Study material for Week-2 on

Partial Differential Equations

S. Sivaji Ganesh, IIT Bombay.

Distributed to students of MA 515, Autumn 2021.

 Study material on
Partial Differential Equations

S. Sivaji Ganesh
Department of Mathematics
Indian Institute of Technology Bombay

July 25, 2021

 Preface

This notes is intended for circulation to students of a course on Partial differential

equations.
Though the notes are proof-read many times, it could still have some misprints.
Please send your comments, corrections etc. to X@gmail.com, where X=msg2sgspde.

S. Sivaji Ganesh, X=msg2sgspde

Mumbai.
July 25, 2021.

v
40 Chapter 2. First order PDE

where F ∈ C 1 (�2 ) is arbitrary.

Further, let us find a particular solution of the given PDE which satisfies
the Cauchy data given by

x = s 2 , y = 2s, z = s, 0 ≤ s ≤ 1.

By imposing the above conditions, the arbitrary F gets fixed. Note that
z2 s2
x − 2 = C1 implies that 2 = C1 . Also note that y − z = C2 implies that
s = C2 . Thus we get a relation between C1 , C2 , namely, C22 = 2C1 . This
gives us the relation

(y − u(x, y))2 = 2x − u 2 (x, y),

which on simiplification reads as 2u 2 (x, y) − 2y u(x, y) + (y 2 − 2x) = 0.

Solving for u(x, y), we get
�
y ± 4x − y 2
u(x, y) = . (2.36)
2
Note that both the functions defined by (2.36) satisfy the Cauchy data.
However neither of the two functions is a solution to the given PDE since
they are not differentiable functions at any point on the projection of the
datum curve to �2 , which lies on the parabola y 2 = 4x. Thus the given
Cauchy problem has no solution. An alternative method to arrive at this
conclusion may be found in Exercise 2.8.

2.4 Cauchy problem for quasilinear PDE

Let I be an interval in �, and the functions f , g , h be continuously differ-
entiable on I such that ( f � (s))2 + ( g � (s))2 �= 0 for all s ∈ I . Consider the
space curve Γ described parametrically by

Γ : x = f (s), y = g (s), z = h(s), s ∈ I. (2.37)

For a given space curve Γ , recall that Cauchy problem for the quasilnear
equation (QL)

a(x, y, u) u x + b (x, y, u) uy = c(x, y, u) (QL)

consists of finding a solution u to the equation (QL) satisfying the Cauchy

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 41

condition
u ( f (s), g (s)) = h(s)
for s belonging to a subinterval of I .
Geometrically speaking, Cauchy problem consists of finding an integral
surface on which at least a piece of the datum curve Γ lies.
The hypotheses and notations introduced in Subsection 2.3.1 will be in
force throughout this section. Further, the projection of Γ to Ω2 is denoted
by Γ2 .
In this section, existence of a solution to the Cauchy problem for the
equation (QL) is proved under the assumption that base characteristic curves
and the curve Γ2 intersect transversally. The proof is constructive but for
a last minute appeal to inverse function theorem whose assertion is only
existential. The constructive part of the proof is known as the method of
characteristics. Cauchy problems for which the transversality condition
fails are also analyzed.
A brief outline of the section: A trial run of the method of character-
istics is carried out for linear and semilinear equations in Subsection 2.4.1.
After understanding the connection between an integral surface and the
family of characteristic curves associated to the quasilinear equation (QL)
in Subsection 2.4.2, a local existence result is proved in Subsection 2.4.3 un-
der a transversality condition. Failure of the transversality condition and
its consequences to Cauchy problems are analyzed in Subsection 2.4.4.
Reasons behind the local nature of the existence result are discussed in
Subsection 2.4.5. Cauchy problem for quasilinear equations in more than
two independent variables is discussed in Subsection 2.4.6.

2.4.1 Method of characteristics for linear, semilinear equations

The system of characteristic ODEs (chara.ODE) for linear equation (L)

and semilinear equation (SL) are

Linear equation Semilinear equation

dx dx
= a(x, y) = a(x, y)
dt dt
dy dy
= b (x, y) = b (x, y)
dt dt
dz dz
= c(x, y)z + d (x, y) = c(x, y, z)
dt dt

July 25, 2021 Sivaji

42 Chapter 2. First order PDE

Note that for linear equation (L) and semilinear equation (SL),the sys-
tem of characteristic equations decouple into two sets of equations: A sys-
tem of two equations for variables x and y which does not involve the vari-
able z, and a single equation for the variable z that can be solved given a
solution (x(t ), y(t )) of the first two equations. In other words, base charac-
teristics which are defined as the projections of characteristic curves to Ω2 ,
can be determined directly without the knowledge of characteristic curves
since the equations for x and y do not involve the variable z.
An idea to solve Cauchy problem for linear and semilinear equations

Step 1 Through every point ( f (s), g (s)) ∈ Γ2 , pass a base characteristic curve
i.e., for each s ∈ I find solutions (x(t ), y(t )) to the initial value prob-
lem
dx dy
= a(x, y), = b (x, y), x(0) = f (s), y(0) = g (s)
dt dt
and take their traces to get base characteristic curves through ( f (s), g (s)).
As we have assumed that a, b ∈ C 1 (Ω3 ), it follows that a, b ∈ C 1 (Ω2 ).
Therefore the functions a, b are locally Lipschitz on Ω2 and therefore
exactly one base characteristic passes through ( f (s), g (s)). Let the so-
lutions be denoted by x = X (t , s) and y = Y (t , s).

Step 2 A solution u can be determined along each of the base characteristics

found in Step 1. Setting Z(t , s) = u(X (t , s), Y (t , s)), equation for
Z(t , s) is given by

dz
= c (X (t , s), Y (t , s)) z + d (X (t , s), Y (t , s))
dt
in the case of linear equation, and by

dz
= c (X (t , s), Y (t , s), z)
dt
in the case of semilinear equation.
In the case of linear equation, the equation for z is linear in z with
variable coefficients, and hence Z(t , s) can be determined for all t
for which X (t , s), Y (t , s) were found. On the other hand, in the
case of semilinear equation, the equation for z is a nonlinear differen-
tial equation and hence Z(t , s) may not be found for all t for which
X (t , s), Y (t , s ) were found. This is because linear ODEs have global

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 43

solutions while nonlinear ODEs do not have global solutions in gen-

eral.
Step 3 At the end of Steps 1 and 2, we have a family of characteristic curves
parametrized by s ∈ I which pass through points of Γ . Each mem-
ber of the family (family is indexed by s ∈ I ) is described parametri-
cally by the 3-tuple of functions (X (t , s ), Y (t , s ), Z(t , s)) defined for
t belonging to a subinterval of �. Thus, we seem to have a ‘surface’
described parametrically by two parameters t and s. We know that
Z(t , s ) was designed to be the value of u (X (t , s), Y (t , s)) where u is
a solution to the PDE.
The question now is how to express a surface described by two param-
eters t , s as the graph of a function u : Ω2 → � which is differentiable
and is also a solution to the given PDEs (L) or (SL). Note that a curve
(a surface) which is parametrically described using smooth functions
need not be a regular (smooth) curve (respectively, a surface).
The question posed in the last paragraph will be answered in Subsec-
tion 2.4.3 where a detailed analysis in a more general setting of quasilinear
equation (QL) is presented.

2.4.2 Integral surface and Characteristic curves

In the last subsection, we presented ideas to solve Cauchy problem for lin-
ear and semilinear equations. We noted in Step 3 therein, that we can form
a parametrized surface. Recall that the parametrized surface was described
using the parameters s and t : the parameter s runs through the datum
curve while the parameter t runs through each of the characteristic curve
passing through a point on the datum curve (correspoinding to each of the
values of the parameter s). The following two questions arise naturally:
(i) Does the parametrized surface correspond to an integral surface ?
(ii) If the answer to (i) above is yes, then how to determine the solution
which defines the integral surface?
In this subsection we answer the question (i) above, and prove that an
integral surface may be constructed using the characteristic curves. We
answer the question (ii) in Subsection 2.4.3.

Theorem 2.29. Let D be an open and connected susbet of Ω2 , and S : z =

u(x, y) be a surface in �3 where u : D → � is a continuously differentiable
function. Then the following two statements are equivalent.

July 25, 2021 Sivaji

44 Chapter 2. First order PDE

(1). The surface S is an integral surface of the equation (QL).

(2). The surface S is a union of characteristic curves for the equation (QL).

Before we proceed to prove, it is important to understand that Theo-

rem 2.29 answers the question “Given a surface S : z = u(x, y) defined by
a function u ∈ C 1 (D), when is it an integral surface of the equation (QL)?”
The Theorem 2.29 also suggests that an integral surface may be constructed
using the family of characteristic curves.

Proof. Proof of (1) =⇒ (2): Mathematical formulation of the statement

(2) is
S = ∪ γP ,
P ∈S

where γP is a characteristic curve passing through P .

As a, b , c ∈ C 1 (Ω3 ), a unique characteristic curve passes through every
point P ∈ Ω3 , and since S ⊆ Ω3 ,we have proved that S ⊆ ∪P ∈S γP .
Thus it remains to show that ∪P ∈S γP ⊆ S. Let P ∈ S, and let γP be the
characteristic curve passing through P . Recall that γP is the trajectory of
the solution (x(t ), y(t ), z(t )) to characteristic system of ODEs (chara.ODE)
satisfying the initial condition (x(0), y(0), z(0)) = P , and defined on an in-
terval J . At this stage we do not know how big the characteristic curve is,
except that such a curve exists through P , near P .
We would like to prove that the entire characteristic curve lies on the
given integral surface S i.e., z(t ) = u (x(t ), y(t )) holds for all t ∈ J . But this
equation is meaningful only if (x(t ), y(t )) ∈ D for all t ∈ J , as u is defined
on D. Therefore entire characteristic curve through P may not lie on S.
However, since (x(t ), y(t )) ∈ D at t = 0, by continuity of the functions
x, y there exists a subinterval J � of J such that (x(t ), y(t )) ∈ D for all t ∈ J � ,
thus making the equation z(t ) = u (x(t ), y(t )) meaningful for all t ∈ J � .
Note that J � could be equal to J .
We are going to show that a part of the characteristic curve γP lies on S.
That is, for all t ∈ J � , the following equality holds

z(t ) = u (x(t ), y(t )) .

Define a function V : J � → � by

V (t ) = z(t ) − u (x(t ), y(t )) .

We will prove that V is the zero function on the interval J � , by showing

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 45

that V is a solution to an initial value problem whose only solution is the

zero function. Since we want V also to be a solution, the initial value
problem may be constructed using V itself. Indeed, firstly we will find an
ODE satisfied by V , and then prescribe an initial condition. To this end,
let us compute the derivative of V as follows.

dx dy
V � (t ) = z � (t ) − u x (x(t ), y(t )) (t ) − uy (x(t ), y(t )) (t )
dt dt
= c (x(t ), y(t ), z(t )) − u x (x(t ), y(t )) a (x(t ), y(t ), z(t ))
− uy (x(t ), y(t )) b (x(t ), y(t ), z(t ))
= c (x(t ), y(t ),V (t ) + u (x(t ), y(t )))
− u x (x(t ), y(t )) a (x(t ), y(t ),V (t ) + u (x(t ), y(t )))
− uy (x(t ), y(t )) b (x(t ), y(t ),V (t ) + u (x(t ), y(t ))) .

Thus the function V : J � → � is a solution to the ordinary differential

equation
U � = f (t , U ), (2.38)
where

f (t , U ) = c (x(t ), y(t ), U + u (x(t ), y(t )))

− u x (x(t ), y(t )) a (x(t ), y(t ), U + u (x(t ), y(t )))
− uy (x(t ), y(t )) b (x(t ), y(t ), U + u (x(t ), y(t ))) .

Thus we have got a formula for f (t , U ), but we need to find its domain
i.e., the values of (t , U ) for which formula for f (t , U ) is meaningful. Since
P ∈ Ω3 , by taking a subinterval of J � if necessary, we find that there exists
a δ > 0 such that f is meaningful on the domain J � × (−δ, δ). Note that
f ∈ C 1 (J � × (−δ, δ)) as the functions a, b , c ∈ C 1 (Ω3 ) and u ∈ C 1 (D).
Note that

f (t , 0) = c (x(t ), y(t ), u (x(t ), y(t )))

− u x (x(t ), y(t )) a (x(t ), y(t ), u (x(t ), y(t )))
− uy (x(t ), y(t )) b (x(t ), y(t ), u (x(t ), y(t )))
= 0,

since u : D → � is a solution to (QL) (since we are assuming that statement

(1) is true), and for each t ∈ J � , the point (x(t ), y(t )) ∈ D. Thus U (t ) ≡ 0
is a solution to the ODE (2.38). Also note that V has the property that

July 25, 2021 Sivaji

46 Chapter 2. First order PDE

V (0) = 0 as P ∈ S. Thus we are led to the initial condition for (2.38),

which is U (0) = 0. Thus the function V is a solution to the initial value
problem
U � = f (t , U ), U (0) = 0. (2.39)
Note that f is a locally Lipschitz function since f ∈ C 1 (J � × (−δ, δ)).
Thus the initial value problem (2.39) has a unique solution. As we already
know that zero function and the function V are solutions to the initial
value problem (2.39), we get V (t ) = 0 for all t ∈ J � . Thus we have shown
that � �
(x(t ), y(t ), z(t )) : t ∈ J � ⊆ S.
In other words, a part of the characteristiv curve γP through P lies on the
integral surface S. This completes the proof of (1) =⇒ (2).

Proof of (2) =⇒ (1): Given that the surface S : z = u(x, y) is a union

of characteristic curves of the equation (QL), we want to show that S is
an integral surface. In other words, we want to show that u solves the
equation (QL) on the domain D. Let (x, y) ∈ D, and let P denote the
point P (x, y, u(x, y)) which lies on the surface S. The function u satisfies
equation (QL) if and only if
� �
u x (x, y), uy (x, y), −1 · (a(P )), b (P ), c(P )) = 0 (2.40)

holds.
Since S is a union of characteristic curves, and P ∈ S, it follows that
there is a characteristic curve γP passing through � P , and γP ⊆ S. Since� nor-
mal to the surface S at P is in the direction of u x (x, y), uy (x, y), −1 , and
(a(x, y, u(x, y)), b (x, y, u(x, y)), c(x, y, u(x, y))) is the direction of tangent
to γP at P , we get (2.40). This completes the proof of (2) =⇒ (1).

A nice illustration of Theorem 2.29 is presented in Example 2.39, see

Figure 2.6 therein.
Remark 2.30. In the search for a solution to (QL), Theorem 2.29 is of great
help. The assertion (2) =⇒ (1) suggests that an integral surface (if exists)
may be constructed as a union of all characteristic curves. However, it
does not assert that the geometric object which is formed as a union of all
characteristic curves is necessarily a surface, leave alone it being an integral
surface. Mathematical analysis takes over in deciding whether the geomet-
rically constructed surface is an integral surface, see Subsection 2.4.3 for
details.

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 47

Mathematical statements when put in plain english sentences, could

lead to imprecise expressions which could be misunderstood. The state-
ment “If two integral surfaces intersect at a point P , then they intersect
along the entire characteristic curve through P " is a classic case of such
a sentence. A corrected formulation of this sentence is stated as Corol-
lary 2.31, and it follows immediately from Theorem 2.29.

Corollary 2.31. Let S1 and S2 be two integral surfaces for the equation (QL)
such that their intersection is non-empty. Let P ∈ S1 ∩ S2 . Then some part of
the characteristic curve passing through P lies on both S1 and S2 .

Another example of a misleading statement that is widely in use is “in-

tersection of two integral surfaces is a characteristic curve". The following
example illustrates that intersection of two distinct integral surfaces is not
necessarily a curve, and any curve on the intersection need not be a char-
acteristic curve. Note however, that through every point of intersection,
there passes a charachteristic curve which lies on both integral surfaces. A
corrected formulation of the original sentence is stated as Corollary 2.33.

Example 2.32. If u : D → � is a solution to the quasilinear equation (QL),

then so is v : D1 → � where D1 is a proper subset of D and v is defined by
v(x, y) = u(x, y). The integral surfaces S u : z = u(x, y) and Sv : z = v(x, y)
are different since domains of the functions u and v are different. But inter-
section of S u and Sv is Sv , which is an integral surface. Through every point
of Sv , we can find a curve which is not a characteristic curve, and another
that is a characteristic curve. This will not be the case if the two integral
surfaces intersect without touching, and is the content of Corollary 2.33.

Two surfaces in �3 are said to touch each other at a point if they have
a common tangent plane at that point. We have the following result con-
cerning curves lying on the intersection of two integral surfaces.

Corollary 2.33. If a curve γ lies on two integral surfaces for the equation
(QL) which intersect without touching at all points of γ , then γ is a character-
istic curve.

Proof. To prove that γ is a characteristic curve, we have to prove that at

any point P on the curve γ , the tangent has the characteristic direction.
The tangential direction to the curve γ at P belongs to the tangent planes

July 25, 2021 Sivaji

48 Chapter 2. First order PDE

to S1 as well as S2 at P , as γ lies on both S1 and S2 . If the direction of

the tangent is not along the characteristic direction (a(P ), b (P ), c(P )), then
it follows that the direction of the tangent to the curve γ at P , and the
characteristic direction (a(P ), b (P ), c(P )) form a linearly independent set
of a two dimensional tangent space. This implies that both the tangent
spaces are same, and hence the tangent planes to S1 and S2 at the point
P coincide. This contradiction proves that tangent to the curve γ at P is
proportional to the characteristic direction at P . Since P is an arbitrary
point on γ , it follows that γ is a characteristic curve.

We did not use Theorem 2.29 in proving Corollary 2.33, and hence a
question on its naming arises. Terming it as corollary is acceptable as one
can prove Corollary 2.33 using Theorem 2.29, and is left as an exercise
(Exercise 2.18).

Remark 2.34. Corollary 2.33 is concerned with two integral surfaces which
intersect without touching. Further if their intersection is a curve, by
Corollary 2.33, the curve must be a characteristic curve for equation (QL).
Thus, we come across such integral surfaces when a Cauchy problem has
more than one solution.
Recall that in Example 2.6, we found that the Cauchy problem has in-
ifinitely many solutions, which are of the form u(x, y) = e x T (y), where
T is a differentiable function such that T (0) = 1. Consider two integral
surfaces S : z = u(x, y) and S̃ : z = ũ(x, y), defined by u : �2 → � and
ũ : �2 → � given by

u(x, y) = e x+y , ũ(x, y) = e x−y .

The two integral surfaces intersect all along the datum curve. In Figure 2.4,
S is depicted in black colour, S̃ is shown in blue colour, and the datum
curve in red colour. In getting these integral surfaces, we used the idea that
if two integral surfaces touch each other at a point, then they have the same
tangent plane at the point of intersection i.e., normal (u x , uy , −1) for both
integral surfaces would be the same. In the examples, we made sure that
normals are not the same.
The reader is encouraged to construct a few examples of pairs of integral
surfaces as above for Example 2.6.

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 49

Figure 2.4. S : z = e x+y (in black), S̃ : z = e x−y (in blue), Datum curve (in red)

2.4.3 Solution via method of characteristics

In this subsection, we outline the method of characteristics and prove the
existence of a solution to the Cauchy problem for the quasilinear equation
(QL).
The method of characteristics is inspired from statement (2) of Theo-
rem 2.29. As the name suggests, the method will rely on using the charac-
teristic curves associated to (QL). Motivation for this method comes from
Theorem 2.29, see Figure 2.6 (in Example 2.39) for a simple illustration.
In the figure the curve in magenta is the datum curve Γ , and characteristic
curves passing through points of Γ are shown in black. The integral sur-
face (shown in blue) may be obtained as a union of all characteristic curves
passing through points of Γ as asserted by Theorem 2.29.
The following are the main steps in the method of characteristics to
solve the Cauchy problem for the quasilinear equation (QL):

Step 1: Passing characteristic curves through points of Γ .

Step 2: Defining a candidate solution u using inverse function theorem.
Step 3: Establishing that u defined in Step 2 solves Cauchy problem.

We start implementing the strategy outlined above. Carrying out Step

2 necessitates assuming additional hypothesis on the data in the Cauchy

July 25, 2021 Sivaji

50 Chapter 2. First order PDE

problem, while Steps 1 and 3 do not pose any new difficulties.

Step 1. Passing characteristic curves through points of Γ .

The requirement of finding an integral surface containing the datum
curve Γ means that we need to weave a surface around Γ in such a way
that the resulting surface is an integral surface for the equation (QL). A
surface could be woven around Γ by passing curves through each point of
Γ . Further if these curves are characteristic curves for the equation (QL),
then the surface is expected to be an integral surface for (QL) in view of
Theorem 2.29.
Let s ∈ I be fixed, and let P ( f (s), g (s), h(s)) be a point on the datum
curve Γ , and let

x = X (t , s), y = Y (t , s), z = Z(t , s) (2.41)

represent solutions to the characteristic system of ODE given by

dx dy dz
= a(x, y, z), = b (x, y, z), = c(x, y, z) (chara.ODE)
dt dt dt
satisfying the initial conditions

X (0, s ) = f (s), Y (0, s) = g (s), Z(0, s) = h(s), (2.42)

defined for t belonging to an interval J s in �, and such that 0 ∈ J s . The exis-

tence of functions in (2.41) with the stated properties follows by applying
Cauchy-Lipschitz-Picard theorem. Figure 2.5 is an illustration of Step 1.

Step 2. Finding a candidate solution.

At the end of Step 1, we have a mathematical object given by equation
(2.41), which is expected to be a surface, namely,

x = X (t , s), y = Y (t , s), z = Z(t , s), (2.41)

described by two parameters: s ∈ I (parametrizes points on Γ ) and t ∈ J s

(parametrizes the characteristic curve through ( f (s), g (s), h(s)) ∈ Γ ).
At best, the parametric surface given by equation (2.41) may be a regular
surface and as a consequence the parametric surface looks like graph near
each of its points. Since our interest lies in finding an integral surface, the
graph must be of a specific form, namely, z = u(x, y). In other words,

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 51

Figure 2.5. Weaving an integral surface using characteristic curves (in blue) around the Da-
tum curve (in red)

“write the third coordinate as a function of the first two coordinates”

This suggests the following strategy for finding u:
(i) Express t and s as functions of x and y, using x = X (t , s), y = Y (t , s).
(ii) Then substitute for t and s in z = Z(t , s) and define u by

u(x, y) := Z (t (x, y), s(x, y))

To implement (i) above, we need to solve for (t , s) in terms of (x, y)

from x = X (t , s), y = Y (t , s). Recall that these equations represent base
characteristic curves. This suggests applying inverse function theorem to
the function
(t , s) �−→ (X (t , s ), Y (t , s)) . (2.43)
In order to apply the inverse function theorem, we need to verify its hy-
pothesis which requires answers to the following questions:
(i) What is the domain of the function?
(ii) Is the function continuously differentiable?
(iii) Is the Jacobian condition satisfied?
The function given in (2.43) by its construction is defined for (t , s) be-
longing to a subset A of the cylinder � × I having the property that, for

July 25, 2021 Sivaji

52 Chapter 2. First order PDE

each fixed s ∈ I , the s-cross section of A is an interval J s . Thus, it is not even

clear if the set A is an open subset of �2 . In spite of this, we may assume
without loss of generality that A is equal to J × I where J is a subinterval
of � due to the following reasons: (i) Lemma 2.19 asserts that we may take
J = �. (ii) by restricting to a small piece of Γ , we can always take J s to be
independent of s belonging to a subinterval of I [56], and is not really a
restriction as we are going to prove a local result.
Thus there is no loss of generality in assuming that the function in (2.43)
is defined for (t , s) ∈ J × I . Moreover the function belongs to the space
C 1 (J × I ), which follows from the differentiable dependence of solutions
to initial value problems on the parameters in the theory of ordinary dif-
ferential equations (for instance, see [56]).

Now we are interested in the invertibility of the function

Φ : J × I → Ω2
(t , s) �−→ (X (t , s ), Y (t , s)) . (2.44)

To apply inverse function theorem, the Jacobian J (t , s) is required to be

non-zero, where J (t , s ) is given by
� �
∂ (X , Y ) � X (t , s ) X (t , s) �
� t s �
J (t , s) := (t , s) = � �.
∂ (t , s ) � Y (t , s) Y (t , s) �
t s

Since we do not know the function Φ given in (2.44) explicitly, we can-

not compute J (t , s) at an arbitrary point (t , s) ∈ J × I . However, using
equations (chara.ODE) and (2.42) we can compute J (0, s) and is given by

� � � �
� X (0, s) X (0, s) � � a ( f (s), g (s ), h(s)) f � (s) �
� t s � � �
J (0, s) = � �=� �.
� Y (0, s) Y (0, s) � � b ( f (s), g (s ), h(s)) g � (s) �
t s

If we assume that J (0, s0 ) �= 0 at an s0 ∈ I , then we can apply the (local)

inverse function theorem [5]. This assumption is known as transversality
condition, which is defined below.

Definition 2.35 (Transversality condition). The quasilinear equation

(QL) and the initial curve Γ given by (2.37) are said to satisfy transversality
condition at s ∈ I if the base characteristic curve corresponding to a charac-
teristic curve passing through the point ( f (s), g (s), h(s)) ∈ Γ , intersects the

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 53

projection Γ2 of the datum curve Γ at the point ( f (s), g (s )) non-tangentially

i.e., the tangent lines do not coincide. Analytically this condition translates to
� �
� a ( f (s), g (s ), h(s)) f � (s) �
� �
� � �= 0.
� b ( f (s), g (s ), h(s)) g (s) �
�

In other words, transversality condition rules out the possibility of base

characteristic curves touching the projected datum curve Γ2 , or intersecting
along a piece of it. It justifies the use of the word ‘transversal’, which means
‘to cut’.
Since conclusions from inverse function theorem are local in nature,
we may assume J (0, s0 ) �= 0. Note that s = s0 corresponds to the point
( f (s0 ), g (s0 ), h(s0 )) ∈ Γ . By inverse function theorem, there exist

(i) an open subset E of J ×I containing the point (0, s0 ) , and an open sub-
set F of Ω2 containing the point (X (0, s0 ), Y (0, s0 )) i.e., ( f (s0 ), g (s0 )),
(ii) a continuously differentiable function Ψ : F → E such that

Ψ ◦ Φ = IE , Φ ◦ Ψ = IF , (2.45)

where IE and IF are the identity functions defined on E and F respec-

tively.

Denoting Ψ(x, y) = (T (x, y), S(x, y)), for (t , s) ∈ E, the equations (2.45)
yield
t = T (x, y), s = S(x, y).
Recall that Z(t , s) was expected to be the value of a solution at the point
(X (t , s), Y (t , s)), we are now in a position to define a candidate for being a
solution to the Cauchy problem (QL)-(2.37).
Define a function

u : F →� given by u(x, y) = Z (T (x, y), S(x, y)) . (2.46)

As the function u defined above is a composition of two continuously dif-

ferentiable functions, it follows that u ∈ C 1 (F ).

Step 3. Candidate solution is indeed a solution.

We now show that the function u defined in Step 2 is a solution to the
equation (QL).

July 25, 2021 Sivaji

54 Chapter 2. First order PDE

Note that from the definition of u in (2.46) we have for (t , s) ∈ E,

Z(t , s ) = u (X (t , s), Y (t , s)) . (2.47)

Differentiating the equation (2.47) w.r.t. the variable t , we get

Z t (t , s ) = u x (X (t , s ), Y (t , s )) X t (t , s)+ uy (X (t , s), Y (t , s)) Y t (t , s ). (2.48)

Since X (· , s), Y (· , s), Z(· , s) are solutions to the characteristic system of

ODEs (chara.ODE), the equation (2.48) becomes

c (X (t , s), Y (t , s), Z(t , s)) = u x (X (t , s), Y (t , s)) a (X (t , s), Y (t , s), Z(t , s))
+ uy (X (t , s), Y (t , s )) b (X (t , s), Y (t , s), Z(t , s)) .
(2.49)

In terms of (x, y) ∈ F , the equation (2.49) reads as

c (x, y, u(x, y)) = u x (x, y) a (x, y, u(x, y)) + uy (x, y) b (x, y, u(x, y)) .

This means that u : F → � solves the equation (QL). The function u

satisfies the Cauchy data i.e., u( f (s), g (s)) = h(s), follows on substituting
t = 0 in the equation (2.47).
In summary, we have obtained an integral surface defined by u : F → �
given by
S : z = u(x, y) (2.50)
containing a piece of the datum curve Γ containing the point ( f (s0 ), g (s0 ), h(s0 )),
in fact that much part of Γ for which ( f (s), g (s)) ∈ F . If we could apply
global inverse function theorem, we would have got a solution defined on
a subset of Ω2 such that the corresponding integral surface would contain
the entire datum curve Γ . Note that even if global inverse function theo-
rem were applicable, the solution may still be not defined on whole of Ω2 ,
some reasons for the same will be explored in Subsection 2.4.5.
Thus we have proved the existence part of the following theorem.

Theorem 2.36 (Local existence and uniqueness theorem).

Hypotheses

1. Let Ω3 ⊆ �3 be an open connected domain, and a, b , c ∈ C 1 (Ω3 ).

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 55

2. Consider the following Cauchy problem for a quasilinear PDE given by

a(x, y, u) u x + b (x, y, u) uy = c(x, y, u) (QL)

and the Cauchy data

Γ : x = f (s), y = g (s), z = h(s), s ∈ I, (2.37)

where I ⊆ � is an interval, f , g , h ∈ C 1 (I ) such that ( f � (s))2 +( g � (s))2 �=

0 for every s ∈ I , and such that Γ ⊂ Ω3 .
3. Let P0 (x0 , y0 , z0 ) = ( f (s0 ), g (s0 ), h(s0 )) ∈ Γ , where s0 ∈ I . Assume that
transversality condition holds at s0 . That is,
� �
� a ( f (s ), g (s ), h(s )) f � (s ) �
� 0 0 0 0 �
� � �= 0. (2.51)
� b ( f (s ), g (s ), h(s )) g � (s ) �
0 0 0 0

Conclusion
(i) The Cauchy problem (QL)-(2.37) has a solution defined on a neighbour-
hood of the point (x0 , y0 ) ∈ Ω2 .
(ii) Let D1 and D2 be open neighbourhoods of the point (x0 , y0 ), and let u1 :
D1 → � and u2 : D2 → � be solutions to the Cauchy problem (QL)-
(2.37). Then u1 ≡ u2 on some subset of D1 ∩ D2 containing the point
(x0 , y0 ). That is, solution to the Cauchy problem is locally unique, near
the datum curve Γ .

Proof. Steps 1 through 3 constitute a proof of assertion in (i). Let us

turn to the proof of (ii) now. Given that S1 : z = u1 (x, y) and S2 : z =
u2 (x, y) are integral surfaces containing the point P0 (x0 , y0 , z0 ) ∈ Γ . Thus
P0 ∈ Γ ∩ S1 ∩ S2 , and as a consequence a part of Γ , say Γ � , lies on S1 ∩ S2 .
By Corollary 2.31, some part of each of the characteristic curves of equa-
tion (QL) passing through points of Γ � will lie on S1 as well as on S2 . Due
to the transversality condition (2.51), as seen in Step 2, any point on a
characteristic curve has the form (x, y, u1 (x, y)) as well as (x, y, u2 (x, y))
for (x, y) ∈ F ∩ D1 ∩ D2 . This proves that

S1 ≡ S2 in a neighbourhood of P0 ,

and hence u1 ≡ u2 on some neighbourhood of (x0 , y0 ).

The method of proof for showing the existence of solutions to Cauchy
problem is known as the method of characteristics.

July 25, 2021 Sivaji

56 Chapter 2. First order PDE

Remark 2.37. (i) Recall that a function u was defined in the equation
(2.46) which was later shown to be a soluton to the equation (QL) in
Step 3 of the proof of existence theorem. We give a geometric proof
of the same.
The surface S given by the equation (2.50) has parametric equations

x = X (t , s), y = Y (t , s), z = Z(t , s), (t , s) ∈ E.

The curves s = constant are the characteristic curves of equation (QL),

and at any point on these curves the tangential direction is given by
(X t , Y t , Z t ). Since characteristic curves lie on S, (X t , Y t , Z t ) is a tan-
gential direction to S as well. Hence the normal to S is perpendicular
to the direction (X t , Y t , Z t ), which is nothing but the characteristic
direction. This implies that S is an integral surface.
(ii) Any proof of the uniqueness assertion in Theorem 2.36 is incomplete,
if the transversality assumption (2.51) is not used. For, the uniqueness
fails with the failure of transversality condition (see Lemma 2.49).
(iii) Even though the condition ( f � (s))2 + ( g � (s))2 �= 0 for every s ∈ I was
part of hypotheses for Theorem 2.36 (due to local nature of the re-
sult it is enough to assume this condition at s = s0 ), it follows from
another hypothesis in Theorem 2.36, namely the transversality con-
dition (2.51).

Remark 2.38. At the end of Step 1 in the proof of Theorem 2.36, a parametrized
surface (2.41) was obtained as the union of characteristic curves. The func-
tions describing the parametrized surface are continously differentiable
functions of the parameters t ∈ J and s ∈ I . This is a consequence of dif-
ferentiable dependence of solutions to initial value problems in the theory
of ODEs: smoothness in t is due to the smoothness of the characteristic
vector field defined by (QL); and the smoothness in s is through the initial
conditions for characteristic system of ODEs which in turn are coming
from the description of the datum curve, as noted in the proof of Theo-
rem 2.36.
As shown in Step 2 in the proof of Theorem 2.36, the parametrized sur-
face (2.41) represents an integral surface locally whenever the transversality
condition is satisfied.
Thus the representation (2.41) misleads us to believe that Cauchy prob-
lems always possess a solution. As we shall see later, in Examples 2.40, 2.42,
and 2.44, parametrized surface (2.41) have singularities at points of the da-
tum curve which correspond to failure of the transversality condition.

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 57

Rest of this subsection is devoted to solving a few Cauchy problems for

(QL) using the method of characteristics.

Example 2.39. Consider the Cauchy problem for the equation

u x = 0,

where the Cauchy data is given by u(0, y) = sin y for y ∈ �.

Let us parametrize the given Cauchy data as

Γ : x = 0, y = s, z = sin s, s ∈ �.

The characteristic system of ODE for the given equation is

dx dy dz
= 1, = 0, = 0.
dt dt dt
Solution of the characteristic system of ODE satisfying the initial con-
ditions
x(0) = 0, y(0) = s, z(0) = sin s
is given by

x = X (t , s) = t , y = Y (t , s) = s, z = Z(t , s) = sin s.

From the first two equations, we get

s = S(x, y) = y, t = T (x, y) = x.

Thus, the solution is given by

u(x, y) = Z(T (x, y), S(x, y)) = sin y,

which is defined for all (x, y) ∈ �2 . The integral surface is depicted in

Figure 2.6.

Example 2.40. Consider the Cauchy problem for the equation

u x + uy + u = 1,

July 25, 2021 Sivaji

58 Chapter 2. First order PDE

Figure 2.6. Example 2.39: Integral surface (in blue), datum curve (in magenta), characteristic
curves (in black)

where the Cauchy data is given by u(x, x + x 2 ) = sin x for x > 0.

Let us parametrize the given Cauchy data as

Γ : x = s, y = s + s 2 , z = sin s, s > 0.

The characteristic system of ODE for the given equation is

dx dy dz
= 1, = 1, = 1 − z.
dt dt dt
Solution of the characteristic system of ODE satisfying the initial con-
ditions
x(0) = s, y(0) = s + s 2 , z(0) = sin s
is given by

x = X (t , s) = t + s, y = Y (t , s) = t + s + s 2 , z = Z(t , s) = 1−e −t +e −t sin s

From the first two equations, we get

� �
s = S(x, y) = y − x, t = T (x, y) = x − y − x.

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 59

Figure 2.7. Example 2.40: Integral surface (in blue), datum curve (in green)

Thus, the solution is given by

� � �� �
y−x−x y−x−x
u(x, y) = Z(T (x, y), S(x, y)) = 1 − e +e sin y−x ,

which is defined on the domain

� �
(x, y) ∈ �2 : y > x .

The integral surface is depicted in Figure 2.7.

Example 2.41. Consider the Cauchy problem for the linear partial differ-
ential equation

x u x − y uy = u − y, for x ∈ �, y > 0

where the Cauchy data is given by u(x, y) = y on the parabola x = y 2

(y > 0).
Let us parametrize the given Cauchy data as

Γ : x = s 2 , y = s, z = s, s > 0.

July 25, 2021 Sivaji

60 Chapter 2. First order PDE

The characteristic system of ODE for the given equation is

dx dy dz
= x, = −y, = z − y.
dt dt dt
Solution of the characteristic system of ODE satisfying the initial con-
ditions
x(0) = s 2 , y(0) = s, z(0) = s
is given by
s � −t �
x = X (t , s) = s 2 e t , y = Y (t , s) = s e −t , z = Z(t , s) = e + et .
2
From the first two equations, we get

s = S(x, y) = (xy)1/3 , e −T (x,y) = x −1/3 y 2/3 , e T (x,y) = x 1/3 y −2/3 .

Thus, the solution is given by

1� �
u(x, y) = Z(T (x, y), S(x, y)) = y + x 2/3 y −1/3 ,
2
which is defined on the domain
� �
(x, y) ∈ �2 : x > 0, y > 0 .

The integral surface is shown in Figure 2.8.

Example 2.42. Consider the Cauchy problem for the linear partial differ-
ential equation
2
u x + 3y 3 uy = 2, (2.52)
where the Cauchy data is given by u(x, 1) = 1 + x for all x ∈ �.
Let us parametrize the given Cauchy data as

Γ : x = s, y = 1, z = 1 + s, s ∈ �.

The characteristic system of ODE for the given equation is

dx dy 2 dz
= 1, = 3y 3 , = 2.
dt dt dt
Sivaji IIT Bombay
2.4. Cauchy problem for quasilinear PDE 61

Figure 2.8. Example 2.41: Integral surface (in blue), datum curve (in green)

Solution of the characteristic system of ODE satisfying the initial con-

ditions
x(0) = s, y(0) = 1, z(0) = 1 + s
is given by

x = X (t , s) = t + s, y = Y (t , s) = (t + 1)3 , z = Z(t , s) = 2t + s + 1.

From the first two equations, we get

t = T (x, y) = y 1/3 − 1, s = S(x, y) = x + 1 − y 1/3 .

Thus, the solution is given by

u(x, y) = Z(T (x, y), S(x, y)) = x + y 1/3 ,

which is defined on the domain

� �
(x, y) ∈ �2 : y > 0 .

Note that the integral surface contains the entire datum curve Γ . The inte-
gral surface is shown in Figure 2.9.

July 25, 2021 Sivaji

62 Chapter 2. First order PDE

Figure 2.9. Example 2.42: Views of integral surface (in green), datum curve (in magenta)

Remark 2.43. In this remark, we record a few observations on the Cauchy

problem presented in Example 2.42 and its solution.
1. The function u(x, y) = x + y 1/3 is differentiable at all points of �2
excluding the points on x-axis. The integral surface depicted in Fig-
ure 2.9 develops a singularity as we approach the line x = z, y =
0. Observe that graph is steeping near y = 0 indicating the non-
differentiability of u on x-axis.

2. The Jacobian J (t , s), which plays an important role in applying in-

verse function theorem, is given by
� �
� 1 1 �
� �
J (t , s) = � � = −3(t + 1)2 ,
� 3(t + 1)2 0 �

which vanishes at all points (t , s) for which t = −1. These are pre-
cisely the points on the line x = z, y = 0. This observation shows
that one may not be able to solve the Cauchy problem if the datum
curve lies on the line x = z, y = 0, and this problem is analyzed in
Example 2.53.
3. The PDE (2.52) is defined for all (x, y) ∈ �2 , however the Cauchy
problem does not have a solution defined on �2 . This example shows
that Cauchy problems for even linear equations need not have solu-
tions defined on the domain Ω2 .
Contrast this situation with an IVP associated to a linear ODE which
always admits global solutions. Thus theory of linear PDEs is more
complicated than that of linear ODEs.

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 63

The method of characteristics was a constructive procedure to a large

extent, except for the existential nature of conclusions from the inverse
function theorem. Despite this deficiency, we could still obtain explicit so-
lutions in all the examples so far. However, the following example shows
that finding an explicit expression for solution is an impossible task in gen-
eral, and in such cases one tries to find an approximate solution like a nu-
merical solution.
Example 2.44. Consider the Cauchy problem for the equation

(sin u) u x + uy = 0,

where the Cauchy data is given by u(0, y) = y for all y ∈ �.

Let us parametrize the given Cauchy data as

Γ : x = 0, y = s, z = s, s ∈ �.

The characteristic system of ODE for the given equation is

dx dy dz
= sin z, = 1, = 0.
dt dt dt
Solution of the characteristic system of ODE satisfying the initial con-
ditions
x(0) = 0, y(0) = s, z(0) = s
is given by

x = X (t , s) = t sin s, y = Y (t , s) = t + s, z = Z(t , s) = s. (2.53)

The constructive part of finding a solution is over, and we cannot obtain

explicitly the functions S(x, y) and T (x, y), from the first two equations of
(2.53), as described in Step 2 of the proof of the local existence theorem.
The only hope is to conclude the existence of a solution to the Cauchy
problem, by showing that the transversality condition is satisfied.
Note that the Jacobian
� �
� sin s 0 �
� �
J (0, s) := � � = sin s
� 1 1 �

is zero whenever s belongs to the set of all integral multiples of π (observe

that all points of this set are isolated points), and non-zero otherwise. Thus

July 25, 2021 Sivaji

64 Chapter 2. First order PDE

by local existence and uniqueness theorem, given any point P0 (0, y0 , y0 ) on

Γ where y0 is not an integral multiple of π there exists an integral surface
for the given PDE containing P0 and a piece of Γ .

Remark 2.45. In this remark, we record a few observations on the Cauchy

problem presented in Example 2.44.
Observe from (2.53) that x, y, z satisfy an implicit relation φ(x, y, z) = 0
where
φ(x, y, z) = (y − z) sin z − x,
and hence the following implicit relation is satisfied by a possible solution
u to the Cauchy problem:

(y − u) sin(u) − x = 0. (2.54)

Observe that the point (0, y0 , y0 ) is a solution to (2.54). We also have

∂φ
(0, y0 , y0 ) = − sin y0 ,
∂z
which is not equal to zero whenever y0 is not an integral multiple of π.
Thus we are in a position to apply implicit function theorem which asserts
that u can be expressed as a function (x, y) in a neighbourhood of (x0 , y0 )
whenever y0 is not an integral multiple of π.
Thus we conclude the existence of a local solution to the Cauchy prob-
lem at all points on the datum curve except at the points (0, mπ, mπ)
where m ∈ �.

Example 2.46. This example illustrates the local nature of solutions guar-
anteed by the existence and uniqueness theorem, namely, the Cauchy prob-
lem may not have a solution that is defined on all of Ω2 . In this example
Ω2 is the domain �2 \ { (0, 0) }.
Two Cauchy problems for the linear partial differential equation

−y u x + x uy = 0 (2.55)

will be considered. Note that the equation is defined for all (x, y) ∈ �2
such that (x, y) �= (0, 0).

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 65

The characteristic system of ODE for the given equation is

dx dy dz
= −y, = x, = 0. (2.56)
dt dt dt
From equations (2.56), it follows that the base characteristics are the family
of circles x 2 +y 2 = C 2 (indexed by C > 0) centered at the origin, and having
radius C . Note that the third equation in (2.56) implies that any solution
to the equation (2.55) is constant on each member of the family of base
characteristics.

(a) Cauchy data 1 is given by u(x, 0) = x for x ∈ �\{0}. Let us parametrize

the given Cauchy data as

Γ : x = s, y = 0, z = s, s ∈ � \ {0}.

Solution of the characteristic system of ODE (2.56) satisfying the ini-

tial conditions
x(0) = s, y(0) = 0, z(0) = s
is given by

x = X (t , s) = s cos t , y = Y (t , s) = s sin t , z = Z(t , s) = s.

From the first two equations, we get for s > 0

� y
s = S(x, y) = x 2 + y 2 , t = T (x, y) = arctan ,
x
and for s < 0
� y
s = S(x, y) = − x 2 + y 2 , t = T (x, y) = arctan .
x
In both the cases, the functions S and T are not defined at points (x, y)
where x = 0. Since z = Z(t , s) = s, the solution is given by

u(x, y) = Z(T (x, y), S(x, y)) = S(x, y),

which leads us to the following expression for u:

� �
− x 2 + y 2 if x < 0, y ∈ �,
u(x, y) = �
x 2 + y 2 if x > 0, y ∈ �.
Note that the function u is defined on the open set �2 \ { y-axis }, and

July 25, 2021 Sivaji

66 Chapter 2. First order PDE

all points of Γ lie on the corresponding integral surface z = u(x, y).

Observe that the function u does not have a continuous extension to
�2 .

� hand the function v : � \ { (0, 0) } → � defined by

2
On the other
v(x, y) = x 2 + y 2 is a solution to the equation (2.55). However v is
not a solution to the Cauchy problem. For, for x ∈ � \ { 0}, v(x, 0) =
|x| on one hand, and v(x, 0) = x on the other. This is not possible for
x < 0.
In either of the cases presented above, a solution to the Cauchy prob-
lem that is defined on �2 \ { (0, 0) } does not exist.

(b) Cauchy data 2 is given by u(x, 0) = h(x) for x ∈ �, where h : � → �

is an even function such that h ∈ C 1 (�). Following the same proce-
dure as in the case of Cauchy data 1, the solution in the present case
is given by �� �
u(x, y) = h x2 + y2 ,
defined for (x, y) ∈ �2 \ { (0, 0) }. Note that u(x, 0) = h(|x|) = h(x)
since h is an even function. Thus, the Cauchy problem has a solution
defined on �2 \ { (0, 0) }.

2.4.4 Failure of the transversality condition and consequences

In the proof of Theorem 2.36, the transversality condition (2.51) played an

important role in proving both existence and uniqueness assertions. Thus
it is natural to ask “What happens if transversality condition is not satis-
fied?" We expect that Cauchy problem may not have a solution or it may
have multiplicity of solutions or it may still have a unique solution.
In the next result we prove that the datum curve must necessarily have
a characteristic direction if the transversality condition is not satisfied and
a solution to the Cauchy problem is expected.

Lemma 2.47. Let z = u(x, y) be an integral surface corresponding to the

Cauchy problem (QL)-(2.37) containing a part Γ � of the datum curve Γ (corre-
sponding to s ∈ I � , where I � is a subinterval of I ). Let P0 ( f (s0 ), g (s0 ), h(s0 )) ∈

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 67

Γ � where s0 ∈ I � . Denote
� �
� a ( f (s), g (s ), h(s)) f � (s) �
� �
J (0, s) := � �.
� b ( f (s), g (s ), h(s)) g � (s) �

(i) If J (0, s0 ) = 0, then the datum curve Γ has characteristic direction at s0 .

That is,
f � (s0 ) g � (s0 ) h � (s0 )
= = . (2.57)
a(P0 ) b (P0 ) c(P0 )
(ii) If J (0, s) = 0 for every s ∈ I � , then the curve Γ � is a characteristic curve
for the equation (QL).

Proof. Note that J (0, s0 ) = 0 if and only if

b (P0 ) f � (s0 ) − a(P0 ) g � (s0 ) = 0.

Since u is a solution to the quasilinear equation (QL), and h(s) = u( f (s), g (s))
holds for s ∈ I � , we have

c(P0 ) g � (s0 ) − b (P0 )h � (s0 )

� �
= a(P0 )u x ( f (s0 ), g (s0 )) + b (P0 )uy ( f (s0 ), g (s0 )) g � (s0 )
� �
−b (P0 ) u x ( f (s0 ), g (s0 )) f � (s0 ) + uy ( f (s0 ), g (s0 )) g � (s0 )
� �
= a(P0 ) g � (s0 ) − b (P0 ) f � (s0 ) u x ( f (s0 ), g (s0 )) = 0.

This proves (2.57). That is, Γ has characteristic direction at P0 . From

(i), we conclude that Γ � is a characteristic curve if J (0, s) = 0 for all s ∈ I � .
This completes the proof of lemma.

We will now analyze the situation where J (0, s) = 0 for every s ∈ I .

Once again Lemma 2.47 asserts that the datum curve Γ must be a character-
istic curve if there is a solution to the Cauchy problem. In the next result,
we will prove that if Γ is a characteristic curve, and J (0, s) = 0 for every
s ∈ I , then there exists infinitely many solutions to the Cauchy problem.

Definition 2.48 (Characteristic Cauchy problem). The Cauchy problem

(QL)-(2.37) where the datum curve Γ is a characteristic curve for the equation
(QL), is called a Characteristic Cauchy problem.

Lemma 2.49. Consider the Cauchy problem (QL)-(2.37). Further assume

July 25, 2021 Sivaji

68 Chapter 2. First order PDE

that for every s ∈ I ,

� �
� a ( f (s), g (s ), h(s)) f � (s) �
� �
J (0, s) := � � = 0.
� b ( f (s), g (s ), h(s)) g � (s) �

(i) Let P0 be a point on Γ . If Γ is a characteristic curve, then there exists an

infinite number of integral surfaces which contain a part of the initial
curve Γ containing the point P0 .
(ii) If Γ does not have a characteristic direction at any of its points, then the
Cauchy problem does not admit a solution.

Proof. Proof of (i): Denote the tangential direction to the datum curve Γ
at the point P0 by v := (v1 , v2 , v3 ). Choose any direction (i.e., a non-zero
vector) w := (w1 , w2 , w3 ) such that the set { (v1 , v2 ), (w1 , w2 )} is linearly in-
dependent in �2 , and choose any curve � Γ ⊂ Ω3 through P0 such that the
�
tangent to Γ at P0 is in the direction of w. As a consequence, J�Γ (P0 ) �= 0.
Then, by Theorem 2.36, there exists a unique integral surface S� for equa-
tion (QL) containing a piece of � Γ in a neighbourhood of P0 . Then the
integral surface S� also contains a part of Γ containing P0 , as Γ is a charac-
teristic curve through P0 . Note that the tangent plane at P0 to the integral
surface S� contains the two vectors v and w. In fact, the tangent plane is
given by the subspace of �3 spanned by the two vectors v and w translated
to the point P0 .
Since there are an infinite number of directions w having the property
that the set { (v1 , v2 ), (w1 , w2 )} is linearly independent in �2 , we get an infi-
nite number of integral surfaces such that a part of Γ lies on each of them.
All of these integral surfaces are pairwise distinct since the tangent planes
to the integral surfaces at the point P0 are distinct.
Proof of (ii): Assume that there exists a solution to the Cauchy problem
(QL)-(2.37). By Lemma 2.47, the datum curve Γ must be a characteristic
curve. This contradicts our assumption on Γ . Thus we conclude that the
Cauchy problem does not admit a solution. This completes the proof of
(ii).

Remark 2.50 (On the three examples of Subsection 2.2.1).

Let us return to the three examples that we considered in Subsection 2.2.1.
We have the following observations regarding them.
(i) In Example 2.5, the Jacobian J (0, s) is given by

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 69

� �
� 1 0 �
� �
J (0, s) := � � = 1,
� 0 1 �
which is non-zero for every s ∈ �. In other words, the transversal-
ity condition is satisfied. Thus Example 2.5 confirms the assertion of
Theorem 2.36 that the Cauchy problem has a unique solution.
(ii) In Example 2.6, the Jacobian J (0, s) is given by
� �
� 1 1 �
� �
J (0, s) := � � = 0,
� 0 0 �
for every s ∈ �. Note that at every point P ( f (s), g (s), h(s)) of Γ (i.e.,
for every s ∈ �), the following equalities hold
� �
(a(P ), b (P ), c(P )) = f � (s), g � (s), h � (s) = (1, 0, c e c s ).
That is, the characteristic direction coincides with the tangential direc-
tion for datum curve. Thus the datum curve is a characteristic curve,
and hence by Lemma 2.49 the characteristic Cauchy problem has in-
finitely many solutions. This was already observed in Example 2.6.
(iii) In Example 2.7, the Jacobian J (0, s) is given by
� �
� 1 1 �
� �
J (0, s) := � � = 0,
� 0 0 �
for every s ∈ �. However, the datum curve is not a characteristic
curve and hence it has no solutions. Even though the transversality
condition is not satisfied for both Examples 2.6 and 2.7, the datum
curve turns out to be a characteristic curve in the case of Example 2.6
and not in the case of Example 2.7. This results in an infinite number
of solutions in the former, while solutions do not exist in the later.

As noted in Remark 2.50, in the three examples of Cauchy problems

presented in Subsection 2.2.1 either J (0, s) �= 0 for every s ∈ � or J (0, s) = 0
for every s ∈ �, for which Lemma 2.49 is also applicable. The observations
in the three examples were tallied with the assertions of the lemma in Re-
mark 2.50.

Remark 2.51 (Possibilities in failure of transversality condition).

Theorem 2.36 asserted the existence of a unique solution for Cauchy

July 25, 2021 Sivaji

70 Chapter 2. First order PDE

problem if the transversality condition (2.51) (denote the Jacobian in (2.51)

by J (0, s0 )) is satisfied at a point on the datum curve corresponding to the
parametric value s0 . Note that if J (0, s0 ) �= 0, then J (0, s) �= 0 for s belonging
to an interval containing s0 . Given the local nature of the assertions in
Theorem 2.36, the following cases exhaust all the possibilities in failure of
transversality condition at s0 .

(i) There exists an interval containing s0 such that the Jacobian J (0, s)
vanishes at all points of the interval.
(ii) The point s0 is an isolated root of the equation J (0, s) = 0 i.e., there
exists an interval containing s0 such that the Jacobian J (0, s) is never
zero on the interval except at s0 .
(iii) The point s0 is not an isolated root of the equation J (0, s) = 0 i.e., there
exists a sequence (sn ) such that J (0, sn ) = 0, and sn → s0 as n → ∞.

Recall that Lemma 2.49 completely answered the questions of existence

and uniqueness in the situation presented in (i) above. A general result
similar to Lemma 2.49 does not hold for the situations described in (ii) and
(iii) above (see Example 2.52, Exercise 2.10).

In the following example, the transversality condition is not satisfied

exactly at one point on the datum curve.

Example 2.52. In this example, we consider the Cauchy problem for the
quasilinear partial differential equation

x 2 u x − y u uy = 0,

satisfying the condition u(x, x) = x ln x for all x ∈ (0, ∞).

Let us parametrize the given Cauchy data as

Γ : x = s, y = s, z = s ln s, s ∈ (0, ∞).

The characteristic system of ODE for the given equation is

dx dy dz
= x 2, = −y z, = 0.
dt dt dt
Solution of the characteristic system of ODE satisfying the initial con-
ditions
x(0) = s, y(0) = s, z(0) = s ln s

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 71

is given by
s
x = X (t , s) = , y = Y (t , s) = s e −(s ln s )t , z = Z(t , s) = s ln s. (2.58)
1− st
z
From the relations (2.58), we get x = ln y
. Thus, a solution is given by

u(x, y) = x ln y,

which is defined on the domain

� �
(x, y) ∈ �2 : y > 0 .

Note that the integral surface contains the entire datum curve Γ .
In this example, note that the Jacobian J (0, s) is given by
� �
� s 2
1 �
� �
J (0, s) = � � = s 2 (1 + ln s).
� −s 2 ln s 1 �

Since s ∈ (0, ∞), the Jacobian J (0, s) is zero only at s = e −1 . Also note that
at s = e −1 , the datum curve has characteristic direction viz. (e −2 , e −2 , 0).

Example 2.53. In this example, we consider the Cauchy problem for the
linear partial differential equation
2
u x + 3y 3 uy = 2, (2.59)

satisfying the condition u(x, 0) = x for all x ∈ �. Note that another

Cauchy problem for the PDE (2.59) was analyzed in Example 2.42. It is
easy to see that the Jacobian J (0, s) in the present case satisfies
� �
� 1 1 �
� �
J (0, s ) = � �=0
� 0 0 �

for every s ∈ �. Thus transversality condition does not hold. Note that the
datum curve Γ does not have a characteristic direction at any of its points.
As a consequence, in view of Lemma 2.49, the Cauchy problem does not
have a solution. In other words, there is no continuously differentiable
function which solves the Cauchy problem.
Despite knowing the answer, we attempt to solve the Cauchy problem
by the method of characteristics, as we are curious to know where it would

July 25, 2021 Sivaji

72 Chapter 2. First order PDE

lead us to.
Let us parametrize the given Cauchy data as

Γ : x = s, y = 0, z = s, s ∈ �.

The characteristic system of ODE for the given equation is

dx dy 2 dz
= 1, = 3y 3 , = 2.
dt dt dt
The characteristic system of ODE along with the initial conditions

x(0) = s, y(0) = 0, z(0) = s

has infinitely many solutions, and are given by

x = X (t , s) = t + s, y = Y (t , s) = 0, z = Z(t , s) = 2t + s, (2.60)

x = X (t , s) = t + s, y = Y (t , s) = t 3 , z = Z(t , s) = 2t + s, (2.61)
�
0 if t < α,
x = X (t , s) = t +s, y = Y (t , s) = , z = Z(t , s) = 2t +s
(t − α)3 if t ≥ α.
where α ∈ [0, ∞), which is not unexpected since the function y �→ y 2/3
is not Lipschitz on any interval containing zero. Also observe that all the
characteristic curves listed above pass through the point (s, 0, s) ∈ Γ .
The equations in (2.60) parametrically represent the z x-plane in �3 .
Since z x-plane is not the graph of any function of the independent vari-
ables x, y, it cannot be an integral surface corresponding to the given PDE.
On the other hand, the equations in (2.61) may be re-written as

z = 2x − s, y = (x − s)3 , (2.62)

where s varies in �. Eliminating the parameter s using the equations in

(2.62), we get
z = x + y 1/3
Thus, a candidate solution may be defined by

u(x, y) = x + y 1/3 .

It satisfies the Cauchy condition, however the function u is not differen-

tiable on any domain in �2 which contains a part of the x-axis on which
the initial condition is prescribed. Thus the candidate solution is not a

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 73

Figure 2.10. Example 2.44: Views of integral surface (in green), datum curve (in magenta)

solution to the Cauchy problem.

Even the surfaces (indexed by α ∈ [0, ∞)) represented parametrically in
(2.61) do not correspond to any integral surface, and checking the same is
left to the reader.
As noted in Remark 2.43, the integral surface obtained in Example 2.42
exhibited singular nature near the line descibed by the equations z = x and
y = 0.

Example 2.54. In Example 2.44, we considered a Cauchy problem and

observed that transversality condition fails whenever the parameter s (de-
scribing the datum curve) takes a value which is an integral multiple of π.
We now discuss what happens to the integral surface determined in Exam-
ple 2.44 near points (0, y0 , y0 ) ∈ Γ where y0 is an integral multiple of π.
The integral surface is depicted in Figure 2.10. The figures on the left
and right sides therein show the non-existence of an integral surface con-
taining the points (0, 0, 0) and (0, −π, −π) which correspond to the para-
metric values s = 0 and s = −π respectively.
See Remark 2.45 for details on reasons for non-existence of a solution
to Cauchy problem near the points (0, 0, 0) and (0, −π, −π) on the datum
curve.

July 25, 2021 Sivaji

74 Chapter 2. First order PDE

2.4.5 On the local nature of solutions to Cauchy problem

Consider the initial value problem

dy
= f (x, y), y(x0 ) = y0 ,
dx
where f : I × � → � (I is an interval in �) is a continuous function. A
solution to the IVP whose domain is the interval I is called a global solution,
otherwise a local solution. Recall that Cauchy-Lipschitz-Picard existence
theorem guarantees the existence of a local solution to the IVP, while there
are sufficient conditions on f under which the IVP has a global solution.
In the context of a Cauchy problem for first order PDE, there are two
notions of local and global solutions. They are
(i) A solution to a Cauchy problem is said to be a global solution (w.r.t.
datum curve) if the corresponding integral surface contains the entire
datum curve, otherwise the solution is called a local solution (w.r.t.
datum curve).
(ii) A solution to a Cauchy problem is said to be a global solution (w.r.t.
domain) if the solution is defined on the domain Ω2 , otherwise the
solution is called a local solution (w.r.t. domain).
Since Γ2 ⊂ Ω2 , if a solution is global (w.r.t. domain), then it is also global
(w.r.t. datum curve). Observe that if a solution is local (w.r.t. datum
curve), then it is also a local solution (w.r.t. domain). Recall that Theo-
rem 2.36 asserted the existence of a local solution (w.r.t. datum curve) to
the Cauchy problem.
We explore the reasons why Cauchy problems may not have global so-
lutions and illustrate them with the help of examples.
1. Recall that base characteristic curves are defined as projection to Ω2
of the characteristic curves in Ω3 . On the other hand, for a semilinear
equation we noted in Subsection 2.4.1 that base characteristic curves
may be found independent of characteristic curves. Further, we ob-
served in Step 2 (in the proof of Theorem 2.36) that the equation for z
may not admit solutions for all t for which base characteristics, which
are given by x = X (t , s), y = Y (t , s), are defined. This might result
in a situation where projection of a characteristic curve may not be
the entire base characteristic curve that was determined directly using
the first two equations from (chara.ODE). The following example
exhibits this possibility, and is a simpler version of Example 2.23.

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 75

Example 2.55. In this example, we are going to solve the Cauchy

problem for the semilinear partial differential equation

u x + uy = u 2 , for x ∈ �, y ∈ � (2.63)

satisfying the condition u(x, 0) = x for all x ∈ �.

Let us parametrize the given Cauchy data as

Γ : x = s, y = 0, z = s, s ∈ �.

The characteristic system of ODE for the given equation is

dx dy dz
= 1, = 1, = z 2.
dt dt dt
Solution of the characteristic system of ODE satisfying the initial con-
ditions
x(0) = s, y(0) = 0, z(0) = s
is given by
1
x = X (t , s) = t + s, y = Y (t , s) = t , z = Z(t , s) = .
s−t
From the first two equations, we get

s = S(x, y) = x − y, t = T (x, y) = y.

Thus, the solution to Cauchy problem is given by

1
u(x, y) = Z(T (x, y), S(x, y)) = ,
x − 2y

defined on the domain

� x�
(x, y) ∈ � : x > 0, y <
2
2
and � x�
(x, y) ∈ �2 : x < 0, y >
2
Both the solutions given above are local (w.r.t. datum curve) solu-
tions. Observe that base characteristic curves are the family of straight
lines x = y + s indexed by s ∈ � which cover the entire plane �2 ,
but still the Cauchy problem does not have a global (even w.r.t. da-

July 25, 2021 Sivaji

76 Chapter 2. First order PDE

tum curve) solution. This is a manifestation of the nonlinearity in the

equation (2.63).

2. Intersecting base characteristic curves is another reason for non-existence

of global (w.r.t. domain) solutions to quasilinear PDEs. In Exam-
ple 2.65, which is a simple and a classical example, we encounter an
initial value problem for Burgers equation where the solution fails to
be a global (w.r.t. domain) solution due to intersecting base character-
istic curves. More details are presented in Subsection 2.6.1
3. In Example 2.46, we considered two Cauchy problems. When we con-
sidered Cauchy data 2, the Cauchy problem has a solution defined on
�2 \ {(0, 0)} which is a global (w.r.t. domain) solution. The function
h describing the datum curve was an even function did play a crucial
role there. On the other hand if h were an odd function, the Cauchy
problem would only have a local (w.r.t. domain) solution as evidenced
by Cauchy problem corresponding to Cauchy data 1 in the same ex-
ample.

2.4.6 Quasilinear equation in more than two independent variables

In this subsection, we discuss method of solution for Cauchy problem as-

sociated to a quasilinear equation in d number of independent variables.
It is instructive to compare the present case with that of d = 2 which was
presented in detail earlier. Since analysis is very similar to that of d = 2,
only main steps are presented here.
Consider a first order quasilinear equation in d -independent variables
x1 , x2 , · · · , xd given by

�
d
ai (x1 , x2 , · · · , xd , u)u xi = c(x1 , x2 , · · · , xd , u), (d-QL)
i =1

where a1 , a2 , · · · , ad , c are functions belonging to the space C 1 (Ωd +1 ), where

Ωd +1 is a domain in �d +1 .
Datum surface Γ is a (d − 1)-dimensional manifold in �d +1 , and the
integral surface S : z = u(x1 , · · · , xd ) corresponding to a solution u to the
equation (d-QL) is a d -dimensional manifold in �d +1 . An integral surface
corresponding to a solution of Cauchy problem is required to contain some
part of Γ .

Sivaji IIT Bombay

2.4. Cauchy problem for quasilinear PDE 77

Let Γ be parametrically described by

xi = fi (s1 , s2 , · · · , sd −1 ), i = 1, 2, · · · , d ,
z = h(s1 , s2 , · · · , sd −1 ),

where fi (i = 1, 2, · · · , d ), and h are functions defined on some open subset

of �d −1 Alternatively, the Cauchy condition can be stated as
� �
u f1 (s1 , s2 , · · · , sd −1 ), · · · , fd (s1 , s2 , · · · , sd −1 ) = h(s1 , s2 , · · · , sd −1 )

The (d + 1)-dimensional system of characteristic ODEs corresponding to

the equation (d-QL) are given by

d xi
= ai (x1 , x2 , · · · , xd , z), i = 1, 2, · · · , d , (2.64a)
dt
dz
= c(x1 , x2 , · · · , xd , z) (2.64b)
dt
Let the solutions to the characteristic system of ODE (2.64) satisfying the
initial conditions

xi (0) = fi (s1 , s2 , · · · , sd −1 ) (i = 1, 2, · · · , d ), z(0) = h(s1 , s2 , · · · , sd −1 )

be given by

xi = Xi (t , s1 , s2 , · · · , sd −1 ) (i = 1, 2, · · · , d ), z = Z(t , s1 , s2 , · · · , sd −1 ). (2.65)

Using base characteristic surfaces, namely xi = Xi (t , s1 , s2 , · · · , sd −1 ) (i =

1, 2, · · · , d ), solve for (t , s1 , s2 , · · · , sd −1 ) in terms of (x1 , x2 , · · · , xd ) near points
of Γ . This can be done under the transversality condition given by

� �
� ∂ f1 ∂ f1 �
� (s) ··· (s) a1 ( f1 (s), f2 (s), · · · , fd (s)) �
� ∂ s1 ∂ sd −1 �
� ∂ f2 ∂ f2 �
� (s) ··· (s) a2 ( f1 (s), f2 (s), · · · , fd (s)) ��
J (0, s) = �� ∂ s1 ∂ sd −1
� �= 0.
� ··· ··· ··· ······ �
� �
� ∂ fd
···
∂ fd
a f f · · · , f �
� ∂ s1
(s) ∂ sd −1
(s) (
d 1 (s), 2 (s), d (s)) �

where s stands for the vector (s1 , s2 , · · · , sd −1 ), or we may assume that J (0, s0 ) �=
0 for some s0 . Under the transversality condition, we can apply inverse
function theorem, and obtain functions Si , T such that

si = Si (x1 , x2 , · · · , xd ) (i = 1, · · · , d ), t = T (x1 , x2 , · · · , xd ).

July 25, 2021 Sivaji

78 Chapter 2. First order PDE

Substituting for s1 , s2 , · · · , sd −1 , t in the last equation of (2.65), we obtain an

implicit expression for the solution to the given Cauchy problem as
� �
u(x1 , x2 , · · · , xd ) = Z T (x1 , x2 , · · · , xd ), S1 (x1 , x2 , · · · , xd ), · · · , Sd −1 (x1 , x2 , · · · , xd ) .

Formulating a local existence and uniqueness theorem like Theorem 2.36

in the present case and proving is left to the reader as an exercise, as we went
through all the necessary ideas for the same.

Example 2.56. Let k ∈ �. Consider the Cauchy problem for the equation

x1 u x1 + x2 u x2 + x3 u x3 = k u,

where the Cauchy data is given by u(x1 , x2 , 1) = h(x1 , x2 ) for all (x1 , x2 ) ∈
�2 , h ∈ C 1 (�2 ).
Let us parametrize the given Cauchy data as

Γ : xi = si (i = 1, 2), x3 = 1, z = h(s1 , s2 ), (s1 , s2 ) ∈ �2 .

The characteristic system of ODE for the given equation is

d xi dz
= xi (i = 1, 2, 3), = k z.
dt dt
For each (s1 , s2 ) ∈ �2 , the solution of the characteristic system of ODE
satisfying the initial conditions

xi (0) = si (i = 1, 2), x3 (0) = 1, z(0) = h(s1 , s2 )

is given by

xi = Xi (t , s1 , s2 ) = si e t (i = 1, 2), x3 = X3 (t , s1 , s2 ) = e t , (2.66a)

z = Z(t , s1 , s2 ) = h(s1 , s2 )e k t . (2.66b)

From the equations (2.66a), we get

xi
si = Si (x1 , x2 , x3 ) = (i = 1, 2), t = T (x1 , x2 , x3 ) = ln x3 .
x3
Sivaji IIT Bombay
2.5. Cauchy problem for general nonlinear equation 79

Thus, the solution to Cauchy problem is given by

u(x1 , x2 , x3 ) = Z(T (x1 , x2 , x3 ), S1 (x1 , x2 , x3 ), S2 (x1 , x2 , x3 ))

= h(S1 (x1 , x2 , x3 ), S2 (x1 , x2 , x3 ))e kT (x1 ,x2 ,x3 )
� �
x1 x2 k
=h , x .
x3 x3 3

Thus the solution is given by

� �
x1 x2
u(x1 , x2 , x3 ) = x3k h , , (2.67)
x3 x3

defined for all (x1 , x2 , x3 ) ∈ �3 for which x3 �= 0. For each λ ∈ � such that
λ �= 0, observe that the function u given by (2.67) satisfies

u(λx1 , λx2 , λx3 ) = λk u(x1 , x2 , x3 ).

whenever x3 �= 0. If the function u is known to be continuous, then the

above equation holds for (x1 , x2 , x3 ) ∈ �3 . Thus the solution to the Cauchy
problem is a homogeneous function of order k.

2.5 Cauchy problem for general nonlinear equation

In this section, we would like to extend the method of characteristics to
solve Cauchy problem for general nonlinear equations.
Let Ω5 be an open and connected subset of �5 , and let F : Ω5 → �
be an arbitrary function. We denote the argument of F by the 5-tuple
(x, y, z, p, q). Such an F gives rise to the most general form of a first order
PDE, which is
F (x, y, u, u x , uy ) = 0. (GE)
Observe that the equation (GE) reduces to a quasilinear equation (QL) if
F is of the form

F (x, y, z, p, q) = a(x, y, z) p + b (x, y, z)q − c(x, y, z)

for some functions a, b , c. Thus we may say that we are considering equa-
tions which are more general than quasilinear equations.
Our analysis of equation (GE) draws inspiration from that of (QL),
and frequent comparisons to it will be made throughout the presentation

July 25, 2021 Sivaji