CHAPTER 3

First Order PDE

First order equations enjoy a special place in theory of PDE, as they can generally be solved
explicitly using the method of characteristics. Although this method applies more generally
to fully nonlinear equations, such as Hamilton-Jacobi equations, we will restrict attention to
linear and quasilinear equations, in which the first order derivatives of the dependent variable u
occur linearly, with coefficients that may depend on u. The method of characteristics reduces
the determination of explicit solutions to solving ODE. We develop the theory in several
stages, with increasing sophistication, but really the idea is the same all along: first order
PDE become ODE when the PDE is regarded as specifying a directional derivative in several
dimensions.

3.1. The Method of Characteristics for Initial Value Problems

Initial value problems in one space variable x and time t, take the form
(1.1) ut + c(x, t, u)ux = r(x, t, u), t > 0, u(x, 0) = f (x).
Lets assume that c and r are given C 1 (continuously differentiable) functions, and the initial
condition f : R R is a given C 1 function. The coefficient c will be established as a wave
speed, and the notation r is simply for the right hand side of the PDE.
We can solve (1.1) at least for short time (and perhaps only locally in space) using the method
of characteristics, which reduces the initial value problem (1.1) to an initial value problem for
a system of ODE. In this method, we depend on the observation that if {(x(t), t) : t 0} is
a smooth curve, then along the curve, u(x(t), t) has rate of change
d dx
(1.2) u(x(t), t) = ut + ux ,
dt dt
given by the chain rule. Comparing (1.2) with the PDE in (1.1), it looks as though we can
make progress by setting dxdt
= c, and interpret c as a speed. The left hand side of the PDE
can also be interpreted as the derivative of u(x, t), in the direction (c, 1) in x,t-space.1
Now the PDE (1.1), can be replaced by the ODE system
dx du
(1.3) = c; = r.
dt dt
1Strictly

speaking, the direction is (c, 1)/ 1 + c2 ; the magnitude 1 + c2 sets the parameterization by t
rather than arclength.
25
26 3. FIRST ORDER PDE

These ODE are called the characteristic equations. Note that the characteristic equations are
autonomous only if c and r are independent of t.
Initial conditions for the ODE system are derived from the initial condition u(x, 0) = f (x)
for the PDE problem (1.1). To see what the ODE initial conditions should be, lets write
x(0) = x0 , and u(t) in place of u(x(t), t). Then the initial conditions for (1.3) are
(1.4) x(0) = x0 ; u(0) = f (x0 ).
From the theory of ODE, we know that the initial value problem (1.3), (1.4) has a unique
solution (x(t), u(t)), at least locally in time for each x0 . To emphasize that we have a solution
for each x0 , lets write the solution as x = x(t; x0 ), u = u(t; x0 ). The semicolon indicates that
x0 is regarded as a parameter in the ODE initial value problem, but now we are going to treat
x0 as a second variable, so that x and u are functions of the two variables t, x0 .

The parameter x0 specifies the curve in the xt-plane C(x0 ) : x = x(t, x0 ), that we refer to
as the characteristic through x = x0 , t = 0. As long as curves with different values of x0 do
not cross, the family of characteristics fills a region of the upper half plane {(x, t) : t 0},
thereby parameterizing points in the region with x0 , t. At each point P : (x, t) of this region,
we know the solution u, since u = u(t; x0 ) on each characteristic. Figure 3.1 illustrates the
characteristic originating at x0 that passes through point P. Once we have identified the value
of x0 , then u(x, t) = u(t; x0 ).

t
C(x0 )

t P : ( x(t;x0 ),t)

0 x0 x x

Figure 3.1. Characteristic C(x0 ) = {(x(t; x0 ), t) : t 0}, along which u =

u(t; x0 ), for the initial value problem (1.1).

There is a nice physical interpretation of this construction. The parameter x0 , called the
Lagrange variable, labels a material point. Then x = x(t; x0 ) is the Eulerian variable describing
the location at time t of that material point. The value u of the variable can be thought of
either in Lagrange variables, for which u = u(t; x0 ), or in Eulerian variables, for which u is
observed at a fixed location: u = u(x, t), the solution we seek.
 3.1. THE METHOD OF CHARACTERISTICS FOR INITIAL VALUE PROBLEMS 27

Mathematically, to get the solution u explicitly at each point (x, t), we need to invert the
change of variables (x, t) = (x(t, x0 ), t). To do so, we eliminate x0 , and write x0 = x0 (x, t) as
the solution of the equation x = x(t; x0 ). Then u(x, t) = u(t; x0 (x, t)) is the solution of (1.1).
For this kind of initial value problem, the method of characteristics is summarized as:

1. Rewrite the initial value problem (1.1) as a system of ODE, the characteristic equa-
tions (1.3), with initial conditions (1.4).
2. Solve the ODE and initial conditions for x(t), u(t), with parameter x0 = x(0) to get
the solution along each characteristic.
3. Solve for x0 as a function of x, t. This effectively changes variables from t, x0 to x, t.
4. Write the solution u = u(x, t).

Example 1: Solve the initial value problem

(1.5) ux + uy = u, u(x, 0) = cos x.
In this example, y is time-like in the sense that the initial condition is posed at y = 0. The
PDE written as u (1, 1) = u shows that the left hand side is the directional derivative of u
in the direction (1, 1). Consider the lines x = y + k parallel to (1, 1), where the parameter k
plays the same role as x0 above. The rate of change of u along each line is:
d u u
dy
u(y + k, y) = x
1+ y
=u

Therefore, u(y + k, y) = A(k)ey ,

where A(k) is an arbitrary function. Thus, since k = x y,
(1.6) u(x, y) = A(x y)ey
is the general solution of the PDE, depending on the arbitrary function A(k) of a single
variable.

To complete the solution, we use the initial condition to determine A(k) : setting y = 0 in
(1.6),
u(x, 0) = A(x) = cos x.
Thus, the solution of the problem is
u(x, y) = cos(x y)ey .

Since the left hand side of (1.5) is a directional derivative, it is an ordinary derivative in that
direction. Thus, u0 = u in this direction, explaining the exponential growth of the solution
along each characteristic x = y + k. Likewise, the solution would be u(x, y) = cos(x y) if
the right hand side of the PDE were zero.
In this example, we found the characteristics before determining the behavior of u along them.
Generally, the characteristics for equation (1.1) will also depend on the solution, if c depends
on u.
28 3. FIRST ORDER PDE

3.2. The Method of Characteristics for Cauchy Problems in Two Variables

In this section, we present a more general version of the method of characteristics for first
order quasilinear PDEs in two independent variables.

First order quasilinear PDE in two independent variables take the form
(2.7) a(x, y, u)ux + b(x, y, u)uy = c(x, y, u),
where a, b, c are given C 1 functions from R2 R to R. In this equation, neither of the variables
necessarily has a special role such as time. Consequently, the notation is somewhat different
from the previous section.
Rather than posing an initial condition, we pose a more general side condition for equation
(2.7) in the form
(2.8) u = z0 (s) on the curve : x = x0 (s), y = y0 (s),
where x0 , y0 , z0 are given C 1 functions on an interval I. This is sometimes referred to as the
initial curve . Problem (2.7), (2.8) is referred to as The Cauchy problem.

z
z = u(x , y )

(x,y,z)(s,)

(a,b,c)

= 0

n = ( u x , uy , - 1 )

x y

Figure 3.2. Initial curve , characteristic curve tangent to (a, b, c) and solu-
tion surface.
We shall show that for C 1 solutions, the partial differential equation is really an ODE in
disguise (as we saw in example 1 above). Suppose u(x, y) is a solution of the Cauchy problem.
Then the graph z = u(x, y) is a two-dimensional surface in xyz-space that includes the curve
. Equation (2.7) states that the vector field
(a(x, y, z), b(x, y, z), c(x, y, z))
 3.2. THE METHOD OF CHARACTERISTICS FOR CAUCHY PROBLEMS IN TWO VARIABLES 29

is tangent to the solution surface z = u(x, y), since the solution surface has normal

(u(x, y) z) = (ux , uy , 1).

The solution surface can therefore be generated by integrating along the vector field, starting
at each point of the curve : x = x0 (s), y = y0 (s), z = z0 (s), s I. (See Fig. 3.2.) If is the
variable of integration along these integral curves, then the surface generated is parameterized
by (s, ) : x = x(s, ), y = y(s, ), z = z(s, ). To recover u(x, y), we transform from (s, )
back to (x, y) in z and set u(x, y) = z(s, ), establishing the existence of the inverse using the
Inverse Function Theorem.

This procedure to solve the Cauchy problem (2.7), (2.8) is divided into three steps:
1. Generate the solution surface from integral curves.
In this step, we solve the one-parameter family of initial value problems
dx dy dz
= a(x, y, z) = b(x, y, z) = c(x, y, z)
(2.9) d d d
x(0) = x0 (s) y(0) = y0 (s) z(0) = z0 (s),

for each s I. Denote the solution (x, y, z)(s, ). From ODE theory, the solution exists, is
C 1 , and is unique, at least in a neighborhood of . The solution curves in R3 are known
as characteristic curves. We reserve the term characteristics to mean the projection of the
characteristic curves onto the (x, y)-plane.

2. Apply the Inverse Function Theorem.

In this step, we solve the equations
x = x(s, )
(2.10)
y = y(s, )

for (s, ) as a function of (x, y) : (s, ) = (s, )(x, y). The solution is guaranteed by the Inverse
Function Theorem.

3. Write the solution surface as a graph z = u(x, y). Now we are able to write the solution as
a function of x, y :
u(x, y) = z((s(x, y), (x, y)).

This procedure will work as long as the transformation (2.10) is invertible. We can guar-
antee this locally by appealing to the Inverse Function Theorem. Specifically, let P =
(x0 (s0 ), y0 (s0 ), z0 (s0 )) be a point on . In order that (2.10) be invertible near (x, y) = (x0 (s0 ), y0 (s0 )),
we require the Jacobian matrix (x, y)/(s, ) to be invertible at this point. That is, we require
30 3. FIRST ORDER PDE

at P,
 x y
 
 
     x0 (s ) y 0 (s ) 
 (x, y)   s s   0 0 0 0

(2.11)  (s, )  = 
=  6= 0,
   
 x y  a(P ) b(P )
  
 

where we have used (2.8), (2.9). This condition means that the tangent (a, b) to the char-
acteristic at (x0 (s0 ), y0 (s0 )), is not parallel to the tangent (x00 (s0 ), y00 (s0 )) of the projection
of at P onto the (x, y)-plane. Consequently, when (2.11) holds, we say that the curve is
non-characteristic at P. Thus, provided the initial data are non-characteristic in the sense of
(2.11), we have a unique C 1 solution u(x, y) of (2.7), (2.8) for (x, y) near (x0 (s0 ), y0 (s0 )).

Example 2: Solve the Cauchy problem

uux + uy = 1, u(x, x) = 0.

Here, a = z, b = 1, c = 1 and the initial condition is u = 0 on the line y = x. We parameterize

the initial condition as follows:
xo (s) = s yo (s) = s zo (s) = 0
Characteristic equations are
x0 = z, y 0 = 1, z 0 = 1,
with corresponding initial conditions x(0) = s, y(0) = s, z(0) = 0.

Thus, z = , so that x0 = z = . Now we can solve for x and y :

2
x= + s, y = + s.
2
2
Eliminating s, we get a quadratic equation for : x y = 2
. Thus,
p
= 1 1 + 2x 2y.
But = z = u(x, y), and to satisfy the initial condition, we have to take the negative square
root:
p
u(x, y) = 1 1 + 2x 2y.

The solution is valid only for 2x 2y < 1, i.e., y > x 21 . In fact, the solution surface
z = u(x, y) is the lower half of the smooth parabolic surface (z 1)2 = 1 2x + 2y, which
has a fold along the line y = x 12 , z = 1. Since the surface becomes vertical at the fold,
the solution u(x, y) has a singularity on the line y = x 12 where the derivative ux uy blows up.
 3.3. METHOD OF CHARACTERISTICS IN Rn . 31

3.3. Method of Characteristics in Rn .

In this subsection, we repeat the method of characteristics for a single quasilinear first order
equation, to show how the method works in any number of independent variables. Character-
istic curves are of course one-dimensional, and thus contribute one dimension to the solution
surface, which is n 1 dimensional. The remaining dimensions in the surface are provided by
the initial conditions.

Consider x Rn ; u = u(x) R. The first order equation we consider has the general form

(3.12) a(x, u) u = c(x, u)

where a : Rn R Rn , a vector of coefficients, and c : Rn R R, a scalar, are given C 1

functions.
The Cauchy problem involves an n 1 dimensional hypersurface Rn that provides initial
conditions for characteristic curves:

(3.13) x = xo (s), u = uo (s) s Rn1 .

Characteristic curves in (x, z) space (Rn+1 ) are solution curves of the system
dx dz
= a(x, z), = c(x, z)
(3.14) d d
x(0) = xo (s), z(0) = uo (s) for each s.

Note that for each s, we have existence and uniqueness of solutions of (3.14) for | | small,
since a and c are C 1 . Moreover, since the data are C 1 , the solutions are C 1 in s also.
As before, we write the solutions in the form

(3.15) (a) x = x(s, ), (b) z = z(s, ).

The solution u(x) = z(s, ) is expressed in physical variables x if we can invert (3.15(a)) to
get s = s(x), = (x). This is guaranteed by the Inverse Function Theorem, at least locally,
if we assume the hypersurface is non-characteristic, i.e., x/(s, ) is invertible on (where
= 0), with z = u0 (s), and recall that x/ = a :
 
xo (s)
(3.16) det , a(xo (s), uo (s)) 6= 0.
s

In components:
xo = (x1o (s1 , , sn1 ), ..., xno (s1 , , sn1 ))T

a = (a1 , , an )T
32 3. FIRST ORDER PDE

x1o x1o



s1 sn1

x2o ..
xo .
=
s1 .

s .. ..

. .

xn xno
o

s1 sn1

Then we have the solution

u(x) = z(s(x), (x)).
More precisely, the method of characteristics and the Inverse Function Theorem have been
used to prove the following result.

Theorem 3.1. Suppose the data xo , uo are C 1 in a neighborhood of s = 0, and are non-
characteristic in the sense of (3.16) at s = 0. Then there exists a neighborhood N of xo (0)
and a C 1 function u : N R that solves the Cauchy problem (3.12), (3.13) in N .

Example 3. Particle size segregation in an avalanche.

Avalanches and rock slides are examples of granular flow, typically involving particles of
different sizes. In this example, we write a PDE for the transport of two sizes of particles (a
bidisperse mixture) that have the same density. We shall assume that that as the avalanche
flows down the hillside, it establishes a constant depth, and that the velocity varies linearly
with depth. We shall ignore all but the component of velocity that is parallel to the hill-
side. In these circumstances, Gray and Thornton [20] formulated a model that describes the
distribution of particles within the avalanche.
Let x, y denote the spatial variables, and let v(y) = y denote the parallel velocity. These are
shown in Figure 3.3. The dependent variable u = u(x, y, t) is the volume fraction of small
particles. In the flow, large particles tend to rise, and small particles tend to fall. Gray and
Thornton argued that small particles fall at a speed proportional to the volume fraction 1 u
of large particles, essentially because they depend on space opened up by the motion of large
particles. Then large particles have to move upwards to balance the motion of the small
particles. With these assumptions, the PDE is

(3.17) ut + yux + S(u(u 1))y = 0,

where S > 0 is a constant of proportionality. lets suppose there is an initial distribution of

small particles given by

(3.18) u(x, y, 0) = u0 (x, y).

 3.3. METHOD OF CHARACTERISTICS IN Rn . 33

v(y)

Figure 3.3. Coordinates and velocity profile v(y) for avalanche flow model

Characteristic equations for this equation can be written

dx dy du
(3.19) = y, = S(2u 1), = 0.
dt dt dt
Thus, u = u0 (x0 , y0 ) is constant on the characteristic curve through (x0 , y0 ) at t = 0. Conse-
quently,
1
(3.20) y = S(2u 1)t + k, x = S(2u 1)t2 + kt + c, u = u0 (x0 , y0 ).
2
At t = 0, we have x = x0 , y = y0 , so that
1
(3.21) y = S(2u 1)t + y0 , x = S(2u 1)t2 + y0 t + x0 .
2
Thus, characteristics are parabolas in the x, t plane. Now we solve for x0 , y0 in terms of
u, x, y, t :
1
(3.22) y0 = y S(2u 1)t, x0 = x + S(2u 1)t2 yt2 .
2
Finally, we have a formula for the solution u = u(x, y, t), defined implicitly by the equation
 
1 2 2
(3.23) u = u0 x + S(2u 1)t yt , y S(2u 1)t .
2
34 3. FIRST ORDER PDE

This solution technique can be used to study the dynamics of avalanche flow with various
initial and boundary conditions.

3.4. Scalar Conservation Laws and the Formation of Shocks

In this section, we consider the initial value problem for the inviscid Burgers equation. We
show that solutions generated by the method of characteristics typically breakdown in finite
time. This nonlinear wave behavior occurs in applications such as gas dynamics, combustion
and detonation problems, and nonlinear elasticity. The breakdown of solutions signals the
formation of a shock wave, across which the solution is discontinuous.
Consider the initial value problem
(4.24) ut + uux = 0, < x < , t > 0,
with initial condition
(4.25) u(x, 0) = u0 (x), < x < .
The method of characteristics of 3.1 is applicable here:
dx du
=u = 0.
dt dt
Thus, u is constant on each characteristic, and characteristics are therefore straight lines with
speed u :
(4.26) x = ut + x0 , u = constant = u0 (x0 ).
Thus, the solution u = u(x, t) is given implicitly by the equation
(4.27) u = uo (x ut).
Let F (u, x, t) = u u0 (x ut). Generally, we cannot solve for u explicitly, but we can use
the equation to prove local existence near any initial point x = x0 , by applying the Implicit
Function Theorem to F.
3.4.1. Breakdown of smooth solutions. As we saw in 1.4.2, the graph of the solution
steepens where it has negative slope, because larger positive values of u travel faster than
smaller values. For negative values of u, the characteristics travel to the left, but the same is
true: the graph steepens where the slope is negative. Mathematically, we find ux at
some x, as t increases to a time t . The notion that some values of u travel faster than others,
leading to steepening, may be expressed in the statement:

Characteristics x = ut + x0 that originate at points x0 in an interval where u00 (x0 ) < 0 cross
in finite time.

In Fig 3.4 we show the characteristics for the solution shown in Fig. 1.2, and see that charac-
teristics ahead of the crest of the wave eventually cross. If this first occurs at a time t = t ,
then the method of characteristics gives a multi-valued function of (x, t), for t > t in the
region where the characteristics cross. We say the solution breaks down at t = t .
 3.4. SCALAR CONSERVATION LAWS AND THE FORMATION OF SHOCKS 35

0 x
Figure 3.4. Inviscid Burgers equation: crossing characteristics associated
with breakdown of smooth solution.

Our goal is to make this argument rigorous, and to find a formula for the breakdown time t .

To do so, we derive an equation for ux by taking x of the PDE, thus deriving an ODE for
the evolution of ux along characteristics.
First we differentiate equation (4.24):

(ut + uux ) = uxt + u2x + uuxx = 0.
x
Let v = ux . Then we have
vt + uvx = v 2 .
Along characteristics x = ut + xo we get the ODE
dv
(4.28) = v 2 .
dt
This equation (known as a Ricatti equation due to the quadratic nonlinearity) is solved easily.
Notice that it states that v decreases in t, and the more it decreases through negative values,
the more rapidly it continues to decrease.
Now we differentiate the initial condition u(x, 0) = u0 (x) to obtain a corresponding initial
condition for v :
(4.29) v(0) = u0o (xo ).
We solve (4.28), (4.29) to find v along the characteristic x = ut + x0 :
u0o (xo )
(4.30) v= .
1 + u0o (xo )t

We distinguish two cases: 1. If u0o (xo ) > 0, then v stays finite for all t > 0. Consequently, if u0
is monotonically increasing, then u(x, t) is defined for all x, t. Note from (4.27) that u(x, t) only
36 3. FIRST ORDER PDE

takes on values of uo (xo ), xo R. Therefore, if uo is bounded by m, M, m uo (x) M, x R,

then m u(x, t) M for all x R, t > 0.
2. If uo is not monotonically increasing, so that u00 (x0 ) < 0 for some values of x0 , then
1
v as t >0
u0o (xo )
in (4.30). Thus, the solution breaks down (ux ) at different times t on each character-
istic (depending on x0 ). Consequently, the solution u(x, t) of the initial value problem breaks
down at the earliest such time t = t :
 
1 0 1
t = min 0 : u0 (x) < 0 = .
<x< uo (x) min u0o (x)
x

Note that the minimum is achieved where uo has minimum slope, which will be at an inflection
point if uo is C 2 .
To continue the solution beyond t = t , we define weak solutions, in which the function u(x, t)
is allowed to be discontinuous. We continue with this topic in Chapter 13, after first consid-
ering solution and analysis techniques for second order equations.

Problems
1. Use the substitution v = uy to solve for u = u(x, y) :
uxy = 5uy , u(x, x) = 0, uy (x, x) = 2.
2. Solve for u = u(x, t) :
(1 + t2 )ut + ux = 0, u(x, 0) = sin x.
3. Solve for u = u(x, t) :
ut + ux + 3u = e2x+t , u(x, 0) = x.
4. Solve (1.5) using the general method of characteristics. You will need to set up the initial
condition with a parameter s. Show that the initial curve is non-characteristic.
5. Verify that u(x, t) constructed in general in 3.1 is indeed a solution of (1.1). Start by
working out what calculation you have to do to carry out this check. You will have to use the
chain rule repeatedly to check carefully.
6. Take an alternative direct approach to Example 1, reversing the roles of x and y, by setting
d
y = x + k. The PDE then becomes the ODE dx u(x, x + k) = u along characteristics. Solve
and incorporate the initial condition, finally obtaining the solution u(x, y).
 3.4. SCALAR CONSERVATION LAWS AND THE FORMATION OF SHOCKS 37

7. For avalanche flow, equation (3.17), suppose an initial distribution of particles is given by
u(x, y, 0) = u0 (x, y) = x + y, 0 < x, 0 < y < 1,
and an inlet boundary condition is specified by
u(0, y, t) = y, 0 < y < 1, t > 0.
Find the solution u(x, y, t), 0 < x, 0 < y < 1, t > 0 by the method of characteristics.
8. (a) Use the method of characteristics to solve the initial value problem
ut + tux = u2 , < x < , 0 < t < 1
1
u(x, 0) = , < x < .
1 + x2
(b) Show that the solution blows up as t 1 :
lim max u(x, t) = .
t%1 x

9. Sketch the graph of the traffic flow flux Q (see equations (4.23),(4.24) in Chapter 2) as a
function of density u. Explain each zero of Q in terms of the physical model.
10. Formulate constitutive laws for the traffic flux Q as a function of density assuming that
traffic speed is a quadratic decreasing function of density. How many parameters are there in
the model? Is it possible to make the flux non-convex as a function of density?
11. Write the details of how to use the Implicit Function Theorem on equation (4.27)to prove:
If u0 is smooth and bounded on (, ) then for each x0 R, there is an interval I R
containing x0 such that the solution u(x, t) exists, is C 1 , and is unique for all x I, and all
small enough t.
12. Let u0 (x) = H(x)x2 , where H(x) = 0 for x < 0 and H(x) = 1 for x 0 is the Heaviside
function. Write the solution u(x, t) of (4.24), (4.25) as an explicit formula for t > 0.
13. (Shock formation) Get the answer (4.30) by differentiating the implicit solution (4.27):
 

u = uo (x ut) , with u = u(x, t).
x
(This simpler approach depends on having the implicit equation for u available, which is not
the case for systems.)
14. Use the method of characteristics to prove global (for all t > 0) existence of a smooth
solution of (4.24), (4.25) when the initial data are given by a strictly increasing but bounded
C 1 function uo .
15. Carry through the analysis for a general scalar conservation law
ut + f (u)x = 0
where f : R R is a given C 2 function. Derive an implicit equation for the solution u(x, t) of
the Cauchy problem, and formulate a condition for the solution to remain smooth for all time.
Likewise, if the condition is violated, find an expression for the time at which the solution
first breaks down.