q:

p il:i consta.nt) using the boundary condition to give

aq
F pox

u(x, y) = \kI1
_ ( x - XO)
p

+ kI . Xo,

along a specific characteristic. The equation for the characteristics ply - YO) =
q(x _ xa) corresponds to the angle of reflection equaling the angle of incidence.
Since p2 + q2 = \kI\2, the more pleasing representation of the phase (solution of the
eikonal equation) follows along a specific characteristic:
u(x, y) = p(x - xo)

+ q(y -

YO)

+ kI

. Xo,

(12.7.20)

where u(xo, YO) = kI . Xo is the phase of the incident wave on the boundary.

12.7.3

First-Order Nonlinear Partial Differential Equations'

oq

+ Fqay

= -Fy

,
ruq.

The characteristic direction is the same as in (12.7.23), so that (12.7.23) is amended

to become
dl:
Fp

dy
Fq

dp
-F" - Fup

-Fy

dq.
- Fuq

(12.7.24)

In order to solve for u, we want to derive a differential equation for u(x, y) along
the characteristics:
du

ou
7>:'dx
(.IX

au

dx

dy

dx

+ ",dy
= pdx + qdy = pFp'F + qFq-F, = (pFp + qFq)'F
uy
p
q
p

The complete system to solve for p, q, and u is

Any first-order nonlinear partial differential equation can be put in the form

ou ou)=o.
F ( x,y,u, ox' oy

As with the eikonal equation example of the previous subsection, we show

p = ~; and q = g~ solve quasi-linear partial differential equations, and hen'
(12.7.21) can be solved by the method of characteristics. Using p and q gives
F(x, y, u,p, q) = O.

dx
Fp

dy
Fq

dp
dq
-F" - Fup = -Fy -

du
~!q

pFp

+ qFq '

(12.7.25)