Professor H. M. Atassi
CLASS NOTES
ON
QUASILINEAR PARTIAL DIFFERENTIAL EQUATIONS
1. INTRODUCTION
Definition 1: An equation containing partial derivatives of the unknown function u
is said to be an n-th order equation if it contains at least one n-th order derivative,
but contains no derivative of order higher than n.
Definition 2: A partial differential equation is said to be linear if it is linear with
respect to the unknown function and its derivatives that appear in it.
Definition 3: A partial differential equation is said to be quasilinear if it is linear
with respect to all the highest order derivatives of the unknown function.
Example 1: The equation
∂
2
u
∂x
2
+ a(x, y)
∂
2
u
∂y
2
− 2u = 0
is a second order linear partial differential equation. However, the following equation
∂u
∂x
∂
2
u
∂x
2
+
∂u
∂y
∂
2
u
∂y
2
+ u
2
= 0
is a second order quasilinear partial differential equation. Finally, the equation
(
∂u
∂x
)
2
+ (
∂u
∂y
)
2
− u = 0
is a first order partial differential equation which is neither linear nor quasilinear.
Definition 4: A solution of a partial differential equation is any function that, when
substituted for the unknown function in the equation, reduces the equation to an
identity in the unknown variables.
Example 2: Let us consider the one dimensional wave equation
(
∂
2
∂t
2
− c
2
∂
2
∂x
2
)u = 0.
It is well known, and will be shown later, that the general solution of this equation
can be cast as
u = f(x − ct) + g(x + ct),
where f and g are arbitrary twice differentiable functions of the single variables ξ =
x − ct and η = x + ct, respectively. It is easy to see, using the chain rule, that
∂
2
u
∂t
2
= c
2
(
d
2
f
dx
2
+
d
2
g
dy
2
),
∂
2
u
∂x
2
=
d
2
f
dx
2
+
d
2
g
dy
2
.
1
Substitution into the wave equation leads to an identity.
Definition 5: Let φ(~x) be a function of the vector ~x = (x
1
, x
2
, . . . , x
n
) having first
order derivatives. Then the vector in R
n
(D
1
, D
2
, . . . , D
n
)φ,
where D
i
=
∂
∂x
i
, is called the gradient of φ and usually denoted as
0
grad φ
0
or
0
∇φ
0
.
Definition 6: Let ~v be a unit vector in R
n
and v measures the distance on ~v. Then
the limit
dφ(~x)
dv
= lim
∆t→0
φ(~x + ~v∆t) − φ(~x)
∆t
,
if it exists, is called the derivative of φ in the ~v direction. It is easy to show that
dφ(~x)
dv
= ∇ φ(~x) ·~v = (v
1
D
1
, v
2
D
2
, . . . , v
n
D
n
)φ(~x),
where ~v = (v
1
, v
2
, . . . , v
n
).
Example 3:
(i) In R
2
, ~x = (x, y). Then
grad φ(x, y) = (D
x
, D
y
)φ = (
∂φ
∂x
,
∂φ
∂y
).
The derivative of φ in the direction ~v = (1/
√
2, 1/
√
2), the first bisector of the plane
x-y, is
dφ
dv
=
1
√
2
(
∂φ
∂x
+
∂φ
∂y
).
(ii) Let φ(x, y, z) be a function having continuous first order derivatives. The equation
φ(x, y, z) = c,
where c ∈ R, represents a surface S. As we move along a path γ(s) on the surface S,
dφ
ds
=
∂φ
∂x
dx
ds
+
∂φ
∂y
dy
ds
+
∂φ
∂z
dz
ds
= 0,
dφ
ds
= grad φ·(
dx
ds
,
dy
ds
,
dz
ds
) = 0.
For an infinitesimal change in ds, the vector (dx/ds, dy/ds, dz/ds) is in the plane
tangent to the surface, at the point M(x, y, z). Consequently, grad φ is orthogonal
to the surface φ = c.
2
2. FIRST ORDER QUASILINEAR PARTIAL DIFFERENTIAL EQUA-
TIONS
We restrict our exposition to first order quasilinear partial differential equations (FO-
QPDE) with two variables, since this case affords a real geometric interpretation.
However, the treatment can be extended without difficulty to higher order spaces.
The general form of FOQPDE with two independent variables is
a(x, y, u)
∂u
∂x
+ b(x, y, u)
∂u
∂y
= c(x, y, u) (2.1)
where a, b, and c are continuous functions with respect to the three variables x, y, u.
Let u = u(x, y) be a solution to equation (2.1). If we identify u with the third
coordinate z in R
3
, then u = u(x, y) represents a surface S. The direction of the
normal to S is the vector (u
x
, u
y
, −1), where u
x
, u
y
are short hand notation of
∂u
∂x
and
∂u
∂y
, respectively. On the other hand, equation (2.1) can be written as the inner
product
(u
x
, u
y
, −1) · (a, b, c) = 0. (2.2)
Thus, (a, b, c) is perpendicular to the normal to S and consequently, must lie in the
plane tangent to S.
Let us consider a path γ(s) on S. The rate of variation of u as we move along γ is
du
ds
=
∂u
∂x
dx
ds
+
∂u
∂y
dy
ds
. (2.3)
The vector (dx/ds, dy/ds, du/ds) is naturally the tangent to the curve γ(s). Equation
(2.3) can be rewritten as the inner product
(u
x
, u
y
, −1) · (dx/ds, dy/ds, du/ds) = 0. (2.4)
Comparing (2.2) and (2.4), we see that there is a particular family of curves on the
surface S, defined by
dx/ds
a
=
dy/ds
b
=
du/ds
c
. (2.5)
These curves are called characteristics and will be denoted by C(s), or simply C.
We note that there are actually only two independent equations in the system (2.5),
therefore, its solutions comprise in all a two-parameter family of curves in space.
Theorem 1: Any one parameter subset of the characteristics generates a solution of
the first order quasilinear partial differential equation (2.1).
Proof: Let u = u(x, y) be the surface generated by a one-parameter family C(s) of the
characteristics whose differential equations are (2.5). By taking the rate of variation
of u along a characteristic curve C(s), we get
du
ds
= u
x
dx
ds
+ u
y
dy
ds
,
or
(u
x
, u
y
, −1) · (dx/ds, dy/ds, du/ds) = 0. (2.6)
3
But equations (2.5) state that the vectors (dx/ds, dy/ds, du/ds) and (a, b, c) are
collinear. Therefore,
(u
x
, u
y
, −1) · (a, b, c) = 0,
or
au
x
+ bu
y
= c,
and we are done.
Corollary 1: The general solution to equation (2.1) is defined by a single relation
between two arbitrary constants occurring in the general solution of the system of
ordinary differential equations
(dx/ds)
a
=
(dy/ds)
b
=
(du/ds)
c
;
or, in other words, by any arbitrary function of one independent variable.
Example 1: Consider the first order partial differential equation,
∂u
∂t
+ u
∂u
∂x
= 0.
The characteristics are defined by
(dt/ds)
1
=
(dx/ds)
u
=
(du/ds)
0
.
The last equation gives immediately u = k
1
. Since u is constant along a given
characteristic, then the first equation can be integrated immediately:
x − k
1
t = k
2
The general solution is then
k
1
= f(k
2
),
where f is an arbitrary function of the single variable k
2
. Substituting k
1
and k
2
by
their expressions, we finally have
u = f(x − ut).
Example 2: Consider the equation
3
∂u
∂x
− 7
∂u
∂y
= 0.
The characteristics are solution of the system
(dx/ds)
3
=
(dy/ds)
−7
=
(du/ds)
0
.
4
By integration, we get
u = k
1
,
3y + 7x = k
2
.
The characteristics are straight lines intersection of the two families of planes defined
by these equations. Any arbitrary relation between k
1
and k
2
is a solution. The
general solution is then
u = f(3y + 7x),
where f is an arbitrary function of the one independent variable 3y + 7x.
Example 3: Consider the equation
y
∂u
∂x
− x
∂u
∂y
= 0.
The characteristics are solution of the system
(dx/ds)
y
=
(dy/ds)
−x
=
(du/ds)
0
.
By integration, we get
u = k
1
,
x
2
+ y
2
= k
2
.
The characteristics are circles located in the plane u = k
1
. The general solution is
u = f(x
2
+ y
2
),
where f is an arbitrary function of the independent variable (x
2
+ y
2
). Geometrically,
the general solution is any surface of revolution around the u-axis.
Remark 1: The previous examples illustrate the application of the theory of char-
acteristics to find the general solution to a first order quasilinear partial differential
equation. Given the simple forms of these examples, the student could become sus-
picious that in the general case it will not be possible to get a closed analytical form
of the solution. A careful examination of system (2.5) suffices to convince you that,
in the general case, i.e., when a, b, and c are arbitrary functions of x, y, and u,
the system of ordinary differential equations (2.5) cannot be integrated analytically.
Nevertheless, for a given initial or boundary value problem, system (2.5) provides a
numerical solution.
Remark 2: When equation (2.1) is a linear homogeneous partial differential equation
a(x, y)u
x
+ b(x, y)u
y
= c(x, y)u. (2.7)
The solutions form a vector space which we call the solution space. This solution
space is naturally the null space of the linear operator
a(x, y)D
x
+ b(x, y)D
y
− c(x, y). (2.8)
5
It should be pointed out that since the general solution to (2.7) is any arbitrary func-
tion of one independent variable, there is a nondenumerable infinity of independent
solutions to equations (2.7). Therefore, the dimension of null space of the operator
(2.8) is infinity. This result appears to be in sharp contrast with what we know
about the first order ordinary linear differential operators where the null space is one-
dimensional. One may also add that this augurs the difficulties we shall encounter in
the study of partial differential operators.
THE BOUNDARY VALUE PROBLEM FOR A FIRST ORDER
PARTIAL DIFFERENTIAL EQUATION
The theory of characteristics enables us to define the solution to FOQPDE (2.1) as
surfaces generated by the characteristic curves defined by the ordinary differential
equations (2.5). However, a physical problem is not uniquely specified if we simply
give the differential equation which the solution must satisfy, for, as we have seen,
there are an infinite number of solutions of every equation. In order to make the
problem a definite one, with a unique answer, we must pick out of the mass of pos-
sible solutions, the one which has certain definite properties along definite boundary
surfaces. These properties represent the boundary conditions which the solution must
satisfy. The first fact which we must notice is that we cannot try to make the solutions
of a given equation satisfy any sort of boundary conditions; for there is a definite set
of boundary conditions which will give nonunique or impossible answers. The study
of the proper boundary conditions to be specified on definite boundary curves or sur-
laid the foundations to our present knowledge in this area.
Statement of the Cauchy problem for FOQPDE
Let γ(t) be a curve in region R of the x-y plane, where the value of the function
u(x, y) which satisfies equation (2.1) is specified as u(t). What are the conditions to
be satisfied by γ(t) and u(t), in order that the boundary value problem so defined
has a unique solution?
Consider an initial curve γ(t) in a region R
ξ = ξ(t),
(2.9)
η = η(t).
Since u(x, y) is specified on γ(t) as
u(x, y) = u(t) (2.10)
then the three functions
ξ = ξ(t), η = η(t), u = u(t) (2.11)
6
of the parameter s, the equations of the characteristics are
dx
ds
= a,
dy
ds
= b,
du
ds
= c, (2.12)
equations
x = x(s, t), y = y(s, t), u = u(s, t). (2.13)
Initial Curve
Γ(t)
C(s) Characteristics
Figure 2.1: Solution of the boundary-value problem of a first-order partial differential
equation by a one-parameter family of characteristics.
Equation (2.13) represent the surface S in a parametric form. When t is kept constant
we move along a characteristic curve C(s). For a given value of s, for example s
0
, we
move on the surface S along the curve Γ.
Let us now assume that the initial curve Γ was a characteristic curve C. It is then
obvious that equations (2.9) satisfy system (2.12) and there will be no surface solution
generated this way. We conclude that in order to generate a surface solution, Γ
should not be a characteristic curve. Mathematically, this means that the two vectors
(dξ/dt, dη/dt, du/dt) and (a, b, c) must be linearly independent. A sufficient condition
to insure this linearly independence is that the determinant
J =
dξ
dt
dη
dt
a b
(2.14)
never vanishes on Γ.
An alternative form to the condition (2.14) can be obtained from equations (2.13).
The vectors tangent to C and Γ are respectively (x
s
, y
s
, u
s
) and (x
t
, y
t
, u
t
). A sufficient
condition for their independence is that the Jacobian
J =
x
s
y
s
x
t
y
t
7
never vanishes on Γ. We are then led to the following theorem:
Theorem 2: The solution u(x, y) may be freely specified along a curve γ in R and the
resulting specification determines a unique solution of (2.1) if and only if γ intersects
the projection on the x-y plane of each characteristic curve exactly once.
Example 4:
Take the simple equation
u
x
+ u
y
= c(x, y, u) (2.15)
and take R as the square region in Figure 2.2. The projection on the x-y plane of
the characteristic curves (often called the characteristics) are the lines x −y = λ, for
x, y ∈ R. A few of these characteristics are shown in Figure 2.2 (a). We see that the
curve γ
1
of Figure 2.2 (b) meets the conditions of theorem 2 so that specification of
u(x, y) along γ
1
converts (2.1) into a boundary-value problem with a unique solution.
However, Curve γ
2
of Figure 2.2(c) while not intersecting any characteristic more
along γ
2
would specify u(x, y) = u(t) only for those characteristics which intersect
γ
2
, i.e., inside a sub region of R between the limiting characteristics shown in Figure
2.2(c), which is called domain of influence of the curve γ
2
. Specification of u(x, y)
along γ
2
, then determines v only in the domain of influence of γ
2
and we would say
such a boundary-value problem was understated for the whole region R. The curve
γ
3
of Figure 2.2(d)intersects some characteristics more than once, so specification of
u(x, y) along v would convert (2.1) to a boundary-value problem with no solution
unless we were lucky enough that the specified values for the several points on the
same characteristics were compatible with the general solution; we say, this boundary-
value problem is overspecified.
(a)
(b) (c) (d)
1 1 1 1
1 1 1 1
γ
1
γ
2
γ
3
Figure 2.2: (a) Characteristics of of (2.15), (b) Well-posed boundary-value problem,
(c) Understated boundary-value problem, (d) Overspecified boundary-value problem.
Theorem 3: The boundary-value problem for FOQPDE
au
x
+ bu
y
= c, x, y, in R
8
u(x, y) = u(t), x, y, on γ
has a unique solution for any specified u(t) if and only if γ intersects the projection
of each characteristic curve in R exactly once.
Remark 3: Due to the nonlinearity of (2.1), the domain of influence has only local
significance. The geometry of the characteristics can become quite complicated to
the point that a well posed boundary-value problem could be very difficult to solve
as we move along the characteristics.
3. SYSTEMS OF FIRST ORDER QUASILINEAR PARTIAL
DIFFERENTIAL EQUATIONS
We denote by u, v the dependent variables (unknown functions) and x, y the inde-
pendent variables. The general form of the system of differential equations is
L
1
= A
1
u
x
+ B
1
u
y
+ C
1
v
x
+ D
1
v
y
+ E
1
= 0,
(3.1)
L
2
= A
2
u
x
+ B
2
u
y
+ C
2
v
x
+ D
2
v
y
+ E
2
= 0,
in which A
1
, A
2
, ···, E
2
are known functions of x, y, u, v. We make the assumption
that all functions occurring in the theory are continuous and possess as many contin-
uous derivatives as may be required. Without restrictions we assume that nowhere
A
1
/A
2
= B
1
/B
2
= C
1
/C
2
= D
1
/D
2
.
If E
1
= E
2
= 0, the system is homogeneous. If the coefficients A
1
, A
2
, ···, E
2
are
functions of x and y only, the equations are linear and consequently much easier to
handle.
Definition 1: If the system is homogeneous, E
1
= E
2
= 0, and the coefficients A
1
,
A
2
, ···, D
2
are functions of u, v alone, the equations are said to be reducible. The
following theorem justifies the above definition.
Theorem 1: For any region where the Jacobian
j = u
x
v
y
− u
y
v
x
(3.2)
is not zero, a reducible system (3.1) can be transformed by interchanging the roles of
the dependent and independent variables into the following equivalent linear system
A
1
y
v
− B
1
x
v
− C
1
y
u
+ D
1
x
u
= 0,
(3.3)
A
2
y
v
− B
2
x
v
− C
2
y
u
+ D
2
x
u
= 0.
9
The transformation of the (x, y)- plane into the (u, v)- plane is called a hodograph
transformation. It is advantageous, in general, to perform the hodograph transfor-
mation whenever the Jacobian j 6= 0 in the region where a solution is desired.
Characteristic curves and characteristic equations
We have shown (Definition 1.6) that an expression such as A
1
u
x
+ B
1
u
y
represents
derivative in a direction dx/dy = A
1
/B
1
. Each equation of (3.1) can be thought of
as a linear relation between a derivative of u in the direction (A, B) and a derivative
of v in the direction (C, D). We now ask for a linear combination
L = λ
1
L
1
+ λ
2
L
2
,
so that in the differential expression L, the derivatives of u and those of v combine
to derivatives in the same direction. Such direction, which depends on the variables
x, y as well as u, v, is called characteristic direction.
Suppose the direction is given by the ratio x
σ
/y
σ
. Then the condition that, in L, u
and v are differentiated in this direction, is simply
λ
1
A
1
+ λ
2
A
2
λ
1
B
1
+ λ
2
B
2
=
λ
1
C
1
+ λ
2
C
2
λ
1
D
1
+ λ
2
D
2
=
x
σ
y
σ
(3.4)
The expression L can be written after multiplication with either x
σ
or y
σ
as
(λ
1
A
1
+ λ
2
A
2
)u
σ
+ (λ
1
C
1
+ λ
2
C
2
)v
σ
+ (λ
1
E
1
+ λ
2
E
2
)x
σ
= x
σ
L
(3.5)
(λ
1
B
1
+ λ
2
B
2
)u
σ
+ (λ
1
D
1
+ λ
2
D
2
)v
σ
+ (λ
1
E
1
+ λ
2
E
2
)y
σ
= y
σ
L
If at the point (x, y), the functions u and v satisfy (3.1), we obtain four homogeneous
linear equations for λ
1
and λ
2
:
λ
1
(A
1
y
σ
− B
1
x
σ
) + λ
2
(A
2
y
σ
− B
2
x
σ
) = 0,
λ
1
(C
1
y
σ
− D
1
x
σ
) + λ
2
(C
2
y
σ
− D
2
x
σ
) = 0, (3.6)
λ
1
(A
1
u
σ
+ C
1
v
σ
+ E
1
x
σ
) + λ
2
(A
2
u
σ
+ C
2
v
σ
+ E
2
x
σ
) = 0,
λ
1
(B
1
u
σ
+ D
1
v
σ
+ E
1
y
σ
) + λ
2
(B
2
u
σ
+ D
2
v
σ
+ E
2
y
σ
) = 0.
In order to have nontrivial solutions to (3.6), all determinants of two rows in the
matrix of the coefficient of λ
1
and λ
2
must vanish. Thus a number of characteristic
relations follow. From the first two equations, we get
A
1
y
σ
− B
1
x
σ
A
2
y
σ
− B
2
x
σ
C
1
y
σ
− D
1
x
σ
C
2
y
σ
− D
2
x
σ
= 0 (3.7)
or
ay
2
σ
− 2bx
σ
y
σ
+ cx
2
σ
= 0, (3.8)
10
where
a = [AC], 2b = [AD] + [BC], c = [BD]
with the abbreviation
[XY ] = X
1
Y
2
− X
2
Y
1
.
Equation (3.8) defines the directions of the characteristics. There are three cases:
(i) ac − b
2
> 0, then (3.8) cannot be satisfied by a real direction. The differential
equations (3.1) are then called elliptic.
(ii) ac − b
2
= 0, then there is only one characteristic direction. The differential
equations (3.1) are then called parabolic.
(iii) ac − b
2
< 0, we have two different characteristic directions through each point.
The differential equations (3.1) are then called hyperbolic.
It is obvious that the notion of characteristics is useful only when they do exist. For
this reason, we shall from now on assume the hyperbolic character of system (3.1)
and accordingly suppose
ac − b
2
< 0.
Let ξ = y
σ
/x
σ
be the slope of the characteristics, then ξ satisfies the quadratic
equation
aξ
2
− 2bξ + c = 0. (3.9)
This equation has two distinct real solutions ξ
+
and ξ
−
. The characteristic curves are
the envelopes to these slopes C
+
for ξ
+
and C
−
for ξ
−
. Their respective equations
are
I
+
:
dy
dα
= ξ
+
dx
dα
along C
+
,
(3.10)
I
−
:
dy
dβ
= ξ
−
dx
dβ
along C
−
.
Let us to return to system (3.6) and write that the determinant of the first and third
equation is nil.
A
1
y
σ
− B
1
x
σ
A
2
y
σ
− B
2
x
σ
A
1
u
σ
+ C
1
v
σ
+ E
1
x
σ
A
2
u
σ
+ C
2
v
σ
+ E
2
x
σ
= 0. (3.11)
After rearrangement, we obtain
T u
σ
+ (aξ − S)v
σ
+ (Kξ − H)x
σ
= 0, (3.12)
in which,
T = [AB], S = [BC], K = [AE], H = [BE]. (3.13)
This relation holds on C
+
if we identify ξ with ξ
+
and σ with α, and likewise, C
−
, if
identify ξ with ξ
−
and σ with β.
11
Thus we arrive at the following four characteristic equations:
I
+
y
α
− ξ
+
x
α
= 0,
I
−
y
β
− ξ
−
x
β
= 0,
II
+
T u
α
+ (aξ
+
− S)v
α
+ (Kξ
+
− H)x
α
= 0, (3.14)
II
−
T u
β
+ (aξ
−
− S)v
β
+ (Kξ
−
− H)x
β
= 0,
It is not overemphasizing to stress again that II
+
is valid only on C
+
whose equation
is I
+
and II
−
on C
−
whose equation is I
−
.
Remark 1: Unless the system (3.1) is linear, ξ
+
and ξ
−
depend on x, y and u, v.
Consequently, the characteristics depend on the individual solutions u, v, and so does
the hyperbolic character of the system (3.1).
Example 1: Let us consider the case of a one-dimensional unsteady isentropic flow.
The unknown functions are the density ρ and the velocity u, and the independent
variables are x, t. The flow equations are:
ρ
t
+ uρ
x
+ ρu
x
= 0,
(3.15)
u
t
+ uu
x
+
c
2
ρ
ρ
x
= 0.
We find that a = 1, b = u, c = u
2
− c
2
, ac − b
2
= −c
2
< 0. Hence there are two
characteristic families:
x
α
= (u + c)t
α
, x
β
= (u − c)t
β
(3.16)
The characteristic equation for ρ and u become
u
α
+
c
ρ
ρ
α
= 0, u
β
−
c
ρ
ρ
β
= 0. (3.17)
Equations (3.16) express the fact that the characteristic curves in the (x, t)-plane
represent motions of possible disturbances called sound waves whose velocities
dx
dt
= u + c,
dx
dt
= u − c
differ from the particle velocity u by the sound velocity ±c.
For an isentropic flow, (γ − 1)
dρ
ρ
= 2
dc
c
. This result can be used to eliminate ρ from
(3.18) and get
u
α
+
2
γ − 1
c
α
= 0, u
β
−
2
γ − 1
c
β
= 0. (3.18)
12
Equations (3.18) can then be integrated to give
u +
2
γ − 1
c = r, u −
2
γ − 1
c = s, (3.19)
where r and s are constant along the characteristics C
+
and C
−
, respectively. r and
s are known as the Riemann invariants. If one of the Riemann invariants is constant
throughout the domain, the solution corresponds to a wave motion in a one direction
only and is said to be a simple wave.
Proposition: A flow in a region adjacent to a region of constant state is always a
simple wave.
Example 2: The equations for steady two-dimensional isentropic irrotational inviscid
flow are
ρu
x
+ ρu
y
+ uρ
x
+ vρ
y
= 0, (3.20)
uu
x
+ vu
y
+
c
2
ρ
ρ
x
= 0, (3.21)
uv
x
+ vv
y
+
c
2
ρ
ρ
y
= 0. (3.22)
The density ρ can be eliminated from these equations by multiplying the first by −
c
2
ρ
,
the second by u and the third by v, and adding them. The resulting equation is
(u
2
− c
2
)u
x
+ uvu
y
+ uvv
x
+ (v
2
− c
2
)v
y
= 0. (3.23)
To this equation we add the condition of irrotational flow
u
y
− vx = 0. (3.24)
The equations of the characteristics directions is
(c
2
− u
2
)ξ
2
+ 2uvξ + (c
2
− v
2
) = 0. (3.25)
The characteristics are real if u
2
+ v
2
> c
2
, or if the fluid is supersonic. In this case,
the system of characteristic equations can be written as
I
+
: y
α
= ξ
+
x
α
,
I
−
: y
β
= ξ
−
x
β
,
II
+
: u
α
= −ξ
+
v
α
, (3.26)
II
−
: u
β
= −ξ
−
v
β
.
The characteristic curves determined from I
+
and I
−
are usually called Mach lines.
13
Theorem 2: Consider a smooth curve γ(t) where u(t) and v(t) are freely specified,
then the boundary-value problem for the system of differential equations
L
1
= A
1
u
x
+ B
1
u
y
+ C
1
v
x
+ D
1
v
y
+ E
1
= 0,
(3.27)
L
2
= A
2
u
x
+ B
2
u
y
+ C
2
v
x
+ D
2
v
y
+ E
2
= 0,
in which A
1
, A
2
, ···, E
2
are known functions of x, y, u, v in the region R bounded by
the two characteristics passing though point P and the domain of dependence cut by
them from the initial curve γ(t) has a unique solution.
3. SECOND ORDER QUASILINEAR PARTIAL DIFFERENTIAL EQUA-
TION
We denote by Φ the dependent variable and x, y the independent variables. The
general form of a second order partial differential equation is
aΦ
xx
+ 2bΦ
xy
+ cΦ
yy
+ d = 0, (3.28)
in which a, b, c, d are known functions of x, y, Φ, Φ
x
, Φ
y
. This problem can be
reduced to that of (3.1) by introducing the variables
u = Φ
x
, v = Φ
y
and the fact that for continuously differentiable functions
u
y
= v
x
The equation for the characteristics is
ay
2
σ
− 2bx
σ
y
σ
+ cx
2
σ
= 0, (3.29)
where a, b, c are defined as previously. Calculating the relationship between u
σ
and
v
σ
as in (3.12), we finally arrive at the following characteristic system
I
+
y
α
− ξ
+
x
α
= 0,
I
−
y
β
− ξ
−
x
β
= 0,
II
+
u
α
+ ξ
−
v
α
+ x
α
d
a
= 0, (3.30)
II
−
u
β
+ ξ
+
v
β
+ x
β
d
a
= 0.
Canonical form of a second order partial differential equation
Note that the two families of characteristic I
+
and I
−
can be defined as
14
x = x(α, β) (3.31)
y = y(α, β) (3.32)
where β is constant along I
+
and α is constant along I
−
. If the Jacobian of the
transformation (3.31), (3.32) is not zero, we can define
α = α(x, y) (3.33)
β = β(x, y). (3.34)
Using the new variables α and β, equation (3.28) becomes
Φ
α,β
= F (α, β, Φ
α
, Φ
α
, Φ). (3.35)
Equation (3.35) in known as the canonical form of (3.28).
15
