Partial Differential Equation (PDE) Part A

Teacher: A Ganguly
• Definition of PDE, Its occurrence in geometry & physics, Order & Degree
A PDE arises in geometry and physics when there will be more than one
independent variable so that the derivatives of any dependent variable will be
partial derivatives instead of ordinary derivatives.
For example, suppose we want to study flow of heat along a 2D body. If we
denote the temperature at any point P(𝑥, 𝑦) of the solid by 𝑧, then naturally 𝑧
will be function of two independent variables 𝑥, 𝑦, as temperature may change
from point to point. Thus when laws of Physics will be applied, we may get a
relation involving 𝑥, 𝑦, 𝑧 and partial derivatives of 𝑧 of order one or higher
yielding a PDE.
For simplicity, let us consider two independent variables 𝑥, 𝑦, and one
dependent variable 𝑧(𝑥, 𝑦), although the terminologies and definitions,
introduced below, may be generalized immediately for more than two
independent variables. We will use notations 𝑝 ≡ 𝜕𝑧/𝜕𝑥, 𝑞 ≡ 𝜕𝑧/𝜕𝑦. Also time
to time we will use suffixes 𝑥, 𝑦 of dependent variable 𝑧 to denote partial
𝜕𝑧 𝜕𝑧 𝜕2 𝑧 𝜕2 𝑧 𝜕2 𝑧
derivatives, i.e. 𝑧𝑥 ≡ , 𝑧𝑦 ≡ , 𝑧𝑥𝑥 ≡ 2
, 𝑧𝑥𝑦 ≡ ≡ 𝑧𝑦𝑥 , 𝑧𝑦𝑦 ≡ etc.
𝜕𝑥 𝜕𝑦 𝜕𝑥 𝜕𝑥𝜕𝑦 𝜕𝑦 2
Now a general form of PDE will be
𝜕𝑧 𝜕𝑧 𝜕2 𝑧 𝜕2 𝑧 𝜕2 𝑧
𝐹 �𝑥, 𝑦, 𝑧, , , 2
, , ,…� = 0 (1.1)
𝜕𝑥 𝜕𝑦 𝜕𝑥 𝜕𝑥𝜕𝑦 𝜕𝑦 2

Order & Degree of PDE

The order of highest order partial derivative term of (1.1) is the order of PDE
(1.1). The exponent of highest order partial derivative term is called the degree
of PDE (1.1).
𝜕2 𝑧
For example, 𝑝𝑥 + 𝑞𝑦 = 0 is of 1st order and of 1st degree. The PDE +
𝜕𝑥 2
𝜕2 𝑧
= 0 is of 2nd order & 1st degree. The PDEs 𝑝2 𝑥 + 𝑞𝑦 = 0, 𝑝2 𝑥 + 𝑞𝑦 +
𝜕𝑦 2
𝜕2 𝑧
𝑥 = 0 are respectively of 1st order, 2nd degree and 2nd order, 1st degree.
𝜕𝑥 2

Linear & Nonlinear PDE

PDE (1.1) will be linear if ∃ no terms with power higher than one w.r.t.
dependent variable 𝑧 and its partial derivatives. Otherwise PDE will be called
nonlinear.

1
 𝜕2 𝑧 𝜕𝑧 2
For example, 𝑝𝑥 + 𝑞𝑦 = 𝑧 2 is non-linear, + � � = 𝑧 is also nonlinear.
𝜕𝑥 2 𝜕𝑥
𝜕2 𝑧 𝜕2 𝑧
2
+ = 0 is linear.
𝜕𝑥 𝜕𝑦 2

Linear, Semi-linear, Quasi-linear, Non-linear : Non-linear PDE, defined above,

may be further sub-categorized into three types:
 Semi-linear: Non-linear terms present w.r.t. 𝑧, e.g. 𝑧 2 , 𝑧 3 etc, but ∃ no
non-linear terms like 𝑧𝑝, 𝑝𝑞, 𝑝2 etc.
𝜕2 𝑧 𝜕𝑧
e.g. 𝑥 + 𝑦2 − 𝑧 3 𝑥 = 0 is semi-linear.
𝜕𝑥 2 𝜕𝑦
 Quasi-linear: Non-linear terms present w.r.t. 𝑧 and product of 𝑧 and its
partial derivatives, e.g. 𝑧 2 , 𝑧 3 , 𝑧𝑝, 𝑧 2 𝑞, 𝑧𝜕 2 𝑧/𝜕𝑥 2 etc, but ∃ no non-linear
terms like 𝑧𝑝, 𝑝𝑞, 𝑝2 , (𝜕 2 𝑧/𝜕𝑦 2 )3 etc.
𝜕2 𝑧 𝜕𝑧
e.g. 𝑧 + 𝑧2 − 𝑧 3 𝑦 = 0 is quasi-linear.
𝜕𝑥 2 𝜕𝑦
 Non-linear: All non-linear terms present, e.g. 𝑧 3 , 𝑝𝑞, (𝜕 2 𝑧/𝜕𝑦 2 )3 , 𝑧 2 𝑞 etc
e.g. 𝑝2 𝑥 + 𝑝𝑞𝑦 = 1 is non-linear.
 Linear: No non-linear terms present.
𝜕𝜓 𝜕2 𝑧
e.g =𝛼 is linear.
𝜕𝑡 𝜕𝑥 2

Remark: In this note, most often I use the term “Linear PDE” to mean
Linear/Semi-linear/Quasi-linear PDE when there is no context of classification.
• Formation of PDE by eliminating arbitrary constants/functions

 First order PDE

Consider a surface of revolution with z-axis as axes of symmetry. It is well-
known from geometry that equation to such surfaces is
𝑧 = 𝑓(𝑥 2 + 𝑦 2 ) [𝑓 is arbitrary function] (2.1)
Let us now eliminate arbitrary function 𝑓. Denoting 𝑢 = 𝑥 2 + 𝑦 2 , using chain
d𝑓
rule, we differentiate 𝑧 in (2.1) partially w.r.t. 𝑥 & 𝑦, and get 𝑝 = 2𝑥 ,𝑞 =
d𝑢
d𝑓
2𝑦 . Eliminating 𝑓 from two equations, we get
d𝑢

𝑝𝑦 − 𝑞𝑥 = 0, (2.2)
which is a 1st order 1st degree linear PDE.

2
In particular, for 𝑓 = 𝑐 ± �𝑥 2 + 𝑦 2 − 𝑎2 , where 𝑎, 𝑐 are arbitrary constants,
equation (2.1) becomes
𝑥 2 + 𝑦 2 + (𝑧 − 𝑐)2 = 𝑎2 , (2.3)
which represents a system of sphere of radius 𝑎 with centre (0,0, 𝑐) on z-axis.
Let us eliminate 𝑎, 𝑐 from (2.3). Treating 𝑧 as dependent variable, we
differentiate 𝑧 w.r.t. 𝑥 & 𝑦, and get 𝑝(𝑧 − 𝑐) + 𝑥 = 0 , 𝑞(𝑧 − 𝑐) + 𝑦 = 0, so
that one arbitrary constant 𝑎 has been automatically eliminated. Next
eliminating 𝑐 from two resulting equations, we get same PDE 𝑝𝑦 − 𝑞𝑥 = 0, as
it should be according to (2.1) and (2.2). But note that in the first case, we
eliminate one arbitrary function, whereas in second case, we eliminate two
arbitrary constants, i.e. contexts were different.
Hence, given a non-differential relation between 𝑥, 𝑦, 𝑧(𝑥, 𝑦) through an
arbitrary function 𝑓 or through two arbitrary constants 𝑎, 𝑐, we will obtain PDE
by eliminating 𝑓 (or 𝑎, 𝑐) from given relation and two relations obtained by
partial differentiations of given relation w.r.t. 𝑥, 𝑦. Essentially, this means that
given non-differential relation, which represents a family of surfaces in space, is
a solution of resulting PDE. This solution surface is called integral surface of
PDE.
Note that the PDE thus obtained may be linear or nonlinear. Consider a general
relation of the form
𝐹(𝑥, 𝑦, 𝑧, 𝑎, 𝑏) = 0, [𝑎, 𝑏 are arbitrary constants, 𝐹 is known function] (2.4)
To eliminate 𝑎, 𝑏 from (2.4), we partially differentiate (2.4) w.r.t. 𝑥, 𝑦, and get
𝜕𝐹 𝜕𝐹 𝜕𝐹 𝜕𝐹
+𝑝 = 0, +𝑞 = 0. (2.5)
𝜕𝑥 𝜕𝑧 𝜕𝑦 𝜕𝑧

From these two and given relation, we can always eliminate 𝑎, 𝑏 to obtain 1st
order PDE of the following general form
𝑓(𝑥, 𝑦, 𝑧, 𝑝, 𝑞) = 0. [𝑓 is a known function] (2.6)
Note that PDE (2.6) may not be linear, in general. See the example below. At
this point, note that the surface (2.4) is a solution of PDE (2.6), and hence is an
integral surface of (2.6). Later we will see that this solution belongs to a class of
integrals of 1st order PDEs.
Example: Eliminate 𝑎, 𝑏 from 𝐹(𝑥, 𝑦, 𝑧, 𝑎, 𝑏) ≡ 𝑧 − (𝑥 + 𝑎)(𝑦 + 𝑏) = 0
Solution: 𝑝 = 𝑦 + 𝑏, 𝑞 = 𝑥 + 𝑎

3
Substituting for 𝑎, 𝑏 into given relation, resulting PDE is 𝑧 = 𝑝𝑞, which is a 1st
order but nonlinear PDE.
Example: Find the PDE of a system of spheres of radius 5 with centre in xy-
plane.
Solution: Let C(𝑎, 𝑏, 0) be centre. Then given equation is
(𝑥 − 𝑎)2 + (𝑦 − 𝑏)2 + 𝑧 2 = 25
To eliminate two arbitrary constants 𝑎, 𝑏, differentiate above equation partially
w.r.t. 𝑥, 𝑦 (treating 𝑧(𝑥, 𝑦) as dependent variable):
(𝑥 − 𝑎) + 𝑧𝑝 = 0, (𝑦 − 𝑏) + 𝑧𝑞 = 0
Eliminating 𝑎, 𝑏, from above relation and given equation, we finally have
following nonlinear 1st order PDE:
𝑧 2 (1 + 𝑝2 + 𝑞 2 ) = 25

Let me now come back again to the context of elimination of arbitrary

functions. Previously, I have shown that elimination of one arbitrary function
𝑓(𝑢) from given relation 𝑧 = 𝑓(𝑢), where 𝑢 is a known function of 𝑥, 𝑦, leads
us to a linear 1st order PDE 𝑝𝑦 − 𝑞𝑥 = 0. Now, let us consider a slight
generalization of given relation as
𝑓(𝑢, 𝑣) = 0 [u , 𝑣 are known functions of 𝑥, 𝑦, 𝑧] (2.7)
Note that 𝑧(𝑥, 𝑦) is dependent variable. Now, to eliminate arbitrary function 𝑓,
we differentiate (2.7) partially w.r.t. 𝑥, 𝑦:
𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑓
(𝑢𝑥 + 𝑢𝑧 𝑝) + (𝑣𝑥 + 𝑣𝑧 𝑝) = 0 , �𝑢𝑦 + 𝑢𝑧 𝑞� + �𝑣𝑦 + 𝑣𝑧 𝑞� = 0. (2.8)
𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
𝜕𝑓 𝜕𝑓
It is straightforward to eliminate , from two linear algebraic homogeneous
𝜕𝑢 𝜕𝑣
equations (2.8) by noticing that for non-zero values of them, coefficient
determinant must vanish. Hence, we get
𝑢𝑥 + 𝑢𝑧 𝑝 𝑣𝑥 + 𝑣𝑧 𝑝
�𝑢 + 𝑢 𝑞
𝑦 𝑧 𝑣𝑦 + 𝑣𝑧 𝑞 � = 0
Quasi-linear
Expanding and collecting terms for 𝑝, 𝑞, we get a 1 order linear PDE:
st

�𝑢𝑧 𝑣𝑦 − 𝑢𝑦 𝑣𝑧 �𝑝 + (𝑢𝑥 𝑣𝑧 − 𝑢𝑧 𝑣𝑥 )𝑞 + �𝑢𝑥 𝑣𝑦 − 𝑢𝑦 𝑣𝑥 � = 0. (2.9)

4
Note at this point that PDE (2.9) is the most general form of quasi-linear
(Definitions are given above for PDE of arbitrary order, and will be given later
separately for 1st order PDE) 1st order PDE of the following form:
𝑃(𝑥, 𝑦, 𝑧)𝑝 + 𝑄(𝑥, 𝑦, 𝑧)𝑞 = 𝑅(𝑥, 𝑦, 𝑧). (2.10)
Note also that the family of surfaces given by (2.7) is a solution of PDE (2.10),
and hence defines Integral Surface. Later we will see that this type of solution
containing one arbitrary function belongs to a class of integrals of 1st order
PDEs.
Example: Eliminate arbitrary function 𝜙 from 𝜙(𝑥 + 𝑦 + 𝑧, 𝑥 2 + 𝑦 2 − 𝑧 2 ) = 0
Solution: Denote 𝑢 = 𝑥 + 𝑦 + 𝑧 , 𝑣 = 𝑥 2 + 𝑦 2 − 𝑧 2 . Differentiate given
relation 𝜙(𝑢, 𝑣) = 0 partially w.r.t. 𝑥, 𝑦:
𝜕𝜙 𝜕𝜙 𝜕𝜙 𝜕𝜙
(1 + 𝑝) + 2(𝑥 − 𝑧𝑝) = 0 , (1 + 𝑞) + 2(𝑦 − 𝑧𝑞) =0
𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
𝜕𝜙 𝜕𝜙
Eliminating , , we get
𝜕𝑢 𝜕𝑣

1+𝑝 2(𝑥 − 𝑧𝑝)

� � = 0 ⟹ (𝑦 + 𝑧)𝑝 − (𝑧 + 𝑥)𝑞 = 𝑥 − 𝑦
1+𝑞 2(𝑦 − 𝑧𝑞)

 Formation of higher order PDE by eliminating arbitrary

functions/constants
Consider following relation involving two arbitrary functions 𝑓, 𝑔
𝑧 = 𝑓(𝑥 − 𝑐𝑦) + 𝑔(𝑥 + 𝑐𝑦) [𝑐 > 0 is known constant] (2.10)
Let us denote 𝑢 = 𝑥 − 𝑐𝑦 , 𝑣 = 𝑥 + 𝑐𝑦. Here, merely differentiating once w.r.t.
𝑥, 𝑦, we can’t eliminate 𝑓, 𝑔, as evident from two resulting relations:
d𝑓 d𝑔 d𝑓 d𝑔
𝑝= + , 𝑞 = −𝑐 +𝑐 , (2.11)
d𝑢 d𝑣 d𝑢 d𝑣

If we now differentiate 1st one of (2.11) w.r.t. 𝑥, and 2nd one w.r.t. 𝑦, we get
𝜕2 𝑧 d2 𝑓 d2 𝑔 𝜕2 𝑧 d2 𝑓 d2 𝑔
= + , = 𝑐2 + 𝑐2 . (2.12)
𝜕𝑥 2 d𝑢 2 d𝑣 2 𝜕𝑦 2 d𝑢 2 d𝑣 2

d2 𝑓 d2 𝑔
We now can eliminate , from two relations (2.12), and get following 2nd
d𝑢2 d𝑣 2
order linear PDE:
𝜕2 𝑧 𝜕2 𝑧
= 𝑐2 . (2.13)
𝜕𝑦 2 𝜕𝑥 2

5
Remark: 2nd order linear PDE (2.13) is a well-known equation, known as “Wave
Equation”. We will study it at later stage of the course. But, for now, we can
definitely say that given relation (2.10) is a solution of Wave Equation (2.13).
Consider now a slight generalization of (2.10) as follows:
𝑧 = 𝑓(𝑢) + 𝑔(𝑣) + 𝑤 . [𝑢, 𝑣, 𝑤 are known functions of 𝑥, 𝑦] (2.14)
𝜕2 𝑧 𝜕2 𝑧 𝜕2 𝑧
Let us use the notations 𝑟 ≡ 2
,𝑠 ≡ ,𝑡 ≡ for 2nd order partial
𝜕𝑥 𝜕𝑥𝜕𝑦 𝜕𝑥 2
derivatives. We will also use the symbol 𝑓 ′ , 𝑓 ′′ , 𝑔′ , 𝑔′′ to denote ordinary
derivatives of 𝑓, 𝑔 w.r.t. their single arguments. Now we will compute
𝑝, 𝑞, 𝑟, 𝑠, 𝑡 by partially differentiating (2.14) w.r.t. 𝑥, 𝑦 twice:
𝑝 = 𝑢𝑥 𝑓 ′ + 𝑣𝑥 𝑔′ + 𝑤𝑥 , 𝑞 = 𝑢𝑦 𝑓 ′ + 𝑣𝑦 𝑔′ + 𝑤𝑦
𝑟 = 𝑢𝑥 2 𝑓 ′′ + 𝑢𝑥𝑥 𝑓 ′ + 𝑣𝑥 2 𝑔′′ + 𝑣𝑥𝑥 𝑔′ + 𝑤𝑥𝑥
(2.15)
𝑠 = 𝑢𝑥 𝑢𝑦 𝑓 ′′ + 𝑢𝑥𝑦 𝑓 ′ + 𝑣𝑥 𝑣𝑦 𝑔′′ + 𝑣𝑥𝑦 𝑔′ + 𝑤𝑥𝑦
𝑡 = 𝑢𝑦 2 𝑓 ′′ + 𝑢𝑦𝑦 𝑓 ′ + 𝑣𝑦 2 𝑔′′ + 𝑣𝑦𝑦 𝑔′ + 𝑤𝑦𝑦
So, we have a system of five non-homogeneous linear algebraic equations for
four unknowns 𝑓 ′ , 𝑔′ , 𝑓 ′′ , 𝑔′′, and hence eliminating these, we will get a
relation, which will be coefficient determinant equal to zero:
𝑝 − 𝑤𝑥 𝑢𝑥 𝑣𝑥 0 0
𝑞 − 𝑤𝑦 𝑢𝑦 𝑣𝑦 0 0
� �
𝑟 − 𝑤𝑥𝑥 𝑢𝑥𝑥 𝑣𝑥𝑥 𝑢𝑥 2 𝑣𝑥 2 = 0 . (2.16)
�𝑠 − 𝑤𝑥𝑦 𝑢𝑥𝑦 𝑣𝑥𝑦 𝑢𝑥 𝑢𝑦 𝑣𝑥 𝑣𝑦 �
𝑡 − 𝑤𝑦𝑦 𝑢𝑦𝑦 𝑣𝑦𝑦 𝑢𝑦 2 𝑣𝑦 2
A close inspection of the determinant in lhs of (2.16) reveals two things:- 1)
second order derivative appears so that PDE (2.16) is of second order, 2)
Derivative terms appear only in 1st column, and all derivative terms in 1st
column are of power 1, so that PDE (2.16) is linear. In fact, expanding the
determinant in terms of the elements of 1st column, PDE (2.16) will be of the
following form:
𝑅𝑟 + 𝑆𝑠 + 𝑇𝑡 + 𝑃𝑝 + 𝑄𝑞 = 𝑊, (2.17)
where 𝑃, 𝑄, 𝑅, 𝑆, 𝑇, 𝑊 are functions of 𝑥, 𝑦 only, and hence it is linear 2nd order
PDE.
Note that there are several occasions when we will end up with nonlinear higher
order PDE by elimination. For instance, in the model (2.14), if we let the known

6
functions 𝑢, 𝑣, 𝑤 to depend on dependent variable 𝑧 also, resulting PDE after
elimination of arbitrary functions 𝑓, 𝑔 will be 2nd order nonlinear PDE. Without
going into lengthy calculation, we may infer that by noticing that in the
equations (2.15) & (2.16), we will have to replace 𝑢𝑥 , 𝑣𝑥 , 𝑢𝑦 , 𝑣𝑦 , 𝑤𝑥 , 𝑤𝑦 by
(𝑢𝑥 + 𝑢𝑧 𝑝), (𝑣𝑥 + 𝑣𝑧 𝑝), �𝑢𝑦 + 𝑢𝑧 𝑞�, �𝑣𝑦 + 𝑣𝑧 𝑞�, (𝑤𝑥 + 𝑤𝑧 𝑝), �𝑤𝑦 + 𝑤𝑧 𝑞�
respectively, so that 𝑢𝑥𝑥 will be replaced by (𝑢𝑥 + 𝑢𝑧 𝑝)𝑥 etc. Thus derivative
terms appear all over in the determinant (2.16) unlike the previous case wherein
these appear only in 1st column. So it is inevitable that after expansion of the
determinant we will get a nonlinear 2nd order PDE.
Formation of PDE by elimination for more than two independent variables
We will just show one examples, as there is no extra theoretical constraint for
more than two variables.
Example: Eliminate 𝑎, 𝑏, 𝑐 from 𝜓(𝑥, 𝑦, 𝑡) = 𝑎(𝑥 + 𝑦) + 𝑏(𝑥 − 𝑦) + 𝑎𝑏𝑡 + 𝑐
𝜕𝜓 𝜕𝜓 𝜕𝜓 𝜕𝜓 2 𝜕𝜓 2 𝜕𝜓
Solution: =𝑎+𝑏, = 𝑎 − 𝑏, = 𝑎𝑏 ⟹ � � + � 𝜕𝑦 � = 4 𝜕𝑡
𝜕𝑥 𝜕𝑦 𝜕𝑡 𝜕𝑥

Note that resulting PDE is 1st order nonlinear.

• Classification of 1st order PDE as linear/quasi-linear/semi-linear/nonlinear
From this topic onward, we will consider two independent variables 𝑥, 𝑦, and
one dependent variable 𝑧(𝑥, 𝑦). A general form of linear 1st order PDE is
z
𝑃(𝑥, 𝑦)𝑝 + 𝑄(𝑥, 𝑦)𝑞 = 𝑅(𝑥, 𝑦)𝑧 + 𝐹(𝑥, 𝑦). (3.1)
Note that RHS is called non-homogeneous term in the sense that there is no
partial derivative term there so that its degree is 0 w.r.t 𝑝, 𝑞, whereas in LHS
degree of each term is one.
In Sec 1, a general definition was given to classify a PDE of arbitrary order as
Linear/Semi-linear/Quasi-linear/Non-linear. For the case of 1st order PDE, I’ve
decided to provide separately the definitions regarding same classifications.
Semi-linear: RHS depends arbitrarily on dependent variable 𝑧. General form:
𝑃(𝑥, 𝑦)𝑝 + 𝑄(𝑥, 𝑦)𝑞 = 𝑅(𝑥, 𝑦, 𝑧). (3.2)
Note that here RHS may contain nonlinear term like 𝑧 2 , but LHS doesn’t
contain terms which are non-linear in 𝑝, 𝑞, i.e. ∃ no term like 𝑧𝑝, 𝑧𝑞, 𝑧 2 𝑝 etc.
Such equation (3.2) is classified as 1st order Semi-linear PDE.
Quasi-linear: Coefficient functions in LHS and RHS depend arbitrarily on
dependent variable 𝑧. General form:

7
 𝑃(𝑥, 𝑦, 𝑧)𝑝 + 𝑄(𝑥, 𝑦, 𝑧)𝑞 = 𝑅(𝑥, 𝑦, 𝑧). (3.3)
Note here in both of RHS & LHS, there may be non-linear term like 𝑧𝑝, 𝑧𝑞, 𝑧 2 𝑝
etc in LHS and 𝑧 2 , 𝑧 3 etc in RHS, but ∃ no term which is non-linear in
derivatives, eg. Terms like 𝑝2 , 𝑝𝑞, 𝑞 2 etc are absent. Such equation (3.3) is
classified as Quasi-linear PDE.
Remark: Note that in spite of presence of certain non-linear terms in semi-
linear/quasi-linear PDE, these two can be clubbed with linear PDE in the sense
that these three categories can be solved by Lagrange’s method, to be discussed
in following section. This is the reason why sometimes I only use the term
“Linear” to refer Linear/Semi-linear/Quasi-linear for brevity when there is no
context of classification.

Classification of Integrals (Solutions) of a 1st order PDE

Consider 1st order PDE in its most general form as follows
𝑓(𝑥, 𝑦, 𝑧, 𝑝, 𝑞) = 0. (3.4)
1. Complete Integral (CI): A two-parameter family of solutions of PDE (3.4) of
the form
𝐹(𝑥, 𝑦, 𝑧(𝑥, 𝑦), 𝑎, 𝑏) = 0, [𝑎, 𝑏 arbitrary constants] (3.5)
is called Complete Solution (or Integral) of PDE. Recall the discussion in Sec 2
[see equations (2.4)-(2.6)], wherein I’ve shown that eliminating 𝑎, 𝑏, one
obtains (3.4), which is now understood as CI. Hence, CI of 1st order PDE of two
independent variables will contain two arbitrary constants.
2. General Integral (GI): A solution of PDE (3.4) involving one arbitrary
function 𝑔 of the form
𝑔(𝑢, 𝑣) = 0, (3.6)
where 𝑢, 𝑣 are known functions of 𝑥, 𝑦, 𝑧, is called General Solution (or
Integral) of PDE (3.4). Recall the discussion in Sec 2 [see equations (2.7)-
(2.10)], wherein I’ve shown that eliminating 𝑔, one obtains a 1st order quasi-
linear PDE (3.3)which is now understood as GI.
The Author of Textbook commented that the name GI is actually ‘illusory’ in
the sense that it is possible to derive from CI. However, the procedure of
derivation of GI from CI is well-known [see Refs 1,2] since long ago. Let me
discuss below the procedure without proof [for proofs, interested reader may see
Text Book and Ref 1].

8
To derive GI from CI (3.5), let 𝑏 = 𝜙(𝑎), so that (3.5) reduces to following
one-parameter family, which is sub-family of surfaces given by CI (3.5):
𝐹(𝑥, 𝑦, 𝑧, 𝑎, 𝜙(𝑎)) = 0. (3.7)
We then consider the envelope of one-parameter family of surface (3.7), which
can be obtained by partially differentiating (3.7) w.r.t. 𝑎:
𝐹𝑎 + 𝐹𝑏 𝜙 ′ (𝑎) = 0. (3.8)
From equation (3.8), we may solve for 𝑎:
𝑎 = 𝑎(𝑥, 𝑦, 𝑧). (3.9)
Substituting for 𝑎 from (3.9) into (3.7), we have GI of PDE (3.4) in the
following form
𝐹(𝑥, 𝑦, 𝑧, 𝑎(𝑥, 𝑦, 𝑧), 𝜙(𝑎(𝑥, 𝑦, 𝑧))) = 0. (3.10)
Note that GI (3.10) contains one arbitrary function 𝜙, which is the property of
GI of 1st order PDE (3.4).
3. Particular Integral (PI): If 𝑏 = 𝜙(𝑎) is prescribed satisfying some given
condition, we will get from (3.10) a solution containing no arbitrary constant
or function. This solution is known as Particular Solution (or Integral).
Indeed ∃ infinitely many such particular solutions. However, in a physical
problem, may be only one of them is acceptable solution.

4. Singular Integral (SI): This is such a solution of PDE (3.4),which, as in the

case for 1st order ODE, can’t be obtained from GI. This is called Singular
Integral (or Solution). Geometrically, SI, if exists, represents envelope of
two-parameter family of surfaces, given by CI (3.5). Equation for envelope
of (3.5) may be obtained by partially differentiating (3.5) w.r.t. 𝑎, 𝑏:
𝐹𝑎 (𝑥, 𝑦, 𝑧, 𝑎, 𝑏) = 0 , 𝐹𝑏 (𝑥, 𝑦, 𝑧, 𝑎, 𝑏) = 0 (3.11)
If envelope exists, then it will be possible to eliminate 𝑎, 𝑏 from three
equations (3.5) and (3.11), and the resulting relation will be SI of PDE (3.4).
Note that SI may or may not exist.

5. Special Solution: For some 1st order PDEs, apart from the class of solutions
discussed above, ∃ special solution, mentioned in Textbook. See the
examples below, taken from Textbook.

Example: Show that 𝐹(𝑥 + 𝑦, 𝑥 − √𝑧) is GI of the PDE 𝑝 − 𝑞 = 2√𝑧,

𝑎𝑥+𝑏𝑦
where 𝐹 is an arbitrary function. Also check whether √𝑧 = + 𝑏 is CI
𝑎−𝑏

9
 of given PDE, whether 𝑎, 𝑏 (𝑎 ≠ 𝑏) are two arbitrary constants. Finally
verify 𝑧 = 0 is yet another solution which can’t be obtained from GI, and
hence 𝑧 = 0 is a solution.
Solution: Denoting 𝑢 = 𝑥 + 𝑦 , 𝑣 = 𝑥 − √𝑧, eliminate arbitrary function 𝐹
from the relation 𝐹(𝑢, 𝑣) = 0 by differentiating partially w.r.t. 𝑥, 𝑦:
𝑝 𝑞
𝐹𝑢 + �1 − � 𝐹𝑣 = 0 , 𝐹𝑢 − 𝐹𝑣 = 0
2 √𝑧 2 √𝑧
So, eliminating 𝐹𝑢 , 𝐹𝑣 from above two equations,
𝑞 𝑝
+ �1 − � = 0 ⟹ 𝑝 − 𝑞 = 2 √𝑧
2 √𝑧 2 √𝑧
Hence, 𝐹(𝑥 + 𝑦, 𝑥 − √𝑧) is GI of the PDE 𝑝 − 𝑞 = 2√𝑧.
𝑎𝑥+𝑏𝑦
Next, from √𝑧 = + 𝑏, eliminate arbitrary constants 𝑎, 𝑏 by
𝑎−𝑏
differentiating partially w.r.t. 𝑥, 𝑦:
𝑝 𝑎 𝑞 𝑏
= , = ⟹ 𝑝 − 𝑞 = 2 √𝑧
2 √𝑧 𝑎 − 𝑏 2 √𝑧 𝑎 − 𝑏
𝑎𝑥+𝑏𝑦
Hence, √𝑧 = + 𝑏 is CI of given PDE
𝑎−𝑏
Last Part: We see that 𝑧 = 0 can’t be obtained from GI or CI, but for 𝑧 = 0,
we have 𝑝 = 0, 𝑞 = 0, and hence given PDE is satisfied by 𝑧 = 0. Note that
it is not a SI, since in the process of finding envelope of two-parameter
family CI, one has to consider the case 𝑧 ≠ 0 to eliminate 𝑎, 𝑏. Hence, 𝑧 = 0
is a special solution.

• Lagrange’s method for solving 1st order linear PDE

Here, for convenience, in the term “Linear”, I’ve clubbed linear, semi-linear and
quasi-linear PDEs, due to the fact that all of three categories can be solved by
Lagrange’s method theoretically (i.e. subject to separability and analytical
integrability). Hence, let us consider following 1st order PDE
𝑃𝑝 + 𝑄𝑞 = 𝑅, (4.1)
where 𝑃, 𝑄, 𝑅 are functions of 𝑥, 𝑦, 𝑧 in general.
Lagrange’s Method: General Solution of PDE (4.1) is given by an arbitrary
function 𝜙 of two known functions 𝑢(𝑥, 𝑦, 𝑧), 𝑣(𝑥, 𝑦, 𝑧) as follows
𝜙(𝑢, 𝑣) = 0 , (4.2)
where
𝑢(𝑥, 𝑦, 𝑧) = 𝑐1 , 𝑣(𝑥, 𝑦, 𝑧) = 𝑐2 (4.3)

10
are solutions of following Differential Equations (DE), known as Lagrange’s
Auxiliary Equation (AE), given by
d𝑥 d𝑦 d𝑧
= = , (4.4)
𝑃 𝑄 𝑅

𝑐1 , 𝑐2 being two arbitrary constants.

The result (4.2)-(4.4) can be proved analytically (by using knowledge of
elimination) or geometrically (using geometric properties of surface generated
by curves in space).
Analytical Proof I’ll not show detail steps as reader can do that using
discussions in previous section [see discussions in Sec 2]. I just mention that
eliminating arbitrary function 𝜙 from GS (4.2) by partial differentiation w.r.t.
𝑥, 𝑦, we will obtain PDE (4.1), where
𝑃 = 𝑢𝑧 𝑣𝑦 − 𝑢𝑦 𝑣𝑧 , 𝑄 = 𝑢𝑥 𝑣𝑧 − 𝑢𝑧 𝑣𝑥 , 𝑅 = 𝑢𝑦 𝑣𝑥 − 𝑢𝑥 𝑣𝑦 . (4.5)
Next we will eliminate 𝑐1 , 𝑐2 from solutions 𝑢(𝑥, 𝑦, 𝑧) = 𝑐1 , 𝑣(𝑥, 𝑦, 𝑧) = 𝑐2 of
DE (4.4). Taking differentials of d𝑢 = 0 , d𝑣 = 0, we conclude that following
two equations must be compatible:
d𝑥 d𝑦 d𝑧
𝑢𝑥 d𝑥 + 𝑢𝑦 d𝑦 + 𝑢𝑧 d𝑧 = 0, = =
𝑃 𝑄 𝑅
i.e. we must have
𝑃𝑢𝑥 + 𝑄𝑢𝑦 + 𝑅𝑢𝑧 = 0. (4.6)
Similarly for the other solution, we have
𝑃𝑣𝑥 + 𝑄𝑣𝑦 + 𝑅𝑣𝑧 = 0. (4.7)
From (4.6) & (4.7), we get 𝑃, 𝑄, 𝑅 given by (4.5). Hence proved.
Geometrical Proof: In the first step, we will show that all integral surfaces (i.e.
solutions) of PDE (4.1) are generated by the integral curves (i.e. solutions) of
DE (4.4). In the second step, we will show that all surfaces generated by
integral curves of DE (4.4) are integral surfaces of PDE (4.1).
Step 1. Equations (4.2) and (4.3) implies that any integral surface of PDE (4.1)
is given by 𝑧 = 𝑓(𝑥, 𝑦), 𝑓 is arbitrary function. Then it is well-known from
differential calculus that d.ratios (direction ratios) of normal to this surface is
�𝐹𝑥 , 𝐹𝑦 , 𝐹𝑧 � = (𝑝, 𝑞, −1), where 𝐹(𝑥, 𝑦, 𝑧) = 𝑓(𝑥, 𝑦) − 𝑧. Now, the PDE (4.1) is
the geometric statement of the fact that the normal to the surface is

11
perpendicular to the direction defined by direction ratios(𝑃, 𝑄, 𝑅). This means
that the direction (𝑃, 𝑄, 𝑅) is tangential to the surface.
On the other hand, any integral curve of DE (4.4) has tangential directions at
each point M(𝑥, 𝑦, 𝑧) as (d𝑥, d𝑦, d𝑧) = (𝑃, 𝑄, 𝑅), which means that the integral
curve lies entirely on the surface.
Hence, if we start from an arbitrary point M(𝑥, 𝑦, 𝑧) on the surface and move
along the direction (𝑃, 𝑄, 𝑅), we will trace out an integral curve of DE (4.4),
which lies entirely on the surface. Furthermore, given 𝑃, 𝑄, 𝑅, integral curve
through a point M is unique. Thus it is proved that integral surface of PDE (1.1)
is generated by integral curves of AE (4.4).
Step 2: Suppose any integral curve of AE (4.4) generates the surface 𝑧 =
𝑓(𝑥, 𝑦), then that curve must be perpendicular to the tangent to the surface.
Clearly, direction ratios of the integral curve of AE (4.4) are (d𝑥, d𝑦, d𝑧) =
(𝑃, 𝑄, 𝑅), and tangential directions of the surface 𝑧 = 𝑓(𝑥, 𝑦) are (𝑝, 𝑞, −1).
Hence, geometrical property of perpendicularity is expressed by the following
relation
𝑃𝑝 + 𝑄𝑞 = 𝑅,
which is given PDE. Hence generated surface 𝑧 = 𝑓(𝑥, 𝑦) is integral surface of
PDE (4.1).
We are yet to prove that any surface generated by the integral curves of AE
(4.4) has an equation of the form (4.2). Let any curve on the surface which is
not a particular member of the system (4.3) have equations
𝜒(𝑥, 𝑦, 𝑧) = 0 , 𝜓(𝑥, 𝑦, 𝑧) = 0. (4.8)
If the curve (4.3) is a generating curve of the surface, it must intersect the curve
(4.8). The condition of this intersection will be obtained by eliminating 𝑥, 𝑦, 𝑧
from four equations (4.3) & (4.8). This relation will be a relation of the form
𝐹(𝑐1 , 𝑐2 ) = 0 . (4.9)
Thus the surface is generated by curve (4.3) which obey condition (4.9), and
hence have equation of the form (4.2).
Conversely, any surface of the form (4.2) is generated by integral curve (4.3) of
AE (4.4), because it is that surface generated by those curves which satisfy
relation (4.9).
This completes the proof.

12
Geometrical Interpretation of Lagrange’s Equation, Method of Characteristics
The 1st order linear PDE (4.1), also called Lagrange’s equation, and the DE
(4.4) in Lagrange’s method is called Lagrange’s auxiliary equation or simply
Auxiliary Equation (AE). Now, geometrical interpretation of Lagrange’s
method is already given in the geometrical proof of the method given above. Let
me point out the take away from above proof:
At any point M(𝑥, 𝑦, 𝑧) on the solution surface (known as Integral
Surface) 𝐹(𝑥, 𝑦, 𝑧) ≡ 𝑓(𝑥, 𝑦) − 𝑧 = 0, the direction (𝑝, 𝑞, −1) is normal
to the solution surface.
Given PDE 𝑃𝑝 + 𝑄𝑞 = 𝑅 implies that the direction (𝑃, 𝑄, 𝑅) is
perpendicular to the normal at each point M(𝑥, 𝑦, 𝑧), and hence the
direction (𝑃, 𝑄, 𝑅) is tangential to the solution surface. This triad
(𝑃, 𝑄, 𝑅) is called Characteristic direction.
Integral curve of AE (4.4) is such a curve in space such that the tangent to
this curve has d.ratios (𝑃, 𝑄, 𝑅). These curves
𝑢(𝑥, 𝑦, 𝑧) = 𝑐1 , 𝑣(𝑥, 𝑦, 𝑧) = 𝑐2 are called Characteristic Curves.
Integral surface of PDE is generated by Integral Curve of AE and
conversely Integral Curve of AE generates Integral surface of PDE.
The method of solving 1st order linear PDE, due to Lagrange, described above,
is also called method of characteristic. Before showing some examples on
Lagrange’s method, let me mention that the GI 𝜙(𝑢, 𝑣) = 0 may also be
expressed equivalently as 𝑢 = 𝑔(𝑣) or 𝑣 = ℎ(𝑣), because of the arbitrariness of
the functions.
Example: Find GI of PDE 𝑦 2 𝑝 − 𝑥𝑦𝑞 = 𝑥(𝑧 − 2𝑦) by the method of
characteristics
Solution: 𝑃 = 𝑦 2 , 𝑄 = −𝑥𝑦, 𝑅 = 𝑥(𝑧 − 𝑧𝑦) are characteristic directions.
Characteristic curves are integrals of the characteristic equations
d𝑥 d𝑦 d𝑧
= =
𝑦 2 −𝑥𝑦 𝑥(𝑧 − 2𝑦)
From 1st two fractions, 𝑥d𝑥 + 𝑦d𝑦 = 0 ⟹ 𝑥 2 + 𝑦 2 = 𝑐1 , a family of circles
with centre at origin and of radius √𝑐1 , one family of characteristic curves.
From last two, we get 1st order linear ODE:
d𝑧 1
+ � � 𝑧 = 2 ⟹IF= 𝑦 ⟹ 𝑦𝑧 − 𝑦 2 = 𝑐2 , 2nd family of characteristic curves
d𝑦 𝑦

Hence, GS is in terms of an arbitrary function: (𝑥 2 + 𝑦 2 , 𝑦𝑧 − 𝑦 2 ) = 0 .

13
• Integral Surfaces through a given Curve: Cauchy Problem
Given a 1st order PDE 𝑓(𝑥, 𝑦, 𝑧, 𝑝, 𝑞) = 0 and a curve Γ: 𝑥 = 𝑥0 (𝑡), 𝑦 =
𝑦0 (𝑡), 𝑧 = 𝑧0 (𝑡), 𝑡 ∈ [𝑎, 𝑏], the Cauchy Problem is to find a solution 𝑧 =
𝑧(𝑥, 𝑦) of the PDE such that on curve Γ, 𝑧 = 𝑧0 (𝑡) = 𝑧�𝑥0 (𝑡), 𝑦0 (𝑡)�, ∀ 𝑡 ∈
[𝑎, 𝑏], i.e. we are to find an integral surface of PDE which contains the given
curve Γ.
Consider 1st order linear PDE in its general form:
𝑃(𝑥, 𝑦, 𝑧(𝑥, 𝑦))𝑝 + 𝑄(𝑥, 𝑦, 𝑧(𝑥, 𝑦)𝑞 = 𝑅(𝑥, 𝑦, 𝑧(𝑥, 𝑦). (5.1)
Cauchy Data: Given curve Γ: 𝑥 = 𝑥0 (𝑡), 𝑦 = 𝑦0 (𝑡), 𝑧 = 𝑧0 (𝑡), 𝑡 ∈ [𝑎, 𝑏], 𝑡 is
just a parameter.
Our problem is to find the integral surface
𝐺�𝑢�𝑥(𝑡), 𝑦(𝑡), 𝑧(𝑥(𝑡), 𝑦(𝑡))�, 𝑣�𝑥(𝑡), 𝑦(𝑡), 𝑧(𝑥(𝑡), 𝑦(𝑡))� � = 0, which
contains the curve Γ, i.e. the surface passes through the point
�𝑥0 (𝑡), 𝑦0 (𝑡), 𝑧0 (𝑡)� for all 𝑡 ∈ [𝑎, 𝑏]. Note that this means that 𝐺 is not an
arbitrary function, it will be of specified form, which we are to find out.
Now, from the theory of 1st order linear PDE, discussed before, we know that
GI of PDE (5.1) is of the form
𝐹(𝑢(𝑥, 𝑦, 𝑧), 𝑣(𝑥, 𝑦, 𝑧)) = 0, [𝐹 is arbitrary function] (5.2a)
where
𝑢(𝑥, 𝑦, 𝑧) = 𝑐1 , 𝑣(𝑥, 𝑦, 𝑧) = 𝑐2 , (5.2b)
are solutions of Lagrange’s AE
d𝑥 d𝑦 d𝑧
= = . (5.2c)
𝑃 𝑄 𝑅

Equations (5.2a) & (5.2b) implies that we have a relation between two arbitrary
constants 𝑐1 , 𝑐2 of the form
𝐹(𝑐1 , 𝑐2 ) = 0. (5.3)
Now, if the integral surface pass through given curve Γ, characterized by single
parameter t in the following form:
Γ: 𝑥 = 𝑥0 (𝑡), 𝑦 = 𝑦0 (𝑡), 𝑧 = 𝑧0 (𝑡), 𝑡 ∈ [𝑎, 𝑏], (5.4)
then the integral curves 𝑢(𝑥, 𝑦, 𝑧), 𝑣(𝑥, 𝑦, 𝑧) must satisfy following two
equations:

14
 𝑢�𝑥0 (𝑡), 𝑦0 (𝑡), 𝑧0 (𝑡)� = 𝑐1 , 𝑣�𝑥0 (𝑡), 𝑦0 (𝑡), 𝑧0 (𝑡)� = 𝑐2 , ∀𝑡 ∈ [𝑎, 𝑏]. (5.5)
From two equations in (5.5), we can eliminate single parameter 𝑡, to obtain a
relation of the type
𝐺(𝑐1 , 𝑐2 ) = 0. (5.6)
Note that GI (5.3) contains (5.6) for 𝐹 ≡ 𝐺, because 𝐹 is an arbitrary function.
Hence, substituting for arbitrary constants 𝑐1 , 𝑐2 from equations (5.2b) into
(5.6), we obtain our desired integral surface as follows
𝐺�𝑢(𝑥, 𝑦, 𝑧), 𝑣(𝑥, 𝑦, 𝑧)� = 0. (5.7)
Note that the surface (5.7) contains given curve because of (5.5). Further, this
must be integral surface of PDE, as this is derived from GI (5.2a). In fact, it is a
PI of PDE.
Example: Find the integral surface of the PDE (𝑥 − 𝑦)𝑝 + (𝑦 − 𝑥 − 𝑧)𝑞 = 𝑧
through the circle 𝑥 2 + 𝑦 2 = 1, 𝑧 = 1.
d𝑥 d𝑦 d𝑧 d𝑥+d𝑦+d𝑧
Solution: Lagrange’s AE: = = ⟹ each ratio=
𝑥−𝑦 𝑦−𝑥−𝑧 𝑧 0

Hence, we get d(𝑥 + 𝑦 + 𝑧) = 0 ⟹ 𝑥 + 𝑦 + 𝑧 = 𝑐1

Next, considering 2nd & 3rd fraction, and using above integral curve, we have
d𝑦 d𝑧 d𝑦 d𝑧 2𝑦 − 𝑐1
= ⟹ = ⟹ = 𝑐2
𝑦−𝑥−𝑧 𝑧 2𝑦 − 𝑐1 𝑧 𝑧2
Substituting for 𝑐1 from 1st integral curve, we get 2nd integral curve
𝑦−𝑥−𝑧
= 𝑐2
𝑧2
Now, from above discussion, it is clear that for Cauchy problem, we are to find
a fixed relation between 𝑐1 , 𝑐2 using Cauchy data.
We see that given curve : 𝑥 2 + 𝑦 2 = 1. Putting 𝑧 = 1 in two integral curves,
2𝑦 = 𝑐1 + 𝑐2 , 𝑦 − 𝑥 = 𝑐2 + 1 ⟹ 2𝑥 = 𝑐1 − 𝑐2 − 2
Substituting for 𝑥, 𝑦 from above into 𝑥 2 + 𝑦 2 = 1, the fixed relation is
(𝑐1 − 𝑐2 − 2)2 + (𝑐1 + 𝑐2 )2 = 4 ⟹ 𝑐1 2 + 𝑐2 2 − 2(𝑐1 + 𝑐2 ) = 0
Finally, we substitute for 𝑐1 , 𝑐2 from the integral curves into above relation:
𝑧 4 (𝑥 + 𝑦 + 𝑧)2 + (𝑦 − 𝑥 − 𝑧)2 − 2𝑧 2 [𝑧 2 (𝑥 + 𝑦 + 𝑧) − (𝑦 − 𝑥 − 𝑧)] = 0

15
Remark: In above example, the curve is given in non-parametric form. See the
example below, where curve is given in parametric form. Parametric form was
taken above for theoretical discussion. Furthermore, non-parametric form of
curve in above problem may be put in parametric form also as follows:
Γ: 𝑥(𝑡) = cos 𝑡 , 𝑦(𝑡) = sin 𝑡 , 𝑧(𝑡) = 1, 0 ≤ 𝑡 ≤ 2𝜋
Thus, from the two integral curves, we must have
𝑐1 = 1 + sin 𝑡 + cos 𝑡 , 𝑐2 = sin 𝑡 − cos 𝑡 − 1
⟹ 2 sin 𝑡 = (𝑐1 + 𝑐2 ), 2 cos 𝑡 = 𝑐1 − 𝑐2 − 2
Eliminating 𝑡 from above equations, we get
(𝑐1 + 𝑐2 )2 + (𝑐1 − 𝑐2 − 2)2 = 4, which is the same fixed relation between
𝑐1 , 𝑐2 , and so substituting for them from integral curves, we get same surface.
Example: Solve the PDE (2𝑥𝑦 − 1)𝑝 + (𝑧 − 2𝑥 2 )𝑞 = 2(𝑥 − 𝑦𝑧) with initial
data : given straight line 𝑥0 (𝑠) = 1, 𝑦0 (𝑠) = 0, 𝑧0 (𝑠) = 𝑠
d𝑥 d𝑦 d𝑧
Solution: AE: = =
2𝑥𝑦−1 𝑧−2𝑥 2 2(𝑥−𝑦𝑧)

zd𝑥+d𝑦+𝑥d𝑧 2xd𝑥+2𝑦d𝑦+d𝑧
⟹ Each ratio= =
0 0

Thus, we get d(𝑥𝑧 + 𝑦) = 0 , d(𝑥 2 + 𝑦 2 + 𝑧) = 0

⟹ 𝑥𝑧 + 𝑦 = 𝑐1 , 𝑥 2 + 𝑦 2 + 𝑧 = 𝑐2
Hence, we may take GI in the form
𝑥 2 + 𝑦 2 + 𝑧 = 𝜙(𝑥𝑧 + 𝑦)
We are to find specified form of 𝜙 such that above integral surface passes
through the line 𝑥0 (𝑠) = 1, 𝑦0 (𝑠) = 0, 𝑧0 (𝑠) = 𝑠. Substituting for 𝑥, 𝑦, 𝑧 from
line to above surface, we get
1 + 0 + 𝑠 = 𝜙(1. 𝑠 + 0) ⟹ 𝜙(𝑠) = 1 + 𝑠
Thus, desired solution is 𝑥 2 + 𝑦 2 + 𝑧 = 1 + (𝑥𝑧 + 𝑦)
Before going to next section, I would like to show an example of finding
particular solution for given functional relation between the parameters from CI,
and also of finding GI and SI.
Example: Recall the example, discussed in Sec 2 [see 2nd example below (2.6)],
where eliminating 𝑎, 𝑏 from (𝑥 − 𝑎)2 + (𝑦 − 𝑏)2 + 𝑧 2 = 25, we get PDE
𝑧 2 (1 + 𝑝2 + 𝑞 2 ) = 25, so that former is CI of PDE.

16
Now, write down GI by taking 𝑏 = 𝜙(𝑎). Then choosing 𝜙(𝑎) = 𝑎, write down
particular solution. Finally find SI, if exists.
Solution: Given CI may be expressed as
𝐹(𝑥, 𝑦, 𝑧, 𝑎, 𝑏) ≡ (𝑥 − 𝑎)2 + (𝑦 − 𝑏)2 + 𝑧 2 − 25 = 0
Taking 𝑏 = 𝜙(𝑎), 𝜙 is arbitrary function, we have one parameter family of
surface:
2
𝐹�𝑥, 𝑦, 𝑧, 𝑎, 𝜙(𝑎)� ≡ (𝑥 − 𝑎)2 + �𝑦 − 𝜙(𝑎)� + 𝑧 2 − 25 = 0
The GI is the envelope of above surface, which is 𝑎-eliminant of above equation
and the following equation obtained by partial differentiation of 𝐹 w.r.t. 𝑎
𝐹𝑎 ≡ 𝑥 − 𝑎 + �𝑦 − 𝜙(𝑎)�𝜙 ′ (𝑎) = 0
Next choosing 𝜙(𝑎) = 𝑎, we get two equations from above two equations:
(𝑥 − 𝑎)2 + (𝑦 − 𝑎)2 + 𝑧 2 − 25 = 0, 𝑥 − 𝑎 + (𝑦 − 𝑎) = 0 ⟹ 2𝑎 = 𝑥 + 𝑦
Hence, particular solution of PDE for 𝜙(𝑎) = 𝑎: (𝑥 − 𝑦)2 + 2𝑧 2 = 50
Next, to see SI, if exists, we differentiate CI w.r.t. 𝑎, 𝑏:
(𝑥 − 𝑎) = 0, (𝑦 − 𝑏) = 0
To eliminate 𝑎, 𝑏 from given CI, we put 𝑎 = 𝑥, 𝑏 = 𝑦: 𝑧 = ±5
This is required SI. It is easy to verify that this is a solution of PDE.

• Orthogonal Surface to a given System of Surfaces

This is an application of the theory of 1st order linear PDE to Geometry. Here
we will determine the system of surfaces orthogonal to a given system of
surfaces. Suppose given system is a one-parameter family of surfaces
𝑓(𝑥, 𝑦, 𝑧) = 𝑐. (6.1)
A system of surfaces will be called orthogonal to given system (6.1), if former
cuts each of given surfaces at right angles. Suppose the orthogonal surfaces are
given by
𝐺(𝑥, 𝑦, 𝑧) ≡ 𝑔(𝑥, 𝑦) − 𝑧 = 0. (6.2)
From the geometric interpretation of Lagrange’s method for 1st order linear
PDE, we know that normal to given surfaces (6.1) at any point M(𝑥, 𝑦, 𝑧) on the

17
surface is the direction given by (𝑃, 𝑄, 𝑅) = �𝑓𝑥 , 𝑓𝑦 , 𝑓𝑧 �, and the normal at
common point M on orthogonal surfaces (6.2) is the direction given by
(𝑝, 𝑞, −1). Now since two systems (6.1) and (6.2) are orthogonal, two normal
must be perpendicular to each other, which transpires in mathematical language:
𝑃𝑝 + 𝑄𝑞 = 𝑅. (6.3)
Note that 𝑃 = 𝑓𝑥 , 𝑄 = 𝑓𝑦 , 𝑅 = 𝑓𝑧 are known, since 𝑓 is given function. Now,
PDE (6.3) is 1st order linear PDE, which can be solved by Lagrange’s method.
Conversely, any solution of linear PDE (6.3) must be orthogonal to every
surface characterized by given system (6.1), since relation (6.3) simply means
that normal to any solution of (6.3) is perpendicular to the normal to that
member of the given system (6.1) which passes through the same point.
Hence, orthogonal surface to given system (6.1) is determined by general
solution of linear PDE (6.3), so that the surfaces orthogonal to the system (6.1)
are the surfaces generated by the integral curves of characteristic equation of
PDE (6.3):
d𝑥 d𝑦 d𝑧 d𝑥 d𝑦 d𝑧
= = ⟹ = =
𝑃 𝑄 𝑅 𝑓𝑥 𝑓𝑦 𝑓𝑧
Example: Find the system of surfaces orthogonal to given system of surfaces
given by 𝑧(𝑥 + 𝑦) = 𝑐(3𝑧 + 1), 𝑐 is a free parameter.
𝑧(𝑥+𝑦) 𝑧 𝑧
Solution: Write 𝑓(𝑥, 𝑦, 𝑧) ≡ = 𝑐, we compute 𝑓𝑥 = , 𝑓𝑦 = ,
3𝑧+1 3𝑧+1 3𝑧+1
(𝑥+𝑦)
𝑓𝑧 = (3𝑧+1)2, so that orthogonal surfaces are general solutions of following PDE:

𝑧 𝑧 (𝑥 + 𝑦)
𝑓𝑥 𝑝 + 𝑓𝑦 𝑞 = 𝑓𝑧 ⟹ 𝑝+ 𝑞=
3𝑧 + 1 3𝑧 + 1 (3𝑧 + 1)2
d𝑥 d𝑦 d𝑧 d𝑥−d𝑦
AE is = = (𝑥+𝑦) ⟹each ratio= ⟹ 𝑢 ≡ 𝑥 − 𝑦 = 𝑐1 is one
𝑧(3𝑧+1) 𝑧(3𝑧+1) 0
integral curve. Also, writing AE as
d𝑥 d𝑦 𝑧(3𝑧+1)d𝑧 𝑥d𝑥+𝑦d𝑦−𝑧(3𝑧+1)d𝑧
= = (𝑥+𝑦)
⟹each ratio=
1 1 0

⟹ 𝑣 ≡ 𝑥 2 + 𝑦 2 − 2𝑧 3 − 𝑧 2 = 𝑐2 , which is our 2nd integral curve.

Hence, orthogonal surfaces are given by the following system
𝑥 2 + 𝑦 2 − 2𝑧 3 − 𝑧 2 = 𝜙(𝑥 − 𝑦), 𝜙 is arbitrary function.
************End of Part A**************

18