2017 Nonlinear Partial Differential Equations

September 26, 2017 11:52 A First Course in Partial Differential Equations 9in x 6in b3044-ch11 page 461
Chapter 11
by 117.98.126.28 on 05/27/24. Re-use and distribution is strictly not permitted, except for Open Access articles.
Nonlinear Partial Differential

Equations
A First Course in Partial Differential Equations Downloaded from www.worldscientific.com
So far this textbook has concentrated on linear partial differential equations

except in Chap. 2 where the solutions to first order, quasilinear partial dif-
ferential equations were found using the method of characteristics. In the
real world, many physical phenomena are the result of complex interactions
of many physical processes such as convection, diffusion, and reaction. To
model such interactions, nonlinearity is unavoidable, and as a result, non-
linear partial differential equations arise. The subject of nonlinear partial
differential equations is much more complicated and challenging than its
linear counterpart. As seen in earlier chapters, the Principle of Superpo-
sition of solutions and the method of separation of variables play crucial
roles in the process of finding solutions of boundary value problems for
linear partial differential equations. However, these fundamental principles
and methods are no longer applicable to most nonlinear partial differential
equations. In fact, there is no general method that can be used to solve a
large class of nonlinear partial differential equations. As Lawrence C. Evans
[Lindenstrauss et al. (1994)] once put it:
Keep in mind that there is in truth no central core theory of nonlinear
PDE, nor can there be. The sources of PDE are so many — physical,
probabilistic, geometric, etc. — that the subject is a confederation of di-
verse subareas, each studying different phenomena for different nonlinear
PDE by utterly different methods.
The lack of a general core theory makes the study of nonlinear partial
differential equations much less systematic and much more challenging. Al-
most all the methods or techniques for nonlinear differential equations are
on an ad hoc basis: different nonlinear partial differential equations often
call for different methods. Furthermore, the method developed for studying
a particular nonlinear partial differential equation is usually closely related
461
462 A First Course in Partial Differential Equations
to the physical application from which the equation arises. The physical
application often motivates the expectation of the existence of certain types
of solutions and the specific properties of such solutions. For example, many
nonlinear wave equations originated from wave or signal propagation and, as
a result, have traveling wave solutions, which will be introduced in Sec. 11.1.
In addition to finding special types of solutions to nonlinear partial dif-

ferential equations, there are methods that apply to certain types of non-
linear partial differential equations. For example, through an appropriate

change of variable, Burgers’1 equation can be transformed to the familiar
homogeneous heat equation. This transformation is detailed in Sec. 11.2.
The goal of this chapter is to introduce readers to a few important
nonlinear partial differential equations and several important techniques in
their study. No attempt will be made for a comprehensive discussion or for
going into depth in the subject of nonlinear partial differential equations. In
Sec. 11.1, some introductory concepts related to traveling wave and solitary
wave solutions are described. Section 11.2 is devoted to the discussion of
Burgers’ equation focusing on the existence of traveling wave solutions and
the Cole2 -Hopf3 transformation which transforms the Burgers’ equation
to a linear homogenous heat equation. This is followed by a discussion
of the existence and properties of soliton solutions to the Korteweg4-de
Vries5 (KdV) and nonlinear Schrödinger equations in Sec. 11.3 and Sec. 11.4
respectively. The chapter ends with some suggestions and comments for
further study and a set of exercises for understanding and extending the
discussions begun in this chapter.
11.1 Waves and Traveling Waves
Traveling wave solutions of nonlinear partial differential equations have

drawn much attention and are one of the main topics of this chapter. In
this context a wave refers to any disturbance or signal that travels through
a medium over time, carrying energy with it as it propagates. It could
refer to a water wave, a sound wave, or an electromagnetic wave, even a
signal passing through a fiber optical cable. In the physical world, wave
propagation is a commonly observed phenomenon. In general, if u(x, t)
represents the disturbance or signal strength at the location x at time t, it
1 Johannes Martinus Burgers, Dutch physicist (1895–1981).
2 JulianD. Cole, American mathematician (1925–1999).
3 Eberhard Hopf, German-American mathematician and astronomer (1902–1983).
4 Diederik Johannes Korteweg, Dutch mathematician (1848–1941).
5 Gustav de Vries, Dutch mathematician (1866–1934).
Nonlinear Partial Differential Equations 463
is a traveling wave if u has the form,

u(x, t) = U (x − c t) (11.1)
for some single-variable function U and a fixed real constant c. For a wave
u(x, t) given by Eq. (11.1), the initial wave or profile u(x, 0) = U (x) is pro-
pagated at speed |c|. For many nonlinear partial differential equations,

bounded traveling wave solutions are of chief interest.
The reader may recall that the idea of traveling wave solutions occurred
in the discussion of solutions to the one-dimensional wave equation found
using d’Alembert’s approach. In Chap. 5, the solution of the standard wave
equation on the real number line with the initial conditions
u(x, 0) = f (x) and ut (x, 0) = 0,
(the plucked string) is given by
1
u(x, t) = (f (x − c t) + f (x + c t)).
2
In this case, u(x, t) is the sum of two simple traveling waves, that is, the
initial disturbance f (x) of the string is propagated to the right and left, re-
spectively, with speed |c|, each at half of the magnitude. Therefore, strictly
speaking, u(x, t) is not a traveling wave itself, rather it is the sum of two
simple traveling waves.
In general, to find the traveling wave solution to a partial differential
equation, the solution u(x, t) is assumed to have the form given by Eq. (11.1)
and substituted into the partial differential equation to identify the constant
c and the function U . The function U will satisfy an ordinary differential
equation and, possibly, some boundary conditions. The following example
demonstrates this basic idea.
Example 11.1. Find all bounded, traveling wave solutions of the (linear)
partial differential equation,
ut − uxxx = 0. (11.2)
Solution. Let c be any constant and assume that u(x, t) = U (x − c t) is a
solution of Eq. (11.2). Setting ξ = x − c t and substituting u(x, t) = U (ξ)
in the equation yield
−c U (ξ) − U (ξ) = 0.
Integrating the above equation once produces
U (ξ) + c U (ξ) = A, (11.3)
where A is an arbitrary constant of integration. If c < 0, the general

solution of Eq. (11.3) is
√ √ A
U (ξ) = A1 e −c ξ + A2 e− −c ξ +
c
where A1 and A2 are arbitrary constants. Unless A1 and A2 are both zero,
U (ξ) is unbounded. If A1 = A2 = 0, then solution U (ξ) = A/c is a constant

which solves the original partial differential equation. If c = 0, Eq. (11.3)
produces no bounded solution other than constant solutions. The last case
to consider is that of c > 0. In this case the general solution of Eq. (11.3)
is
√ √ A
U (ξ) = A1 cos( c ξ) + A2 sin( c ξ) +
c
where A1 and A2 are again arbitrary constants. The function U (ξ) is
bounded and leads to the following bounded traveling wave solution of
Eq. (11.2),
√ √ A
u(x − c t) = A1 cos( c(x − c t)) + A2 sin( c(x − c t)) + (11.4)
c
for any choices of constants A1 , A2 , and A.
Equation (11.2) is a linear, dissipative partial differential equation. The
nonconstant, bounded traveling wave solutions occur only for c > 0. This
implies that all traveling wave solutions travel to the right only. Further-
more, disregarding the constant A/c, all nonconstant traveling wave solu-
tions are linear combinations of
√ √
u1 (x, t) = cos( c(x − c t)) and u2 (x, t) = sin( c(x − c t)).
√
Setting k = c and ω = k 3 , these two solutions can be written as
u1 (x, t) = cos(kx − ωt) and u2 (x, t) = sin(kx − ωt).
Such traveling wave solutions are often called one-dimensional plane wave
solutions and are often conveniently represented in the form of complex
exponentials as u(x, t) = ei(kx−ωt) and its complex conjugate. The constant
k is called the wave number and ω the angular frequency. The equality
ω = k 3 describes the relationship between the angular frequency and the
wave number and is often referred to as the dispersion relation for the
partial differential equation.
As another simple example, consider the partial differential equation
ψt = i ψxx (11.5)
Equation (11.5) is often referred to as the free Schrödinger equation for free
particles. Free particles are particles which are not bound by any external
force or equivalently exist in a region of space where the potential energy

is constant (or more specifically zero). If ψ(x, t) = ei(kx−ωt) is a solution
to Eq. (11.5), the following dispersion relation holds,
−i ω = −i k 2 ⇐⇒ ω = k 2 .
The traveling wave solutions found in the examples above are periodic.
Another class of traveling wave solution, of particular importance for non-

linear partial differential equations, is the solitary wave or soliton, solu-
tion which is a special type of traveling wave solution. In general, a solitary

wave solution has two important features. First, it must have a permanent
form, meaning that its profile will not change as it propagates with time.
Next it is localized, meaning that it decays to 0 or approaches a constant
at infinity. A soliton solution, or simply a soliton, is a solitary wave solu-
tion that is stable in the sense that it maintains its wave profile even after
colliding with another soliton. Therefore, strictly speaking, solitons and
solitary wave solutions are different concepts though some authors choose
not to distinguish between them. The exact definition or classification is
not of importance for the discussion here. Readers interested in the details
of these concepts may see [Newell (1985)] or [Drazin and Johnson (1989)]
for more information.
Traveling wave solutions do not exist for every nonlinear partial dif-
ferential equation. For the equations for which traveling wave solutions
do exist, only certain initial profiles (initial conditions) lead to traveling
wave solutions. Of course, similar remarks hold for solitary wave solu-
tions. However, traveling (solitary) wave solutions are of importance since
they often reveal the interaction between different physical processes, es-
pecially between nonlinearity and the effects of reaction and/or dispersion.
Note that when seeking a traveling wave solution, the wave speed c must
be determined so that the function defined by Eq. (11.1) is a solution to
the underlying partial differential equation. When the function defined by
Eq. (11.1) is substituted into the partial differential equation, an ordinary
differential equation is produced with, possibly, some specified behavior at
infinity. Therefore, in many cases, to find the traveling wave solutions of
a partial differential equation is essentially a problem of finding the solu-
tion or solutions to a boundary value problem of an ordinary differential
equation.
The existence and properties of travelling wave solutions for a few non-
linear partial differential equations are discussed in later sections. The
discussion will be restricted to one spatial dimension only, although many
concepts can be adapted to higher-dimensional cases.
11.2 Burgers’ Equation
The discussion of nonlinear partial differential equations starts with the

well-known Burgers’ equation. This is one of the simplest nonlinear par-
tial differential equations. In Sec. 2.3 a simple traffic flow model for traffic
on a straight stretch of highway was developed. The discussion there leads

naturally to Burgers’ equation. Recall that ρ(x, t) was used to denote the
vehicle density at position x at time t and q(x, t) the traffic flux, the num-
ber of vehicles passing location x per unit time at time t. The conservation
of vehicles leads to Eq. (2.41) (repeated here for convenience),
∂ρ ∂q
(x, t) + (x, t) = 0.
∂t ∂x
The vehicle velocity is assumed to be a linear function of the traffic density.
This leads to the assumption that the traffic flux is proportional to the
vehicle density and produced the model in Eq. (2.43). In reality, the traffic
flux can be related to the vehicle density in a more complicated fashion
or can even depend on other factors. For example, the flux q(x, t) may
depend on the rate of change of ρ with respect to x. Suppose q(x, t) =
Q(ρ(x, t)) − µρx for some function Q(ρ) and constant µ > 0. If Q(ρ) is
taken to be the simple quadratic function Q(ρ) = umax ρ2 /ρmax , then the
traffic flux is given by
umax 2 ∂ρ
q ≡ q(ρ) = ρ −µ
ρmax ∂x
where umax is the maximum speed limit of vehicles (for example, the posted
speed limit) and ρmax is the maximum vehicle density. Under these assump-
tions, the traffic density ρ(x, t) can be modeled by the following nonlinear
partial differential equation,
∂ρ umax ∂ρ ∂2ρ
+2 ρ − µ 2 = 0. (11.6)
∂t ρmax ∂x ∂x
By setting ξ = ρmax x/(2umax ) and u(ξ, t) = ρ(x, t) and rewriting ξ as x
thereafter, Eq. (11.6) becomes the following
∂u ∂u ∂2u
+u − ν 2 = 0, (11.7)
∂t ∂x ∂x
where the diffusion coefficient ν > 0 is a constant. This equation is com-
monly referred to as the standard form of Burgers’ equation, which is non-
linear due to the presence of the term u ∂u/∂x.
Although Burgers’ equation is derived from the traffic flow model, its
applications go far beyond that context. Historically Burgers’ equation
was first proposed to model the motion of turbulent flow [Burgers (1948)].
In fact, Burgers’ equation is commonly considered a simplified version of
the Navier6 -Stokes7 equations, the fundamental equations of viscous fluid
dynamics. In the context of fluid flow, the nonlinear term u ux models a
convection process and the term ν uxx models the effect of viscosity or the
diffusion process. For this reason the constant ν is often referred to as the
kinetic viscosity or simply, viscosity. Some authors refer to Eq. (11.7) as
the viscous form of Burgers’ equation. Burgers’ equation arises from many
other physical situations as well, such as the propagation of sound waves
in a viscous medium and the study of magnetohydrodynamics. For more
details, see [Hamilton and Blackstock (1998)] and [Olesen (2003)]. If the
viscosity ν = 0, Burgers’ equation becomes one of the simplest nonlinear
conservation laws (often referred to as the inviscid Burgers’ equation).
Mathematically, Burgers’ equation describes the balance between the
effects of nonlinearity and diffusion. If u and/or ux are so small that the
nonlinear term of Eq. (11.7) can be ignored, the solutions of Burgers’ equa-
tion should be similar to those of the heat equation. On the other hand, if
|u| and/or |ux | are large, the nonlinear term plays a more dominant role in
the equation, and the balance of the nonlinear term and the diffusion term
results in nonlinear waves — traveling wave solutions. It is also worthwhile
to point out that Burgers’ equation is parabolic if ν > 0 and it is hyperbolic
if ν = 0. If ν > 0, under relatively mild conditions on the initial data, the
solution of Burgers’ equation is smooth and is defined for all t ≥ 0. How-
ever, if ν = 0, the solution is not smooth and a shock wave can develop as
indicated in the discussion of the traffic flow model in Sec. 2.3 and later in
this section.
Before solving Eq. (11.7) a brief discussion of conservation laws will be
presented. Burgers’ equation is one example of a conservation law.
11.2.1 Conservation Laws and Burgers’ Equation

A careful approach to the solution of the inviscid Burgers’ equation will
make use of the assumption that the viscous Burgers’ equation can be
interpreted as a mathematical model of fluid flow. As such, physical quan-
tities like mass should be conserved. In fact Burgers’ equation (and other
similar mathematical models) are often categorized as conservation laws.
In some cases the solution to Burgers’ equation is not unique or is multi-
6 Claude-Louis Navier, French engineer and physicist (1785–1836).
7 George Gabriel Stokes, Irish mathematician (1819–1903).
valued. Once the quantity that must be conserved in Burgers’ equation is

determined, that quantity can be used to determine which of the multiple
values (if any) of u is the “correct” solution to Burgers’ equation. This will
be important later in Sec. 11.2.4 when the solution to the inviscid Burgers’
equation admits a shock.
To determine the conserved quantity, integrate the viscous Burgers’

equation in Eq. (11.7) with respect to x over an arbitrary interval [a, b].
b b b
ut dx + u ux dx − ν uxx dx = 0
a a a
b
1 1
ut dx + (u(b, t))2 − (u(a, t))2 − νux (b, t) + νux (a, t) = 0. (11.8)
a 2 2
Note that by Leibniz’s rule,
b
d b db da
u(x, t) dx = ut dx + u(b, t) − u(a, t) . (11.9)
dt a a dt dt
Using Eq. (11.8) to make a replacement in Eq. (11.9) produces

d b db 1 2
u(x, t) dx = u(b, t) − (u(b, t)) + νux (b, t)
dt a dt 2

da 1 2
− u(a, t) − (u(a, t)) + νux (a, t) . (11.10)
dt 2
b
Hence the rate of change of a u(x, t) dx is the quantity,
x=b
dx 1
u(x, t) − (u(x, t))2 + ν ux (x, t) . (11.11)
dt 2 x=a
Since the rate of change of the integral of u over [a, b] is dependent only
on the values at the endpoints of the integral (namely a and b) then the
integral of u is said to be conserved.
In the following sections, attention will be focused on the existence and
properties of traveling wave solutions of the viscous Burgers’ equation and
the Cole-Hopf transformation technique for determining solutions to the
viscous Burgers’ equation before turning to the inviscid Burgers’ equation.
11.2.2 Traveling Wave Solutions of Burgers’ Equation

As mentioned earlier, a traveling wave solution is a solution of the form
U (x − c t) for some constant c. Setting ξ = x − c t and substituting
u(x, t) = U (ξ) into Eq. (11.7) produce the equation,
dU dU d2 U
−c +u −ν = 0.
dξ dξ dξ 2
Integrating both sides of the equation with respect to ξ yields

1 dU
−c U + U 2 − ν = A, (11.12)
2 dξ
where A is a constant. Therefore,
dU 1
= (U 2 − 2c U − 2A). (11.13)
dξ 2ν
Equation (11.13) can be solved using the separation of variables technique
for first-order ordinary differential equations. The solutions of interest in

this case are the bounded solutions. It can be shown that there is no
bounded solution of Eq. (11.13) if the quadratic equation U 2 −2c U −2A = 0
has only one real root (other than the constant solution) or two complex
roots, see Exercise 6. Assume that the equation U 2 − 2c U − 2A = 0 has
two distinct real roots, that is, the inequality
√ c2 + 2A > 0 is satisfied,
√ and
2
denote the two real roots as U1 = c − c + 2A and U2 = c + c + 2A. 2
The constant functions U (ξ) = U1 and U (ξ) = U2 are equilibrium solutions

to Eq. (11.13). If U (ξ) is a solution to Eq. (11.13) with initial value between
U1 and U2 , then by the uniqueness of solutions to the ordinary differential
equation, U (ξ) will be bounded and stay between U1 and U2 . Separating
the variables in Eq. (11.13) and rewriting the equation as
1 dU 1
=
(U − U1 )(U − U2 ) dξ 2ν
allow both sides to be integrated with respect to ξ. This results in

U − U1 (U1 − U2 )
ln = ξ + B,
U − U2 2ν
where B is an arbitrary constant of integration. Therefore using the fact
that U1 < U (ξ) < U2 ,
U − U1
= eαξ+B ,
U2 − U
where α = (U1 − U2 )/(2ν) < 0. Solving this equation for U yields
U1 + U2 eαξ+B
U (ξ) = . (11.14)
1 + eαξ+B
As expected, the solution U satisfies
lim U (ξ) = U2 and lim U (ξ) = U1 .
ξ→−∞ ξ→∞
This leads to the following traveling wave solution for Burgers’ equation
U1 + U2 eα(x−c t)+B
u(x, t) = U (x − c t) = , (11.15)
1 + eα(x−c t)+B
which satisfies the following limits for each fixed t,

lim u(x, t) = U2 and lim u(x, t) = U1 .
x→−∞ x→∞
Note that if B = 0, the corresponding traveling wave solution becomes

U1 + U2 eα(x−c t)
u(x, t) = U (x − c t) =
1 + eα(x−c t)
and the initial wave profile is obtained by setting t = 0,
U1 + U2 eαx
u(x, 0) = U (x) = .
1 + eαx
Only when the initial condition is given by the function above or one of
its horizontal translations does this lead to a traveling wave solution. The
traveling wave solutions have the following properties:
(1) For any fixed t, a traveling wave solution decreases from U2 to U1 as x
increases from −∞ to ∞. Note that U1 and U2 are equilibrium solutions
of Eq. (11.13). In the language of dynamical systems, the solution U (ξ)
given by Eq. (11.14) is a homoclinic orbit of the ordinary differential
equation in Eq. (11.13) [Guckenheimer and Holmes (1983)].
(2) The traveling waves travel from the left to right with a speed c =
(U1 + U2 )/2 since U1 and U2 are solutions of the quadratic equation
U 2 − 2c U − 2A = (U − U1 )(U − U2 ) = 0, see Exercise 8.
(3) As the viscosity ν approaches 0, for fixed t and x, if x < c t, the solution
u(x, t) → U2 and if x > c t, u(x, t) → U1 . In the limit, a shock wave
develops.
Figure 11.1 illustrates the traveling wave solution to Burgers’ equation. In
the upper plot the viscosity ν = 1 and the surface plot appears smooth. In
the lower plot the viscosity has been decreased to ν = 1/4 and the surface
appears to drop more abruptly.
11.2.3 The Cole-Hopf Transformation

The existence of traveling wave solutions of Burgers’ equation was estab-
lished in the last section. A natural question to ask then, is whether it is
possible to find all solutions to Burgers’ equation, not merely the travel-
ing wave solutions. More specifically, motivated by the analysis of the heat
equation and wave equation, does Burgers’ equation have a solution for any
given initial condition,
u(x, 0) = f (x) for −∞ < x < ∞. (11.16)

Fig. 11.1 Surface plots of the traveling wave solutions to Burgers’ equation (Eq. (11.7)).
In the upper plot the viscosity parameter ν = 1. When the viscosity is lowered (to
ν = 1/4, as in the lower plot) the solution drops more abruptly. In the limit as ν → 0 it
may be expected that the solution develops a discontinuity (called a shock).
If there exists a solution to the initial value problem, is the solution unique
and is there a way to find the solution explicitly? This section attempts
to partially answer these questions. As a matter of fact, Burgers’ equation
can be transformed to a homogeneous heat equation through a change of
variable, called the Cole-Hopf transformation. To introduce the trans-
formation, Burgers’ equation is rewritten as

∂u ∂ 1 2 ∂u
+ u −ν = 0 for −∞ < x < ∞ and t > 0. (11.17)
∂t ∂x 2 ∂x
The Cole-Hopf transformation is defined by
vx ∂
u = −2ν ⇐⇒ u = −2ν [ln v] , (11.18)
v ∂x
where it is implicitly assumed that the unknown function v(x, t) > 0. The
chain rule implies
∂u v vxt − vx vt v vtx − vt vx ∂ vt
= −2ν = −2ν = −2ν ,
∂t v2 v2 ∂x v
∂u v vxx − (vx )2
= −2ν ,
∂x v2
and therefore in Eq. (11.17) the term u2 /2 − ν ∂u/∂x can be written as
(vx )2 2 v vxx − (vx )
2
vxx
2ν 2 2
+ 2ν 2
= 2ν 2 ,
v v v
and as a result Burgers’ equation can be written in the form
∂ vt ∂ 2 vxx
−2ν + 2ν = 0,
∂x v ∂x v
which simplifies to the following:

∂ vt − ν vxx
= 0. (11.19)
∂x v
Clearly, if a function v(x, t) > 0 is a solution the linear, homogeneous heat
equation
vt = ν vxx , (11.20)
the numerator of the term inside the partial derivative in Eq. (11.19) will
be identically zero and thus v(x, t) is a solution to Eq. (11.19). Hence the
function u(x, t) defined by Eq. (11.18) is a solution to Burgers’ equation. On
the other hand, if u(x, t) is any solution to Eq. (11.7), then by integrating
Eq. (11.18) and solving for v, the function
1
x
v(x, t) = e− 2ν 0
u(y,t) dy
(11.21)
is a solution of Eq. (11.19). Of course, the main interest is to obtain solu-

tions of Burgers’ equation from solutions to the heat equation. Since many
solutions of the heat equation are known, by using the Cole-Hopf transfor-
mation, many solutions of Burgers’ equation can be found. For example,
since v(x, t) = A + Be−νt sin x is a solution of the heat equation for any
constants A and B, if A is large enough to guarantee that v(x, t) is positive

everywhere,
−2νBe−νt cos x
u(x, t) =
A + Be−νt sin x
is a solution to Burgers’s equation.
Next consider the initial value problem of Burgers’ equation. Assume
that Eq. (11.7) is given an initial condition defined by Eq. (11.16), then the
function v will have a corresponding initial condition given by Eq. (11.21),
1
x
v(x, 0) = g(x) = e− 2ν 0
f (y) dy
. (11.22)
By Theorem 4.3, if g(x) satisfies the condition specified there, the solution
to Eq. (11.20) is
∞
1 (x−y)2
v(x, t) = √ g(y)e− 4νt dy for t > 0. (11.23)
4πνt −∞
This implies
∞
−1 x−y (x−y)2
vx (x, t) = √ g(y) e− 4νt dy for t > 0. (11.24)
4πνt −∞ 2νt
Substituting Eqs. (11.23) and (11.24) in Eq. (11.18) yields a solution of the
initial value problem of the Burgers’ equation:
∞ − (x−y)
2
2ν −∞ g(y) x−y2νt e
4νt dy
u(x, t) = ∞ (x−y)2
− 4νt dy
−∞ g(y)e
∞ 1
− 2ν

( 0y f (s) ds+ (x−y)
2
)
(x − y)e 2t dy
= −∞ 2 . (11.25)
∞ 1 y (x−y)
t −∞ e− 2ν ( 0 f (s) ds+ 2t ) dy
According to Theorem 4.5 the solution to the initial value problem consist-
ing of Eqs. (11.7) and (11.16) is unique provided f is piecewise continuous
and bounded.
Example 11.2. Consider the initial value problem consisting of the viscous
Burgers’ equation and the initial condition:
ut + u ux − ν uxx = 0,
0 if x < 0,
u(x, 0) =
1 if x ≥ 0.
Find a solution to this initial value problem using the Cole-Hopf transfor-
mation.
Solution. According the formula given in Eq. (11.22),
1 if x < 0,
g(x) =
e−x/(2ν) if x ≥ 0.
Using this in Eq. (11.25), the solution to the initial value problem is
∞ − (x−y)
2
−∞ g(y)(x − y)e dy
4νt
u(x, t) = ∞ (x−y)2
t −∞ g(y)e− 4νt dy
0 (x−y)2
− 4νt
∞ y (x−y)2
−∞ (x − y)e dy + 0 (x − y)e− 2ν − 4νt dy

= 0 (x−y)2 ∞ y (x−y)2
t −∞ e− 4νt dy + t 0 e− 2ν − 4νt dy
∞ 2 t−2x ∞ √ 2
4νt √x ze−z dz + te 4ν √t−x ( 4νt − 4νz)e−z dz
4νt 4νt
= √ ∞ t−2x √ ∞
t 4νt √x e−z2 dz + te 4ν 4νt √t−x e−z2 dz
4νt 4νt
t−2x

t−x
e 4ν 1 − erf √4νt
= t−2x
e 4ν 1 − erf √t−x 4νt
+ 1 − erf √4νtx
for t > 0. In order to evaluate the integrals involved in the solution,√the

technique of completing√ the square and the substitutions z = (x − y)/ 4νt
and z = (y + t − x)/ 4νt were used. The error function introduced in
Eq. (4.43) is used to express the values of the integrals. Figure 11.2 illus-
trates the solution to this example.
11.2.4 Inviscid Burgers’ Equation and Shock Waves

If the viscosity ν = 0, then Burgers’ equation (Eq. (11.7)) becomes
∂u ∂u
+u = 0, (11.26)
∂t ∂x
known as the inviscid Burgers’ equation. This equation is a first order
partial differential equation of hyperbolic type as opposed to the viscous
Burgers’ equation which is parabolic. As a result, the solutions of the
inviscid Burgers’ equation have dramatically different properties.
The dependence on ν is apparent in the traveling wave solution to
Eq. (11.15) for the viscous Burgers’ equation. As ν approaches 0, the
traveling wave solutions becomes steeper for x near the value c t as can be
seen in the surface plots of the solutions in Fig. 11.1. In the following, the

Fig. 11.2 Surface plot of the solution to the viscous Burgers’ equation with unit step
function as the initial condition. For the purposes of plotting the viscosity parameter is
ν = 1. Note that as t increases the surface becomes smoother.
traveling wave solution given by Eq. (11.15) will be denoted as uν (x, t). To
solve the inviscid Burgers’ equation in general requires the method of char-
acteristics described in Sec. 2.2. The characteristic system for Eq. (11.26)
can be expressed as:
dx dt du
= u, = 1, and = 0.
ds ds ds
The characteristics are parameterized by s. The third characteristic equa-
tion indicates that u is constant along characteristics. The second charac-
teristic equation allows the parameter s to be identified with the indepen-
dent variable t. Hence the first characteristic equation may be rewritten
as dx/dt = u and thus the solution to the characteristic system may be
expressed as
x = u t + C1 ,
u = C2 ,
where C1 and C2 are arbitrary constants which may be determined from
initial conditions imposed on the inviscid Burgers’ equation. If the initial
condition is given as u(x, 0) = f (x), then along the characteristic through

(x0 , 0) it is the case that C1 = x0 and thus x0 = x−u t on the characteristic.
Therefore the solution can be written in implicit form as
u(x, t) = f (x0 ) = f (x − u(x, t)t). (11.27)
2
For example if u(x, 0) = e−x , then by Exercise 14 of Chap. 2, the implicit
form of the solution is
2
u(x, t) = e−(x−u(x,t)t) .
However, there is a complication in determining the solution. Since the
characteristics are straight lines with slopes that depend on u, unless u is
a constant, these characteristics may intersect for some t > 0. If the t-axis
t
2.0
1.5
1.0
0.5
0.0 x
0.0 0.5 1.0 1.5 2.0
Fig. 11.3 The intersection of characteristics for the inviscid Burgers’ equation (with
2
initial condition u(x, 0) = e−x ) indicate that a more careful approach to the solution
must be followed in the vicinity of the intersections. The thick curve is the locus of
points where characteristics intersect and is called a shock. The shock is the location of
a discontinuity in the solution to the inviscid Burgers’ equation.
is oriented vertically and the x-axis is oriented horizontally as in Fig. 11.3,

the slope of the characteristics is 1/u(x, 0), thus if u(x, 0) is a decreasing
function of x in some interval, the slopes of the characteristics will increase,

leading to the intersection of the characteristics at some t > 0. Figure 11.3
illustrates characteristics which intersect at some time t, with 1 ≤ t ≤ 2.
At the location and time of intersection of two characteristics with different
values of u, a phenomenon known as a shock occurs. If the inviscid Burgers’
equation is thought of as a model of fluid flow, then the downstream fluid

is moving slower than the upstream fluid, or equivalently, the upstream
fluid is overtaking the downstream fluid. Since the dependent variable u is

constant along each characteristic, the intersection of characteristics implies
u is multivalued, which means the solution is not well-defined. Figure 11.4
illustrates a typical multivalued situation for the solution to the inviscid
Burgers’ equation.
xs
1.0 u3
0.8
0.6
u2
u
0.4
0.2
u1
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
x
Fig. 11.4 When characteristics of the inviscid Burgers’ equation intersect, the contin-
uous solution to the equation become undefined due to it being multivalued. Hence
the solution in the region where characteristics intersect must be discontinuous. In this
2
figure t = 2 and the initial condition for the inviscid Burgers’ equation is u(x, 0) = e−x .
The shock will form at the earliest time the surface u(x, t) has a vertical
tangent. Using implicit differentiation on the general solution found in
Eq. (11.27) to the inviscid Burgers’ equation, the slope of the solution in
the x-direction is
f (z)
ux = .
1 + t f (z)
A vertical tangent occurs when ux is undefined, or equivalently when t =
−1/f (z). Thus the earliest time this occurs is the time for which f (z) takes
on its negative minimum. Thereader will be asked to show in Exercise 11

2
that the shock forms at t0 = e/2 when f (x) = e−x . The x-coordinate
at which the shock forms is found by solving the equation 1 + t0 f (x0 ) = 0
which makes ux undefined and gives the coordinate on the t = 0 line where
the characteristic passing through the beginning of the shock passes. When
2
f (x) = e−x , solving the equation,

e 2 1
0=1−2 x0 e−x0 =⇒ x0 = √ .
2 2
Hence the shock begins at the point with coordinates:

√ e
(x, t) = (x0 + t0 f (x0 ), t0 ) = 2,
2
in the specific example explored here. Figure 11.5 depicts the solution to
2
the inviscid Burgers’ equation with initial condition u(x, 0) = e−x . Notice
Fig. 11.5 A plot of the solution surface to the inviscid Burgers’ equation with initial
2
condition u(x, 0) = e−x . A shock forms at t = e/2 and to avoid becoming multivalued
the solution surface must develop a discontinuity across the shock.

the discontinuity across the shock, after the shock forms at t = e/2. The
type of solution presented in Fig. 11.5 is known as an entropy solution.
The term “entropy” comes from the field of thermodynamics and implies
that solutions may be lost along shocks.
Suppose the shock is located at position xs (the s-subscript denoting
shock) at time t. The integral of u is conserved even when passing through

the shock and since the shock has infinitesimal width, the definite integral
must be zero. This observation yields a method for determining the value of
the discontinuous solution to the inviscid Burgers’ equation along a shock.
Note that in Fig. 11.4 the function u(x, t) is multivalued for an interval on
the x-axis. The correct value of u can be found by placing the shock (the
vertical line labeled xs in Fig. 11.4) at such as position that the shaded area
to the right of xs and left of the curve u(x, t) = f (x − u(x, t)t) equals the
shaded area to the left of xs and right of the curve. Since one region is on
either side of the vertical line and the regions are of equal area, their sum
is 0. The value of u (labeled u2 in Fig. 11.4) where xs intersects the curve
u(x, t) = f (x − u(x, t)t) and which separates the left and right regions is
then the “correct” value of u(x, t).
More features of the shock can be determined as well. The position of
the shock is a function of time. Thus by choosing a < xs and b > xs in
Eq. (11.10) and setting ν = 0, in the limit as a → x− +
s and b → xs ,
dxs 1 dxs 1
u(xs +, t) − (u(xs +, t))2 = u(xs −, t) − (u(xs −, t))2
dt 2 dt 2
dxs 1
= (u(xs +, t) + u(xs −, t))
dt 2
where u(xs +, t) and u(xs −, t) denote the values of the solution. There-
fore the shock moves at a velocity which is the average of the values of u
immediately on either side of the shock.
This section concludes with an examination of the relationship between
the solution to the viscous Burgers’ equation, the formula for which was
given in Eq. (11.25), and the solution to the inviscid Burgers’ equation.
More details relevant to this discussion and more examples may be found
in [Whitham (1974)]. To see the connection between the entropy solution
to Eq. (11.26) and the solution to Eq. (11.7) as ν → 0+ , let uν (x, t) be the
solution to Eq. (11.7) and let u(x, t) be the entropy solution to Eq. (11.26)
with the same initial conditions, u(x, 0) = f (x). During this discussion
(x, t) will remain fixed. The solution to the viscous Burgers’ equation given
by Eq. (11.25) is single-valued and continuous for all t while the solution
to Eq. (11.26) may be discontinuous and contain shocks depending on the
initial condition f (x). The limit ν → 0+ should produce the same shock
structure and values as the viscous solution.
The solution uν (x, t) is given by Eq. (11.25). To study the limit as
ν → 0+ , consider the dominant contributions to the improper integrals
in the numerator and denominator of Eq. (11.25). Since these integrals
converge, the presence of the term,

y
1
f (s) ds+(x−y)2 /(2t))
e− 2ν ( 0
in both integrals suggests the dominant contributions occur in intervals

y
where the function h(y) ≡ 0 f (s) ds + (x − y)2 /(2t) reaches a minimum.
For a fixed x, h(y) reaches a minimum when
x−y
h (y) = f (y) − = 0, (11.28)
t
and h (y) = f (y) + 1/t > 0. For fixed (x, t), let ξ be a solution to
Eq. (11.28) and call ξ a stationary point. The contribution to the im-
proper integral from a neighborhood of the stationary point can be found
by Laplace’s method (sometimes called the method of steepest descent).
Using Taylor’s8 theorem to expand about ξ implies

1 1
h(y) ≈ h(ξ) + f (ξ) + (y − ξ)2 ,
2 t
where the truncation error is O((y − ξ)3 ). Thus the contribution to the
improper integral from a neighborhood of a stationary point is
∞ ∞
1 1 2
e −h(y)/(2ν)
dy ≈ e −h(ξ)/(2ν)
e− 2ν (f (ξ)+ t )(y−ξ) /2 dy
−∞ −∞

4πνt 4πν −h(ξ)/(2ν)
=
e−h(ξ)/(2ν) = e .
1 + tf (ξ) h (ξ)
Likewise, in a neighborhood of a stationary point,

∞
x − y −h(y)/(2ν) x−ξ 4πν −h(ξ)/(2ν)
e dy ≈ (ξ)
e .
−∞ t t h
If there is only one stationary point then substituting these two approxi-
mations into Eq. (11.25) gives the approximation
x−ξ
u(x, t) ≈ = f (ξ) (11.29)
t
8 Brook Taylor, English mathematician (1685–1731).
using Eq. (11.28). Treating the approximation in Eq. (11.29) as an equation

and solving for x yield the pair of equations:
x = ξ + f (ξ)t,
u = f (ξ).
Note that this pair of equations is merely the parametric representation of

the characteristics for the inviscid Burgers’ equation. The stationary point
is the parameter in this representation.

Now suppose there are two stationary points, ξ3 and ξ1 with ξ3 < ξ1 .
The dominant contributions to the improper integrals in Eq. (11.25) come
from both ξ3 and ξ1 , thus
−h(ξ )/(2ν) −h(ξ )/(2ν)
(x − ξ3 ) e √ + (x − ξ1 ) e √
3 1
h (ξ3 ) h (ξ1 )
u(x, t) ≈ . (11.30)
−h(ξ )/(2ν) −h(ξ )/(2ν)
t e √ 3 + e √ 1
h (ξ3 ) h (ξ1 )
In the case where h(ξ3 ) < h(ξ1 ) then h(ξ1 )/(2ν) is very much larger than
h(ξ3 )/(2ν) as ν → 0+ . The argument above shows that as ν approaches
zero from the right u(x, t) ≈ (x − ξ3 )/t. When h(ξ3 ) > h(ξ1 ) then u(x, t) ≈
(x − ξ1 )/t as ν → 0+ . The case of h(ξ3 ) = h(ξ1 ) implies
ξ3 ξ1
(x − ξ3 )2 (x − ξ1 )2
f (s) ds + = f (s) ds + .
0 2t 0 2t
By Eq. (11.28), which defines a stationary point, t f (ξi ) = x− ξi for i = 1, 2
and thus,
ξ1 ξ3
f (s) ds − f (s) ds
0 0
(x − ξ3 )2 (x − ξ1 )2
= −
2t 2t
f (ξ3 ) f (ξ1 )
= (x − ξ3 ) − (x − ξ1 )
2 2
f (ξ3 ) f (ξ1 )
= (tf (ξ1 ) + ξ1 − ξ3 ) − (tf (ξ3 ) + ξ3 − ξ1 )
2 2
ξ1
1
f (s) ds = (f (ξ3 ) + f (ξ1 )) (ξ1 − ξ3 ). (11.31)
ξ3 2
Readers will recognize that right-hand side of Eq. (11.31) as the trape-
zoidal rule for approximating a definite integral. See Fig. 11.6. Hence the
area under the chord connecting (ξ3 , u3 ) to (ξ1 , u1 ) is the value of the defi-
nite integral in Eq. (11.31). This implies the area under the curve u = f (x)
ξ1 ξ2 ξ3
1.0 u1
0.8
0.6
u2
f(x)
0.4
0.2
u3
0.0
-1 0 1 2 3
x
Fig. 11.6 The points on the solution curve u(x, t) = f (x − u(x, t)t) of the inviscid
Burgers’ equation are shifted horizontally to the left by an amount u(x, t)t to produce
the initial data. Thus the breaking wave of Fig. 11.4 is shifted into the initial condition.
The vertical line which denoted the position of the shock becomes the chord seen in this
figure. The shaded areas are equal. For the purpose of drawing this curve, the function
2
f (x) = e−x .
and above the chord over the interval [ξ3 , ξ2 ] equals the area under the
chord and above the curve in the interval [ξ2 , ξ1 ]. This equal area notion
was encountered earlier when determining the value of the solution to the
inviscid Burgers’ equation along the shock. Equation (11.31) provides a
way to determine the value of the solution to the inviscid Burgers’ equation
along a shock. If ξ3 < ξ1 are values of x such that Eq. (11.31) is satis-
fied then the point (ξ2 , u2 ) where the chord connecting (ξ3 , u3 ) to (ξ1 , u1 )
intersects the graph of u = f (x) provides the value u2 of the solution to
the inviscid Burgers’ equation on the shock. Since the stationary points
ξ1 and ξ3 are functions of (x, t) then the direction of the inequality be-
tween h(ξ1 ) and h(ξ3 ) causes the value of u(x, t) to abruptly “switch” when
crossing the shock. This analysis has assumed that the initial condition is
a “one-humped” function, but the idea can be generalized.
11.3 The Korteweg-de Vries Equation
The Korteweg-de Vries (often abbreviated, KdV) equation is another

important nonlinear partial differential equation and may be one of the most
studied nonlinear partial differential equations. Historically, the Korteweg-
de Vries equation, named after Korteweg and de Vries, originated from the
study of waves in shallow waters. The investigation started with observa-
tions made by John Scott Russell9 during experiments to determine the
most efficient hull design for canal boats. He reported [Russell (1845)]:
I was observing the motion of a boat which was rapidly drawn along a narrow
channel by a pair of horses, when the boat suddenly stopped — not so the
mass of water in the channel which it had put in motion; it accumulated
round the prow of the vessel in a state of violent agitation, then suddenly
leaving it behind, rolled forward with great velocity, assuming the form of a
large solitary elevation, a rounded, smooth and well-defined heap of water,
which continued its course along the channel apparently without change of
form or diminution of speed. I followed it on horseback, and overtook it
still rolling on at a rate of some eight or nine miles an hour, preserving its
original figure some thirty feet long and a foot to a foot and a half in height.
Its height gradually diminished, and after a chase of one or two miles I lost it
in the windings of the channel. Such, in the month of August 1834, was my
first chance interview with that singular and beautiful phenomenon which I
have called the “Wave of Translation”.
However, the Korteweg-de Vries equation was not formulated using a

mathematically rigorous argument until the work of Korteweg and de Vries
in 1845 and the true importance of the Korteweg-de Vries equation was not
fully realized until the last half of century. This recognition began with a
discovery in a numerical simulation carried out by Zabusky10 and Kruskal11
in 1965 of a self-enforcing solitary wave that maintains its shape as it prop-
agates with constant velocity. Such a solution is called a soliton solution,
see [Zabusky and Kruskal (1965)]. The discovery of soliton solutions of the
Korteweg-de Vries equation shed light on many physical phenomena that
could not be explained before. Readers are referred to [de Jager (2011)] for
the derivation of the Korteweg-de Vries equations.
9 John Scott Russell, Scottish engineer (1808–1882).
10 Norman J. Zabusky, American physicist (1929–)
11 Martin D. Kruskal, American mathematician and physicist (1925–2006).
There are several commonly used equivalent formulations of the

Korteweg-de Vries equation:
ut + 6u ux + uxxx = 0, (11.32)
ut + u ux + k uxxx = 0, (11.33)
ut − 6u ux + uxxx = 0, (11.34)
ut + u ux + uxxx = 0, (11.35)
ut + αu ux + βuxxx = 0. (11.36)
All of these forms are equivalent and may be obtained from each other
by rescaling the independent and dependent variables. Suppose x = A X,
t = B T , and u = C V then
C 6C 2 C
ut + 6u ux + uxxx = VT + V VX + 3 VXXX = 0,
B A A
which is equivalent to
6B C B
VT + V VX + 3 VXXX = 0.
A A
By choosing the constants A, B, and C appropriately, any form of the
Korteweg-de Vries equation given in Eqs. (11.33)–(11.36) can be obtained.
The details are left for the reader in Exercise 14.
Mathematically, the Korteweg-de Vries equation describes the interac-
tion among the temporal evolution, the nonlinearity, and the dispersion
effect (the triple partial derivative term). Solitons are created as a result
of this complicated interaction, especially between the nonlinear term and
dispersion effects in the medium. A study of the Korteweg-de Vries equa-
tion can easily fill several monographs, but the remainder of this section
will discuss only the existence of the simplest solitons.
Just as was the case for Burgers’ equation, the investigation starts with
determining if Eq. (11.32) has a solution of the form given in Eq. (11.1).
Setting ξ = x − c t and substituting u(x, t) = U (ξ) into the Korteweg-de
Vries equation produce
d3 U dU dU
3
+ 6U −c = 0.
dξ dξ dξ
Integrating both sides with respect to ξ yields
d2 U
+ 3U 2 − c U = A,
dξ 2
where A is an arbitrary constant of integration. To further simplify the
problem, assume that U → 0, U → 0, and U → 0 as ξ → ∞ and as
ξ → −∞, then
d2 U
+ 3U 2 − c U = 0.
dξ 2
This is a second-order (nonlinear) ordinary differential equation with miss-

ing first derivative U . Multiplying the ordinary differential equation by
U = dU/dξ and integrating both sides of the resulting equation yield
2
1 dU 1
+ U 3 − cU 2 = B,
2 dξ 2
where B is again an arbitrary constant of integration. The assumptions

that U → 0 and U → 0 as ξ → ∞ and ξ → −∞ imply that B = 0, and
therefore, after solving for dU/dξ,

dU √
= ±U c − 2U , (11.37)
dξ
where the ± sign should be chosen according to the sign of dU/dξ. In
the following, the ± sign will be dropped since it makes no difference in
the integration steps to follow. To solve for U , separate the variables in
Eq. (11.37) and integrate,

1
√ dU = ξ + A, (11.38)
U c − 2U
where A is again a constant. To evaluate the integral on the left side, make
a change of variable which, while not obvious, is motivated by identities
from hyperbolic trigonometry,
c sinh ζ
U = sech2 ζ =⇒ dU = −c dζ.
2 cosh3 ζ
Performing the substitution in Eq. (11.38) results in

1 2 −c sinh ζ
√ dU = 3 dζ
U c − 2U 2 2
c sech ζ c − c sech ζ cosh ζ

2 2 2 −1 2U
= −√ 1 dζ = − √ ζ = − √ sech .
c c c c
Thus the equation,

2 −1 2U
− √ sech =ξ+A
c c
is solved for U = U (ξ) to give
√
c c
sech2
U (ξ) = ξ+A .
2 2
Therefore, the Korteweg-de Vries equation has the following traveling wave
solution:
√
c 2 c
u(x, t) = sech (x − c t) + A .
2 2
√
The presence of the term c and the desire to find a real-valued solution,
suggest that c > 0. If the constant c is replaced by 4c2 for convenience,
then
u(x, t) = 2c2 sech2 (c(x − 4c2 t) + A). (11.39)
The solution given by Eq. (11.39) is a soliton solution of the Korteweg-de

Vries equation. Note that this soliton solution is positive for all x and t,
for each fixed t has a single maximum, and u(x, t) → 0 as |x| → ∞. These
features are “particle like”. Thus the name “soliton” was used to describe
such traveling wave solutions of nonlinear partial differential equations. In
addition, all such solutions travel to the right with a speed depending on the
amplitude of the solution. A solution with larger amplitude travels faster
with a speed equal to twice its amplitude as demonstrated in Fig. 11.7.
Fig. 11.7 The soliton solution expressed in Eq. (11.39) travels to the right as t increases
at a wave speed twice the amplitude on the wave. Two solitons are graphed above. The
dashed soliton has c = 3/2 and A = 1 while the solid curve has c = 1 and A = 0. It is
apparent the dashed soliton is traveling faster to the right than the solid curve.
The soliton solutions given by Eq. (11.39) are not the only traveling wave
solutions of the Korteweg-de Vries equation. There are many other solitons
and solitary wave solutions of the Korteweg-de Vries equation. For exam-
ple, if the assumption that the traveling wave solution and its derivatives
vanishes at infinity is relaxed, then the function U (ξ) satisfies:

2
1 dU c
= −U 3 + U 2 + A U + B ≡ F (U ) (11.40)
2 dξ 2
instead of Eq. (11.37), where A and B are arbitrary constants. Equation
(11.40) has real solutions if and only if F (U ) is positive. Since the function
F (U ) is a cubic polynomial (with a negative leading coefficient), it may
have one or three real zeros. From the properties of the zeros of F (U ), the
conditions under which Eq. (11.40) has real solutions and the properties of
the solutions can be determined. For example, if F (U ) has three distinct
real zeros, say U1 < U2 < U3 , then Eq. (11.40) has real bounded solutions
if the initial value is between U2 and U3 and all such solutions oscillate
between U2 and U3 . In fact, the solution in this case can be represented by
elliptic integrals and leads to traveling wave solutions of the Korteweg-de
Vries equation, called cnoidal waves. For details, see [Drazin and Johnson
(1989)]. The soliton solutions given by Eq. (11.39) are called one solitons
to indicate they have one peak. There are also two-solitons, three-solitons,
and more generally, N -solitons. For a more complete treatment of the topic,
see [Drazin and Johnson (1989)] or [Newell (1985)].
One of the most prominent features of soliton solutions is their remark-
able stability property. Soliton solutions maintain their form even after
two solitons propagating at different speeds collide. The waves seem to
simply pass through each other and each keeps its wave profile with only a
slight phase shift in wave position (the peak of the wave profile). Consider
the initial value problem of the Korteweg-de Vries equation with initial
condition,

9 3 9 3
u(x, 0) = sech2 x + 8 + sech2 x .
2 2 8 4
If the Korteweg-de Vries equation was linear then the solution to the initial
value problem would be

9 2 3 9 2 3
u(x, t) = sech (x − 9t) + 8 + sech (x − 9t/4) .
2 2 8 4
However, due to the nonlinearity of the Korteweg-de Vries equation, the
solitons interact in such a way as to produce a phase shift so that for t
much larger than the time of interaction,

9 2 3 9 2 3
u(x, t) ≈ sech (x − 9t) + 8 + θ1 + sech (x − 9t/4) + θ2 ,
2 2 8 4
for some θ1 and θ2 [Tao (2009)]. Figure 11.8 shows the interaction of the two
solitons over time. The overall shapes of the two solitons appear unchanged
after the faster wave passes the slower wave. The shift in phase of the
solitons is illustrated in Fig. 11.9. The solid curve is the actual solution of
the initial value problem while the dashed curve shows the profile of the
sum of the two solitons if there had been no interaction between them. For
more details, see [Toda (1989)] or [Drazin and Johnson (1989)].
Fig. 11.8 Two solitons interact without constructive or destructive interference. The
shapes of the solitons are largely unchanged after the faster soliton passes the slower
soliton. The stability of solitons is one of their interesting properties.
11.4 The Nonlinear Schrödinger Equation
This section introduces the nonlinear Schrödinger equation (NLS) also re-
ferred to as the Schrödinger wave equation since it describes how the wave
function of a particle evolves over time. The wave function of a particle
when multiplied by its complex conjugate can be regarded as the probability
u(x,1.5)
4
3
x
-10 -5 5 10
Fig. 11.9 When solitons interact they experience a phase shift. The solid curve depicts
the profile of the solution to an initial value problem in which interaction takes place
while the dashed curve shows the sum of the solitons had there been no interaction.
density function for the location of a particle at any given time. The
Schrödinger equation is the fundamental equation of quantum mechanics,
just as Newton’s second law of motion is the foundation of classical mechan-
ics. However, along with technological advances, the Schrödinger equation
has become even more important due to its connection with the way that
signals are propagated in fiber optics, see [Stolen (2008)] and [Agrawal
(2013)]. This section will only discuss the existence of soliton solutions to
the Schrödinger equation in the one dimensional case.
The standard form of the nonlinear Schrödinger equation in one spatial
dimension is
i ψt + ψxx + γ|ψ|2 ψ = 0, (11.41)
√
where i = −1 is the imaginary unit and ψ is a complex-valued function of
t and x. To motivate the approach below, assume Eq. (11.41) has a plane
wave solution of the form
ψ(x, t) = A ei(kx−ωt) , (11.42)
where A, k, and ω are constants to be determined. The real or the complex
part of ψ(x, t) could have been chosen as the plane wave solution as well.
One motivation to look for solutions in the form above is that the method
of undetermined coefficients familiar from the study of ordinary differential
equation may enable the constants to be found. As mentioned in Sec. 11.1,
the constant k in Eq. (11.42) is called the wave number and ω is the fre-
quency. Substituting the plane wave function defined in Eq. (11.42) for ψ
in Eq. (11.41) produces the equation
i(−i ω) + i2 k 2 + γ A2 = 0,
which can be simplified as

ω = k 2 − γ A2 . (11.43)
Therefore, for any k, if ω is defined by Eq. (11.43), the function ψ(x, t) is

a solution to the nonlinear Schrödinger equation. Equation (11.43) deter-
mines the relationship among the velocity of the solution, the wave number,
and the amplitude.
A more general solution to the nonlinear Schrödinger equation takes the
form
ψ(x, t) = f (ξ)ei(kx−ωt) (11.44)
where ξ = x − c t and the symbols k, c, and ω are constants. Note that this
solution incorporates features of a traveling wave and a plane wave. Sub-
stituting Eq. (11.44) into the nonlinear Schrödinger equation of Eq. (11.41)
yields
0 = −c if + ωf + f + 2i kf − k 2 f + γf 3
= f + i(2k − c)f + (ω − k 2 )f + γf 3 .
If f is a real-valued function of ξ, the term in the equation above involving
i must vanish, which implies c = 2k and the equation simplifies to
f + (ω − k 2 )f + γf 3 = 0.
Multiplying the equation above by 2f and antidifferentiating both sides
yield
γ
(f )2 = A − (ω − k 2 )(f )2 − (f )4 , (11.45)
2
where A is an arbitrary constant of integration. If the solution ψ(x, t) and
its derivative ψx (x, t) vanish as x → ±∞, then f → 0 and f → 0 as
x → ±∞. This implies A = 0 and function f satisfies
γ
(f )2 = (k 2 − ω)(f )2 − (f )4 .
2

Solving for f produces

df γ
= ±f (k 2 − ω) − (f )2 .
dξ 2
Separating the variables in this first-order nonlinear ordinary differential

equation and integrating both sides imply (taking the negative root),

−1
ξ+A= df
f (k − ω) − γ2 (f )2
2
where A is an arbitrary constant of integration. The indefinite integral here

is similar to the integral seen in Eq. (11.38) during the discussion of the
Korteweg-de Vries equation. A similar integration by substitution step can
be used here. Making the change of variable,

2(k 2 − ω) 2(k 2 − ω)
f= sech ζ =⇒ df = − sech ζ tanh ζ dζ
γ γ
enables the indefinite integral to be rewritten as

−1 1 1
df = √ dζ = √ ζ
f (k 2 − ω) − γ2 (f )2 k2 − ω k2 − ω

1 −1 γ
= √ sech f .
k2 − ω 2(k 2 − ω)
Hence

1 γ
√ sech−1 f =ξ+A
k2 − ω 2(k 2 − ω)
or solving for f (ξ),

2(k 2 − ω)
f (ξ) = sech((k 2 − ω)1/2 (ξ + A)). (11.46)
γ
Therefore, substituting Eq. (11.46) into Eq. (11.44) yields the following
solution of the nonlinear Schrödinger equation,

2(k 2 − ω) i(kx−ωt)
ψ(x, t) = e sech((k 2 − ω)1/2 (x − 2k t + A)). (11.47)
γ
The solution found above is called an envelope soliton solution of the
nonlinear Schrödinger equation. There are many other interesting solutions
of the nonlinear Schrödinger equation. In fact, if the assumption that ψ(x, t)
vanishes as x → ±∞ is dropped, then the function f may satisfy the more
general equation given by Eq. (11.45). In this case, the solution will be
represented by elliptic functions of the first kind. The details of this case
are beyond the scope of this text. Readers interested in exploring more of
this case may consult [Hasegawa and Matsumoto (2012)].
11.5 Further Comments
Nonlinear partial differential equations arise from a great variety of fields

ranging from electrostatics, fluid mechanics, and quantum mechanics to
signal processing, finance and many more. Not surprisingly, the study
of nonlinear partial differential equations has drawn a great amount of

attention from many researchers. The material presented in this chapter
is merely a brief introduction, which touched only upon a tiny part of
this huge area. Only the one-dimensional cases of Burgers’ equation, the
Korteweg-de Vries equation, and the nonlinear Schrödinger equation are
discussed and the discussion is, in large part, concentrated on the existence
and basic properties of traveling wave solutions of the partial differential
equations. There are many more important nonlinear partial differential
equations. One of these equations is the sine-Gordon equation which
originated from differential geometry and relativistic field theory. This
equation, along with several solution techniques, was known even in the
nineteenth century. However, it has gained much attention since the 1970s
with the discovery that it possesses soliton solutions (the so-called “kink”
and “antikink”). Readers will be asked to explore the existence of such
solutions in the exercises using the techniques presented in this chapter,
see Exercise 12.
Even though there seems to be no general theory for solving nonlinear
partial differential equations, there are methods that can be employed to
solve certain types of nonlinear partial differential equations. For example,
the initial value problems of Burgers’ equation can be solved using the Cole-
Hopf transformation and certain traveling wave solutions of some nonlinear
partial differential equations can be obtained by solving the corresponding
ordinary differential equations. The inverse scattering transformation and
the Bäcklund12 transformation stand out among other well-known meth-
ods. The inverse scattering method originated from quantum mechanics in
the 1930s. The method was applied to solve Korteweg-de Vries equation
in 1967, see [Gardener (1967)] and later it was applied to other nonlinear
partial differential equations. The inverse scattering transformation repre-
sents one of the most important advances in the study of nonlinear partial
differential equations. The method of the inverse scattering transformation
sometimes is called the nonlinear Fourier transform method since, in some
generalized sense, the method is analogous to the method of the Fourier
12 Albert Victor Bäcklund, Swedish mathematician (1845–1922).

transform used to solve linear partial differential equations. The Bäcklund

transformation has its origins in differential geometry and may be consid-
ered as a transformation among solution surfaces. The basic idea is, through
such a transformation, the solutions of a nonlinear partial differential equa-
tion can be connected with those of another equation whose solutions can
be found and/or better understood. Readers interested in studying these

and other methods can find more details in [Ablowitz (1981)] and [Rogers
(1982)], or [Debnath (2012)] and the references therein.
11.6 Exercises
(1) Consider the linear partial differential equation,

∂ 2u ∂u ∂u
2
+α +β = 0.
∂x ∂t ∂x
Find all bounded traveling wave solutions of the equation.
(2) The partial differential equation
∂2u ∂2u
2
− α2 2 + u = 0
∂t ∂x
is called a (linear) Klein -Gordon14 equation, which can be considered
13
as a modification of the wave equation. Find all bounded traveling

wave solutions of the equation.
(3) Consider the linear partial differential equation,
∂ 3u ∂u ∂u
+α +β = 0.
∂x3 ∂t ∂x
Find all the bounded traveling wave solutions of the equation.
(4) Assume the following equations have plane wave solutions of the form
u(x, t) = Aei(kx−ωt) and find the dispersion relation for each equation.
(a) The linear Klein-Gordon equation: utt + u = α2 uxx .
(b) The beam equation: utt + α2 uxxxx = 0.
(c) The linear Korteweg-de Vries equation: ut + α ux + β uxxx = 0.
(5) An example of the bistable equation is
ut − uxx = −u + H(u − 1/2), (11.48)
where H(z) is the Heaviside function. Find the ordinary differential
equation satisfied by a traveling wave solution to Eq. (11.48).
13 Oskar Benjamin Klein, Swedish theoretical physicist (1894–1977).
14 Walter Gordon, German theoretical physicist (1893–1939).
(6) Consider the ordinary differential equation in Eq. (11.13).

(a) Show that if the equation U 2 − 2c U − 2A = 0 has complex roots
Eq. (11.13) has no bounded solutions.
(b) Show that if the equation U 2 − 2c U − 2A = 0 has a repeated real
root, the only bounded solution to Eq. (11.13) is constant.
(7) Verify graphically if Eq. (11.13) has two distinct equilibrium solutions,
say U1 < U2 , that U1 is stable and U2 is unstable. Plot several typical
integral curves.
(8) Show that if a function of the form given in Eq. (11.15) is a traveling
wave solution to Burgers’ equation as given in Eq. (11.7), then c =
(U1 + U2 )/2.
(9) A slight generalization of Burgers’ equation is the following:
ut + (F (u))x − ν uxx = 0.
Assume that F (u) is a smooth function with F (u) > 0 for all u.
Derive the ordinary differential equation satisfied by the traveling wave
solution u(x, t) = U (x − c t) and discuss the solution properties of the
ODE.
(10) Show that Fisher’s15 equation,
u
ut − uxx = γ u 1 −
K
can be written in the nondimensional form,
ut − uxx = u(1 − u)
and derive the ordinary differential equation satisfied by U (ξ) = U (x−
c t) if U (x − c t) = u(x, t) is a traveling wave solution of Fisher’s
equation (in the nondimensional form).
(11) Show that if the initial condition of the inviscid Burgers’ equation is
2
f (x) = e−x , a shock wave forms at t = e/2.
(12) The sine-Gordon equation is another well-known nonlinear partial dif-
ferential equation:
∂ 2 u ∂ 2u
− 2 + sin u = 0. (11.49)
∂t2 ∂x
√
Let c be any constant and γ = 1/ 1 − c2 . Show that the sine-Gordon
equation is invariant under the change of variables:
ξ = γ(x − c t) and η = γ(t − c x).
15 Ronald Aylmer Fisher, English statistician (1890–1962).
(13) Consider the sine-Gordon equation (SGE) shown in Eq. (11.49).

(a) Assume that the sine-Gordon equation has a traveling wave solu-
tion in the form of u(x, t) = U (x − c t) for some constant c and
derive the ordinary differential equation for U .
(b) Assume U → 0 and U → 0 as ξ → ±∞ and solve the ordinary
differential equation in the previous part.

(14) Let x = A X, t = B T , and u = C V and show that the alternative
forms of the Korteweg-de Vries equation given in Eqs. (11.33)–(11.36)

can be derived from Eq. (11.32) by the appropriate choices of constants
A, B, and C.
(15) Show that, if u(x, t) is a solution of the Korteweg-de Vries equation,
then
w(x, t) = A2 u(Ax + 6ABt + C, A3 t + D) − B
is also a solution.
(16) Let c > 0 and A be any constant. Verify directly that
√
c c
u(x, t) = sech2 (x − c t) + A
2 2
is a solution the Korteweg-de Vries equation (found in Eq. (11.32))
and use technology to graph the function for various values of c as a
function of x with different values of t.
(17) Let f (t) be any continuous function on (−∞, ∞). Show that if v(x, t)
is a solution of the equation,
vt + 3(vx )2 + vxxx = f (t),
t
then w(x, t) = v(x, t) − 0 f (s) ds is a solution to
ut + 3(ux )2 + uxxx = 0
and that u = vx and u = wx are solutions of the Korteweg-de Vries
equation (Eq. (11.32)).
(18) The partial differential equation,
vt − 6v 2 vx + vxxx = 0 (11.50)
is called the modified Korteweg-de Vries equation. Assume that

V → 0, V → 0, and V → 0 as |x| → ∞ and follow the procedure of
finding the solution to the Korteweg-de Vries equation to show that
v(x, t) = − csch(x − t)
is a solution of the equation.
(19) Assume that V → A (a constant), V → 0, and V → 0 as |x| →

∞ for any fixed t in the modified Korteweg-de Vries equation (see
Exercise 18) and show that
2
v(x, t) = A +
e−(x+(6A2 −1)t) − 4A + (4A2 − 1)ex+(6A2 −1)t
is a solution of the equation.

(20) Show that if v(x, t) is a solution of the modified Korteweg-de Vries
equation (Eq. (11.50)), then u = v 2 + vx is a solution of the Korteweg-

de Vries equation (Eq. (11.32)).
(21) The Boussinesq16 equation is the following:
utt − α2 uxx − β uxxxx − γ(u2 )xx = 0. (11.51)
Assume that u(x, t) = U (x − c t) is a solution of the Boussinesq equa-
tion and that ξ = x − c t. Find the ordinary differential equation that
U (ξ) satisfies.
16 Joseph Valentin Boussinesq, French mathematician (1842–1929).

2017 Nonlinear Partial Differential Equations

Uploaded by

Copyright:

Available Formats

2017 Nonlinear Partial Differential Equations

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2017 Nonlinear Partial Differential Equations

Uploaded by

Copyright:

Available Formats

September 26, 2017 11:52 A First Course in Partial Diﬀerential Equations 9in x 6in b3044-ch11 page 461

Nonlinear Partial Diﬀerential

So far this textbook has concentrated on linear partial diﬀerential equations

462 A First Course in Partial Diﬀerential Equations

In addition to ﬁnding special types of solutions to nonlinear partial dif-

linear partial diﬀerential equations. For example, through an appropriate

11.1 Waves and Traveling Waves

Traveling wave solutions of nonlinear partial diﬀerential equations have

Nonlinear Partial Diﬀerential Equations 463

is a traveling wave if u has the form,

pagated at speed |c|. For many nonlinear partial diﬀerential equations,

464 A First Course in Partial Diﬀerential Equations

where A is an arbitrary constant of integration. If c < 0, the general

U (ξ) is unbounded. If A1 = A2 = 0, then solution U (ξ) = A/c is a constant

Nonlinear Partial Diﬀerential Equations 465

force or equivalently exist in a region of space where the potential energy

Another class of traveling wave solution, of particular importance for non-

tion which is a special type of traveling wave solution. In general, a solitary

466 A First Course in Partial Diﬀerential Equations

11.2 Burgers’ Equation

The discussion of nonlinear partial diﬀerential equations starts with the

on a straight stretch of highway was developed. The discussion there leads

Nonlinear Partial Diﬀerential Equations 467

11.2.1 Conservation Laws and Burgers’ Equation

468 A First Course in Partial Diﬀerential Equations

valued. Once the quantity that must be conserved in Burgers’ equation is

To determine the conserved quantity, integrate the viscous Burgers’

11.2.2 Traveling Wave Solutions of Burgers’ Equation

Nonlinear Partial Diﬀerential Equations 469

Integrating both sides of the equation with respect to ξ yields

for ﬁrst-order ordinary diﬀerential equations. The solutions of interest in

The constant functions U (ξ) = U1 and U (ξ) = U2 are equilibrium solutions

470 A First Course in Partial Diﬀerential Equations

which satisﬁes the following limits for each ﬁxed t,

Note that if B = 0, the corresponding traveling wave solution becomes

11.2.3 The Cole-Hopf Transformation

Nonlinear Partial Diﬀerential Equations 471

472 A First Course in Partial Diﬀerential Equations

formation, Burgers’ equation is rewritten as

Nonlinear Partial Diﬀerential Equations 473

is a solution of Eq. (11.19). Of course, the main interest is to obtain solu-

constants A and B, if A is large enough to guarantee that v(x, t) is positive

474 A First Course in Partial Diﬀerential Equations

−∞ (x − y)e dy + 0 (x − y)e− 2ν − 4νt dy

for t > 0. In order to evaluate the integrals involved in the solution,√the

11.2.4 Inviscid Burgers’ Equation and Shock Waves

Nonlinear Partial Diﬀerential Equations 475

476 A First Course in Partial Diﬀerential Equations

condition is given as u(x, 0) = f (x), then along the characteristic through

is oriented vertically and the x-axis is oriented horizontally as in Fig. 11.3,

Nonlinear Partial Diﬀerential Equations 477

function of x in some interval, the slopes of the characteristics will increase,

equation is thought of as a model of fluid flow, then the downstream fluid

ﬂuid is overtaking the downstream ﬂuid. Since the dependent variable u is

478 A First Course in Partial Diﬀerential Equations

on its negative minimum. Thereader will be asked to show in Exercise 11

Nonlinear Partial Diﬀerential Equations 479

shock) at time t. The integral of u is conserved even when passing through

480 A First Course in Partial Diﬀerential Equations

converge, the presence of the term,

on its negative minimum. Thereader will be asked to show in Exercise 11