Dr.

Sanjay Yadav, Associate professor at K L University, has a significant experience of working

with national and international students, has a significant experience of working for 13 years in
academics and research space. His specialized delivery expertise in his areas of interest such as
partial differential equation, data securities and cryptography, have won his many accolades. A
Ph.D. in applied mathematics with specialized in Partial differential equations, resulting 14
research publication in SCI & Scopus journals and guided 2 PhD and 15 master projects. He has
attended about 13 national and international conference besides being an integral part of various
workshops and FDPs as a coordinator and recourse person.
Partial Differential Equations
Course Description.

The study of partial differential equations (PDEs)is a part of mathematics that has vast applications
in the field of Applied Science and technology. This course aims to teach the basics of Partial
differential equations (PDEs), a subject that touches on many branches of pure mathematics,
applied mathematics, as well as physics and applied science. It is addressed to first and second-
year graduate students, or anyone with an interest in the topic. Partial differential equations are a
very rich subject; so much so that at a research level most workers in the field specialize in one of
the many sub-fields. The aim of this one-semester course is both to give an overview of the subject
as much as possible and introduce some tools that are used throughout. This should prepare
students adequately for the many more advanced courses in PDEs that are (and will be) offered in
the department. In this course as we know PDEs appear as mathematical models for many physical
phenomena. Closed-forum solutions to most of the PDEs cannot found. One of the possible ways
to understand the models is by studying the qualitative properties exhibited by their solutions.
Overall, In this course, we study first order linear and nonlinear PDEs, and the properties of the
three important types of second order linear PDEs (Wave, Laplace, Heat) would be studied and
compared.

Pre-requisites: The main pre-requisite is some basic real analysis, primarily Lebesgue measure on
Rn, the basics of the Fourier transform on Rn and some functional analysis, along with
multivariable calculus (Stokes' theorem). Familiarity with the general theory of Ordinary
Differential equations is desirable, but not a prerequisite. Exposure to partial derivates will be
helpful.

Focus: The focus on this course is on the general theory of PDEs. There will be some discussion
of finding explicit solutions to specific equations with specific methods. There will be no
discussion of numerical analysis related to PDEs.
The Partial Differential Equations course contains Four Module

MODULE 1: BASICS AND FIRST ORDER PDE

Basics of partial differential equations, Modeling with partial differential equations,
Partial differential equations of first order, Linear first order P.D.E., Method of
characteristics, Lagrange, Charpit's method.

MODULE 2: SECOND ORDER PDE

Partial differential equation of second order, Classification of second order equation,

Hyperbolic, Parabolic and Elliptic equations, Linear second order partial differential
equations with constant coefficients.
MODULE 3: ELLIPTIC PDES AND ITS SOLUTIONS

Elliptic Equations: Laplace equation in Cartesian, polar, spherical and cylindrical

coordinates and its solution by Fourier series method, Poisson equation in 2D.

MODULE 4 ; HYPERBOLIC AND PARABOLIC:

Hyperbolic differential equations, One dimensional wave equation, Solution of the wave
equation by Method of separation of variables, Boundary and initial value problem of two
dimensional wave equation. Parabolic differential equations, One dimensional diffusion
equation, Boundary conditions; Method of separation of variables, Solutions in cylindrical
and spherical equation
Table of contents
Module -1
Unit 1.1 Basics
Unit 1.2 Definitions

Module -2
Second order partial differential equations and its classifications
Unit 2.1 Classification of Second order PDEs
Unit 2.2 Constant coefficient to find homogenous and non-homogenous solution

Module -3
One dimensional elliptic equation
Unit 3.1 One dimensional Elliptic formation
Unit 3.2 Solution of Elliptic equations

Module -4
ONE DIMENSIONAL HYPERBOLIC AND PARABOLIC EQUATIONS
Unit 4.1 Formation of one dimensional hyperbolic and parabolic equations
Unit 4.2 Solution of One dimensional Hyperbolic PDEs
Unit-4.3 Solution of One dimensional parabolic PDEs
Partial Differential Equations

The reader has, already been introduced to the notion of partial differential equations.
Here, we shall begin by studying the ways in which partial differential equations are formed.
Then we shall investigate the solutions of special types of partial differential equations of the
first and higher orders.
In what follows x and y will, usually be taken as the independent variables and z, the
dependent variable so that z = f (x, y) and we shall employ the following notation:

𝜕𝑧 𝜕𝑧 𝜕 2𝑧 𝜕𝑧2 𝜕 2𝑧
= 𝑝, = 𝑞, 2 = 𝑟 = 𝑠, =𝑠
𝜕𝑥 𝜕𝑦 𝜕𝑥 𝜕𝑥𝜕𝑦 𝜕𝑥𝜕𝑦

1.1.1 FORMATION OF PARTIAL DIFFERENTIAL EQUATIONS

Unlike the case of ordinary differential equations which arise from the elimination of
arbitrary constants; the partial differential equations can be formed either by the elimination of
arbitrary constants or by the elimination of arbitrary functions from a relation involving three or
more variables. The method is best illustrated by the following examples:
Example 17.1. Derive partial differential equation (by eliminating the constants) from the
equation

Solution. Differentiating (i) partially with respect to x and y, we get-

𝜕𝑧 2𝑥 1 1 𝜕𝑧 𝑝
2 = 2 𝑜𝑟 2 = = 𝑥
𝜕𝑥 𝑎 𝑎 𝑥 𝜕𝑥
2𝜕𝑧 2𝑦 1 1 𝜕𝑧 𝑞
= 2 𝑜𝑟 2 = =
𝜕𝑦 𝑏 𝑏 𝑦 𝜕𝑦 𝑦
and
Substituting these values of I/a2 and I/b2 in (i), we get

2z=x p + yq

as the desired partial differential equation of the first order.

Example 1. Form the partial differential equations (by eliminating the arbitrary constants) from
(𝑎)𝑧 = (𝑥 + 𝑦)𝜙(𝑥 2 + 𝑦 2 )
(𝑏)𝑧 = 𝑓(𝑥 + 𝑎𝑡) + 𝑔(𝑥 − 𝑎𝑡)
(𝑐)𝑓(𝑥 2 + 𝑦 2 , 𝑧 − 𝑥𝑦) = 0

Solution. (a) We have 𝑧 = (𝑥 + 𝑦)𝜙(𝑥 2 + 𝑦 2 )

Differentiating z partially with respect to x and y,

𝜕𝑧
𝑝= = (𝑥 + 𝑦)𝜙 ′ (𝑥 2 − 𝑦 2 ) ⋅ 2𝑥 + 𝜙(𝑥 2 + 𝑦 2 )
𝜕𝑥
𝜕𝑧
𝑞= = (𝑥 + 𝑦)𝜙 ′ (𝑥 2 − 𝑦 2 ). (−2𝑦) + 𝜙(𝑥 2 − 𝑦 2 )
𝜕𝑥
2
𝑝− = 2𝑥(𝑥 + 𝑦)𝜙 ′ (𝑥 2 − 𝑦 2 )
𝑥+𝑦
2
𝑞− = −2𝑦(𝑥 + 𝑦)𝜙 ′ (𝑥 2 − 𝑦 2 )
𝑥+𝑦

𝑝−2⁄(𝑥+𝑦) 𝑥
Division gives 𝑞−𝑧⁄(𝑥+𝑦) = 𝑦

i.e., [p(x+y)-z]y+[q(x+y)-z]x
(x+y)(py+qx)-z(x+y)=0
Hence, py+qz=z is required equation.

(b) We have 𝑧 = 𝑓(𝑥 + 𝑎𝑡) + 𝑔(𝑥 − 𝑎𝑡)

Differentiating z partially with respect to x and t,
𝜕𝑧 ′ (𝑥 ′ (𝑥
𝜕 2𝑧
=𝑓 + 𝑎𝑡) + 𝑔 − 𝑎𝑡), 2 = 𝑓 ′′ (𝑥 + 𝑎𝑡) + 𝑔′′ (𝑥 − 𝑎𝑡)
𝜕𝑥 𝜕𝑥
𝜕𝑧 ′ (𝑥 ′ (𝑥 𝜕2𝑧 2 ′′ (𝑥 𝜕2𝑧
= 𝑎𝑓 + 𝑎𝑡) − 𝑎𝑔 − 𝑎𝑡), 2 = 𝑎 𝑓 + 𝑎𝑡) + 𝑎2 𝑔′′ (𝑥 − 𝑎𝑡)=𝑎2 2
𝜕𝑡 𝜕𝑡 𝜕𝑥

𝜕2𝑧 𝜕2𝑧
Thus, the desired partial differential equation is𝜕𝑡 2 = 𝑎2 𝜕𝑥 2

which is an equation of the second order and (i) is its solution.

(c) Let 𝑥 2 + 𝑦 2 = 𝑢 and 𝑧 − 𝑥𝑦 = 𝑣 so that 𝑓(𝑢, 𝑣) = 0.

Differentiating partially with respect to x and y, we have
 𝜕𝑓 𝜕𝑢 +𝜕𝑢 𝜕𝑓 𝜕𝑣 𝜕𝑣
( 𝑝) + ( + 𝑝) = 0
𝜕𝑢 𝜕𝑥 𝜕𝑧 𝜕𝑣 𝜕𝑥 𝜕𝑧
or
𝜕𝑓 𝜕𝑓
(2𝑥) + (−𝑦 + 𝑝) = 0
𝜕𝑢 𝜕𝑣
(i)
And
𝜕𝑓 𝜕𝑢 +𝜕𝑢 𝜕𝑓 𝜕𝑣 𝜕𝑣
( 𝑞) + ( + 𝑞) = 0
𝜕𝑢 𝜕𝑥 𝜕𝑧 𝜕𝑣 𝜕𝑥 𝜕𝑧
or
𝜕𝑓 𝜕𝑓
(2𝑦) + (−𝑥 + 𝑞) = 0
𝜕𝑢 𝜕𝑣
(ii)
𝜕𝑓 𝜕𝑓
Eliminating and from (i) and (ii), we get
𝜕𝑢 𝜕𝑣
2𝑥 −𝑦 + 𝑝
| | = 0 𝑜𝑟 𝑥𝑞 − 𝑦𝑝 = 𝑥 2 − 𝑦 2
2𝑦 −𝑥 + 𝑞

Example 2 Find the differential equation of all planes which are at a constant distance a from the
origin.
Solution. The equation of the plane in ‘normal form’ is
𝑙𝑥 + 𝑚𝑦 + 𝑛𝑧 = 𝑎

where l, m, n are the d.c.s of the normal! from the origin to the plane.
Then
𝑙 2 + 𝑚2 + 𝑛2 = 𝑙 𝑜𝑟 𝑛 = √(1 − 𝑙 2 − 𝑚2 )
(i) becomes
𝑙𝑥 + 𝑚𝑦 + √(1 − 𝑙 2 − 𝑚2 )𝑧 = 𝑎

Differentiating partially with respect to x, we get

𝑙 + √(1 − 𝑙 2 − 𝑚2 ). 𝑝 = 0
Differentiating partially with respect to y, we get
𝑚 + √(1 − 𝑙 2 − 𝑚2 ). 𝑞 = 0

Now we have to eliminate l,m from(ii),(iii) and (iv).

From (iii),l=−√(1 − 𝑙 2 − 𝑚2 ). 𝑝 and m=−√(1 − 𝑙 2 − 𝑚2 ). 𝑞
Squaring and adding , 𝑙 2 + 𝑚2 = (1 − 𝑙 2 − 𝑚2 )(𝑝2 + 𝑞 2 )
or

𝑝2 +𝑞 2 1
(1 + 𝑝2 + 𝑚2 )(𝑙 2 + 𝑚2 ) = 𝑝2 + 𝑞 2 𝑜𝑟 1 − 𝑙 2 − 𝑚2 = 1 − = 1+𝑝2+𝑞2
1+𝑝2 +𝑞2
580

Also
−𝑝 −𝑞
𝑙= 𝑎𝑛𝑑 𝑚 =
√1 + 𝑝2 + 𝑞 2 √1 + 𝑝2 + 𝑞 2
Substituting the values of l,m and 1 − 𝑙 2 − 𝑚2 in (ii), we obtain

−𝑝𝑥 𝑞 1
− + 𝑧=𝑎
√1 + 𝑝2 + 𝑞 2 √1 + 𝑝2 + 𝑞 2 √1 + 𝑝2 + 𝑞 2

Or 𝑧 = 𝑝𝑥 + 𝑞𝑦 + 𝑎√1 + 𝑝2 + 𝑞 2 which is the required partial differential equation.

PROBLEMS 3
Form the partial differential equations (by eliminating the arbitrary constants) from:
5.Find the differential equation of all spheres of fixed radius having their centres in the xy-plane,
(Madras, 2000 S )
6. Find the differential equation of all spheres whose centres lie on the z-axis. {Kerela, 2005)
Form the partial differential equations (by eliminating the arbitrary constants) from

1.2 SOLUTIONS OF A PARTIAL DIFFERENTIAL EQUATION

It is clear from the above examples that a partial differential equation can result both from
elimination of arbitrary constants and from the elimination of arbitrary functions.
The solution ...(1) of a first order partial differential equation which
contains two arbitrary constants is called a complete integral.
A solution obtained from the complete integral by assigning particular values to the
arbitrary constants is called a particular integral.
If we put b = (a) in (1) and find the envelope of the family of surfaces fix, y, z, = O,
then we get a solution containing an arbitrary function 4, which is called the general integral.
The envelope of the family of surfaces (I), with parameters a and b, if it exists, is called
single integral. The singular integral differs from the particular integral in that it is not
obtained from the complete integral by giving particular values to the constants.

1.1.3 EQUATIONS SOLVABLE BY DIRECT INTEGRATION

We now consider such partial differential equations which can be solved by direct
integration. In place Of the usual constants of integration, we must, however, use arbitrary
functions of the variable held fixed.
 𝜕3𝑦
Example 1. Solve 𝜕𝑥 2 𝜕𝑦 + 18𝑥𝑦 2 + 𝑠𝑖𝑛(2𝑥 − 𝑦) = 0

Solution. Integrating twice with respect to x (keeping y fixed),

𝜕 2𝑧 1
+ 9𝑥 2 𝑦 2 − 𝑐𝑜𝑠(2𝑥 − 𝑦) = 𝑓(𝑦)
𝜕𝑥𝜕𝑦 2
𝜕𝑧 1
+ 3𝑥 3 𝑦 3 − 𝑠𝑖𝑛(2𝑥 − 𝑦) = 𝑥𝑓(𝑦) + 𝑔(𝑦)
𝜕𝑦 4
Now integrating with respect toy (keeping* fixed)
1
𝑧 + 𝑥 3 𝑦 3 − 𝑐𝑜𝑠(2𝑥 − 𝑦) = 𝑥∫ 𝑓(𝑦) 𝑑𝑦 + ∫ 𝑔(𝑦) 𝑑𝑦 + 𝑤(𝑥)
4
The result may be simplified by writing

∫ 𝑓(𝑦) 𝑑𝑦 = 𝑢(𝑦) 𝑎𝑛𝑑 ∫ 𝑔(𝑦)𝑑𝑦 = 𝑢(𝑦)

1
Thus 𝑧 = 4 𝑐𝑜𝑠(2𝑥 − 𝑦) − 𝑥 3 𝑦 3 + 𝑥𝑢(𝑦) + 𝑣(𝑦) = 𝑤(𝑥) where u,v,w are arbitrary functions.

𝜕2 𝑧 𝜕𝑧
Example . Solve𝜕𝑥 2 + 𝑧 = 0 ,given that when x=0, z= ⅇ 𝑦 and 𝜕𝑥 = 1

Solution. If z were function of x alone, the solution would have been z = A sin x + B cos x,
where A and B are constants. Since z is a function of x and y, A and B can be arbitrary functions
of y. Hence the solution of the given equation is 𝑧 = 𝑓(𝑦) 𝑠𝑖𝑛 𝑥 + 𝜙(𝑦) 𝑐𝑜𝑠 𝑥

𝜕𝑧
= 𝑓(𝑦) 𝑐𝑜𝑠 𝑥 − 𝜙(𝑦) 𝑠𝑖𝑛 𝑥
𝜕𝑥
When x=0;z=ⅇ 𝑦 ,ⅇ 𝑦 = ϕ(y).
𝜕𝑧
When x=0, 𝜕𝑥 = 1,1= 𝑓(𝑦)

Hence the desired solution is 𝑧 = 𝑠𝑖𝑛 𝑥 + ⅇ 𝑦 𝑐𝑜𝑠 𝑥.

𝜕2𝑧 𝜕𝑧
Example Solve = 𝑠𝑖𝑛𝑥𝑠𝑖𝑛𝑦 ,for which = −2𝑠𝑖𝑛𝑦 when x=0 and z=0 when y is an odd
𝜕𝑥𝜕𝑦 𝜕𝑦
multiple of π/2.

𝜕2𝑧
Solution. Given equation is 𝜕𝑥𝜕𝑦 = 𝑠𝑖𝑛𝑥𝑠𝑖𝑛𝑦

Integrating with respect to x (keeping y fixed), we get

 𝜕𝑧
= −𝑐𝑜𝑠𝑥𝑠𝑖𝑛𝑦 + 𝑓(𝑦)
𝜕𝑦
𝜕𝑧
When x=0, 𝜕𝑦 = −2𝑠𝑖𝑛𝑦,

−2𝑠𝑖𝑛𝑦 = −𝑠𝑖𝑛𝑦 + 𝑓(𝑦) 𝑜𝑟 𝑓(𝑦) = −𝑠𝑖𝑛𝑦

𝜕𝑧
(i). becomes 𝜕𝑦 = −𝑐𝑜𝑠𝑥𝑠𝑖𝑛𝑦 − 𝑠𝑖𝑛𝑦

Integrating with respect to y(keeping x fixed), we get

𝑧 = 𝑐𝑜𝑠𝑥𝑐𝑜𝑥𝑦 + 𝑐𝑜𝑠𝑦 + 𝑔(𝑥)
When y is the odd multiple of π/2 ,z=0.

0 = 0 + 0 + 𝑔(𝑥) 𝑜𝑟 𝑔(𝑥) = 0

Hence from (ii), the complete solution is 𝑧 = (1 + 𝑐𝑜𝑠𝑥)𝑐𝑜𝑠𝑦

PROBLEM-17.2

LINEAR EQUATIONS OF THE FIRST ORDER

A linear partial differential equation of the first order, commonly known as Lagrange's
Linear equation* , is of the form
Pp + Qq = R
Where P, Q and R are functions of x, y, z. This equation is called a quasi-linear equation.
When P, Q and R are independent of z it is known as linear equation.
Such an equation is obtained by eliminating an arbitrary function from +(u,
v) = O where u, v are some functions of x, y, z.
Differentiating (2) partially with respect to x and y.
𝜕𝜙 𝜕𝑢 𝜕𝑢 𝜕𝜙 𝜕𝑣 𝜕𝑣 𝜕𝜙 𝜕𝑢 𝜕𝑢 𝜕𝜙 𝜕𝑣 𝜕𝑣
( + 𝑝) + ( + 𝑝) = 0 𝑎𝑛𝑑 ( + 𝑞) + ( + 𝑞) = 0
𝜕𝑢 𝜕𝑥 𝜕𝑧 𝜕𝑣 𝜕𝑥 𝜕𝑧 𝜕𝑢 𝜕𝑦 𝜕𝑧 𝜕𝑣 𝜕𝑦 𝜕𝑧
 𝜕𝑢 𝜕𝑢 𝜕𝑣 𝜕𝑣
𝜕𝜙 𝜕𝜙
+ 𝜕𝑧 𝑝 + 𝜕𝑧 𝑝
𝜕𝑥 𝜕𝑥
Eliminating 𝜕𝑢 and , we get |𝜕𝑢 𝜕𝑢 𝜕𝑣 𝜕𝑣 |=0
𝜕𝑣
+ 𝜕𝑧 𝑞 + 𝜕𝑧 𝑞
𝜕𝑦 𝜕𝑦

𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
Which simplifies to(𝜕𝑦 𝜕𝑧 − 𝜕𝑧 𝜕𝑦) 𝑝 + ( 𝜕𝑧 𝜕𝑥 − 𝜕𝑥 𝜕𝑧 ) 𝑞 = (𝜕𝑥 𝜕𝑦 − 𝜕𝑥 𝜕𝑦)

This is of the same form as (1).

Now suppose u=a and v=b, where a,b are constants , so that
𝜕𝑢 𝜕𝑢 𝜕𝑢
𝑑𝑥 + 𝑑𝑦 + 𝑑𝑧 = 𝑑𝑢 = 0
𝜕𝑥 𝜕𝑦 𝜕𝑧
𝜕𝑣 𝜕𝑣 𝜕𝑣
𝑑𝑥 + 𝑑𝑦 + 𝑑𝑧 = 𝑑𝑣 = 0
𝜕𝑥 𝜕𝑦 𝜕𝑧
By cross-multiplication, we have
𝑑𝑥 𝑑𝑦 𝑑𝑧
= =
𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
− − −
𝜕𝑦 𝜕𝑧 𝜕𝑧 𝜕𝑦 𝜕𝑧 𝜕𝑥 𝜕𝑥 𝜕𝑧 𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑥
𝑑𝑥 𝑑𝑦 𝑑𝑧
= =
𝑃 𝑄 𝑅
The solutions of these equations are u = a and v = b.
𝜙 (u, v) = O is the required solution of
Thus to solve the equation Pp+Qq=R
(i)form the subsidiary equation
𝑑𝑥 𝑑𝑦 𝑑𝑧
= =
𝑃 𝑄 𝑅
(ii) solve these simultaneous equations by the method of 16.10 giving u=a and v=b as its
solution
(iii) write the complete solution as ϕ(u,v)=0 or u= 𝑓 (v)

𝑦2𝑧 2 𝑧𝑞
Example Solve 𝑝+𝑥 = 𝑦2
𝑥

Solution. Rewriting the given equation as

𝑦 2 𝑧𝑝 + 𝑥 2 𝑧𝑞 = 𝑦 2 𝑥
The subsidiary equation are
𝑑𝑥 𝑑𝑦 𝑑𝑧
2
= 2 = 2
𝑦 𝑧 𝑥 𝑧 𝑦 𝑥
The first two fractions give 𝑥 2 𝑑𝑥 = 𝑦 2 𝑑𝑦
Integrating we get, 𝑥 3 − 𝑦 3 = 𝑎
Again, the first and third fractions give 𝑥𝑑𝑥 = 𝑧𝑑𝑧
Integrating, we get 𝑥 2 − 𝑧 2 = 𝑏
Hence from (i) and (ii) , the complete solution is
𝑥 3 − 𝑦 3 = 𝑓(𝑥 2 − 𝑧 2 )

𝜕𝑧 𝜕𝑧
Example Solve (𝑚𝑧 − 𝑛𝑦) 𝜕𝑥 + (𝑛𝑥 − 𝑙𝑧) 𝜕𝑦 = 𝑙𝑦 − 𝑚𝑥
Solution. Here, the subsidiary equations are
𝑑𝑥 𝑑𝑦 𝑑𝑧
= =
𝑚𝑧 − 𝑛𝑦 𝑛𝑥 − 𝑙𝑧 𝑙𝑦 − 𝑚𝑥
Using multipliers x, y, and z, we get each fraction =
𝑥 𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧
0

𝑥 𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧 =0 which on integration gives 𝑥 2 + 𝑦 2 + 𝑧 2 = 𝑎

Again, using multipliers l, m and n, we get each fraction =
𝑙𝑑𝑥 + 𝑚𝑑𝑦 + 𝑛𝑑𝑧
0

𝑙𝑑𝑥 + 𝑚𝑑𝑦 + 𝑛𝑑𝑧 = 0 which on integration gives 𝑙𝑥 + 𝑚𝑦 + 𝑛𝑧 = 𝑏

Hence from (i) and (ii), the required solution is
𝑥 2 + 𝑦 2 + 𝑧 2 = 𝑓(𝑙𝑥 + 𝑚𝑦 + 𝑛𝑧)

Example Solve (𝑥 2 − 𝑦 2 − 𝑧 2 )𝑝 = 2𝑥𝑦𝑞 = 2𝑥𝑧

Solution. Here the subsidiary equations are
𝑑𝑥 𝑑𝑦 𝑑𝑧
2 2 2
= =
𝑥 −𝑦 −𝑧 2𝑥𝑦 2𝑥𝑧
ⅆ𝑦 ⅆ𝑧
From the last two fractions, we have =
𝑦 𝑧
Which on integration gives log y=log z + log a or y/z=a
Using multipliers x, y, and z, we have each fraction=
 𝑥 𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧
𝑥(𝑥 2 + 𝑦 2 + 𝑧 2 )

2𝑥 𝑑𝑥 + 2𝑦𝑑𝑦 + 2𝑧𝑑𝑧 𝑑𝑧
=
𝑥(𝑥 2 + 𝑦 2 + 𝑧 2 ) 𝑧
Which on integration gives log(𝑥 2 + 𝑦 2 + 𝑧 2 ) = 𝑙𝑜𝑔𝑧 + 𝑙𝑜𝑔𝑏
𝑥2 + 𝑦2 + 𝑧2
=𝑏
𝑧
Hence from (i) and (ii), the required solution is
𝑦
𝑥 2 + 𝑦 2 + 𝑧 2 = 𝑧𝑓 ( )
𝑧

Example Solve 𝑥 2 (𝑦 − 𝑧)𝑝 + 𝑦 2 (𝑧 − 𝑥)𝑞 = 𝑧 2 (𝑥 − 𝑦)

Solution. Here the subsidiary equations are
𝑑𝑥 𝑑𝑦 𝑑𝑧
= =
𝑥 2 (𝑦 − 𝑧) 𝑦 2 (𝑧 − 𝑥) 𝑧 2 (𝑥 − 𝑦)
Using multipliers 1/x, 1/y, and 1/z, we have each fraction=
1 1 1
𝑥 . 𝑑𝑥 + 𝑦 . 𝑑𝑦 + 𝑧 . 𝑑𝑧
0
𝑑𝑥 𝑑𝑦 𝑑𝑧
= = =0
𝑥 𝑦 𝑧
Which on integration gives 𝑙𝑜𝑔𝑥 + 𝑙𝑜𝑔𝑦 + 𝑙𝑜𝑔𝑧 = 𝑙𝑜𝑔𝑎 𝑜𝑟 𝑥𝑦𝑥 = 0
Using multipliers 1/𝑥 2 , 1/𝑦 2 , and 1/𝑧 2 , we have each fraction=
1 1 1
2 . 𝑑𝑥 + 2 . 𝑑𝑦 + 2 . 𝑑𝑧
𝑥 𝑦 𝑧
0
𝑑𝑥 𝑑𝑦 𝑑𝑧
= = =0
𝑥2 𝑦2 𝑧2
Which on integration gives -
1 1 1
+ + =0
𝑥 𝑦 𝑧

Hence from (i) and (ii), the required solution is

1 1 1
𝑥𝑦𝑧 = 𝑓 ( + + )
𝑥 𝑦 𝑧
Example Solve (𝑥 2 − 𝑦𝑧)𝑝 − (𝑦 2 − 𝑧𝑥)𝑞 = 𝑧 2 − 𝑥𝑦
 Solution. Here the subsidiary equations are
𝑑𝑥 𝑑𝑦 𝑑𝑧
2
= 2 = 2
𝑥 − 𝑦𝑧 𝑦 − 𝑧𝑥 𝑧 − 𝑥𝑦

ⅆ𝑥−ⅆ𝑦 ⅆ𝑦−ⅆ𝑧
Each of these equations=𝑥 2 −𝑦 2−(𝑦−𝑥)𝑧 = 𝑦 2−𝑧 2−𝑥(𝑧−𝑦)
𝑑(𝑥 − 𝑦) 𝑑(𝑦 − 𝑧) 𝑑(𝑥 − 𝑦) 𝑑(𝑦 − 𝑧)
= 𝑜𝑟 =
(𝑥 − 𝑦)(𝑥 + 𝑦 + 𝑧) (𝑦 − 𝑧)(𝑥 + 𝑦 + 𝑧) (𝑥 − 𝑦) (𝑦 − 𝑧)
𝑥−𝑦
Which on integration gives, log(x-y)=log(y-z)+log c or 𝑦−𝑧 = 𝑐
Each of the subsidiary equations (i)=
𝑥 𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧
𝑥3 + 𝑦 3 + 𝑧 3 − 3𝑥𝑦𝑧
𝑥 𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧
(𝑥 + 𝑦 + 𝑧)(𝑥 2 + 𝑦 2 + 𝑧 2 − 𝑦𝑧 − 𝑧𝑥 − 𝑥𝑦)
Also, each of the subsidiary equations =
𝑥 𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧
𝑥2 + 𝑦 2 + 𝑧 2 − 𝑦𝑧 − 𝑧𝑥 − 𝑥𝑦
Equating (iii) and (iv) and cancelling the common factor, we get
𝑥 𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧
= 𝑑𝑥 + 𝑑𝑦 + 𝑑𝑧
𝑥+𝑦+𝑧

Or ∫ (𝑥 𝑑𝑥 + 𝑦𝑑𝑦 + 𝑧𝑑𝑧) = ∫ (𝑥 + 𝑦 + 𝑧)𝑑(𝑥 + 𝑦 + 𝑧) + 𝑐 ′

Or 𝑥 2 + 𝑦 2 + 𝑧 2 = (𝑥 + 𝑦 + 𝑧)2 + 2𝑐 ′ 𝑜𝑟 𝑥𝑦 + 𝑦𝑧 + 𝑧𝑥 + 𝑐 ′ = 0

Combining (ii) and (v), the general solution is

𝑥−𝑦
= 𝑓(𝑥𝑦 + 𝑦𝑧 + 𝑧𝑥)
𝑦−𝑧

NON-LINEAR EQUATIONS OF THE FIRST ORDER

Those equations in which p and q occur other than in the first degree are called non-linear
partial differential equations of the first order. The complete solution Of such an equation
contains only two arbitrary constants (i.e., equal to the number of independent variables
involved) and the particular integral is obtained by giving particular values to the constants.]
Here we shall discuss four standard forms of these equations.
 Form I. f(p, q) = 0, i.e., equations containing p and q only.
Its complete solution is z = ax + by + c
where a and b are connected by the relation f (a, b) = O ...(2)
𝜕𝑧 𝜕𝑧
[Since from (1).p=𝜕𝑥 = 𝑎 𝑎𝑛𝑑 𝑞 = 𝜕𝑦 = 𝑏 which when substituted in (2) give f (p, q) = 0]

Expressing (2) as b = ϕ (a) and substituting this value of b in (I), we get the required
solution as z = ax + ϕ (a)y+c in which a and c are arbitrary constants.

Example 17.12 Solve 𝑝 − 𝑞 = 1

Solution. The complete solution is z = ax + by + c where a - b = I
Hence z = ax + a -1y + c is the desired solution

Example Solve 𝑥 2 𝑝2 + 𝑦 2 𝑞 2 = 𝑧 2
Solution. Given equation can be reduced to the above form by writing it as-
𝑥 𝜕𝑧 2 𝑦 𝜕𝑧 2
( ⋅ ) +( ⋅ ) =1
𝑧 𝜕𝑥 𝑧 𝜕𝑦
ⅆ𝑥 ⅆ𝑦 ⅆ𝑧
And setting = 𝑑𝑢, = 𝑑𝑣, = 𝑑𝜔 𝑠𝑜 𝑡ℎ𝑎𝑡 𝑢 = 𝑙𝑜𝑔𝑥, 𝑣 = log 𝑦, 𝑤 = log 𝑧
𝑥 𝑦 𝑧
𝜕𝑤 2 𝜕𝑤 2
Then (i) becomes ( 𝜕𝑢 ) + ( 𝜕𝑣 ) = 1
𝜕𝜔 𝜕𝜔
𝑃2 + 𝑄 2 = 1 𝑤ℎⅇ𝑟ⅇ 𝑃 = ,𝑄 =
𝜕𝑢 𝜕𝑣
Its complete solution is 𝜔 = 𝑎𝑢 + 𝑏𝑣 + 𝑐
Where 𝑎2 + 𝑏 2 = 1 𝑜𝑟 𝑏 = √(1 − 𝑎2 )
(ii) becomes 𝜔 = 𝑎𝑢 + √(1 − 𝑎2 )𝑣 + 𝑐
Or 𝑙𝑜𝑔 𝑧 = 𝑎 𝑙𝑜𝑔 𝑥 + √(1 − 𝑎2 ) 𝑙𝑜𝑔 𝑦 + 𝑐 which is the required solution

Form 11. f(z, p, q) = 0, i.e. , equations not containing x and y.

As a trial solution, assume that z is a function of u = x + ay, where a is an arbitrary constant.
𝜕𝑧 𝑑𝑧 𝜕𝑢 𝑑𝑧 𝜕𝑧 𝑑𝑧 𝜕𝑢 𝑑𝑧
𝑝= = ⋅ = 𝑞= = ⋅ =𝑎
𝜕𝑥 𝑑𝑢 𝜕𝑥 𝑑𝑢 𝜕𝑦 𝑑𝑢 𝜕𝑦 𝑑𝑢
Substituting the values of p and q in f (z, p, q) = 0, we get
𝜕𝑧 𝑑𝑧
𝑓 (𝑧, ,𝑎 ) = 0
𝜕𝑢 𝑑𝑢

Which is an ordinary differential equation of the first order

 ⅆ𝑧
Rewriting it as ⅆ𝑢= ϕ (z, a) it can be easily integrated
𝑓 (z, a) = u + b, or x + ay + b = 𝑓 (z, a) which is the desired complete solution.
Thus, to solve 𝑓 (z, p, q) = O,
(i) assume u = x + ay and substitute p = dz/du, q = a dz/du in the given equation;
(ii) solve the resulting ordinary differential equation in z and u;
(iii) replace u by x + aye

Example Solve p(1+q)=qz

Solution. Let u = x + ay, so that p = dz/du and q = a dz/du.
Substituting these values of p and q in the given equation, we have

𝑑𝑧 𝑎 𝑑𝑧 𝑑𝑧 𝑎 𝑑𝑧 𝑎 𝑑𝑧
(1 + ) = 𝑎𝑧 𝑜𝑟 = 𝑎𝑧 − 1 𝑜𝑟 ∫ = ∫ 𝑑𝑢 + 𝑏
𝑑𝑢 𝑑𝑢 𝑑𝑢 𝑑𝑢 𝑎𝑧 − 1

𝑙𝑜𝑔(𝑎𝑧 − 1) = 𝑢 + 𝑏 𝑜𝑟 𝑙𝑜𝑔(𝑎𝑧 − 1) = 𝑥 + 𝑎𝑦 + 𝑏

which is the required complete solution

Example Solve 𝑞 2 = 𝑧 2 𝑝2 (1 − 𝑝2 )
Solution. Setting 𝑢 = 𝑦 + 𝑎𝑥 𝑎𝑛𝑑 𝑧 = 𝑓(𝑢) , 𝑤ⅇ 𝑔ⅇ𝑡
𝜕𝑧 𝑑𝑧 𝜕𝑢 𝑑𝑧 𝑑𝑧 𝜕𝑢 𝑑𝑧
𝑝= = ⋅ =𝑎 𝑎𝑛𝑑 𝑞 = ⋅ =
𝜕𝑥 𝑑𝑢 𝜕𝑥 𝑑𝑢 𝑑𝑢 𝜕𝑦 𝑑𝑢
ⅆ𝑧 2 ⅆ𝑧 2 ⅆ𝑧 2
The given equation becomes (ⅆ𝑢) = 𝑎2 𝑧 2 (ⅆ𝑢) {1 − 𝑎2 (ⅆ𝑢) }

𝑑𝑧 2 𝑑𝑧 √(𝑎2 𝑧 2 ) − 1
𝑎 𝑧 ( ) = 𝑎2 𝑧 2 − 1
4 2
𝑜𝑟 =
𝑑𝑢 𝑑𝑢 𝑎2 𝑧
Integrating,

√(𝑎2 𝑧 2 ) − 1 1
∫ 𝑑𝑧 = ∫ 𝑑𝑢 + 𝑐 𝑜𝑟 (𝑎2 𝑧 2 − 1)2 = 𝑢 + 𝑐
𝑎2 𝑧
𝑎2 𝑧 2 = (𝑦 + 𝑎𝑥 + 𝑐)2 + 1
ⅆ𝑧
The second factor (i) is ⅆ𝑢=0. Its solution is z=c’

Example Solve 𝑧 2 (𝑝2 𝑥 2 + 𝑞 2 ) = 1

Solution. Given equation can be reduced to the above form by writing it as
 2
𝜕𝑧 𝑧 𝜕𝑧 2
𝑧 [(𝑥 ) + ( ) ] = 1
𝜕𝑥 𝜕𝑦

ⅆ𝑧 ⅆ𝑧
Putting X = log x, so that x ⅆ𝑥 = ⅆ𝑋 takes the standard form

𝜕𝑧 𝑧
2
𝜕𝑧 2
𝑧 [( ) + ( ) ] = 1
𝜕𝑋 𝜕𝑦
𝜕𝑧 𝜕𝑧 𝜕𝑧 𝜕𝑧
Let u=X+ay and put 𝜕𝑋 = 𝜕𝑢 and 𝜕𝑦 = 𝑎 𝜕𝑢 in (ii), so that

ⅆ𝑧 𝑧 𝜕𝑧 2
𝑧 2 [(ⅆ𝑢) + 𝑎2 (𝜕𝑢) ] = 1 or √(1 + 𝑎2 )𝑧𝑑𝑧 = ±𝑑𝑢

Integrating, √(1 + 𝑎2 )𝑧 2 = ±2𝑢 + 𝑏 = ±2(𝑋 + 𝑎𝑦) + 𝑏

√(1 + 𝑎2 )𝑧 2 = ±2(𝑙𝑜𝑔𝑥 + 𝑎𝑦) + 𝑏

which is the complete solution required.

Form 111. f(x, p) = F(y, q), i.e., equations in which z is absent and the terms containing x and p
can be separated from those containing y and q.
As a trial solution assume that 𝑓 (x, p) = 𝑓 (y, q) = a, say
Then solving for p, we get p=ϕ(x)
And solving for q, we get q=ѱ(y)
𝜕𝑧 𝜕𝑧
Since 𝑑𝑧 = 𝜕𝑥 𝑑𝑥 + 𝜕𝑦 𝑑𝑦 = 𝑝𝑑𝑥 + 𝑞𝑑𝑦

𝑑𝑧 = ϕ(x)𝑑𝑥 + ѱ(y)𝑑𝑦
Integrating,

𝑧 = ∫ ϕ(x)𝑑𝑥 + ∫ ѱ(y)𝑑𝑦 + 𝑏
which is the desired complete solution containing two constants a and b.
Example Solve 𝑧 2 (𝑝2 + 𝑞 2 ) = 𝑥 + 𝑦
Solution. Given equation is 𝑝2 − 𝑥 = 𝑦 − 𝑞 2 = 𝑎, 𝑠𝑎𝑦

𝑝2 − 𝑥 = 𝑎 gives p=√(𝑎 + 𝑥)

𝑦 − 𝑞 2 = 𝑎 gives q=√(𝑦 − 𝑎)
Substituting these values of p and q in dz=pdx+qdy, we get

𝑑𝑧 = √(𝑎 + 𝑥)𝑑𝑥 + √(𝑦 − 𝑎)𝑑𝑦

2 3⁄ 2 3⁄
Integrating, 𝑧 = 3 (𝑎 + 𝑥) 2 + 3 (𝑦 − 𝑎) 2 +𝑏

Example Solve 𝑧 2 (𝑝2 + 𝑞 2 ) = 𝑥 2 + 𝑦 2

Solution. This equation can be reduced to the above form by writing it as

𝜕𝑧 2 𝜕𝑧 2
(𝑧 ) + (𝑧 ) = 𝑥 2 + 𝑦 2
𝜕𝑥 𝜕𝑦
1
And putting 𝑧𝑑𝑥 = 𝑑𝑍, 𝑖. ⅇ. , 𝑍 = 2 𝑧 2

𝜕𝑍 𝜕𝑍 𝜕𝑧 𝜕𝑧
= . =𝑧 =𝑃
𝜕𝑥 𝜕𝑧 𝜕𝑥 𝜕𝑥
𝜕𝑍 𝜕𝑍 𝜕𝑧 𝜕𝑧
= . =𝑧 =𝑄
𝜕𝑦 𝜕𝑧 𝜕𝑦 𝜕𝑦
(i)becomes
𝑃2 + 𝑄 2 = 𝑥 2 + 𝑦 2
𝑃2 − 𝑥 2 = 𝑦 2 − 𝑄 2

𝑃 = √(𝑥 2 + 𝑎) 𝑎𝑛𝑑 𝑄 = √(𝑦 2 − 𝑎)

𝑑𝑍 = 𝑃𝑑𝑥 + 𝑄𝑑𝑦 𝑔𝑖𝑣ⅇ𝑠

𝑑𝑍 = √(𝑥 2 + 𝑎)𝑑𝑥 + √(𝑦 2 − 𝑎)𝑑𝑦

Integrating, we have
1 1 1 1
𝑍 = 2 . 𝑥. √(𝑥 2 + 𝑎) + 2 . 𝑎. log [𝑥 + √(𝑥 2 + 𝑎)] + 2 . 𝑦. √(𝑦 2 − 𝑎) − 2 . 𝑎. log[𝑦 +
√(𝑦 2 − 𝑎)] + 𝑏

𝑥 + √(𝑥 2 + 𝑎)
𝑧 2 = 𝑥√(𝑥 2 + 𝑎) + √(𝑦 2 − 𝑎) + 𝑎 𝑙𝑜𝑔 + 2𝑏
𝛾 + √(𝑦 2 − 𝑎)
This is the required solution.
Example Solve (𝑥 + 𝑦)(𝑝 + 𝑞)2 + (𝑥 − 𝑦)(𝑝 − 𝑞)2 = 1
Solution This equation can be reduced to the form 𝑓 (x,q)=F(y,q) by putting u=x+y, v=x-y and
taking z=(z(u,v)
𝜕𝑧 𝜕𝑧 𝜕𝑢 𝜕𝑧 𝜕𝑣
Then, 𝑝 = 𝜕𝑥 = 𝜕𝑢 . 𝜕𝑥 + 𝜕𝑣 . 𝜕𝑥 = 𝑃 + 𝑄
𝜕𝑧 𝜕𝑧 𝜕𝑢 𝜕𝑧 𝜕𝑣 𝜕𝑧 𝜕𝑧
And 𝑄 = 𝜕𝑦 = 𝜕𝑢 . 𝜕𝑦 + 𝜕𝑣 . 𝜕𝑦 = 𝑃 − 𝑄, 𝑤ℎⅇ𝑟ⅇ 𝑃 = 𝜕𝑢 , 𝑄 = 𝜕𝑣

Substituting these, the given equation reduces to

𝑢(2𝑃)2 + 𝑣(2𝑄)2 = 1 𝑜𝑟 4𝑃2 𝑢 = 1 − 𝑄 2 𝑣 = 𝑎(𝑠𝑎𝑦)

1 𝑎 1 1−𝑎
𝑃 = ± .√ ,𝑄 = ± .√
2 𝑢 2 𝑣

𝜕𝑧 𝜕𝑧
𝑑𝑧 = 𝑑𝑢 + = 𝑝𝑑𝑢 + 𝑄𝑑𝑣
𝜕𝑢 𝜕𝑣
1 √𝑎 𝑑𝑢 √1 − 𝑎 𝑑𝑣
=± . . ± .
2 2 √𝑢 2 √𝑣
Integrating , we have 𝑧 = ±√𝑎√𝑢 ± √1 − 𝑎√𝑣 + 𝑏

Or 𝑧 = ±√𝑎(𝑥 + 𝑦) ± √(1 − 𝑎)(𝑥 − 𝑦) + 𝑏

Which is the required solution.
Form IV. z = PX + qy + f(p, q): an equation analogous to the Clairaut's equation (S 11.14).
Its complete solution is z = ax + by + f (a, b) which is obtained by writing a for p and b for q in
the given equation.

Example Solve 𝑧 = 𝑝𝑥 + 𝑞𝑦 + √(1 + 𝑝2 + 𝑞 2 )

Solution . Given equation is of the form z=px+qy+ 𝑓 (p,q) where 𝑓 (p,q)= √(1 + 𝑝2 + 𝑞 2 )

Its complete solution is z=ax+by+√(1 + 𝑎2 + 𝑏 2 )

CHARPIT'S METHOD*
We now explain a general method for finding the complete integral of a non-linear partial
differential equation which is due to Charpit.
Consider the equation
𝑓(𝑥, 𝑦, 𝑧, 𝑝, 𝑞) = 0
 ...(1)
Since z depends on x and y, we have
𝜕𝑧 𝜕𝑧
𝑑𝑧 = 𝑑𝑥 + 𝑑𝑦 = 𝑝𝑑𝑥 + 𝑞𝑑𝑦
𝜕𝑢 𝜕𝑦
...(2)
Now if we can find another relation involving x, y, z, p, q such as +ϕ(x, y, z, p, q) = O
...(3) then we can solve (I) and (3) for p and q and substitute in (2). This will give the solution
provided (2) is integrable.
To determine ϕ, we differentiate (1) and (3) w.r.t x and y giving ,
𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑝 𝜕𝑓 𝜕𝑞
+ 𝑝+ + =0
𝜕𝑥 𝜕𝑧 𝜕𝑝 𝜕𝑥 𝜕𝑞 𝜕𝑥
𝜕ϕ 𝜕ϕ 𝜕ϕ 𝜕𝑝 𝜕ϕ 𝜕𝑞
+ 𝑝+ + =0
𝜕𝑥 𝜕𝑧 𝜕𝑝 𝜕𝑥 𝜕𝑞 𝜕𝑥
𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑝 𝜕𝑓 𝜕𝑞
+ 𝑝+ + =0
𝜕𝑦 𝜕𝑧 𝜕𝑝 𝜕𝑦 𝜕𝑞 𝜕𝑦
𝜕ϕ 𝜕ϕ 𝜕ϕ 𝜕𝑝 𝜕ϕ 𝜕𝑞
+ 𝑝+ + =0
𝜕𝑦 𝜕𝑧 𝜕𝑝 𝜕𝑦 𝜕𝑞 𝜕𝑦
𝜕𝑝
Eliminating 𝜕𝑥 between the equations (4) and (5), we get

𝜕𝑓 𝜕ϕ 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕ϕ 𝜕𝑓 𝜕𝑞
( − )+( − )𝑝 + ( − ) =0
𝜕𝑥 𝜕𝑝 𝜕𝑥 𝜕𝑝 𝜕𝑧 𝜕𝑝 𝜕𝑧 𝜕𝑝 𝜕𝑞 𝜕𝑝 𝜕𝑞 𝜕𝑝 𝜕𝑥
𝜕𝑞
Also eliminating 𝜕𝑦 between the equations (6) and (7) , we obtain

𝜕𝑓 𝜕ϕ 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕ϕ 𝜕𝑓 𝜕𝑞
( − )+( − )𝑞 + ( − ) =0
𝜕𝑦 𝜕𝑝 𝜕𝑦 𝜕𝑞 𝜕𝑧 𝜕𝑞 𝜕𝑧 𝜕𝑞 𝜕𝑝 𝜕𝑞 𝜕𝑝 𝜕𝑞 𝜕𝑦
𝜕𝑞 𝜕2𝑧 𝜕𝑞
Adding (8) and (9) by using = =
𝜕𝑥 𝜕𝑥𝜕𝑦 𝜕𝑦

We mean that the last terms in both cancel and the other terms , on rearrangement , give
𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕𝑓 𝜕ϕ 𝜕𝑓 𝜕ϕ
( +𝑝 ) +( +𝑞 ) + (−𝑝 −𝑞 ) + (− ) + (− ) =0
𝜕𝑥 𝜕𝑧 𝜕𝑝 𝜕𝑦 𝜕𝑧 𝜕𝑞 𝜕𝑝 𝜕𝑞 𝜕𝑧 𝜕𝑝 𝜕𝑥 𝜕𝑞 𝜕𝑦
 𝜕𝑓 𝜕ϕ 𝜕𝑓 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ 𝜕𝑓 𝜕𝑓 𝜕ϕ
(− ) + (− ) + (−𝑝 −𝑞 ) +( +𝑝 ) +( +𝑞 ) =0
𝜕𝑝 𝜕𝑥 𝜕𝑞 𝜕𝑦 𝜕𝑝 𝜕𝑞 𝜕𝑧 𝜕𝑥 𝜕𝑧 𝜕𝑝 𝜕𝑦 𝜕𝑧 𝜕𝑞

This is Lagrange's linear equation (S 17.5) with x, y, z, p, q as independent variables and

as the dependent variable. Its solution will depend on the solution of the subsidiary equations.

𝑑𝑥 𝑑𝑦 𝑑𝑧 𝑑𝑝 𝑑𝑞 𝜕ϕ
= = = = =
𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑓 0
− − −𝑝 −𝑞 +𝑝 +𝑞
𝜕𝑝 𝜕𝑞 𝜕𝑝 𝜕𝑞 𝜕𝑥 𝜕𝑧 𝜕𝑦 𝜕𝑧

An integral of these equations involving p or q or both, can be taken as the required

relation (3), which along with (1) will give the values of p and q to make (2) integrable. Of
course, we should take the simplest of the integrals so that it may be easier to solve for p and q.
Example Solve (𝑝2 + 𝑞 2 )𝑦 = 𝑞𝑧
Charpit’s subsidiary equations are
𝑑𝑥 𝑑𝑦 𝑑𝑧 𝑑𝑝 𝑑𝑞
= = = = 2
−2𝑝𝑦 𝑧 − 2𝑞𝑦 −𝑞𝑧 −𝑝𝑞 𝑝

The last of these equations give 𝑝𝑑𝑝 + 𝑞𝑑𝑞 = 0

Integrating, 𝑝2 + 𝑞 2 = 𝑐 2
𝑐 2𝑦
Now to solve (i) and (ii) put 𝑝2 + 𝑞 2 = 𝑐 2 in (i) so that 𝑞 = 𝑧

Substituting this value of q in (ii), we get 𝑝 = 𝑐√(𝑧 2 − 𝑐 2 𝑦 2 )⁄𝑧

𝑐 𝑐2𝑦
Hence 𝑑𝑧 = 𝑝𝑑𝑥 + 𝑞𝑑𝑦 = 𝑧 . √(𝑧 2 − 𝑐 2 𝑦 2 )𝑑𝑥 + . 𝑑𝑦
𝑧

1
. 𝑑(𝑧 2 − 𝑐 2 𝑦 2 )
𝑧𝑑𝑧 = 𝑐 𝑦𝑑𝑦 = 𝑐√(𝑧 2 − 𝑐 2 𝑦 2 )𝑑𝑥 𝑜𝑟 2
2
= 𝑐𝑑𝑥
√(𝑧 2 − 𝑐 2 𝑦 2 )

Integrating , we get √(𝑧 2 − 𝑐 2 𝑦 2 ) = 𝑐𝑥 + 𝑎 𝑜𝑟 𝑧 2 = (𝑎 + 𝑐𝑥)2 + 𝑐 2 𝑦 2 which is

required complete solution.

Example Solve 2𝑥𝑧 − 𝑝𝑥 2 − 2𝑞𝑥𝑦 + 𝑝𝑞 = 0

Solution. 𝑓(𝑥, 𝑦, 𝑧, 𝑝, 𝑞) = 2𝑥𝑧 − 𝑝𝑥 2 − 2𝑞𝑥𝑦 + 𝑝𝑞 = 0

Charpit’s subsidiary equations are

 𝑑𝑥 𝑑𝑦 𝑑𝑧 𝑑𝑝 𝑑𝑞
= = 2 = =
𝑥2 − 𝑞 2𝑥𝑦 − 𝑝 𝑝𝑥 − 2𝑝𝑞 + 2𝑞𝑥𝑦 2𝑧 − 2𝑞𝑦 0

𝑑𝑞 = 0 𝑜𝑟 𝑞 = 𝑎
2𝑥(𝑧−𝑎𝑦)
Putting q=a in (i) we get p= 𝑥 2 −𝑎

2𝑥(𝑧 − 𝑎𝑦) 𝑑𝑧 − 𝑎𝑑𝑦 2𝑥

𝑑𝑧 = 𝑝𝑑𝑥 + 𝑞𝑑𝑦 = 𝑑𝑥 + 𝑎𝑑𝑦 𝑜𝑟 = 2 𝑑𝑥
𝑥2 − 𝑎 𝑧 − 𝑎𝑦 𝑥 −𝑎
Integrating, 𝑙𝑜𝑔(𝑧 − 𝑎𝑦) = log(𝑥 2 − 𝑎) + 𝑙𝑜𝑔𝑏
𝑧 − 𝑎𝑦 = 𝑏(𝑥 2 − 𝑎) 𝑜𝑟 𝑧 = 𝑎𝑦 + 𝑏(𝑥 2 − 𝑎)
which is required complete solution.

Example 2𝑧 + 𝑝2 + 𝑞𝑦 + 2𝑦 2 = 0
Solution. 𝑓(𝑥, 𝑦, 𝑧, 𝑝, 𝑞) = 2𝑧 + 𝑝2 + 𝑞𝑦 + 2𝑦 2
Charpit’s subsidiary equations are
𝑑𝑥 𝑑𝑦 𝑑𝑧 𝑑𝑝 𝑑𝑞
= = = =
−2𝑝 −𝑦 (−2𝑝2 + 𝑞𝑦) 2𝑝 4𝑦 + 3𝑞

From first and fourth ratios,

𝑑𝑝 = −𝑑𝑥 𝑜𝑟 𝑝 = −𝑥 + 𝑎
Substituting p=a-x in the given equation, we get
1
𝑞= . [−2𝑧 − 2𝑦 2 − (𝑎 − 𝑥)2 ]
𝑦
1
𝑑𝑧 = 𝑝𝑑𝑥 + 𝑞𝑑𝑦 = (𝑎 − 𝑥)𝑑𝑥 − . [2𝑧 + 2𝑦 2 + (𝑎 − 𝑥)2 ]𝑑𝑦
𝑦
Multiplying both sides by 2𝑦 2 ,
2𝑦 2 𝑑𝑧 + 4𝑥𝑦𝑧𝑑𝑦 = 2𝑦 2 (𝑎 − 𝑥)𝑑𝑥 − 4𝑦 3 𝑑𝑦 − 2𝑦(𝑎 − 𝑥)2 𝑑𝑦
Integrating,
2𝑧𝑦 2 = −[𝑦 2 (𝑎 − 𝑥)2 + 𝑦 4 ] + 𝑏
𝑦 2 [(𝑥 − 𝑎)2 + 2𝑧 + 𝑦 2 ] = 𝑏

Which is the required solution.

Module : 2
HOMOGENEOUS LINEAR EQUATIONS WITH CONSTANT COEFFICIENTS

An equation of the form

𝜕𝑛 𝑧 𝜕𝑛 𝑧 𝜕𝑛 𝑧
+ 𝑘1 𝜕𝑥 𝑛−1 𝜕𝑦 + ⋯ + 𝑘𝑛 𝜕𝑦 𝑛 = 𝐹(𝑥, 𝑦) (1)
𝜕𝑥 𝑛
in which 𝑘’s are constants, is called a homogenous linear partial differential equation of the nth
order with constant coefficients. It is called homogenous because all terms contain derivatives of
the same order.
On writing,
𝜕𝑟 𝜕𝑟
= 𝐷𝑟 and = 𝐷′𝑟
𝜕𝑥 𝑟 𝜕𝑦 𝑟
becomes
(𝐷𝑛 + 𝑘1 𝐷𝑛−1 𝐷′𝑟 + 𝐷′ + ⋯ + 𝑘𝑛 𝐷′𝑛 )𝑧 = 𝐹(𝑥, 𝑦)
or briefly
𝑓(𝐷, 𝐷′ )𝑧 = 𝐹(𝑥, 𝑦)
As in the case of ordinary linear equations with constant coefficients the complete solution
of (1) consists of two parts, namely : the complementary function and the particular integral.
The complementary function is the complete solution of the equation 𝑓(𝐷, 𝐷’)𝑧 = 0,
which must contain n arbitrary functions. The particular integral is the particular solution of
equation (2).

RULES FOR FINDING THE COMPLEMENTARY FUNCTION

𝜕2𝑧 𝜕2𝑧 𝜕2𝑧

Consider the equation + 𝑘1 𝜕𝑥𝜕𝑦 + 𝑘2 𝜕𝑦 2 = 0 (1)
𝜕𝑥 2

which in symbolic form is

(𝐷2 + 𝑘1 𝐷𝐷′ + 𝑘2 𝐷2 )𝑧 = 0

Its symbolic operator equated to zero, 𝑖. ⅇ., 𝐷2 + 𝑘1 𝐷𝐷′ + 𝑘2 𝐷2 = 0 is called the auxiliary
equation (A.E)
Let its root be 𝐷/𝐷’ = 𝑚1 , 𝑚2 .

Case I. If the roots be real and distinct then (2) is equivalent to

(𝐷 − 𝑚1 𝐷′ )(𝐷 − 𝑚2 𝐷′ )𝑧 = 0

It will be satisfied by the solution of

(𝐷 − 𝑚2 𝐷′ )𝑧 = 0, 𝑖. ⅇ. , 𝑝 − 𝑚2 𝑞 = 0

This is a Lagrange's linear and the subsidiary equations are

 𝜕𝑥 ⅆ𝑦 𝜕𝑧
= −𝑚 = , whence 𝑦 + 𝑚2 𝑥 = 𝑎 and 𝑧 = 𝑏.
1 2 0

∴ its solution is 𝑧 = 𝜙(𝑦 + 𝑚2 𝑥).

Similarly (3) will also be satisfied by the solution of

(𝐷 − 𝑚1 𝐷′ )𝑧 = 0, 𝑖. ⅇ., by 𝑧 = 𝑓(𝑦 + 𝑚1 𝑥)

Hence the complete solution of (1) is 𝑧 = 𝑓(𝑦 = 𝑚1 𝑥) + 𝜙(𝑦 + 𝑚2 𝑥)

Case II. If the roots be equal (𝑖. ⅇ. , 𝑚1 = 𝑚2 ), then (2) is equivalent to

(𝐷 − 𝑚1 𝐷′)2 𝑧 = 0

Putting (𝐷 − 𝑚1 𝐷′)𝑧 , it becomes (𝐷 − 𝑚1 𝐷′ )𝑢 = 0 which gives

𝑢 = 𝜙(𝑦 + 𝑚1 𝑥)

∴ takes the form (𝐷 − 𝑚1′ 𝐷)𝑧 = 𝜙(𝑦 + 𝑚1 𝑥) or 𝑝 − 𝑚1 𝑞 = 𝜙(𝑦 + 𝑚1 𝑥)

This is again Lagrange's linear and the subsidiary equations are

𝜕𝑥 𝜕𝑦 𝜕𝑧
= −𝑚 = 𝜙(𝑦+𝑚
1 1 1 𝑥)
giving

𝑦 + 𝑚1 𝑥 = 𝑎 and 𝜕𝑧 = 𝜙(𝑎)𝜕𝑥, 𝑖. ⅇ. , 𝑧 = 𝜙(𝑎)𝑥 + 𝑏

Thus the complete solution of (1) is

𝑧 − 𝑥𝜙(𝑦 + 𝑚1 𝑥) = 𝑓(𝑦 + 𝑚1 𝑥). 𝑖. ⅇ. , 𝑧 = 𝑓(𝑦 + 𝑚1 𝑥) + 𝑥𝜙(𝑦 + 𝑚1 𝑥).

𝜕2𝑧 𝜕2𝑧 𝜕2𝑧

Example . Solve 2 𝜕𝑥 2 + 5 𝜕𝑥𝜕𝑦 + 2 𝜕𝑦 2 = 0.

Solution. Given equation in symbolic form is (2𝐷2 + 5𝐷𝐷′ + 2𝐷′2 )𝑧 = 0.

Its auxiliary equation is 2𝑚2 + 5𝑚 + 2 = 0, where 𝑚 = 𝐷⁄𝐷′ .

which gives
𝑚 = −2, −1

1
Here the complete solution is 𝑧 = 𝑓1 (𝑦 − 2𝑥) + 𝑓2 (𝑦 − 2 𝑥)
which may be written as 𝑧 = 𝑓1 (𝑦 − 2𝑥) + 𝑓2 (2𝑦 − 𝑥).
 Example Solve 4𝑟 + 12𝑠 + 9𝑡 = 0.

Solution. Given equation in symbolic form is (4𝐷2 + 12𝐷𝐷′ + 9𝐷′2 )𝑧 = 0

for
𝜕2𝑧 𝜕2𝑧 𝜕2𝑧
𝑟 = 𝜕𝑥 2 = 𝐷2 𝑧, 𝑠 = 𝜕𝑥𝜕𝑦 = 𝐷𝐷′𝑧 and 𝑡 = 𝜕𝑦 2 = 𝐷′2 𝑧.

∴ Its auxiliary equation is 4𝑚2 + 12𝑚 + 9 = 0 , whence 𝑚 = − 3⁄2, − 3⁄2

Hence the complete solution is 𝑧 = 𝑓1 (𝑦 − 1.5𝑥) + 𝑥𝑓2 (𝑦 − 1.5𝑥).

RULES FOR FINDING THE PARTICULAR INTEGRAL

Consider the equation (𝐷2 + 𝑘1 𝐷𝐷′ + 𝑘2 𝐷′2 )𝑧 = 𝐹(𝑥, 𝑦) 𝑖. ⅇ. , 𝑓(𝐷, 𝐷′ )𝑧 = 𝐹(𝑥, 𝑦).

1
∴ P.I. = 𝑓(𝐷,𝐷′ ) 𝐹(𝑥, 𝑦)

Case I. When 𝐹(𝑥, 𝑦) = ⅇ 𝑎𝑥+𝑏𝑦

𝑎𝑥=𝑏𝑦
Since 𝐷ⅇ 𝑎𝑥+𝑏𝑦 = 𝑎ⅇ 𝑎𝑥+𝑏𝑦 ; 𝐷′𝑒 = 𝑏ⅇ 𝑎𝑥+𝑏𝑦

𝑎𝑥=𝑏𝑦
∴ 𝐷2 ⅇ 𝑎𝑥+𝑏𝑦 = 𝑎2 ⅇ 𝑎𝑥+𝑏𝑦 ; 𝐷𝐷′𝑒 = 𝑎𝑏ⅇ 𝑎𝑥+𝑏𝑦

and 𝐷′2 ⅇ 𝑎𝑥+𝑏𝑦 = 𝑏 2 ⅇ 𝑎𝑥+𝑏𝑦

∴ (𝐷2 + 𝑘1 𝐷𝐷′ + 𝑘2 𝐷′2 )ⅇ 𝑎𝑥+𝑏 = (𝑎2 + 𝑘1 𝑎𝑏 + 𝑘2 𝑏 2 )ⅇ 𝑎𝑥+𝑏𝑦

𝑖. ⅇ., 𝑓(𝐷, 𝐷′ )ⅇ 𝑎𝑥+𝑏𝑦 = 𝑓(𝑎, 𝑏)ⅇ 𝑎𝑥+𝑏𝑦

Operating both sides by 1⁄𝑓 (𝐷, 𝐷′) , we get

1 1
P.I. = 𝑓(𝐷,𝐷′) ⅇ 𝑎𝑥+𝑏𝑦 = 𝑓(𝑎,𝑏) ⅇ 𝑎𝑥+𝑏𝑦

Case II. When 𝐹(𝑥, 𝑦) = sin(𝑚𝑥 + 𝑛𝑦) 𝑜𝑟 cos (𝑚𝑥 + 𝑛𝑦)

Since 𝐷2 sin(𝑚𝑥 + 𝑛𝑦) = −𝑚2 sin (𝑚𝑥 + 𝑛𝑦)

𝐷𝐷′ sin(𝑚𝑥 + 𝑛𝑦) = −𝑚𝑛 sin (𝑚𝑥 + 𝑛𝑦)

and 𝐷′2 sin(𝑚𝑥 + 𝑛𝑦) = −𝑛2 sin (𝑚𝑥 + 𝑛𝑦)

∴ 𝑓(𝐷2 , 𝐷𝐷′ , 𝐷′2 ) sin(𝑚𝑥 + 𝑛𝑦) = 𝑓(−𝑚2 , −𝑚𝑛, −𝑛2 )sin (𝑚𝑥 + 𝑛𝑦)

Operating both sides by 1⁄𝑓(𝐷2 , 𝐷𝐷′ , 𝐷′2 ) , we get

1 1
P.I. = sin(𝑚𝑥 + 𝑛𝑦) = 𝑓(−𝑚2 ,−𝑚𝑛,−𝑛2) sin (𝑚𝑥 + 𝑛𝑦)
𝑓(𝐷 2 ,𝐷𝐷 ′ ,𝐷′2 )

Similarly about the P.I. for 𝑐𝑜𝑠 (𝑚𝑥 + 𝑛𝑦)

Case III. When 𝐹(𝑥, 𝑦) = 𝑥 𝑚 𝑦 𝑛 , 𝑚 and 𝑛 being constants

1
∴ P.I. = 𝑥 𝑚 𝑦 𝑛 = [𝑓(𝐷, 𝐷′)]−1 𝑥 𝑚 𝑦 𝑛 .
𝑓(𝐷,𝐷′)

To evaluate it, we expand [𝑓(𝐷, 𝐷′)]−1 in ascending powers of D or D’ by Binomial

theorem and then operate on 𝑥 𝑚 𝑦 𝑛 term by term.

Case IV. When 𝐹(𝑥, 𝑦) is any function of x and y.

1
∴ P.I. = 𝐹(𝑥, 𝑦)
𝑓(𝐷,𝐷 ′ )

To evaluate it, we resolve 1/𝑓(𝐷, 𝐷′) into partial fractions treating 𝑓(𝐷, 𝐷′ ) as a function
of D alone and operate each partial fraction on 𝐹(𝑥, 𝑦) remembering that

1
𝐹(𝑥, 𝑦) = ∫ 𝐹(𝑥, 𝑐 − 𝑚𝑥)𝑑𝑥
𝐷−𝑚𝐷 ′

Where c is replaced by 𝑦 + 𝑚𝑥 after integration.

WORKING PROCEDURE TO SOLVE THE EQUATION

𝜕𝑛 𝑧 𝜕𝑛 𝑧 𝜕𝑛 𝑧
+ 𝑘1 𝜕𝑥 𝑛−1 𝜕𝑦 + ⋯ + 𝑘𝑛 𝜕𝑦 𝑛 = 𝐹(𝑥, 𝑦).
𝜕𝑥 𝑛

Its symbolic form is

(𝐷𝑛 + 𝑘1 𝐷𝑛−1 𝐷′ + ⋯ + 𝑘𝑛 𝐷′𝑛 )𝑧 = 𝐹(𝑥, 𝑦)

or briefly

𝑓(𝐷, 𝐷 ′ )𝑧 = 𝐹(𝑥, 𝑦)
 Step I. To find the C.F.

(i) Write the A.E.

𝑖. ⅇ., 𝑚𝑛 + 𝑘1 𝑚𝑛−1 + ⋯ + 𝑘𝑛 = 0 and solve it for m.
(ii) Write the C.F. as follows
Roots of A.E. C.F.
1. 𝑚1 , 𝑚2 , 𝑚3 … (distinct roots) 𝑓1 + 𝑚1 𝑥) + 𝑓2 + 𝑚2 𝑥) + 𝑓3 (𝑦 + 𝑚3 𝑥)
(𝑦 (𝑦
2. 𝑚1 , 𝑚1 , 𝑚3 … (two equal roots) +⋯
3. 𝑚1 , 𝑚1 , 𝑚1 … (three equal roots) 𝑓1 + 𝑚1 𝑥) + 𝑥𝑓2 (𝑦 + 𝑚1 𝑥) + 𝑓3 (𝑦 + 𝑚3 𝑥)
(𝑦
+⋯
𝑓1 (𝑦 + 𝑚1 𝑥) + 𝑥𝑓2 (𝑦 + 𝑚1 𝑥) + 𝑥 2 𝑓3 (𝑦 + 𝑚1 𝑥)
+⋯

Step II. To find the P.I.

1
From the symbolic form, P.I. = 𝑓(𝐷,𝐷′ ) 𝐹(𝑥, 𝑦).

1
(i) When 𝐹(𝑥, 𝑦) = ⅇ 𝑎𝑥+𝑏𝑦 P.I. = 𝑓(𝐷,𝐷′) ⅇ 𝑎𝑥+𝑏𝑦 [𝑃𝑢𝑡 𝐷 = 𝑎 𝑎𝑛𝑑 𝐷’ = 𝑏]

(ii) When 𝐹(𝑥, 𝑦) = sin(𝑚𝑥 + 𝑛𝑦) 𝑜𝑟 cos (𝑚𝑥 + 𝑛𝑦)

1
P.I. = 𝑓(𝐷2,𝐷𝐷′ ,𝐷′2) sin 𝑜𝑟 cos (𝑚𝑥 + 𝑛𝑦)
[𝑃𝑢𝑡 𝐷2 = −𝑚2 , 𝐷𝐷′ = −𝑚𝑛, 𝐷′2 = −𝑛2]

1
(iii) When 𝐹(𝑥, 𝑦) = 𝑥 𝑚 𝑦 𝑛 , P.I. = 𝑓(𝐷,𝐷′) 𝑥 𝑚 𝑦 𝑛 = [𝑓(𝐷, 𝐷′)]−1 𝑥 𝑚 𝑦 𝑛 .

Expand [𝑓(𝐷, 𝐷′)]−1 in ascending powers of D or D’ and operate on 𝑥 𝑚 𝑦 𝑛 term by term.

1
(iv) When 𝐹(𝑥, 𝑦) is any function of x and y P.I. = 𝐹(𝑥, 𝑦).
𝑓(𝐷,𝐷 ′ )

Resolve 1⁄𝑓(𝐷, 𝐷 ′ ) into partial fractions considering 𝑓(𝐷, 𝐷′ ) as a function of D alone

and operate each partial fraction on 𝐹(𝑥, 𝑦) remembering that

1
𝐹(𝑥, 𝑦) = ∫ 𝐹(𝑥, 𝑐 − 𝑚𝑐)𝑑𝑥
𝐷−𝑚𝐷 ′

where c is replaced by 𝑦 + 𝑚𝑥 after integration.

Example . Solve (𝐷2 + 4𝐷𝐷′ − 5𝐷′2 )𝑧 = sin(2𝑥 + 3𝑦).

 Solution. A.E. of the given equation is 𝑚2 + 4𝑚 − 5 = 0 𝑖. ⅇ., 𝑚 = −1, −5

∴ C.F. = 𝑓1 (𝑦 + 𝑥) + 𝑓2 (𝑦 − 5𝑥)
1
P.I. = 𝐷2+4𝐷𝐷′ −5𝐷′2 sin (2𝑥 + 3𝑦)
[𝑃𝑢𝑡 𝐷2 = −22 , 𝐷𝐷′ = −2 × 3, 𝐷′2 = −32 ]

1 1
= −4+4(−6)−5(−9) sin (2𝑥 + 3𝑦) = 17 sin (2𝑥 + 3𝑦)

1
Hence the C.S. is 𝑧 = 𝑓1 (𝑦 + 𝑥) + 𝑓2 (𝑦 − 5𝑥) + 17 sin(2𝑥 + 3𝑦).

𝜕2𝑧 𝜕2𝑧
Example Solve 𝜕𝑥 2 − 𝜕𝑥𝜕𝑦 = 𝑐𝑜𝑠 𝑥 𝑐𝑜𝑠 2𝑦.

Solution. Given equation in symbolic form is (𝐷2 − 𝐷𝐷′ )𝑧 = 𝑐𝑜𝑥 𝑐𝑜𝑠 2𝑦.

It’s A.E. is 𝑚2 − 𝑚 = 0, whence 𝑚 = 0, 1.

∴ C.F. = 𝑓1 (𝑦) + 𝑓2 (𝑦 + 𝑥)

1 1 1
P.I. = 𝐷2−𝐷𝐷′ cos 𝑥 cos 2𝑦 = 2 𝐷2−𝐷𝐷′ [cos(𝑥 + 2𝑦) + cos (𝑥 − 2𝑦)]
1 1
= 2 [𝐷2−𝐷𝐷′ cos(𝑥 + 2𝑦)

[Put 𝐷2 = −1, 𝐷𝐷 ′ = −2]

1
+ 𝐷2−𝐷𝐷′ cos(𝑥 − 2𝑦)] [Put 𝐷2 = −1, 𝐷𝐷′ = 2]

1 1 1 1 1
= 2 [−1+2 cos(𝑥 + 2𝑦) + −1−2 cos(𝑥 − 2𝑦)] = 2 cos(𝑥 + 2𝑦) − 6 cos (𝑥 − 2𝑦)

1 1
Hence the C.S. is 𝑧 = 𝑓1 (𝑦) + 𝑓2 (𝑦 + 𝑥) + cos(𝑥 + 2𝑦) − cos(𝑥 − 2𝑦).
2 6

𝜕3𝑧 𝜕3 𝑧
Example. Solve 𝜕𝑥 3 − 2 𝜕𝑥 2𝜕𝑦 = 2ⅇ 2𝑥 + 3𝑥 2 𝑦.

Solution. Given equation in symbolic form is

(𝐷2 − 2𝐷2 𝐷′ )𝑧 = 2ⅇ 2𝑥 + 3𝑥 2 𝑦

It’s A.E. is 𝑚3 − 2𝑚2 = 0 , whence 𝑚 = 0, 0, 2.

 ∴ C.F. = 𝑓1 (𝑦) + 𝑥𝑓2 (𝑦) + 𝑓3 (𝑦 + 2𝑥)

1 1 1
P.I. = 𝐷3−2𝐷2𝐷′(2ⅇ 2𝑥 + 3𝑥 2 𝑦) = 2𝐷3−2𝐷2𝐷′ ⅇ 2𝑥 + 3 𝐷3 (1−2𝐷′ /𝐷) 𝑥 2 𝑦
1 3 1 3 2𝐷′ 4𝐷 ′2
= 2 23 −2.22(0) ⅇ 2𝑥 + 𝐷3 (1 − 2𝐷′ /𝐷)−1 𝑥 2 𝑦 = 4 ⅇ 2𝑥 + 𝐷3 (1 + + + ⋯ ) 𝑥2𝑦
𝐷 𝐷2

1 3 2 1 3 2
= 4 ⅇ 2𝑥 + 𝐷3 (𝑥 2 𝑦 + 𝐷 𝑥 2 . 1) = 4 ⅇ 2𝑥 + 𝐷2 (𝑥 2 𝑦 + 3 𝑥 3 )
1
[∵ 𝑓(𝑥) = ∫ 𝑓(𝑥)𝑑𝑥]
𝐷

1 𝑥5 𝑥6
= 4 ⅇ 2𝑥 + 3𝑦 3 . + 2. 4 .
4. 5 5. 6
1
[∵ 𝑓(𝑥) = ∫[∫(∫ 𝑓(𝑥)𝑑𝑥)𝑑𝑥]𝑑𝑥]
𝐷3
𝑒 2𝑥 𝑥5𝑦 𝑥6
= + + 60
4 20

𝑒 2𝑥 𝑥5𝑦 𝑥6
Hence the C.S. is 𝑧 = 𝑓1 (𝑦) + 𝑥𝑓2 (𝑦) + 𝑓3 (𝑦 + 2𝑥) + + + 60
4 20

Example Solve 𝑟 − 4𝑠 + 4𝑡 = ⅇ 2𝑥+𝑦

𝜕2𝑧 𝜕2𝑧 𝜕2𝑧

Solution. Given equation is 2 −4
+4 = ⅇ 2𝑥+𝑦 .
𝜕𝑥 𝜕𝑥𝜕𝑦 𝜕𝑦 2

𝑖. ⅇ., in symbolic form (𝐷 2 − 4𝐷𝐷′ + 4𝐷′2 )𝑧 = ⅇ 2𝑥+𝑦

It’s A.E. is (𝑚 − 2)2 = 0, whence 𝑚 = 2, 2.

∴ C.F. = 𝑓1 (𝑦 + 2𝑥) + 𝑥𝑓2 (𝑦 + 2𝑥)

1
P.I. = (𝐷−2𝐷′)2 ⅇ 2𝑥+𝑦

The usual rule fails because (𝐷 − 2𝐷′ )2 = 0 for 𝐷 = 2 and 𝐷′ = 1

∴ to obtain the P.I., we find from (𝐷 − 2𝐷′ )𝑢 = ⅇ 2𝑥+𝑦 , the solution

𝑢 = ∫ 𝐹(𝑥, 𝑐 − 𝑚𝑥)𝑑𝑥 = ∫ ⅇ 2𝑥+(𝑐−2𝑥) 𝑑𝑥 = 𝑥ⅇ 𝑐 = 𝑥ⅇ 2𝑥+𝑦

[∵ 𝑦 = 𝑐 − 𝑚𝑥 = 𝑐 − 2𝑥]

and from (𝐷 − 2𝐷′ )𝑧 = 𝑢 = 𝑥ⅇ 2𝑥+𝑦 , the solution

 1 1
𝑧 = ∫ 𝑥ⅇ 2𝑥+(𝑐−2𝑥) 𝑑𝑦 = 2 𝑥 2 ⅇ 𝑐 = 2 𝑥 2 ⅇ 2𝑥+𝑦
[∵ 𝑦 = 𝑐 − 𝑚𝑥 = 𝑐 − 2𝑥]

1
Hence the C.S. is 𝑧 = 𝑓1 (𝑦 + 2𝑥) + 𝑥𝑓2 (𝑦 + 2𝑥) + 2 𝑥 2 ⅇ 2𝑥+𝑦 .

𝜕2𝑧 𝜕2𝑧 𝜕2𝑧

Example Solve 𝜕𝑥 2 + 𝜕𝑥𝜕𝑦 − 6 𝜕𝑦 2 = cos(2𝑥 + 𝑦).

Solution. Given equation in symbolic form is (𝐷2 + 𝐷𝐷′ − 6𝐷′2 )𝑧 = 𝑐𝑜𝑠 (2𝑥 + 𝑦)

Its A.E. is 𝑚2 + 𝑚 − 6 = 0 whence 𝑚 = −3, 2.

∴ C.F. = 𝑓1 (𝑦 − 3𝑥) + 𝑓2 (𝑦 + 2𝑥).

Since 𝐷2 + 𝐷𝐷′ − 6𝐷′2 = −22 − (2)(1) − 6(−1)2 = 0

∴ It is a case of failure and we have to apply the general method.

1 1
P.I. = 𝐷2−𝐷𝐷′ −6𝐷′2 cos(2𝑥 + 𝑦) = (𝐷+3𝐷′ )(𝐷−2𝐷′ ) cos (2𝑥 + 𝑦)

1 1
= (𝐷+3𝐷′ ) [∫ cos(2𝑥 + ̅̅̅̅̅̅̅̅̅
𝑐 − 2𝑥 ) 𝑑𝑥]𝑐→𝑦−3𝑥 = (𝐷+3𝐷′) [∫ cos 𝑐 𝑑𝑥]𝑐→𝑦−3𝑥

[∵ 𝑦 = 𝑐 − 𝑚𝑥 = 𝑐 − 2𝑥]

1
= (𝐷+3𝐷′ ) 𝑥𝑐𝑜𝑠(𝑦 + 2𝑥) = [∫ 𝑥 cos(𝑐̅̅̅̅̅̅̅̅̅
+ 3𝑥 + 2𝑥) 𝑑𝑥]𝑐→𝑦−3𝑥 = [∫ 𝑥 cos(5𝑥 + 𝑐) 𝑑𝑥]𝑐→𝑦−3𝑥

x sin(5𝑥+𝑐) cos (5𝑥+𝑐)

=[ + ]
5 25 𝑐→𝑦−3𝑥
[Integrating by parts]

𝑥 1 𝑥 1
= 5 sin(5𝑥 + ̅̅̅̅̅̅̅̅̅
𝑦 − 3𝑥 ) + 25 cos(5𝑥 + ̅̅̅̅̅̅̅̅̅
𝑦 − 3𝑥) = 5 sin(2𝑥 + 𝑦) + 25 cos (2𝑥 + 𝑦)

Hence the C.S. is

𝑥 1
𝑧 = 𝑓1 (𝑦 − 3𝑥) + 𝑓2 (𝑦 + 2𝑥) + 5 sin(2𝑥 + 𝑦) + 25 cos (2𝑥 + 𝑦)

𝑥 1
𝑧 = 𝑓1 (𝑦 − 3𝑥) + 𝑓2 (𝑦 + 2𝑥) + 5 sin(2𝑥 + 𝑦) + 25 cos(2𝑥 + 𝑦).

𝜕2𝑧 𝜕2𝑧 𝜕2𝑧

Example Solve 𝜕𝑥 2 + 𝜕𝑥𝜕𝑦 − 6 𝜕𝑦 2 = 𝑦 cos 𝑥.
or 𝑟 + 𝑠 − 6𝑡 = 𝑦 cos 𝑥.

Solution. Its symbolic form is (𝐷2 + 𝐷𝐷′ − 6𝐷′2 )𝑧 = 𝑦 cos 𝑥

and the A.E. is 𝑚2 + 𝑚 − 6 = 0, whence 𝑚 = −3, 2.

∴ C.F. = 𝑓1 (𝑦 − 3𝑥) + 𝑓2 (𝑦 + 2𝑥).

1 1
P.I. = (𝐷−2𝐷′ )(𝐷+3𝐷′ ) 𝑦 cos 𝑥 = (𝐷−2𝐷′ ) [∫(𝑐 + 3𝑥) cos 𝑥 𝑑𝑥]𝑐→𝑦−3𝑥

[∵ 𝑦 = 𝑐 − 𝑚𝑥 = 𝑐 + 3𝑥]

1
= (𝐷−2𝐷′ ) [(𝑐 + 3𝑥) sin 𝑥 + 3 cos 𝑥]𝑐→𝑦−3𝑥
[Integrating by parts]

1
= (𝐷−2𝐷′ ) (𝑦 sin 𝑥 + 3 cos 𝑥) = [∫{(𝑐 + 3𝑥) sin 𝑥 + 3 cos 𝑥} 𝑑𝑥]𝑐→𝑦−2𝑥

= [(𝑐 − 2𝑥)(− cos 𝑥) − (−2)(− sin 𝑥) + 3 sin 𝑥)]𝑐→𝑦+2𝑥

= −𝑦 cos 𝑥 + sin 𝑥

Hence the C.S.is

𝑧 = 𝑓1 (𝑦 − 3𝑥) + 𝑓2 (𝑦 + 2𝑥) + sin 𝑥 − 𝑦 cos 𝑥.

𝜕2𝑧 𝜕2𝑧 𝜕2𝑧

Example Solve 4 𝜕𝑥 2 − 4 𝜕𝑥𝜕𝑦 + 𝜕𝑦 2 = 16 log(𝑥 + 2𝑦).

Solution. Its symbolic form is 4𝐷2 − 4𝐷𝐷′ + 𝐷′2 = 16 𝑙𝑜𝑔(𝑥 + 2𝑦)

and the A.E. is

4𝑚2 − 4𝑚 + 1 = 0, 𝑚 = 1/2, 1/2.
∴
1 1
C.F. = 𝑓1 (𝑦 + 2 𝑥) + 𝑓2 (𝑦 + 2 𝑥)

1 1 1
P.I. = (2𝐷−𝐷′ )2 16 𝑙𝑜𝑔(𝑥 + 2𝑦) = 4 1 { 1 𝑙𝑜𝑔(𝑥 + 2𝑦)}
(𝐷− 𝐷 ′ ) (𝐷− 𝐷 ′ )
2 2

1 𝑥
=4 1 [∫ 𝑙𝑜𝑔 {𝑥 + 2 (𝑐 − 2)} 𝑑𝑥]
𝐷− 𝐷 ′ 𝑐→𝑦+𝑥⁄2
2
 [∵ 𝑦 = 𝑐 − 𝑚𝑥 = 𝑐 + 3𝑥]

1 1
=4 1 [∫ log (2𝑐)𝑑𝑥]𝑐→𝑦+𝑥⁄2 = 4 1 [𝑥 log (𝑥 + 2𝑦)]
𝐷− 𝐷 ′ 𝐷− 𝐷 ′
2 2
𝑥
= 4 [∫ {𝑥𝑙𝑜𝑔 [𝑥 + 2 (𝑐 − 2)]} 𝑑𝑥] = 4[log 2𝑐 ∫ 𝑥 𝑑𝑥]𝑐→𝑦+𝑥⁄2 = 2𝑥 2 𝑙𝑜𝑔(𝑥 + 2𝑦)
𝑐→𝑦+𝑥⁄2

1 1
Hence the C.S. is 𝑧 = 𝑓1 (𝑦 + 2 𝑥) + 𝑓2 (𝑦 + 2 𝑥) + 2𝑥 2 𝑙𝑜𝑔(𝑥 + 2𝑦)

PROBLEMS 17.6

Solve the following equations :

𝜕3𝑧 𝜕3𝑧 𝜕3𝑧 𝜕3𝑧 𝜕3𝑧 𝜕3𝑧

1. 𝜕𝑥 3 − 4 𝜕𝑥 2 𝜕𝑦 + 4 𝜕𝑥𝜕𝑦 2 = 0. 2. 𝜕𝑥 3 − 3 𝜕𝑥 2 𝜕𝑦 + 4 𝜕𝑦 2 = ⅇ 𝑥+2𝑦 .
𝜕3 𝑧 𝜕3𝑧 𝜕3𝑧
3. (𝐷2 − 2𝐷𝐷′ + 𝐷′2 )𝑧 = ⅇ 𝑥+𝑦 . 4. 𝜕𝑥 3 − 4 𝜕𝑥 2 𝜕𝑦 + 5 𝜕𝑥𝜕𝑦 2 −
𝜕3 𝑧
2 𝜕𝑦 2 = ⅇ 2𝑥+𝑦 .
𝜕2𝑧 𝜕2𝑧 𝜕2𝑧 𝜕2𝑦 𝜕2𝑦
5. 𝜕𝑥 2 − 2 𝜕𝑥𝜕𝑦 + 𝜕𝑦 2 = sin 𝑥. 6. − 𝑎2 𝜕𝑥 2 = 𝐸 sin 𝑝𝑡.
𝜕𝑡 2
𝜕3𝑧 𝜕3𝑧 𝜕3𝑧
7. 𝜕𝑥 3 − 4 𝜕𝑥 2 𝜕𝑦 + 4 𝜕𝑥𝜕𝑦 2 = 2 sin (3𝑥 + 2𝑦).
8. (𝐷3 − 7𝐷𝐷′2 − 6𝐷′3 )𝑧 = cos(𝑥 + 2𝑦) + 4.
𝜕2𝑧 𝜕2𝑧 𝜕2𝑧
9. 𝜕𝑥 2 − 3 𝜕𝑥𝜕𝑦 + 2 𝜕𝑦 2 = ⅇ 2𝑒−𝑦 + ⅇ 𝑥+𝑦 + cos(𝑥 + 2𝑦).
𝜕2𝑧 𝜕2𝑧
10. 𝜕𝑥 2 − 𝜕𝑥𝜕𝑦 = sin 𝑥 cos 2𝑦. 11. (𝐷2 − 𝐷𝐷′ )𝑧 =
cos 2𝑦(sin 𝑥 + 𝑐𝑜𝑠 𝑥).
12. (𝐷2 − 𝐷′2 )𝑧 = ⅇ 𝑥−𝑦 sin (𝑥 + 2𝑦). 13. (𝐷2 + 3𝐷𝐷′ + 2𝐷′2 )𝑧 = 24𝑥𝑦.
𝜕2𝑧 𝜕2𝑧 𝜕2𝑧
14. 𝜕𝑥 2 + 2 𝜕𝑥𝜕𝑦 + 𝜕𝑦 2 = 𝑥 2 + 𝑥𝑦 + 𝑦 2 . 15. (𝐷2 − 𝐷𝐷′ − 2𝐷′2 )𝑧 =
(𝑦 − 1)ⅇ 𝑥 .
16. (𝐷3 + 𝐷2 𝐷′ − 𝐷𝐷′2 − 𝐷′3 )𝑧 = ⅇ 𝑥 𝑐𝑜𝑠 2𝑦. 17. (𝐷2 + 2𝐷𝐷′ + 𝐷′2 )𝑧 =
2 cos 𝑦 − 𝑥 sin 𝑦.

NON-HOMOGENEOUS LINEAR EQUATIONS

If in the equation

𝑓(𝐷, 𝐷′ )𝑧 = 𝐹(𝑥, 𝑦)

the polynomial expression 𝑓(𝐷, 𝐷′) is not homogeneous, then (1) is a non-homogeneous linear
partial differential equation. As in the case of homogeneous linear partial differential equations,
its complete solution = C.F. + P.I.
The methods to find P.I. are the same as those for homogeneous linear equations.

To find the C.F., we factorize 𝑓(𝐷, 𝐷′) into factors of the form 𝐷 − 𝑚𝐷′ − 𝑐. To find the
solution of (𝐷 − 𝑚𝐷′ − 𝑐)𝑧 = 0, we write it as

𝑝 − 𝑚𝑞 = 𝑐𝑧

The subsidiary equations are

ⅆ𝑥 ⅆ𝑦 ⅆ𝑧
= =
1 −𝑚 𝑐𝑧

Its integrals are 𝑦 + 𝑚𝑥 = 𝑎 and 𝑧 = 𝑏ⅇ 𝑐𝑥 .

Taking 𝑏 = 𝜙(𝑎), we get 𝑧 = ⅇ 𝑐𝑥 𝜙(𝑦 + 𝑚𝑥)

as the solution of (2). The solution corresponding to various factors added up, give the C.F. of
(1).

Example Solve (𝐷2 + 2𝐷𝐷′ + 𝐷′2 − 2𝐷 − 2𝐷′ )𝑧 = 𝑠𝑖𝑛 (𝑥 + 2𝑦).

Solution. Here 𝑓(𝐷, 𝐷′ ) = (𝐷 + 𝐷′ )(𝐷 + 𝐷′ − 2)

Since the solution corresponding to the factor 𝐷 − 𝑚𝐷′ − 𝑐 is known to be

𝑧 = ⅇ 𝑐𝑥 𝜙(𝑦 + 𝑚𝑥)

∴ C.F. = 𝜙1 (𝑦 − 𝑥) + ⅇ 2𝑥 𝜙2 (𝑦 − 𝑥)

1
∴ P.I. = 𝐷2+2𝐷𝐷′ +𝐷′2−2𝐷−2𝐷′ sin (𝑥 + 2𝑦)

1
= −1+2(−2)+(−4)−2𝐷−2𝐷′ sin (𝑥 + 2𝑦)

1 2(𝐷+𝐷 ′ )−9
= − 2(𝐷+𝐷′)+9 sin(𝑥 + 2𝑦) = − 4(𝐷2+2𝐷𝐷′ +𝐷′2)−81 sin (𝑥 + 2𝑦)

1
= 39 [2 cos(x + 2y) − 3 sin (𝑥 + 2𝑦)]

Hence the complete solution is

1
𝑧 = 𝜙1 (𝑦 − 𝑥) + ⅇ 2𝑥 𝜙2 (𝑦 − 𝑥) + 39 [2 cos(x + 2y) − 3 sin (𝑥 + 2𝑦)].
 PROBLEMS

Solve the following equations :

𝜕2𝑧 𝜕2𝑧 𝜕𝑧
1. 𝜕𝑥 2 + 𝜕𝑥𝜕𝑦 + 𝜕𝑦 2 − 𝑧 = ⅇ −𝑥 . 2. (𝐷 − 𝐷′ − 1)(𝐷 − 𝐷′ −
2)𝑧 = ⅇ 2𝑥−𝑦 .
𝜕2 𝑧 𝜕2𝑧 𝜕𝑧
3. (𝐷 + 𝐷′ − 1)(𝐷 + 2𝐷′ − 3)𝑧 = 4 + 3𝑥 + 6𝑦. 4. 𝜕𝑥 2 − 𝜕𝑥𝜕𝑦 + 𝜕𝑦 = 𝑥 2 + 𝑦 2 .
5. (𝐷2 + 𝐷𝐷′ + 𝐷′ − 1)𝑧 = sin(𝑥 + 2𝑦). 6. (2𝐷𝐷′ + 𝐷′2 + 3𝐷′)𝑧 =
3 cos(3𝑥 − 2𝑦).

NON-LINEAR EQUATIONS OF THE SECOND ORDER

We now give a method due to Monge*, for integrating the equation

𝑅𝑟 + 𝑆𝑠 + 𝑇𝑡 = 𝑉

in which 𝑅, 𝑆, 𝑇, 𝑉 are functions of 𝑥, 𝑦, 𝑧, 𝑝 and 𝑞.

Since
𝜕𝑝 𝜕𝑝
𝑑𝑝 = 𝜕𝑥 𝑑𝑥 + 𝜕𝑦 𝑑𝑦 = 𝑟𝑑𝑥 + 𝑡𝑑𝑦, and 𝑑𝑞 = 𝑠𝑑𝑥 + 𝑡𝑑𝑦,

we have

𝑟 = (𝑑𝑝 − 𝑡𝑑𝑦)/𝑑𝑥 and 𝑡 = (𝑑𝑞 − 𝑠𝑑𝑥)/𝑑𝑦.

Substituting these values of 𝑟 and 𝑡 in (1), and rearranging the terms, we get

(𝑅𝑑𝑝𝑑𝑦 𝑇𝑑𝑞𝑑𝑥 − 𝑉𝑑𝑎𝑑𝑦) − 𝑅𝑑𝑦 𝑆𝑑𝑦𝑑𝑟 + 𝑇𝑑𝑟) = 0

Let us consider the equations

𝑅𝑑𝑦 2 − 𝑆𝑑𝑦𝑑𝑥 + 𝑇𝑑𝑥 2 = 0

𝑅𝑑𝑝𝑑𝑦 + 𝑇𝑑𝑞𝑑𝑥 − 𝑉𝑑𝑥𝑑𝑦 = 0

which are known as Monge's equations.

 Since (3) can be factorized, we obtain its integral first. In case the factors are different,
we may get two distinct integrals of (3). Either of these together with (4) will give an integral of
(4). If need be, we may also use the relation 𝑑𝑧 = 𝑝𝑑𝑥 + 𝑞𝑑𝑦 while solving (3) and (4).
Let 𝑢(𝑥, 𝑦, 𝑧, 𝑝, 𝑞) = 𝑎 and 𝑣(𝑥, 𝑦, 𝑧, 𝑝, 𝑞) = 𝑏 be the integrals of (3) and (4) respectively.
Then 𝑢 = 𝑎, 𝑢 = 𝑏 evidently constitute a solution of (2) and therefore, of (1) also. Taking 𝑏 =
𝜙(𝑎), we find a general solution of (1) to be 𝑣 = 𝜙(𝑢), which should be further integrated by
methods of first order equations.

Example Solve (𝑥 − 𝑦)(𝑥𝑟 − 𝑥𝑠 − 𝑦𝑠 + 𝑦𝑡) = (𝑥 + 𝑦)(𝑝 − 𝑞).

Solution. Monge's equations are

𝑥𝑑𝑦 2 + (𝑥 + 𝑦)𝑑𝑦 𝑑𝑥 + 𝑦𝑑𝑥 2 = 0

𝑥+𝑦
𝑥𝑑𝑝𝑑𝑦 + 𝑦𝑑𝑞𝑑𝑥 − 𝑥−𝑦 (𝑝 − 𝑞) 𝑑𝑦𝑑𝑥 = 0

(i) may be factorized as (𝑥𝑑𝑦 + 𝑦𝑑𝑥)(𝑑𝑥 + 𝑑𝑦) = 0

whose integrals are 𝑥𝑦 = 𝑐 and 𝑥 + 𝑦 = 𝑐.

Taking 𝑥𝑦 = 𝑐 and dividing each term of (ii) by 𝑥𝑑𝑦 or its equivalent −𝑦𝑑𝑥, we get

ⅆ𝑥−ⅆ𝑦 ⅆ(𝑝−𝑞) ⅆ(𝑥−𝑦)

𝑑𝑝 − 𝑑𝑞 − (𝑝 − 𝑞) = 0 𝑜𝑟 − =0
𝑥−𝑦 𝑝−𝑞 𝑥−𝑦

This gives on integration (𝑝 − 𝑞)/(𝑥 − 𝑦) = 𝑐.

Hence a first integral of the given equation is 𝑝 − 𝑞 = (𝑥 − 𝑦) 𝜙(𝑥𝑦) which is a

Lagrange's linear equation. Its subsidiary equations are

ⅆ𝑥 ⅆ𝑦 ⅆ𝑧
= −1 = (𝑥−𝑦)𝜙(𝑥𝑦)
1

From the first two equations, we have 𝑥 + 𝑦 = 𝑎

Using this, we have

𝑑𝑧 = −𝜙(𝑎𝑥 − 𝑥 2 ). (𝑎 − 2𝑥)𝑑𝑥

which gives 𝑧 = 𝜙1 (𝑎𝑥 − 𝑥 2 ) + 𝑏

Writing 𝑏 = 𝜙2 (𝑎) and 𝑎 = 𝑥 + 𝑦, we get

𝑧 = 𝜙1 (𝑥𝑦) + 𝜙2 (𝑥 + 𝑦).
 Obs. Had we started with the integral 𝑥 + 𝑦 = 𝑐 and divided each term of (ii) by
𝑑𝑥 𝑜𝑟 − 𝑑𝑦, we would have arrived at the same solution.

Example Solve 𝑦 2 𝑟 − 2𝑦𝑠 + 𝑡 = 𝑝 + 6𝑦.

Solution. Monge's equations are 𝑦 2 𝑑𝑦 2 + 2𝑦𝑑𝑦𝑑𝑥 + 𝑑𝑥 2 = 0

and 𝑦 2 𝑑𝑝𝑑𝑦 + 𝑑𝑞𝑑𝑥 − (𝑝 + 6𝑦)𝑑𝑦𝑑𝑥 = 0

(i) gives (𝑦𝑑𝑦 + 𝑑𝑥)2 = 0 𝑖. ⅇ. 𝑦 2 + 2𝑥 = 𝑐

Putting 𝑦𝑑𝑦 = −𝑑𝑥 in (ii), we get

𝑦𝑑𝑝 − 𝑑𝑝 + (𝑝 + 6𝑦)𝑑𝑦 = 0 or (𝑦𝑑𝑝 + 𝑝𝑑𝑦) − 𝑑𝑞 + 6𝑦𝑑𝑦 = 0

whose integral is 𝑝𝑦 − 𝑞 + 3𝑦 2 = 𝑎

Combining this with (iii), we get the integral 𝑝𝑦 − 𝑞 + 3𝑦 2 = 𝜙(𝑦 2 + 2𝑥)

The subsidiary equations for this Lagrange's linear equation are

ⅆ𝑥 ⅆ𝑦 ⅆ𝑧
= −1 = 𝜙(𝑦 2+2𝑥)−3𝑦2
𝑦

From the first two equations, we have 𝑦 2 + 2𝑥 = 𝑐

Using this, we have 𝑑𝑧 + [𝜙(𝑐) − 3𝑦 2 ]𝑑𝑦 = 0

whose solution in 𝑧 + 𝑦𝜙(𝑐) − 𝑦 3 = 𝑏.

Hence the required solution is 𝑧 = 𝑦 3 − 𝑦𝜙(𝑦 2 + 2𝑥) + 𝛹2)

PROBLEMS

Solve :
1. (𝑞 + 1)𝑠 = (𝑝 + 1)𝑡.
2. 𝑟 − 𝑡 𝑐𝑜𝑠 2 𝑥 + 𝑝 tan 𝑥 = 0.
3. 2𝑥 2 2𝑟 − 5𝑥𝑦𝑠 + 2𝑦 2 𝑡 + 2(𝑝𝑥 + 𝑞𝑦) = 0.
4. 𝑥𝑦(𝑡 − 𝑟) + (𝑥 2 − 𝑦 2 )(𝑠 − 2) = 𝑝𝑦 − 𝑞𝑥.
5. 𝑞 2 𝑟 − 2𝑝𝑞𝑠 + 𝑝2 𝑡 + 𝑝𝑞 2 .
6. (1 + 𝑞)2 𝑟 − 2(1 + 𝑝 + 𝑞 + 𝑝𝑞)𝑠 + (1 + 𝑝)2 𝑡 = 0.
OBJECTIVE TYPE OF QUESTIONS

PROBLEMS

Fill up the blanks or choose the correct answer in each of the following problems:

𝜕2𝑧 𝜕𝑧 2 𝜕𝑧
1. The equation 𝜕𝑥 2 + 2𝑥𝑦 (𝜕𝑥) + 𝜕𝑦 = 5 is of order ………. and degree ………..
2. The complementary function of (𝐷2 − 4𝐷𝐷′ + 4𝐷′2 )𝑧 = 𝑥 + 𝑦 is ……..
𝜕2 𝑧
3. The solution of 𝜕𝑦 2 = sin (𝑥𝑦) is ……..
4. A solution of (𝑦 − 𝑧)𝑝 + (𝑧 − 𝑥)𝑞 = 𝑥 − 𝑦 is ……….
5. The particular integral of (𝐷2 + 𝐷𝐷′)𝑧 = sin (𝑥 + 𝑦) is ………
6. The partial differential equation obtained from 𝑧 = 𝑎𝑥 + 𝑏𝑦 + 𝑎𝑏 by eliminating 𝑎
and 𝑏 is ………..
7. Solution of √𝑝 + √𝑞 = 1 is ………
8. Solution of 𝑝√𝑥 + 𝑞 √𝑦 = √𝑧 is ………
9. Solution of 𝑝 − 𝑞 = log(𝑥 + 𝑦).
10. The order of the partial differential equation obtained by eliminating 𝑓 from
𝑧 = 𝑓(𝑥 2 + 𝑦 2 ), is ……….
𝜕𝑧
11. The solution of 𝑥 𝜕𝑥 = 2𝑥 + 𝑦 is
12. By eliminating 𝑎 and 𝑏 from 𝑧 = 𝑎(𝑥 + 𝑦) + 𝑏, the p.d.e. formed is ……….
13. The solution of (𝐷3 − 3𝐷2 𝐷′ + 2𝐷𝐷′2 )𝑧 = 0 is ………
14. By eliminating the arbitrary constants from 𝑧 = 𝑎2 𝑥 + 𝑎𝑦 2 + 𝑏, the partial
differential equation formed is.
15. A solution of 𝑢𝑥𝑦 = 0 is of the form …………
𝜕2 𝑢 𝜕2𝑢
16. If 𝑢 = 𝑥 2 + 𝑡 2 is a solution of 𝑐 2 𝜕𝑥 2 = , then 𝑐 = …….
𝜕𝑡 2
17. The general solution of 𝑢𝑥𝑦 = 𝑥𝑦 is ………….
18. The complementary function of 𝑟 − 7𝑠 + 6𝑡 = ⅇ 𝑥+𝑦 is …………
19. The solution of 𝑥𝑝 + 𝑝𝑞 = 𝑧 is
(i) 𝑓(𝑥 2 , 𝑦 2 ) = 0 (ii) 𝑓(𝑥𝑦, 𝑦𝑧) =0
𝑥 𝑦
(iii) 𝑓(𝑥, 𝑦) = 0 (iv) 𝑓 (𝑦 , 𝑧 ) = 0.
20. The solution of (𝑦 − 𝑧)𝑝 + (𝑧 − 𝑥)𝑞 = 𝑥 − 𝑦, is
(i) 𝑓(𝑥 2 + 𝑦 2 + 𝑧 2 ) = 𝑥𝑦𝑧
(ii) 𝑓(𝑥 + 𝑦 + 𝑧) = 𝑥𝑦𝑧
(iii) 𝑓(𝑥 + 𝑦 + 𝑧) = 𝑥² + 𝑦² + 𝑧²
(iv) 𝑓(𝑥 2 + 𝑦 2 + 𝑧 2 , 𝑥𝑦𝑧) = 0.
21. The partial differential equation from 𝑧 = (𝑐 + 𝑥)² + 𝑦 is
𝜕𝑧 2 𝜕𝑧 2
(i) 𝑧 = (𝜕𝑥) + 𝑦 (ii) 𝑧 = (𝜕𝑦) + 𝑦
1 𝜕𝑧 2 1 𝜕𝑧 2
(iii) 𝑧 = 4 (𝜕𝑥) + 𝑦 (iv) 𝑧 = 4 (𝜕𝑦) + 𝑦
22. The solution of 𝑝 + 𝑞 = 𝑧 is
(i) 𝑓(𝑥𝑦, 𝑦 𝑙𝑜𝑔 𝑧) = 0
(ii) 𝑓(𝑥 + 𝑦, 𝑦 + 𝑙𝑜𝑔 𝑧) = 0

(iii) 𝑓(𝑥 − 𝑦, 𝑦 − log 𝑧) = 0

(iv) None of these

23. Particular integral of (2𝐷2 − 3𝐷𝐷′ + 𝐷′2 )𝑧 = ⅇ 𝑥+2𝑦 is

1 𝑥
(i) 2 ⅇ 𝑥+2𝑦 (ii) − 2 ⅇ 𝑥+2𝑦
(iii) 𝑥ⅇ 𝑥+2𝑦 (iv) 𝑥 2 ⅇ 𝑥+2𝑦

𝜕3𝑧
24. The solution of 𝜕𝑥 3 = 0 is
(i) 𝑧 = (1 + 𝑥 + 𝑥 2 )𝑓(𝑦)
(ii) 𝑧 = (1 + 𝑦 + 𝑦 2 )𝑓(𝑥)
(iii) 𝑧 = 𝑓1 (𝑥) + 𝑦𝑓2 (𝑥) + 𝑦 2 𝑓3 (𝑥)
(iv) 𝑧 = 𝑓1 (𝑦) + 𝑥𝑓2 (𝑦) + 𝑥 2 𝑓3 (𝑦).

25. Particular integral of (𝐷2 − 𝐷′2 )𝑧 = 𝑐𝑜𝑠(𝑥 + 𝑦) is

𝑥
(i) 𝑥 𝑐𝑜𝑠 (𝑥 + 𝑦) (ii) 2 𝑐𝑜𝑠 (𝑥 + 𝑦)
𝑥
(iii) 𝑥 𝑠𝑖𝑛 (𝑥 + 𝑦) (iv) 2 𝑠𝑖𝑛 (𝑥 + 𝑦)
26. The solution of 𝜕 2 𝑧/𝜕𝑥 2 = 𝜕 2 𝑧/𝜕𝑦 2 is
(i) 𝑧 = 𝑓1 (𝑦 + 𝑥) + 𝑓2 (𝑦 − 𝑥)
(ii) 𝑧 = 𝑓1 (𝑦 + 𝑥) + 𝑓1 (𝑦 − 𝑥)
(iii) 𝑧 = 𝑓(𝑥 2 − 𝑦 2 )
(iv) 𝑧 = 𝑓(𝑥 2 + 𝑦 2 )
27. 𝑥𝑢𝑥 + 𝑦𝑢𝑦 = 𝑢2 is a non-linear partial differential equation.
(True or False)
28. 𝑥𝑢𝑥 + 𝑢𝑥𝑥 = 0 is a non-linear partial differential equation.
(True or False)
2 2
29. 𝑢 = 𝑥 − 𝑦 a solution of 𝑢𝑥𝑥 + 𝑢𝑦𝑦 = 0.
(True or False)
𝜕2𝑢 𝜕2𝑢
30. 𝑢 = ⅇ −𝑡 𝑠𝑖𝑛𝑥 is a solution of 𝜕𝑡 2 + 𝜕𝑥 2 = 0.
(True or False)
𝜕𝑢 𝜕𝑢
31. 𝑥 𝜕𝑥 + 𝑡 𝜕𝑡 = 2𝑢 is an ordinary differential equation.
(True or False)
 CO 4. Separation of Variables

At this point we are ready to now resume our work on solving the three
main equations: the heat equation, Laplace’s equation and the wave equa-
tion using the method of separation of variables.

4.1 The heat equation

Consider, for example, the heat equation

ut = u xx , 0 < x < 1, t>0 (4.1)

subject to the initial and boundary conditions

u(x, 0) = x − x2 , u(0, t) = u(1, t) = 0. (4.2)

Assuming separable solutions

u(x, t) = X(x)T(t), (4.3)

shows that the heat equation (4.1) becomes

XT 0 = X 00 T,

which, after dividing by XT and expanding gives

T0 X 00
= , (4.4)
T X

implying that
T 0 = λT, X 00 = λX, (4.5)

where λ is a constant. From (4.2) and (4.3), the boundary conditions be-
comes
X(0) = X(1) = 0. (4.6)
Integrating the X equation in (4.5) gives rise to three cases depending on
the sign of λ but as seen in the last chapter, only the case where λ = −k2
for some constant k is applicable which we have as the solution

X(x) = c1 sin kx + c2 cos kx. (4.7)

Imposing the boundary conditions (4.6) shows that

c1 sin 0 + c2 cos 0 = 0, c1 sin k + c2 cos k = 0, (4.8)

which leads to

c2 = 0, c1 sin k = 0 ⇒ k = 0, π, 2π, . . . nπ, (4.9)

where n is an integer. From (4.5), we further deduce that

2 π2 t
T(t) = c3 e−n ,

giving the solution

∞
2 2
u(x, t) = ∑ bn e−n π t sin nπx,
n=1

where we have set c1 c3 = bn . Using the initial condition gives

∞
u(x, 0) = x − x2 = ∑ bn sin nπx.
n=1

At this point, we recognize that we have a Fourier sine series and that the
coefficients bn are chosen such that
Z 1
bn = 2 (x − x2 ) sin nπx dx
0
· µ ¶ ¸¯1
1 − 2x x2 − x 2 ¯
= 2 cos nπx + − 3 3 cos nπx ¯¯
n2 π 2 nπ n π 0
4
= (1 − (−1)n ).
n π3
3

Thus, the solution of the PDE as

∞
4 1 − (−1)n −n2 π2 t
u(x, t) =
π3 ∑ n3 e sin nπx. (4.10)
n=1
Figure 1 shows the solution at times t = 0, 0.1 and 0.2.
Example 1
Solve
ut = u xx , 0 < x < 1, t>0 (4.11)

subject to
u(x, 0) = x − x2 , u x (0, t) = u x (1, t) = 0. (4.12)

This problem is similar to the proceeding problem except the boundary

conditions are different. The last problem had the boundaries fixed at zero
whereas in this problem, the boundaries are insulated (i.e. no flux bound-
ary conditions). Assuming that the solutions are separable

u(x, t) = X(x)T(t), (4.13)

then from the heat equation, we obtain

T 0 = λT, X 00 = λX, (4.14)

where λ is a constant. The boundary conditions in (4.12) become, accord-

ingly
X 0 (0) = X 0 (1) = 0. (4.15)

Integrating the X equation in (4.14) gives rise to again three cases depend-
ing on the sign of λ but as seen earlier, only the case where λ = −k2 for
some constant k is relevant. Thus, we have

X(x) = c1 sin kx + c2 cos kx. (4.16)

Imposing the boundary conditions (4.15) shows that

c1 k cos 0 − c2 k sin 0 = 0, c1 k cos k − c2 k sin k = 0, (4.17)

which leads to

c1 = 0, sin k = 0 ⇒ k = 0, π, 2π, . . . , nπ, (4.18)

where n is an integer. From (4.14), we also deduce that

2 π2 t
T(t) = c3 e−n ,

giving the solution

∞
2 2
u(x, t) = ∑ an e−n π t cos nπx,
n=0

where we have set c1 c3 = an . Using the initial condition gives

∞
2
u(x, 0) = x − x = ∑ an cos nπx
n=0
∞
a0
= + ∑ an cos nπx.
2 n=1

We again recognize that we have a Fourier cosine series and that the coef-
ficients an are chosen such that
Z 1 ¸¯1
·
2 x2 x3 ¯¯ 1
a0 = 2 (x − x ) dx = 2 − ¯ = ,
0 2 3 0 3
Z 1
an = 2 (x − x2 ) cos nπx dx
0
· µ ¶ ¸¯1
1 − 2x x2 − x 2 ¯
= 2 sin nπx − − 3 3 sin nπx ¯¯
n2 π 2 nπ n π 0
2
= − 3 3 (1 + (−1)n ).
n π

Thus, the solution of the PDE as

∞
1 2 (−1)( n + 1) − 1 −n2 π2 t
u(x, t) = + 3
3 π ∑ n3
e cos nπx. (4.19)
n=1

Figure 1 shows the solution at times t = 0, 0.25 and 0.5. It is interesting

to note that even though that same initial condition are used for each of
the two problems, fixing the boundaries and insulated them gives rise two
totally different behaviors after t ≥ 0.
 5

0.3 y 0.3 y
t=0 t=0

0.2 0.2
t = .05

t = 0.1
0.1 0.1
t = .025

t = 0.2

0.0 0.0

0.0 0.5
x 1.0 0.0 0.5
x 1.0

Figure 1. The solution of the heat equation with the same initial condition with
fixed and no flux boundary conditions.

Example 2
Solve
ut = u xx , 0 < x < 2, t>0 (4.20)

subject to
(
x if 0 < x < 1,
u(x, 0) = u(0, t) = u x (2, t) = 0. (4.21)
2−x if 1 < x < 2,

In this problem, we have a mixture of both fixed and no flux boundary

conditions. Again, assuming separable solutions

u(x, t) = X(x)T(t), (4.22)

gives rise to
T 0 = λT, X 00 = λX, (4.23)

where λ is a constant. The boundary conditions in (4.21) becomes, accord-

ingly
X(0) = X 0 (1) = 0. (4.24)

Integrating the X equation in (4.23) with λ = −k2 for some constant k gives

X(x) = c1 sin kx + c2 cos kx. (4.25)

Imposing the boundary conditions (4.24) shows that

c1 sin 0 + c2 cos 0 = 0, c1 k cos 2k − c2 k sin 2k = 0,

which leads to
π 3π 5π (2n − 1)π
c2 = 0, cos 2k = 0 ⇒ k = , , ,..., ,
4 4 4 4
for integer n. From (4.23), we then deduce that
(2n−1)2 2
T(t) = c3 e− 16 π t
,

giving the solution

∞ (2n−1)2 2 (2n − 1)
u(x, t) = ∑ bn e − 16 π t
sin
4
πx,
n=1

where we have set c1 c3 = bn . Using the initial condition give

∞
(2n − 1)
u(x, 0) = ∑ bn sin 4
πx.
n=1

Recognizing that we have a Fourier sine series, we obtain the coefficients

bn as
Z 1 Z 2
(2n − 1) (2n − 1)
bn = 2 x sin πx dx + (2 − x) sin πx dx
0 4 1 4
· ¸¯1
32 (2n − 1) 8x 2n − 1 ¯
= 2 2
sin πx − cos πx ¯¯
(2n − 1) π 4 (2n − 1)π 4 0
· ¸¯2
32 (2n − 1) 8(x − 2) 2n − 1 ¯
+ − sin πx + cos πx ¯
(2n − 1)2 π 2 4 (2n − 1)π 4 ¯
1
µ ¶
32 (2n − 1)π
= sin + cos nπ .
(2n − 1)2 π 2 4
Hence, the solution of the PDE is
³ ´
(2n−1)π
32 ∞ sin 4 + cos nπ (2n−1)2 2n − 1
− 16 π 2 t
u(x, t) = 2 ∑ 2
e sin πx. (4.26)
π n=1 (2n − 1) 4

Figure 2 shows the solution both in short time t = 0, 0.1, 0.2 and long time
t = 10, 20, 30.
 2 y 0.8 y
t=0 t = 10

t=2 t = 20
1 0.4

t = 30
t=1

x
0 0.0
x
0 1 2 0 1 2

Figure 2. Short time and long time behavior of the solution (4.26).

Example 3
Solve
ut = u xx , 0 < x < 2, t>0 (4.27)

subject to

u(x, 0) = 2x − x2 u(0, t) = 0, u x (2, t) = −u(2, t). (4.28)

In this problem, we have a fixed left endpoint and a radiating right end
point. Assuming separable solutions

u(x, t) = X(x)T(t), (4.29)

gives rise to
T 0 = λT, X 00 = λX, (4.30)

where λ is a constant. The boundary conditions in (4.28) becomes, accord-

ingly
X(0) = 0, X 0 (2) = −X(2). (4.31)

Integrating the X equation in (4.30) with λ = −k2 for some constant k gives

X(x) = c1 sin kx + c2 cos kx. (4.32)

Imposing the first boundary condition of (4.31) shows that

c1 sin 0 + c2 cos 0 = 0, ⇒ c2 = 0
and the second boundary condition of (4.31) gives

tan 2k = −k. (4.33)

It is very important that we recognize that the solutions of (4.33) are not
equally spaced as seen in earlier problems. In fact, there are an infinite
number of solutions of this equation. Figure 3 shows graphically the curves
y = −k and y = tan 2k. The first three intersection points are the first three
solutions of (4.33).

0 1 2 3 4
k

−2

k k3
1
−4
k 2

Figure 3. The graph of y = tan 2k and y = −k.

Thus, it is necessary that we solve for k numerically. The first 10 solutions

are given in table 1.
n kn n kn
1 1.144465 6 8.696622
2 2.54349 7 10.258761
3 4.048082 8 11.823162
4 5.586353 9 13.389044
5 7.138177 10 14.955947
Table 1. The first ten solution of tan 2k = −k.

Therefore, we have
X(x) = c1 sin k n x. (4.34)

Further, integrating (4.30) for T gives

2
T(t) = c3 e−kn t (4.35)
and together with X, we have the solution to the PDE as
∞
2
u(x, t) = ∑ cn e−kn t sin kn x, (4.36)
n=1
Imposing the boundary conditions (4.28) gives
∞
u(0, t) = 2x − x2 = ∑ cn sin kn x, (4.37)
n=1
It is important to know that the cn ’s are not given by the formula
Z
2 2
cn = (2x − x2 )sink n x, dx
2 0
as usual. The reason for this is that the k n ’s are not equally spaced. So
it is necessary to examine (4.37) on its own. Multiplying by sin k m x and
integrating over [0,2] gives
Z 2 ∞ Z 2
2
0
(2x − x ) sin k m x dx = ∑ cn
0
sin k m x sin k n x dx,
n=1
For n 6= m, we have
Z 2
k m sin 2k n cos 2k m − k n sin 2k m cos 2k n
sin k m x sin k n x dx = (4.38)
0 k2n − k2m
and imposing (4.33) for each of k m and k n shows (4.38) to be identically
satisfied. Therefore, we obtain the following when n = m
Z 2 Z 2
2
(2x − x ) sin k n x dx = cn sin2 k n x dx,
0 0
or R2
0 (2x − x2 ) sin k n x dx
cn = R2 (4.39)
2
0 sin k n x dx,

Table 2 gives the first ten cn ’s that correspond to each k n .

n cn n cn
1 0.873214 6 0.028777
2 0.341898 7 -.016803
3 -.078839 8 0.015310
4 0.071427 9 -.010202
5 -.032299 10 0.009458
Table 2. The coefficients cn from (4.77).
Having obtain k n and cn , the solution to the problem is found in (4.36).
Figure 4 show plots at time t = 0, 1, and 2 when 20 terms are used.

y
1.0 t = 0

0.5

t = 1

t = 2
0.0
x
0 1 2

Figure 4. The solution (4.36).

4.1.1 Nonhomogeneous boundary conditions

In the preceding examples, the boundary conditions where either fixed

to zero, insulted or radiating. Often, we encounter boundary condition
which are non standard or nonhomogeneous. For example, the boundary
may be fixed to particular constant or the flux is maintained at a constant
value. The following examples illustrate.

Example 4
Solve
ut = u xx , 0 < x < 3, t>0 (4.40)

subject to
u(x, 0) = 4x − x2 u(0, t) = 0, u(3, t) = 3. (4.41)

If we seek separable solutions u(x, t) = X(x)T(t), then from (4.41) we have

X(0)T(t) = 0, X(3)T(t) = 3, (4.42)

and we have a problem − the second boundary condition doesn’t separate.

To over come this we introduce the transformation u = v + ax + b and ask,
”Can we choose the constants a and b as to fix both boundary conditions
to zero. Upon substitution of both boundary conditions (4.41), we obtain

0 = v(0, t) + a(0) + b, 3 = v(3, t) + 3a + b (4.43)

Now we require that v(0, t) = 0 and v(3, t) = 0 which implies that we

must choose a = 1 and b = 0. Therefore, we have

u = v + x. (4.44)

We notice that under the transformation (4.44), the original equation doesn’t
change form, i.e.
ut = u xx ⇒ vt = v xx

however, the initial condition does change, it becomes

v(x, 0) = 3x − x2 .

Thus, we have the new problem to solve

vt = v xx , 0 < x < 3, t>0 (4.45)

subject to
v(x, 0) = 3x − x2 v(0, t) = 0, v(3, t) = 3. (4.46)

At this point, we seek the usual separable solutions V(x, t) = X(x)T(t)

which lead to the and the systems X 00 = −k2 X and T 0 = −k2 T with the
boundary conditions X(0) = 0 and X(3) = 0. Solving for X gives

X(x) = c1 sin kx + c2 cos kx, (4.47)

and imposing both boundary conditions gives

nπ
X(x) = c1 sin x, (4.48)
3

and
n2 π 2
T(t) = c3 e− 9 t
where n is an integer. Therefore, we have the solution of (4.45) as
∞ n2 π 2 nπ
v(x, t) = ∑ bn e − 9 t sin
3
x.
n=1

Recognizing that we have a Fourier sine series, we obtain the coefficients

bn as

Z
2 3 nπ
bn = (3x − x2 ) sin x dx
3 0 3
· ¸¯3
6(2x − 3) nπ 3(n2 π 2 x2 − 3n2 π 2 x − 18) nπ ¯¯
= − sin x+ cos x ¯
n2 π 2 3 n3 π 3 3 0
32(1 − (−1) ) n
= .
n3 π 3
This gives
∞
32 (1 − (−1)n ) − n2 π2 t nπ
v(x, t) =
π3 ∑ n 3
e 9 sin
3
x.
n=1
and since u = v + x, we obtain the solution for u as
∞
32 (1 − (−1)n ) − n2 π2 t nπ
u(x, t) = x +
π3 ∑ n 3
e 9 sin
3
x. (4.49)
n=1

4
y
t = 0

t = 1

2 t = 2

x
0

0 1 2 3

Figure 5. The solution (4.49) at time t = 0, 1 and 2.

Example 5
Solve
ut = u xx , 0 < x < 1, t>0 (4.50)

subject to
u(x, 0) = 0 u x (0, t) = −1, u x (1, t) = 0. (4.51)
Unfortunately, the trick u = v + ax + b won’t work since u x = v x + a and
choosing a to fix the right boundary to zero only makes the left boundary
nonzero. To overcome this we might try u = v + ax2 + bx but the original
equation changes
ut = u xx , ⇒ vt = v xx + 2a (4.52)

As a second attempt, we try

u = v + a(x2 + 2t) + bx (4.53)

noting now
ut = u xx , ⇒ vt = v xx . (4.54)

Since u x = v x + 2ax + b, then choosing a = 1/2 and b = −1 gives the

the new boundary conditions as v x (0, t) = 0 and v x (2, t) = 0 and the
transformation becomes
1
u = v + (x2 + 2t) − x, (4.55)
2
Finally, we consider the initial condition. From (4.55), we have v(x, 0) =
x2
u(x, 0) − 21 x2 + x = x − 2 and our problem is transformed to the new
problem
vt = v xx , 0 < x < 1, t > 0 (4.56)

subject to
1
v(x, 0) = x − x2 , v x (0, t) = v x (1, t) = 0 (4.57)
2
A separation of variables v = XT leads to X 00 = −k2 X and T 0 = −k2 T
from which we obtain

X = c1 sin kx + c2 cos kx, X = c1 k cos kx − c2 k sin kx, (4.58)

and imposing the boundary conditions (4.57) gives

c1 = 0, k = nπ, (4.59)

where n is an integer. This then leads to

X(x) = c2 cos nπx (4.60)

and further
2 π2 t
T(t) = c3 e−n . (4.61)

Finally we arrive at
∞
a0 2 2
v(x, t) = + ∑ an e−n π t cos nπx.
2 n=1

noting that we have chosen an = c1 c3 . Upon substitution of t = 0 and

using the initial condition (4.57), we have
∞
1 2 a0
x− x = + ∑ an cos nπx.
2 2 n=1

a Fourier cosine series. The coefficients are obtained by

Z 1µ ¶ ¯1
2 1 2 1 ¯ 2
a0 = x − x , dx = x2 − x3 ¯¯ = ,
1 0 2 3 0 3
Z
2 1 1
an = (x − x2 ) cos nπx dx
1 0 2
· µ ¶ ¸¯1
−2(x − 1) (2x − x2 ) 2 ¯
= cos nπx − + sin nπx ¯
n2 π 2 nπ n3 π 3 ¯
0
2
= − 2 2.
n π
Thus, we obtain the solution for v as
∞
1 2 1 −n2 π2 t
v(x, t) = − 2
3 π ∑ n2
e cos nπx.
n=1

and this, together with the transformation (4.55) gives

∞
1 1 2 1 −n2 π2 t
u(x, t) = (x2 + 2t) − x + − 2 ∑ e cos nπx. (4.62)
2 3 π n=1
n2

Figure 6 shows plots at time t = 0.01, 0.5, 1.0 and 1.5. It is interesting to
note that at the left boundary u x = −1 and since the flux φ = −ku x implies
that φ = k > 0 which gives that the flux is position and that heat is being
added at the left boundary. Hence the profile increase at the left while
insulated at the right boundary (no flux).
 2 y

t = 1.5

1
t = 1.0

t = 0.5

t = 0.1
0
x
0.0 0.5 1.0

Figure 6. The solution (4.62) at time t = 0, 1 and 2.

A natural question is, can we transform

ut = u xx , 0 < x < L, t > 0, (4.63)

u(x, 0) = f (x), u(0, t) = p(t), u(L, t) = q(t) (4.64)

to a problem with standard boundary conditions. The answer is yes. We

seek a transformation of the form

u = v + A(t)x + B(t) (4.65)

as to transform the nonstandard boundary conditions to standard ones.

On substitution of the u and v boundary conditions, we obtain

p(t) = 0 + A(t)0 + B(t), q(t) = 0 + A(t)L + B(t) (4.66)

and solving for A(t) and B(t) gives

q(t) − p(t)
A(t) = , B(t) = p(t), (4.67)
L
which results in the transformation
q(t) − p(t)
u = v+ x + p(t)
L
x (L − x)
= v + q(t) + p(t) . (4.68)
L L
However in doing so, we change not only the original equation but also
the initial condition. They becomes, respectively
x (L − x)
vt = v xx − q0 (t) + p0 (t) ,
L L
x (L − x)
v(x, 0) = f (x) − q(0) − p(0) .
L L
The new initial condition doesn’t pose a problem but how do we solve the
heat equation with a term added to the equation.

4.1.2 Nonhomogeneous equations

We now focus our attention to solving the heat equation with a source
term
ut = u xx + Q(x), 0 < x < L, t > 0, (4.69)

u(0, t) = 0, u(L, t) = 0, u(x, 0) = f (x).

To investigate this problem, we will considered a particular example where

L = 2, f (x) = 2x − x2 and Q(x) = 1 − |x − 1|. If we were to consider this
problem without a source term we would have solutions of the form
∞
nπx
u(x, t) = ∑ Tn (t) sin 2
, (4.70)
n=1

where
16(1 − (−1)n ) − n2 π2 t
Tn (t) = e 4 (4.71)
n3 π 3
We note that even if this Tn wasn’t known, we could find it as substitution
of (4.70) into the heat equation and isolating coefficients of sin nπx
2 , would
lead to
n2 π 2
Tn0 (t) = − Tn (t),
4
leading to the solution (4.71). For the problem with a source term we look
for solutions of the same form i.e. (4.70). However, in order that this tech-
nique works, it is necessary to expand the source term also in terms of a
Fourier sine series, i.e.
∞
nπx
Q(x) = ∑ qn sin 2
. (4.72)
n=1

For (
x if 0 < x < 1,
Q(x) = (4.73)
2−x if 1 < x < 2,
where
Z 2
nπx
qn = Q(x) sin dx.
0 2
Z 1 Z 2
nπx nπx
= x sin dx + (2 − x) sin dx,
0 2 x 2
· ¸¯
4 nπx nπx ¯¯1
= sin − 2xnπ cos
n2 π 2 2 2 ¯0
· ¸¯
4 nπx 2x4 nπx ¯¯2
+ − 2 2 sin + cos ,
n π 2 nπ 2 ¯1
8 nπx
= sin . (4.74)
n2 π 2 2
Substituting both (4.70) and (4.72) into (4.69) gives
∞
nπx ∞ ³ nπx ´2 nπx ∞
nπx
∑ Tn0 (t) sin = ∑− Tn (t) sin + ∑ qn sin ,
n=1
2 n=1
2 2 n=1
2

and re-grouping and isolating the coefficients of sin nπx

2 gives

n2 π 2
Tn0 (t) + Tn (t) = qn , (4.75)
4
a linear ODE in Tn (t)! On solving (4.75) we obtain
4 nπ 2
Tn (t) = qn + bn e−( 2 ) t ,
n2 π 2
giving the final solution
∞ µ ¶
4 −( nπ )2 t nπx
u(x, t) = ∑ 2
n π 2
q n + bn e 2 sin
2
. (4.76)
n=1

Imposing the initial condition gives (4.70)

∞ µ ¶
2 4 nπx
2x − x = ∑ 2 2
qn + bn sin .
n=1
n π 2

If we set
4
cn = qn + bn ,
n2 π 2
then we have
∞
2 nπx
2x − x = ∑ cn sin 2
.
n=1
a regular Fourier sine series. Therefore
Z 2³ ´ nπx
cn = 2x − x2 sin dx.
0 2
16 ³ nπ ´
= 1 − cos , (4.77)
n3 π 3 2
which in turn, gives
4
bn = c n − qn ,
n2 π 2
and finally, the solution as
∞ µ µ ¶ ¶
4 4 −( nπ ) 2t nπx
u(x, t) = ∑ 2 2
qn + cn − 2 2 qn e 2 sin , (4.78)
n=1
n π n π 2

where qn and cn are given in (4.74) and (4.77), respectively. Typical plots
are given in figure 7 at times t = 0, 1, 2 and 3.

1.0 y
t = 0

t = 1
0.5
t = 2

t = 3

0.0
x
0 1 2

Figure 7. The solution (4.78) at time t = 0, 1, 2 and 3.

It is interesting to note that if we let t → ∞ the solution approaches the

same curve at t = 3. This is what is called steady state (no changes in
time). It is natural to ask “Can we find this steady state solution.?” The
answer is yes. For the steady state, ut → 0 as t → ∞ and the original PDE
becomes
u xx + Q(x) = 0. (4.79)

Integrating twice with Q(x) given in (4.73) gives

( 3
− x + c1 x + c2 if 0 < x < 1,
u = x3 6 2 (4.80)
6 − x + k 1 x + k 2 if 1 < x < 2,
where c1 , c2 , k1 and k2 are constants of integration. Imposing that the solu-
tion and its first derivative are continuous at x = 1 and that the solution is
zer0 at the endpoints gives
2
c1 − c1 + 1 = 0, c1 + c2 − k1 − k2 + = 0,
3
8
c2 = 0, 2k1 + k2 − = 0,
3
which gives, upon solving
1 3 1
c1 = , c2 = 0, k1 = , k2 = − (4.81)
2 2 3
This, in turn gives the steady state solution as
( 3
− x6 + 2x if 0 < x < 1,
u = x3 2 3x 1
(4.82)
6 − x + 2 − 3 if 1 < x < 2.

4.1.3 Equations with a solution dependent source term

We now consider the heat equation with a solution dependent source term.
For simplicity we will consider a source term that is linear. Take, for ex-
ample

ut = u xx + αu, 0 < x < 1, t > 0,

u(0, t) = 0, u(1, t) = 0, u(x, 0) = x − x2 , (4.83)

where α is some constant. We could try a separation of variables to ob-

tain solutions for this problem, but for more complicated source terms like
Q(x, t)u, a separation of variables is unsuccessful. Therefore, we try a dif-
ferent technique. Here we will try and transform the PDE to one that has
no source term. In attempting to do so, we seek a transformation of the
form
u(x, t) = A(x, t)v (4.84)

and ask “Is is possible to find A such that the source term in (4.83) can be
removed?” Substituting of (4.84) in (4.83) gives

Avt + At v = Av xx + 2A x v x + A xx v + αAv,
and dividing by A and expanding and regrouping gives
µ ¶
Ax A xx At
vt = v xx + 2 v x + − + α v.
A A A
In order to target the standard heat equation, we choose
A xx At
A x = 0, − + α = 0. (4.85)
A A
From the first we obtain that A = A(t) and from the second we obtain
A0 = αA which has the solution A(t) = A0 eαt for some constant A0 . The
boundary conditions becomes

u(0, t) = 0 ⇒ A0 eαt v(0, t) = 0 ⇒ v(0, t) = 0,

v(1, t) = 0 ⇒ A0 eαt v(1, t) = 0 ⇒ v(1, t) = 0,

so the boundary conditions become unchanged. Next, we consider the

initial condition, so

u(x, 0) = x − x2 ⇒ A0 eα0 v(x, 0) = 0 ⇒ A0 v(x, 0) = x − x2 ,

and as to leave the initial condition unchanged, we choose A0 = 1. Thus,

under the transformation
u = eαt v (4.86)

the problem (4.83) becomes

vt = v xx , 0 < x < 1, t > 0,

v(x, 0) = 0, v(1, t) = 0, v(x, 0) = x − x2 , (4.87)

This particular problem was considered at the beginning of this chapter,

(4.1) where the solution was given in (4.26), namely
∞
4 1 − (−1)n −n2 π2 t
v(x, t) =
π3 ∑ n3 e sin nπx,
n=1

and so, from (4.86) we obtain the solution of (4.83) as

4 αt ∞ 1 − (−1)n −n2 π2 t
u(x, t) = 3 e ∑ e sin nπx. (4.88)
π n=1
n3
 Figure 8 shows plots at times t = 0, 0.1 and 0.2 when α = 5 and 12. It
is interesting to note that in the case where α = 5, the diffusion is slower
in comparison with no source term (i.e α = 0 see Figure 1) and there is no
diffusion at all when α = 12.

t = 0.2
0.3 0.4 y
t= 0
t = 0.1

0.2 t= 0
t = 0.1
0.2
t = 0.2
0.1

0.0 0.0
x x
0.0 0.5 1.0 0.0 0.5 1.0

Figure 8. The solution (4.88) of the heat equation with a source (4.83) with α = 5
and α = 12.

It is natural to ask, for what value of α do we achieve a steady state

solution. To answer this consider the first few terms of the solution (4.88)
µ ¶
8 αt −π2 t 1 −9π2 t
u = 3e e sin πx + e sin 3πx + . . . . (4.89)
π 27
Now clearly the exponential terms in (·) will decay to zero with the first
term decaying the slowest. Therefore, it is the balance between eαt and
2
e−π t which determine whether the solution will decay to zero or not. It is
equality α = π 2 that gives the steady state solution.

Example 6
Solve

ut = u xx + αu, 0 < x < 2, t>0

u(x, 0) = 4x − x3 u x (0, t) = 0, u x (2, t) = 0. (4.90)

Established already is that the transformation (4.86) will transform the

equation to the heat equation and will leave the initial condition unchanged.
It is now necessary to determine what happens to the boundary condi-
tions. Using (4.86) we have

u x (0, t) = 0 ⇒ A0 eαt v x (0, t) = 0 ⇒ v x (0, t) = 0, (4.91)

v x (2, t) = 0 ⇒ A0 eαt v x (2, t) = 0 ⇒ v x (2, t) = 0, (4.92)

and so the insulted boundary conditions also remain insulated! Thus, the
problem (4.90) becomes

vt = v xx , 0 < x < 2, t > 0

v(x, 0) = 4x − x2 , v x (0, t) = 0, v x (2, t) = 0. (4.93)

Using a separation of variables and imposing the boundary conditions

gives (see example 1)
∞
1 n2 π 2 nπ
v= a0 + ∑ an e− 4 t cos x, (4.94)
2 n=1
2

where
Z 2 · ¸¯2
2 3 x4 ¯¯
2
a0 = (4x − x ) dx = 2x − = 4,
2 0 4 ¯0
Z
2 2 nπ
an = (4x − x3 ) cos x dx (4.95)
2 0 2
·µ ¶ µ ¶ ¸¯2
2(4x − x3 ) 48 nπ 4(4 − 3x2 ) 96 nπ ¯¯
= + 3 3 sin x+ + 4 4 cos x ¯
nπ n π 2 n2 π 2 n π 2 0
µ ¶ µ ¶
16 96 6 48 nπ 4 96 nπ
= − 2 2− 4 4+ + 3 3 sin + 2 2
+ 4 4 cos .
n π n π nπ n π 2 n π n π 2

Together with the transformation (4.86), the solution of (4.90) is

∞ n2 π 2 nπ
u = 2eαt + eαt ∑ an e− 4 t cos x, (4.96)
n=1
2

where an is given in (4.95). Figure 9 show plots at t = 0, 0.2, 0.4 and 0.6
when α = −2 and 2. It is interesting to note that the sign of α will deter-
mine whether the solution will grow or decay exponentially.
 y y
3 t= 0 8

t = 0.6

6
2
t = 0.4
4
t = 0.2
t = 0.2
1 t = 0.4
2
t = 0.6
t= 0

0 x 0
x
0 1 2
0 1 2
Figure 9. The solution (4.96) of the heat equation with a source (4.90) with no
flux boundary condition with α = −2 and α = 2.

4.1.4 Equations with a solution dependent convective term

We now consider the heat equation with a solution dependent linear con-
vection term, i.e.

ut = u xx + β u x , 0 < x < 1, t > 0

u(0, t) = 0, u(1, t) = 0, u(x, 0) = x − x2 , (4.97)

where β is some constant. We consider the same initial and boundary

conditions as in the previous section as it provides a means of comparing
the two respective problems. Again, we could try a separation of variables
to obtain solutions for this problem, but for more complicated convection
terms like P(x, t)u x , a separation of variable is unsuccessful. Therefore,
we again try a different technique. We will try and transform the PDE
to one that has no convection term. In attempting to do so, we seek a
transformation of the form

u(x, t) = A(x, t)v (4.98)

and ask “Is is possible to find A such that the convection term can be re-
moved?” Substituting of (4.98) in (4.97) gives

Avt + At v = Av xx + 2A x v x + A xx v + β (Av x + A x v) ,
and dividing by A and expanding and regrouping gives
2A x + βA A xx − At + βA x
vt = v xx + vx + v. (4.99)
A A
In order to target the standard heat equation, we choose

2A x + βA = 0, A xx − At + βA x = 0. (4.100)
1
From the first we obtain that A(x, t) = C(t)e− 2 βx and from the second
β2 1 2
we obtain C 0 + 4 C = 0 which has the solution C(t) = A0 e− 4 β t for some
constant A0 . This then gives
1 1 2
A(x, t) = A0 e− 2 βx− 4 β t
(4.101)

The boundary conditions becomes

1 2
u(0, t) = 0 ⇒ C0 e− 4 β t v(0, t) = 0 ⇒ v(0, t) = 0, (4.102)
− 21 − 14 β2 t
v(1, t) = 0 ⇒ C0 e v(1, t) = 0 ⇒ v(1, t) = 0, (4.103)

so the boundary conditions become unchanged. Next, we consider the

initial condition, so
1
u(x, 0) = x − x2 ⇒ v(x, 0) = (x − x2 )e 2 βx , (4.104)

where we have chosen A0 = 1. So here, the initial condition actually

changes. Thus, under the transformation
1 1 2
u = e− 2 βx− 4 β t v, (4.105)

the problem (4.97) becomes

vt = v xx , 0 < x < 1, t > 0,

1
v(x, 0) = 0, v(1, t) = 0, v(x, 0) = (x − x2 )e 2 βx . (4.106)

As in the previous section, the solution is given by

∞
2 2
v(x, t) = ∑ bn e−n π t sin nπx, (4.107)
n=1
where bn is now given by
Z 1
2 1
bn = (x − x2 )e 2 βx sin nπx dx, (4.108)
1 0

and from (4.105)

1 1 2
∞
2 2
u(x, t) = e− 2 βx− 4 β t
∑ bn e−n π t sin nπx, (4.109)
n=1

At this point we consider two particular examples: β = 6 and β = −12.

For β = 6
Z 1
bn = 2 (x − x2 )e3x sin nπx dx
0
27 + n2 π 2 + 2n2 π 2 e3 cos nπ
= −4nπ . (4.110)
(9 + n2 π 2 )3

For β = −12
Z 1
bn = 2 (x − x2 )e−6x sin nπx dx
0
7n2 π 2 + 108 + (5n2 π 2 + 324)e−6 cos nπ
= 2nπ . (4.111)
(36 + n2 π 2 )3

The respective solutions for each are

1 1 2
u(x, t) = 4πe− 2 βx− 4 β t (4.112)
∞ 2 2 2 2 3
27 + n π + 2n π e cos nπ −n2 π2 t
× ∑n e sin nπx,
n=1
(9 + n2 π 2 )3

1 1 2
u(x, t) = 2πe− 2 βx− 4 β t (4.113)
∞ 2 2 2 2 −6
7n π + 108 + (5n π + 324)e cos nπ −n2 π2 t
× ∑n 2 π 2 )3
e sin nπx.
n=1
(36 + n

Figure 10 shows graphs at a variety of times for β = 6 and β = −12.

 0.3 y 0.3 y
t= 0 t= 0

t = 0.025
0.2 0.2
t = 0.05

t = 0.05
0.1 t = 0.1 0.1

0.0 x 0.0 x
0.0 0.5 1.0 0.0 0.5 1.0

Figure 10. The solution (4.112) of the heat equation with convection with fixed
boundary conditions with β = 6 and β = −12.

Example 7
As a final example, we consider

ut = u xx + βu x , 0 < x < 2, t>0 (4.114)

subject to
u(x, 0) = 4x − x3 u x (0, t) = 0, u x (2, t) = 0. (4.115)

This problem is like problem 6 but with insulated boundary conditions.

Here, we will simply transform the problem to one that is in standard form
as to contrast the differences between the two problems. The transforma-
tion (4.105) transforms (4.114) to the heat equation so we will primarily fo-
cus on the boundary and initial conditions. For the boundary conditions,
upon differentiating (4.105) with respect to x gives

1 1 2 β − 1 βx− 1 β2 t
u x = e− 2 βx− 4 β t v x − e 2 4 v, (4.116)
2

and so

β
u x (0, t) = 0 ⇒ v x (0, t) − v(0, t) = 0, (4.117)
2
β
u x (2, t) = 0 ⇒ v x (2, t) − v(2, t) = 0. (4.118)
2
Thus, the insulated boundary condition become radiating boundary con-
ditions. As for the initial condition
1
u(x, 0) = 4x − x3 ⇒ v(x, 0) = (4x − x3 )e 2 βx , (4.119)

so again, the initial condition changes.

4.2 Laplace’s equation

The two dimensional Laplace’s equation is

u xx + uyy = 0 (4.120)

We will show that a separation of variables also works for this equation
As an example consider the boundary conditions

u(x, 0) = 0, u(x, 1) = x − x2 (4.121a)

u(0, y) = 0, u(1, 0) = 0. (4.121b)

If we assume separable solutions of the form

u(x, y) = X(x)Y(y), (4.122)

then substituting this into (4.120) gives

X 00 Y + XY 00 = 0. (4.123)

Dividing by XY and expanding gives

X 00 Y 00
+ = 0, (4.124)
X Y
and since each term is only a function of x or y, then each must be constant
giving
X 00 Y 00
= λ, = −λ. (4.125)
X Y
From the first of (4.121a) and both of (4.121b) we deduce the boundary
conditions
X(0) = 0, X(1) = 0, Y(0) = 0. (4.126)
The remaining boundary condition in (4.121a) will be used later. As seen
the in previous section, in order to solve the X equation in (4.125) subject
to the boundary conditions (4.126), it is necessary to set λ = −k2 . The X
equation (4.125) as the general solution

X = c1 sin kx + c2 cos kx (4.127)

To satisfy the boundary conditions in (4.126) it is necessary to have c2 = 0

and k = nπ, k ∈ Z+ so

X(x) = c1 sin nπx. (4.128)

From (4.125), we obtain the solution to the Y equation

Y(y) = c3 sinh nπy + c4 cosh nπy (4.129)

Since Y(0) = 0 this implies c4 = 0 so

X(x)Y(y) = an sin nπx sinh nπy (4.130)

where we have chosen an = c1 c3 . Therefore, we obtain the solution to

(4.120) subject to three of the four boundary conditions in (4.121)
∞
u= ∑ an sin nπx sinh nπy. (4.131)
n=1

The remaining boundary condition is (4.121a) now needs to be satisfied,

thus
∞
2
u(x, 1) = x − x = ∑ an sin nπx sinh nπ. (4.132)
n=1
This looks like a Fourier sine series and if we let An = an sinh nπ, this
becomes
∞
∑ An sin nπx = x − x2. (4.133)
n=1
which is precisely a Fourier sine series. The coefficients An are given by
Z
2 2³ ´
An = x − x2 sin nπx dx
1 0
16
= 3 3 (1 − cos nπ), (4.134)
n π
and since An = an sinh nπ, this gives

4(1 − (−1)n )
an = . (4.135)
n3 π 3 sinh nπ
Thus, the solution to Laplace’s equation with the boundary conditions
given in (4.121) is
∞
4 1 − (−1)n sinh nπy
u(x, y) = 3
π ∑ n 3
sin nπx
sinh nπ
. (4.136)
n=1

y u
1.0 0.2
u = 0.2

u=
0.1
0.5
u = 0.05

u = 0.025 0.0

u = 0.001 0.5
1.0
0.0 1.0
x
0.0 0.5 1.0
y
x

Figure 11. The solution (4.120) with the boundary conditions (4.121)

In general, using separation of variables, the solution of

u xx + uyy = 0, 0 < x < L x , 0 < y < Ly , (4.137)

subject to

u(x, 0) = 0, u(x, 1) = f (x),

u(0, y) = 0, u(1, y) = 0,

is nπy
∞
nπx sinh Ly
u = ∑ bn sin , (4.139)
n=1
L x sinh nπ
Ly
30

where Z Lx
2 nπx
bn = f (x) sin dx. (4.140)
Lx 0 Lx

In the next three examples, we will construct solutions of Laplace’s equa-

tion when we have nonzero boundary condition on each of the remaining
three sides of the region 0 < 1, 0 < y < 1.

Example 8
Solve
u xx + uyy = 0, (4.141)

subject to

u(x, 0) = 0, u(x, 1) = 0, (4.142a)

u(0, y) = 0, u(1, y) = y − y2 . (4.142b)

Assume separable solutions of the form

u(x, y) = X(x)Y(y) (4.143)

Then substituting this into (4.141) gives

X 00 Y + XY 00 = 0. (4.144)

Dividing by XY and expanding gives

X 00 Y 00
+ = 0, (4.145)
X Y

from which we obtain

X 00 Y 00
= λ, = −λ (4.146)
X Y

From (4.186) we deduce the boundary conditions

X(0) = 0, Y(0) = 0, Y(1) = 0. (4.147)

The remaining boundary condition in (4.186) will be used later. As seen
the in previous problem, in order to solve the Y equation in (4.146) subject
to the boundary conditions (4.147), it is necessary to set λ = k2 . The Y
equation (4.146) as the general solution

Y = c1 sin ky + c2 cos ky (4.148)

To satisfy the boundary conditions in (4.147) it is necessary to have c2 = 0

and k = nπ so
Y(y) = c1 sin nπy. (4.149)

From (4.146), we obtain the solution to the X equation

X(x) = c3 sinh nπx + c4 cosh nπx (4.150)

Since X(0) = 0 this implies c4 = 0. This gives

X(x)Y(y) = an sinh nπx sin nπy (4.151)

where we have chosen an = c1 c3 . Therefore, we obtain

∞
u= ∑ an sinh nπx sin nπy. (4.152)
n=1

The remaining boundary condition is (4.186) now needs to be satisfied,

thus
∞
u(1, y) = y − y2 = ∑ an sinh nπ sin nπy. (4.153)
n=1

If we let An = an sinh nπ, this becomes

∞
∑ An sin nπy = y − y2. (4.154)
n=1

Comparing with previous problem, we find that interchanging x and y

interchanges the two problems and thus we con conclude that

16
An = (1 − cos nπ), (4.155)
n3 π 3
and further that the solution to Laplace’s equation with the boundary con-
ditions given in (4.141) subject to (4.186) is
∞
4 1 − (−1)n sinh nπx
u= 3
π ∑ n3 sinh nπ
sin nπy. (4.156)
n=1

Figure 12 show both a top view and a 3 − D view of the solution. In com-
paring the solutions (4.136) and (4.156) shows that if we interchange x and
y they are the same. This should not be surprising because if we con-
sider Laplace equations with the boundary conditions given in (4.121) and
(4.186), that if we interchange x and y, the problems are transformed to
each other.
In general, using separation of variables, the solution of

u xx + uyy = 0, 0 < x < L x , 0 < y < Ly (4.157)

subject to

u(x, 0) = 0, u(x, 1) = 0
u(0, y) = 0, u(1, y) = g(y).

is
∞ sinh nπx
Lx nπy
u= ∑ bn sinh nπ sin
Ly
(4.159)
n=1 Lx
where Z Ly
2 nπy
bn = g(y) sin dy (4.160)
Ly 0 Ly

Example 9
Solve
u xx + uyy = 0 (4.161)

subject to

u(x, 0) = x − x2 , u(x, 1) = 0 (4.162a)

u(0, y) = 0, u(1, y) = 0. (4.162b)
Again, assuming separable solutions of the form

u(x, y) = X(x)Y(y) (4.163)

leads to
X 00 Y + XY 00 = 0, (4.164)

when substituted into (4.161). Dividing by XY and expanding gives

X 00 Y 00
+ = 0. (4.165)
X Y
Since each term is only a function of x or y then each must be constant
giving
X 00 Y 00
= λ, = −λ (4.166)
X Y
From the first of (4.162a) and both of (4.162b) we deduce the boundary
conditions
X(0) = 0, X(1) = 0, Y(1) = 0, (4.167)

noting that the last boundary condition is different than the boundary con-
dition considered at the beginning of this section (i.e. Y(0) = 0). The re-
maining boundary condition in (4.162a) will be used later. In order to solve
the X equation in (4.166) subject to the boundary conditions (4.167), it is
necessary to set λ = −k2 . The X equation (4.166) as the general solution

Y = c1 sin kx + c2 cos kx (4.168)

To satisfy the boundary conditions in (4.167) it is necessary to have c2 = 0

and k = nπ so
X(x) = c1 sin nπx. (4.169)

From (4.166), we obtain the solution to the Y equation

Y(y) = c3 sinh nπy + c4 cosh nπy (4.170)

Since Y(1) = 0 this implies

sinh nπ
c3 sinh nπ + c4 cosh nπ = 0 ⇒ c4 = −c2 . (4.171)
cosh nπ
From (4.172) we have
sinh nπ
Y(y) = c3 sinh nπy − c3 cosh nπy
cosh nπ
sinh nπ(1 − y)
= −c3 . (4.172)
sinh nπ
so
X(x)Y(y) = an sin nπx sinh nπ(1 − y) (4.173)

where we have chosen an = −c1 c3 / sinh nπ. Therefore, we obtain

∞
u= ∑ an sin nπx sinh nπ(1 − y). (4.174)
n=1

The remaining boundary condition is (4.162a) now needs to be satisfied,

thus
∞
u(x, 0) = x − x2 = ∑ an sin nπx sinh nπ. (4.175)
n=1
At this point we recognize that this problem is now identically to the first
problem in this section where we obtained
4(1 − (−1)n )
an = (4.176)
n3 π 3 sinh nπ
so the solution to Laplace’s equation with the boundary conditions given
in (4.179) is
∞
4 1 − (−1)n sinh nπ(1 − y)
u(x, y) =
π3 ∑ n3 sin nπx sinh nπ . (4.177)
n=1

Figure 12 show both a top view and a 3 − D view of the solution.

In general, using separation of variables, the solution of

u xx + uyy = 0, 0 < x < L x , 0 < y < Ly (4.178)

subject to

u(x, 0) = f (x), u(x, 1) = 0

u(0, y) = 0, u(1, y) = 0.
is
nπ(1−y)
∞
nπx sinh Ly
u = ∑ bn sin (4.180)
n=1
Lx sinh nπ
Ly
where Z Lx
2 nπx
bn = f (x) sin dx (4.181)
Lx 0 Lx

Example 10
As the final example, we consider

u xx + uyy = 0 (4.182)

subject to

u(x, 0) = 0, u(x, 1) = 0 (4.183a)

u(0, y) = y − y2 , u(1, y) = 0. (4.183b)

We could go through a separation of variables to obtain the solution but

we can avoid many of the steps by considering the previous three prob-
lems. One will notice to transform between the first and second problem,
the variables x and y only need to be interchanged. This can also be seen in
their respective solutions. One will also notice to transform between this
problem and problem 9, we only need to transform x and y again. Thus,
to obtain the solution for this final problem, we will transform the solution
given in (4.177) giving
∞
4 1 − (−1)n sinh nπ(1 − x)
u(x, y) =
π3 ∑ n3 sinh nπ
sin nπy. (4.184)
n=1

In general, using separation of variables, the solution of

u xx + uyy = 0, 0 < x < L x , 0 < y < Ly (4.185)

subject to

u(x, 0) = 0, u(x, 1) = 0
u(0, y) = g(y), u(1, y) = 0.
is
∞ sinh nπ(1−x)
Ly nπy
u= ∑ bn sinh nπ sin
Ly
. (4.187)
n=1 Ly

where Z Ly
2 nπy
bn = g(y) sin dy (4.188)
Ly 0 Ly
Second-Order Partial Differential Equations
The most general case of second-order linear partial differential equation (PDE) in two inde-
pendent variables is given by

∂ 2u ∂ 2u ∂ 2u ∂u ∂u
A + B + C + D + E + Fu = G (1)
∂ x2 ∂ x∂ y ∂ y2 ∂x ∂y
where the coefficients A, B, and C are functions of x and y and do not vanish simultaneously,
because in that case the second-order PDE degenerates to one of first order. Further, the
coefficients D, E, and F are also assumed to be functions of x and y. We shall assume that the
function u(x, y) and the coefficients are twice continuously differentiable in some domain Ω.
The classification of second-order PDE depends on the form of the leading part of the
equation consisting of the second order terms. So, for simplicity of notation, we combine the
lower order terms and rewrite the above equation in the following form

∂ 2u ∂ 2u ∂ 2u ∂u ∂u
 
A(x, y) 2 + B(x, y) + C(x, y) 2 = Φ x, y, u, , (2a)
∂x ∂ x∂ y ∂y ∂x ∂y

or using the short-hand notations for partial derivatives,

A(x, y)uxx + B(x, y)uxy + C(x, y)uyy = Φ (x, y, u, ux , uy ) (2b)

As we shall see, there are fundamentally three types of PDEs – hyperbolic, parabolic, and
elliptic PDEs. From the physical point of view, these PDEs respectively represents the wave
propagation, the time-dependent diffusion processes, and the steady state or equilibrium pro-
cesses. Thus, hyperbolic equations model the transport of some physical quantity, such as
fluids or waves. Parabolic problems describe evolutionary phenomena that lead to a steady
state described by an elliptic equation. And elliptic equations are associated to a special state
of a system, in principle corresponding to the minimum of the energy.
Mathematically, these classification of second-order PDEs is based upon the possibility of
reducing equation (2) by coordinate transformation to canonical or standard form at a point. It
may be noted that, for the purposes of classification, it is not necessary to restrict consideration
to linear equations. It is applicable to quasilinear second-order PDE as well. A quasilinear
second-order PDE is linear in the second derivatives only.
The type of second-order PDE (2) at a point (x0 , y0 ) depends on the sign of the discriminant
defined as  
 B 2A 
∆(x0 , y0 ) ≡   = B(x0 , y0 )2 − 4 A(x0 , y0 )C(x0 , y0 ) (3)
2C B 
The classification of second-order linear PDEs is given by the following: If ∆(x0 , y0 ) > 0, the
equation is hyperbolic, ∆(x0 , y0 ) = 0 the equation is parabolic, and ∆(x0 , y0 ) < 0 the equation
is elliptic. It should be remarked here that a given PDE may be of one type at a specific point,
and of another type at some other point. For example, the Tricomi equation

∂ 2u ∂ 2u
+ x =0
∂ x2 ∂ y2
is hyperbolic in the left half-plane x < 0, parabolic for x = 0, and elliptic in the right half-plane
x > 0, since ∆ = −4x. A PDE is hyperbolic (or parabolic or elliptic) in a region Ω if the PDE
is hyperbolic (or parabolic or elliptic) at each point of Ω.
The terminology hyperbolic, parabolic, and elliptic chosen to classify PDEs reflects the anal-
ogy between the form of the discriminant, B2 −4AC, for PDEs and the form of the discriminant,
B2 − 4AC, which classifies conic sections given by

Ax2 + Bxy + Cy2 + Dx + Ey + F = 0

The type of the curve represented by the above conic section depends on the sign of the
discriminant, ∆ ≡ B2 − 4AC. If ∆ > 0, the curve is a hyperbola, ∆ = 0 the curve is an parabola,
and ∆ < 0 the equation is a ellipse. The analogy of the classification of PDEs is obvious. There
is no other significance to the terminology and thus the terms hyperbolic, parabolic, and elliptic
are simply three convenient names to classify PDEs.
In order to illustrate the significance of the discriminant ∆ and thus the classification of the
PDE (2), we try to reduce the given equation (2) to a canonical form. To do this, we transform
the independent variables x and y to the new independent variables ξ and η through the change
of variables
ξ = ξ (x, y), η = η (x, y) (4)
where both ξ and η are twice continuously differentiable and that the Jacobian

∂ (ξ , η )  ξx ξy 
 
J= =  6= 0 (5)
∂ (x, y) ηx ηy 


in the region under consideration. The nonvanishing of the Jacobian of the transformation
ensure that a one-to-one transformation exists between the new and old variables. This simply
means that the new independent variables can serve as new coordinate variables without any
ambiguity. Now, define w(ξ , η ) = u(x(ξ , η ), y(ξ , η )). Then u(x, y) = w(ξ (x, y), η (x, y)) and,
apply the chain rule to compute the terms of the equation (2) in terms of ξ and η as follows:

ux = wξ ξx + wη ηx
uy = wξ ξy + wη ηy
uxx = wξ ξ ξx2 + 2wξ η ξx ηx + wηη ηx2 + wξ ξxx + wη ηxx (6)
uyy = wξ ξ ξy2 + 2wξ η ξy ηy + wηη ηy2 + wξ ξyy + wη ηyy
uxy = wξ ξ ξx ξy + wξ η (ξx ηy + ξy ηx ) + wηη ηx ηy + wξ ξxy + wη ηxy

Substituting these expressions into equation (2) we obtain the transformed PDE as

awξ ξ + bwξ η + cwηη = φ ξ , η , w, wξ , wη

(7)

where Φ becomes φ and the new coefficients of the higher order terms a, b, and c are expressed
via the original coefficients and the change of variables formulas as follows:

a = Aξx2 + Bξx ξy + Cξy2

b = 2Aξx ηx + B(ξx ηy + ξy ηx ) + 2Cξy ηy (8)
c = Aηx2 + Bηx ηy + Cηy2

At this stage the form of the PDE (7) is no simpler than that of the original PDE (2), but this
is to be expected because so far the choice of the new variable ξ and η has been arbitrary.
However, before showing how to choose the new coordinate variables, observe that equation
(8) can be written in matrix form as
T
ξx ξy ξx ξy
    
a b/2 A B/2
=
b/2 c ηx ηy B/2 C ηx ηy

Recalling that the determinant of the product of matrices is equal to the product of the de-
terminants of matrices and that the determinant of a transpose of a matrix is equal to the
determinant of that matrix, we get
   
 a b/2   A B/2  2
=
 b/2 c   B/2 C  J
 

where J is the Jacobian of the change of variables given by (5). Expanding the determinant
and multiplying by the factor, −4, to obtain

b2 − 4ac = J 2 (B2 − 4AC) =⇒ δ = J 2∆ (9)

where δ = b2 − 4ac is the discriminant of the equation (7). This shows that the discriminant
of the transformed equation (7) has the same sign as the discriminant of the original equation
(2) and therefore it is clear that any real nonsingular (J 6= 0) transformation does not change
the type of PDE. Note that the discriminant involves only the coefficients of second-order
derivatives of the corresponding PDE.
Canonical forms
Let us now try to construct transformations, which will make one, or possibly two of the
coefficients of the leading second order terms of equation (7) vanish, thus reducing the equation
to a simpler form called canonical from. For convenience, we reproduce below the original PDE

A(x, y)uxx + B(x, y)uxy + C(x, y)uyy = Φ (x, y, u, ux , uy ) (2)

and the corresponding transformed PDE

a(ξ , η )wξ ξ + b(ξ , η )wξ η + c(ξ , η )wηη = φ ξ , η , w, wξ , wη

(7)

We again mention here that for the PDE (2) (or (7)) to remain a second-order PDE, the
coefficients A, B, and C (or a, b, and c) do not vanish simultaneously.
By definition, a PDE is hyperbolic if the discriminant ∆ = B2 − 4AC > 0. Since the sign of
discriminant is invariant under the change of coordinates (see equation (9)), it follows that for
a hyperbolic PDE, we should have b2 − 4ac > 0. The simplest case of satisfying this condition
is a = c = 0. So, if we try to chose the new variables ξ and η such that the coefficients a and
c vanish, we get the following canonical form of hyperbolic equation:

wξ η = ψ ξ , η , w, wξ , wη


(10a)

where ψ = φ /b. This form is called the first canonical form of the hyperbolic equation. We
also have another simple case for which b2 − 4ac > 0 condition is satisfied. This is the case
when b = 0 and c = −a. In this case (9) reduces to

wαα − wβ β = ψ α , β , w, wα , wβ


(10b)

which is the second canonical form of the hyperbolic equation, where ψ = φ /a.
By definition, a PDE is parabolic if the discriminant ∆ = B2 − 4AC = 0. It follows that for
a parabolic PDE, we should have b2 − 4ac = 0. The simplest case of satisfying this condition is
c (or a) = 0. In this case another necessary requirement b = 0 will follow automatically (since
b2 − 4ac = 0). So, if we try to chose the new variables ξ and η such that the coefficients b
and c vanish, we get the following canonical form of parabolic equation:

wξ ξ = ψ ξ , η , w, wξ , wη


(11)

where ψ = φ /a.
By definition, a PDE is elliptic if the discriminant ∆ = B2 − 4AC < 0. It follows that for
a elliptic PDE, we should have b2 − 4ac < 0. The simplest case of satisfying this condition is
b = 0 and c = a. So, if we try to chose the new variables ξ and η such that b vanishes and
c = a, we get the following canonical form of elliptic equation:

wξ ξ + wηη = ψ ξ , η , w, wξ , wη


(12)

where ψ = φ /a.
In summary, equation (7) can be reduced to a canonical form if the coordinate transformation
ξ = ξ (x, y) and η = η (x, y) can be selected such that:
 • a = c = 0 corresponds to the first canonical form of hyperbolic PDE given by

wξ η = ψ ξ , η , w, wξ , wη


(10a)

• b = 0, c = −a corresponds to the second canonical form of hyperbolic PDE given by

wαα − wβ β = ψ α , β , w, wα , wβ


(10b)

• b = c = 0 corresponds to the canonical form of parabolic PDE given by

wξ ξ = ψ ξ , η , w, wξ , wη


(11)

• b = 0, c = a corresponds to the canonical form of elliptic PDE given by

wξ ξ + wηη = ψ ξ , η , w, wξ , wη


(12)

We will now examine the kind of transformation required to reduce the PDE to its canonical
form.

Hyperbolic equations
For a hyperbolic PDE the discriminant ∆(= B2 − 4AC) > 0. In this case, we have seen that, to
reduce this PDE to canonical form we need to choose the new variables ξ and η such that the
coefficients a and c vanish in (7). Thus, from (8), we have

a = Aξx2 + Bξx ξy + Cξy2 = 0 (13a)

c= Aηx2 + Bηx ηy + Cηy2 =0 (13b)

Dividing equation (13a) and (13b) throughout by ξy2 and ηy2 respectively to obtain

ξx 2 ξx
   
A +B +C = 0 (14a)
ξy ξy
ηx ηx
 2  
A +B +C = 0 (14b)
ηy ηy

Equation (14a) is a quadratic equation for (ξx /ξy ) whose roots are given by
√
−B + B2 − 4AC
µ1 (x, y) =
√2A
−B − B2 − 4AC
µ2 (x, y) =
2A
The roots of the equation (14b) can also be found in an identical manner, so as only two distinct
roots are possible between the two equations (14a) and (14b). Hence, we may consider µ1 as
the root of (14a) and µ2 as that of (14b). That is,
√
ξx −B + B2 − 4AC
µ1 (x, y) = = (15a)
ξy 2A
√
ηx −B − B2 − 4AC
µ2 (x, y) = = (15b)
ηy 2A

The above equations lead to the following two first-order differential equations

ξx − µ1 (x, y)ξy = 0 (16a)

ηx − µ2 (x, y)ηy = 0 (16b)

These are the equations that define the new coordinate variables ξ and η that are necessary
to make a = c = 0 in (7).
Along the coordinate line ξ (x, y) = constant, we have the total derivative of ξ , d ξ = 0. It
follows that
d ξ = ξx dx + ξy dy = 0
and hence, the slope of such curves is given by

dy ξx
=−
dx ξy

We also have a similar result along coordinate line η (x, y) = constant, i.e.,
dy ηx
=−
dx ηy

Using these results, equation (14) can be written as

 2  
dy dy
A −B +C = 0 (17)
dx dx

This is called the characteristic polynomial of the PDE (2) and its roots are given by
√
dy B + B2 − 4AC
= = λ1 (x, y) (18a)
dx √ 2A
dy B − B2 − 4AC
= = λ2 (x, y) (18b)
dx 2A
The required variables ξ and η are determined by the respective solutions of the two ordinary
differential equations (18a) and (18b), known as the characteristic equations of the PDE (2).
They are ordinary differential equations for families of curves in the xy-plane along which ξ =
constant and η = constant. Clearly, these families of curves depend on the coefficients A, B,
and C in the original PDE (2).
 Integration of equation (18a) leads to the family of curvilinear coordinates ξ (x, y) = c1 while
the integration of (18b) gives another family of curvilinear coordinates η (x, y) = c2 , where c1
and c2 are arbitrary constants of integration. These two families of curvilinear coordinates
ξ (x, y) = c1 and η (x, y) = c2 are called characteristic curves of the hyperbolic equation (2) or,
more simply, the characteristics of the equation. Hence, second-order hyperbolic equations have
two families of characteristic curves. The fact that ∆ > 0 means that the characteristic are real
curves in xy-plane.
If the coefficients A, B, and C are constants, it is easy to integrate equations (18a) and
(18b) to obtain the expressions for change of variables formulas for reducing a hyperbolic PDE
to the canonical form. Thus, integration of (18) produces
√ √
B + B2 − 4AC B − B2 − 4AC
y= x + c1 and y= x + c2 (19a)
2A 2A
or √ √
B + B2 − 4AC B − B2 − 4AC
y− x = c1 and y− x = c2 (19b)
2A 2A
Thus, when the coefficients A, B, and C are constants, the two families of characteristic curves
associated with PDE reduces to two distinct families of parallel straight lines. Since the families
of curves ξ = constant and η = constant are the characteristic curves, the change of variables
are given by the following equations:
√
B + B2 − 4AC
ξ = y− x = y − λ1 x (20)
√ 2A
B − B2 − 4AC
η = y− x = y − λ2 x (21)
2A
The first canonical form of the hyperbolic is:

wξ η = ψ ξ , η , w, wξ , wη


(22)

where ψ = φ /b and b is calculated from (8)

b = 2Aξx ηx + B(ξx ηy + ξy ηx ) + 2Cξy ηy

 2
B − (B2 − 4AC)
  
B B
= 2A +B − − + 2C
4A2 2A 2A
B2 ∆
= 4C − =− (23)
A A
Each of the families ξ (x, y) = constant and η (x, y) = constant forms an envelop of the domain
of the xy-plane in which the PDE is hyperbolic.
The transformation ξ = ξ (x, y) and η = η (x, y) can be regarded as a mapping from the
xy-plane to the ξ η -plane, and the curves along which ξ and η are constant in the xy-plane
become coordinates lines in the ξ η -plane. Since these are precisely the characteristic curves,
we conclude that when a hyperbolic PDE is in canonical form, coordinate lines are characteristic
curves for the PDE. In other words, characteristic curves of a hyperbolic PDE are those curves
to which the PDE must be referred as coordinate curves in order that it take on canonical form.
We now determine the Jacobian of transformation defined by (20) and (21). We have
 −λ1 1 
 
J =   = λ2 − λ1
−λ2 1 

We know that λ1 = λ2 only if B2 − 4AC = 0. However, for an hyperbolic PDE, B2 − 4AC 6= 0.

Hence Jacobian is nonsingular for the given transformation. A consequence of λ1 6= λ2 is that
at no point can the particular curves from each family share a common tangent line.
It is easy to show that the hyperbolic PDE has a second canonical form. The following
linear change of variables
α = ξ + η, β = ξ −η
converts (22) into
wαα − wβ β = ψ α , β , w, wα , wβ


(24)
which is the second canonical form of the hyperbolic equations.
For PDE with constant coefficients, the required transformation is given by

α = ξ + η = (y − λ1 x) + (y − λ2 x)
= 2y − (λ1 + λ2 )x
β = ξ − η = (y − λ1 x) − (y − λ2 x)
= (λ2 + λ1 )x

Example 1
Show that the one-dimensional wave equation

∂ 2u 2∂ u
2
− c =0
∂ t2 ∂ x2
is hyperbolic, find an equivalent canonical form, and then obtain the general solution.
Solution To interpret the results for (2) that involve the independent variables x and y in
terms of the wave equation utt − c2 uxx = 0, where the independent variables are t and x, it will
be necessary to replace x and y in (2) and (6) by t and x. It follows that the wave equation is
a constant coefficient equation with

A = 1, B = 0, C = −c2

We calculate the discriminant, ∆ = 4c2 > 0, and therefore the PDE is hyperbolic. The roots of
the characteristic polynomial are given by
√ √
B+ ∆ B− ∆
λ1 = =c and λ2 = = −c
2A 2A
Therefore, from the characteristic equations (18a) and (18b), we have
dx dx
= c, = −c
dt dt
Integrating the above two ODEs to obtain the characteristics of the wave equation

x = ct + k1 , x = −ct + k2

where k1 and k2 are the constants of integration. We see that the two families of characteristics
for the wave equation are given by x − ct = constant and x + ct = constant. It follows, then,
that the transformation
ξ = x − ct, η = x + ct
reduces the wave equation to canonical form. We have,
∆
a = 0, c = 0, b=− = −4c2
A
So in terms of characteristic variables, the wave equation reduces to the following canonical
form
wξ η = 0
For the wave equation the characteristics are found to be straight lines with negative and
positive slopes as shown in Fig. 1. The characteristics form a natural set of coordinates for the
hyperbolic equation.
t

x + ct = a x − ct = a

b
x
a

Figure 1: The pair of characteristic curves for wave equation.

The canonical forms are simple because they can be solved directly by integrating twice.
For example, integrating with respect to ξ gives
Z
wη = 0 d ξ = h(η )

where the ‘constant of integration’ h is an arbitrary function of η . Next, integrating with

respect to η to obtain
Z
w(ξ , η ) = h(η ) d η + f (ξ ) = f (ξ ) + g(η )
where f and g are arbitrary twice differentiable functions and g is just the integral of the
arbitrary function h. The form of the general solutions of the wave equation in terms of its
original variable x and t are then given by

u = f (x − ct) + g(x + ct)

Note that f is constant on “wavefronts” x = ct + ξ that travel towards right, whereas g is

constant on wavefronts x = −ct + η that travel towards left. Thus, any general solution can
be expressed as the sum of two waves, one travelling to the right with constant velocity c and
the other travelling to the left with the same velocity c. This is one of the few cases where the
general solution of a PDE can be found.
As mentioned earlier, hyperbolic PDE has an alternate canonical form with the following
linear change of variables α = ξ + η and β = ξ − η , given by

wαα − wβ β = 0

Example 2
In steady or unsteady transonic flow around wings and airfoils with thickness to chord ratios
of a few percent, we can generally consider that the flow is predominantly directed along
the chordwise direction, taken as the x-direction. In this case, the velocities in the transverse
direction can be neglected and the potential equation reduces to the so-called small disturbance
potential equation:

∂ 2φ ∂ 2φ
1 − M∞2 + =0
∂ x2 ∂ y2
Historically, this was the form of equation used by Murman and Cole (1961) to obtain the first
numerical solution for a transonic flow around an airfoil with shocks.
Show that, depending on the Mach number, the small disturbance potential equation is
elliptic, parabolic, or hyperbolic. Find the characteristic variables for the hyperbolic case and
hence write the equation in canonical form.
Solution The given equation is of the form (2) where

A = 1 − M∞2 , B = 0, C=1

The discriminant, ∆ = B2 − 4AC = −4(1 − M∞2 ). Therefore, the PDE is hyperbolic for M > 1,
elliptic for M < 1, and parabolic for M = 1 (along the sonic line). For the case of supersonic
flow (M > 1), the roots of the characteristic polynomial are given by
√
B+ ∆
p
4(M∞2 − 1) 1
λ1 = = 2
= −p
2A 2(1 − M∞ ) M∞2 − 1
√
B− ∆
p
4(M∞2 − 1) 1
λ2 = =− 2
= p
2A 2(1 − M∞) M∞2 − 1
Therefore, from the characteristic equations (18a) and (18b), we have
dy 1 1
= p , −p
dx M∞2 − 1 M∞2 − 1
Integrating the above two ODEs to obtain the characteristics of the wave equation
x x
y= p + c1 , y = −p + c2
M∞2 − 1 M∞2 − 1
where c1 and c2 are the constants of integration.
p We see that the two families
p of characteristics
for the wave equation are given by y − x/ M∞ − 1 = constant and y + x/ M∞2 − 1 = constant.
2

It follows, then, that the transformation

x x
ξ = y− p , η = y+ p
M∞2 − 1 M∞2 − 1
reduces the wave equation to canonical form. From the relations (6), we have

φxx = wξ ξ ξx2 + 2wξ η ξx ηx + wηη ηx2 + wξ ξxx + wη ηxx

1 2 1
= 2 wξ ξ − 2 wξ η + 2 wηη
M∞ − 1 M∞ − 1 M∞ − 1
φyy = wξ ξ ξy2 + 2wξ η ξy ηy + wηη ηy2 + wξ ξyy + wη ηyy
= wξ ξ + 2wξ η + wηη

Substituting these relations in the given PDE to obtain

wξ η = 0

This is the canonical form of the given hyperbolic PDE.p Here ξ = const. and η = const. lines
represent two families of straight lines with slopes, ±1/ M∞2 − 1.

Parabolic equations
For a parabolic PDE the discriminant ∆ = B2 − 4AC = 0. In this case, we have seen that, to
reduce this PDE to canonical form we need to choose the new variables ξ and η such that the
coefficients a and b vanish in (7). Thus, from (8), we have

a = Aξx2 + Bξx ξy + Cξy2 = 0

Dividing the above equation throughout by ξy2 to obtain

ξx ξx
 2  
A +B +C = 0 (25)
ξy ξy
As the total derivative of ξ along the coordinate line ξ (x, y) = const., d ξ = 0. It follows that

d ξ = ξx dx + ξy dy = 0
and hence, the slope of such curves is given by
dy ξx
=−
dx ξy
Using this result, equation (25) can be written as
 2  
dy dy
A −B +C = 0 (26)
dx dx
This is called the characteristic polynomial of the PDE (2). Since B2 − 4AC = 0 in this case,
the characteristic polynomial (25) has only one root, given by
dy B
= = λ (x, y) (27)
dx 2A
Hence we see that for a parabolic PDE there is only one family of real characteristic curves.
The required variables ξ is determined by the ordinary differential equation (27), known as the
characteristic equations of the PDE (2). This is an ordinary differential equation for families of
curves in the xy-plane along which ξ = constant.
To determine the second transformation variable η , we set b = 0 in (8) so that
2Aξx ηx + Bξx ηy + ξy ηx ) + 2Cξy ηy =0
ξx ξx
 
2A ηx + B ηy + ηx + 2Cηy =0
ξy ξy
    
B B
2A − ηx + B − ηy + ηx + 2Cηy =0
2A 2A
B2
−Bηx − ηy + Bηx + 2Cηy =0
2A
B2 − 4AC ηy


=0
Since B2 − 4AC = 0 for a parabolic PDE, ηy could be an arbitrary function of (x, y) and con-
sequently the transformation variable η can be chosen arbitrarily, as long as the change of
coordinates formulas define a non-degenerate transformation.
If the coefficients A, B, and C are constants, it is easy to integrate equation (27) to obtain
the expressions for change of variable formulas for reducing a parabolic PDE to the canonical
form. Thus, integration of (27) produces
B
y= x + c1 (28a)
2A
or
B
y− x = c1 (28b)
2A
Since the families of curves ξ = constant are the characteristic curves, the change of variables
are given by the following equations:
B
ξ = y− x (29)
2A
η =x (30)
where we have set η = x. The Jacobian of this transformation is
 ξx ξy 
   
 −B/2A 1 
J =  =  = −1 6= 0
ηx ηy   1 0 

Now, we have from (8)

b = 2Aξx ηx + B(ξx ηy + ξy ηx ) + 2Cξy ηy

 
B
= 2A − +B+0 = 0
2A

In these new coordinate variables given by (29) and (30), equation (7) reduces to following
canonical form:
wηη = ψ ξ , η , w, wξ , wη


(31)
where ψ = φ /c. As the choice of η is arbitrary, the form taken by ψ will depend on the choice
of η . We have from (8)
c = Aηx2 + Bηx ηy + Cηy2 = A (32)
Equation (7) may also assume the form

wξ ξ = ψ ξ , η , w, wξ , wη


(33)

if we choose c = 0 instead of a = 0.

Example 3
Show that the one-dimensional heat equation

∂ 2u ∂u
α =
∂x 2 ∂t
is parabolic, choose the appropriate characteristic variables, and write the equation in equivalent
canonical form.
Solution It follows that the heat equation is a constant coefficient equation with

A = α, B = 0, C=0

We calculate the discriminant, ∆ = 0, and therefore the PDE is parabolic. The single root of
the characteristic polynomial is given by

λ = B/2A = 0

Therefore, from the characteristic equation (27), we have

dt
=0
dx
Integrating the above ODE to obtain the characteristics of the wave equation
t =k
where k is the constant of integration. Here t = k lines represents the characteristics. Since the
families of curves ξ = constant are the characteristic curves, the change of variables are given
by the following equations:
ξ = t, η =x
where we have set η = x. This shows that the given PDE is already expressed in canonical form
and thus no change of variable is needed to simplify the structure. Further, we have from (6)
ut = wξ ξt + wη ηt = wξ
and c = A = α . It follows that the canonical form of the heat equation is given by
1
wηη = w
α ξ

Elliptic equations
For an elliptic PDE the discriminant ∆ = B2 − 4AC < 0. In this case, we have seen that, to
reduce this PDE to canonical form we need to choose the new variables ξ and η to produce
b = 0 and a = c, or b = 0 and a − c = 0. Then, from (8) we obtain the following equations:
A ξx2 − ηx2 + B(ξx ξy − ηx ηy ) + C ξy2 − ηy2 = 0



(34a)
2Aξx ηx + B(ξx ηy + ξy ηx ) + 2Cξy ηy = 0 (34b)
For hyperbolic and parabolic PDEs, ξ and η are satisfied by equations that are not coupled
each other (see (13) and (25)). However, equations (34) are coupled since both unknowns ξ
and η appear in both equations. In an attempt to separate them, we add the first of these
equation to complex number i times the second to give
A (ξx + iηx )2 + B(ξx + iηx )(ξy + iηy ) + C (ξy + iηy )2 = 0
Dividing the above equation throughout by (ξy + iηy )2 to obtain
ξx + iηx 2 ξx + iηx
   
A +B +C = 0 (35)
ξy + iηy ξy + iηy
This equation can be solved for two possible values of the ratio
√ √
ξx + iηx −B ± B2 − 4AC −B ± i 4AC − B2
= = (36)
ξy + iηy 2A 2A
Clearly, these two roots are complex conjugates and are given by
√
αx −B + i 4AC − B2
= (37a)
αy 2A
√
βx −B − i 4AC − B2
= (37b)
βy 2A
where β (x, y) is the complex conjugate of α (x, y). They are given by
α (x, y) = ξ (x, y) + iη (x, y) (38a)
β (x, y) = ξ (x, y) − iη (x, y) (38b)
We will now proceed in a purely formal fashion. As the total derivative of α along the coordinate
line α (x, y) = constant, d α = 0. It follows that
d α = αx dx + αy dy = 0
and hence, the slope of such curves is given by
dy αx
=−
dx αy
We also have a similar result along coordinate line β (x, y) = constant, i.e.,
dy βx
=−
dx βy
From the foregoing discussion it follows that
√
dy B − i 4AC − B2
= λ1 = (39a)
dx √2A
dy B + i 4AC − B2
= λ2 = (39b)
dx 2A
Equations (39a) and (39b) are called the characteristic equation of the PDE (2). Clearly, the
solution of this differential equations are necessarily complex-valued and as a consequence there
are no real characteristic exist for an elliptic PDE.
The complex variables α and β are determined by the respective solutions of the two
ordinary differential equations (39a) and (39b). Integration of equation (39a) leads to the
family of curvilinear coordinates α (x, y) = c1 while the integration of (39b) gives another family
of curvilinear coordinates β (x, y) = c2 , where c1 and c2 are complex constants of integration.
Since α and β are complex function the characteristic curves of the elliptic equation (2) are
not real.
Now the real and imaginary parts of α and β give the required transformation variables ξ
and η . Thus, we have
α +β α −β
ξ = η = (40)
2 2i
With the choice of coordinate variables (40), equation (7) reduces to following canonical form:
wξ ξ + wηη = ψ ξ , η , w, wξ , wη


(41)
where ψ = φ /a.

Note: It may be noted that the quasilinear second-order equations in two independent variables
can also be classified in a similar way according to rule analogous to those developed above for
semilinear equations. However, since A, B, and C are now functions of ux , uy , and u its type
turns out to depend in general on the particular solution searched and not just on the values
of the independent variables.
Example 4
Show that the equation
uxx + x2 uyy = 0
is elliptic everywhere except on the coordinate axis x = 0, find the characteristic variables and
hence write the equation in canonical form.
Solution The given equation is of the form (2) where

A = 1, B = 0, C = x2

The discriminant, ∆ = B2 − 4AC = −4x2 < 0 for x 6= 0, and therefore the PDE is elliptic. The
roots of the characteristic polynomial are given by
√ √
B − i 4AC − B2 B + i 4AC − B2
λ1 = = −ix and λ2 = = ix
2A 2A
Therefore, from the characteristic equations (18a) and (18b), we have
dy dy
= −ix, = ix
dx dx
Integrating the above two ODEs to obtain the characteristics of the wave equation
x2 x2
y = −i + c1 , y=i + c2
2 2
where c1 and c2 are the complex constants. We see that the two families of complex charac-
teristics for the elliptic equation are given by y + ix2 /2 = constant and y − ix2 /2 = constant. It
follows, then, that the transformation
x2 x2
α = y+i , β = y−i
2 2
The real and imaginary parts of α and β give the required transformation variables ξ and η .
Thus, we have
α +β α −β x2
ξ = =y η = =
2 2i 2
With this choice of coordinate variables, equation (7) reduces to following canonical form. From
the relations (6), we have

uxx = wξ ξ ξx2 + 2wξ η ξx ηx + wηη ηx2 + wξ ξxx + wη ηxx

= x2 wηη + wη
uyy = wξ ξ ξy2 + 2wξ η ξy ηy + wηη ηy2 + wξ ξyy + wη ηyy
= wξ ξ

Substituting these relations in the given PDE and noting that x2 = 2η , we obtain
1
wξ ξ + wηη = − wη
2η
This is the canonical form of the given hyperbolic PDE. Therefore, the PDE
uxx + x2 uyy = 0
in rectangular coordinate system (x, y) has been transformed to PDE
1
wξ ξ + wηη = − wη
2η
in curvilinear coordinate system (ξ , η ). Here ξ = const. lines represents a family of straight
lines parallel to x axis and η = const. lines represents family of parabolas.

Example 5
Consider the Tricomi equation
uxx − xuyy = 0
This is simple model of a second-order PDE of mixed elliptic-hyperbolic type with two inde-
pendent variables. The Tricomi equation is a prototype of the Chaplygin’s equation for study
of transonic flow.
The Tricomi equation is of the form (2) where
A = 1, B = 0, C = −x
The discriminant, ∆ = B2 − 4AC = 4x. Therefore, the Tricomi equation is hyperbolic for x > 0,
elliptic for x < 0 and degenerates to an equation of parabolic type on the line x = 0. Assuming
x > 0, the roots of the characteristic polynomial are given by
√ √
B+ ∆ √ B− ∆ √
λ1 = = x and λ2 = =− x
2A 2A
Therefore, from the characteristic equations (18a) and (18b), we have
dy √ dy √
= x, =− x
dx dx
Integrating the above two ODEs to obtain the characteristics of the wave equation
y = 32 x3/2 + c1 , y = − 32 x3/2 + c2
where c1 and c2 are the constants of integration. We see that the two families of characteristics
for the wave equation are given by y − 2/(3x3/2 ) = constant and y + 2/(3x3/2 ) = constant. It
follows, then, that the transformation
ξ = y − 32 x3/2 , η = y + 32 x3/2
reduces the wave equation to canonical form. The derivatives of ξ and η are given by
√
ξx = − x ξy = 1
ξxx = − 2√ 1
ξyy = 0
√ x
ηx = x ηy = 1
ηxx = 2√ 1
x
ηyy = 0

17
From the relations (6), we have

uxx = wξ ξ ξx2 + 2wξ η ξx ηx + wηη ηx2 + wξ ξxx + wη ηxx

1 1
= xwξ ξ − 2xwξ η + xwηη − √ wξ + √ wη
2 x 2 x
uyy = wξ ξ ξy2 + 2wξ η ξy ηy + wηη ηy2 + wξ ξyy + wη ηyy
= wξ ξ + 2wξ η + wηη

We also have the following relation

3(η − ξ )
x3/2 =
4
Substituting these relations in the given PDE to obtain
1 wξ − wη
wξ η =
6 ξ −η

This is the canonical form of the Tricomi equation in the hyperbolic region.

Example 6
An interesting example is provided by the stationary potential flow equation in two dimensions,
defined by equation (where c designates the speed of sound):

u2 ∂ 2 φ uv ∂ 2 φ v2 ∂ 2 φ
   
1− 2 −2 2 + 1− 2 =0
c ∂ x2 c ∂ x∂ y c ∂ y2

with
u2 v2
   
uv
A = 1− 2 , B = −2 2 , C = 1− 2
c c c
we can write the potential equation under the form (2). In this
√ particular case the discriminant
2
(B − 4AC) becomes, introducing the Mach number, M(= u2 + v2 /c)
 2
u + v2

2 2


B − 4AC = 4 − 1 = 4 M − 1
c2

and hence the stationary potential equation is elliptic for subsonic flows and hyperbolic for
supersonic flows. Along the sonic line M = 1, the equation is parabolic. This mixed nature of the
potential equation has been a great challenge for the numerical computation of transonic flows
since the transition line between the subsonic and the supersonic regions is part of the solution.
An additional complication arises from the presence of shock waves which are discontinuities of
the potential derivatives and which can arise in the supersonic regions.
Classification of Second-Order Equations in n Variables
We now consider the classification to second-order PDEs in more than two independent vari-
ables. To extend the examination of characteristics for more than two independent variables is
less useful. In n dimension, we need to consider (n − 1) dimensional surfaces. In three dimen-
sions, it is necessary to obtain transformations, ξ = ξ (x, y, z), η = η (x, y, z), and ζ = ζ (x, y, z)
such that all cross derivatives in (ξ , η , ζ ) disappear. However, this approach will fail for more
than three independent variables and hence it is not usually possible to reduce the equation to
a simple canonical form. Consider a general second-order semilinear partial differential equation
in n independent variables
∂ 2u ∂ 2u ∂ 2u ∂ 2u
a11 + a12 + a13 + · · · + a1n +
∂ x1 ∂ x1 ∂ x1 ∂ x2 ∂ x1 ∂ x3 ∂ x1 ∂ xn
∂ 2u ∂ 2u ∂ 2u ∂ 2u
a21 + a22 + a23 + · · · + a2n +
∂ x2 ∂ x1 ∂ x2 ∂ x2 ∂ x2 ∂ x3 ∂ x2 ∂ xn
.. .. .. .. ..
. . . . .
∂ 2u ∂ 2u ∂ 2u ∂ 2u
an1 + an2 + an3 + · · · + ann +
∂ xn ∂ x1 ∂ xn ∂ x2 ∂ xn ∂ x3 ∂ xn ∂ xn
∂u ∂u ∂u ∂u
b1 + b2 + b3 + ··· + bn + cu + d = 0
∂ x1 ∂ x2 ∂ x3 ∂ xn
For more than three independent variables it is convenient to write the above PDE in the
following form:
n n
∂ 2u n
∂u
∑ ∑ i j ∂ xi ∂ x j ∑ bi ∂ xi + cu + d = 0
a + (42)
i=1 j=1 i=1

where the coefficients ai j , bi , c, d are functions of x = (x1 , x2 , . . . , xn ), u = u(x1 , x2 , . . ., xn ), and

n is the number of independent variables. Equation (42) can be written in matrix form as
  ∂u   ∂u 
i a11 · · · a1n ∂ x1
  ∂ .x1 
∂ ∂  .. . . .   ..  
h
∂ x1 · · · ∂ xn  . .
. .   .  + b1 · · · bn  ..  + cu + d = 0
an1 · · · ann ∂u ∂u
∂x n ∂x n

We assume that the coefficient matrix A = (ai j ) to be symmetric. If A is not symmetric, we

can always find a symmetric matrix ai j = 12 (ai j + a ji ) such that (42) can be rewritten as
n n
∂ 2u n
∂u
∑ ∑ ai j + ∑ bi
∂ xi ∂ x j i=1 ∂ xi
+ cu + d = 0
i=1 j=1

For example, consider the equation

∂ 2u ∂ 2u ∂ 2u
− + = f (x1 , x2 )
∂ x21 ∂ x1 ∂ x2 ∂ x22
Since
∂ 2u ∂ 2u
=
∂ x1 ∂ x2 ∂ x2 ∂ x1
we may write
∂ 2u 1 ∂ 2u 1 ∂ 2u ∂ 2u
− − + = f (x1 , x2 )
∂ x21 2 ∂ x1 ∂ x2 2 ∂ x2 ∂ x1 ∂ x22
or in matrix form
" ∂u
#
i 1 −1/2
∂ ∂
h
∂ x1
∂ x1 ∂ x2 ∂u = f (x1 , x2 )
−1/2 1 ∂ x2
Comparing to the general equation in matrix form above, we can see that the coefficient matrix
A is now symmetric.
Now consider the transformation
ξ = Qx
where ξ = (ξ1 , ξ2 , . . . , ξn ) and Q = (qi j ) is an n × n arbitrary matrix. Using index notation, this
transformation can be written as
n
ξi = ∑ qi j x j
j=1
Repeated application of chain rule in the forms
∂ n
∂ ∂ ξk
∂ xi
= ∑ ∂ ξk ∂ xi
k=1
and
∂2 n
∂ 2 ∂ ξk ∂ ξl
∂ xi ∂ x j
= ∑
k,l=1 ∂ ξk ∂ ξl ∂ xi ∂ x j
to the derivatives of u(x1 , x2 , . . ., xn ) in (42) with respect to x1 , x2 , . . . , xn transforms them into
derivatives of w(ξ1 , ξ2 , . . . , ξn ) with respect to ξ1 , ξ2 , . . . , ξn . This allows equation (42) to be
expressed as !
n n
∂ 2u
∑ ∑ ki i j l j ∂ ξk ∂ ξl + lower-order terms = 0
q a q (43)
k,l=1 i, j=1

The coefficient matrix of the terms ∂ 2 u/(∂ ξk ∂ ξl ) in this transformed expression is seen to be
equal to QT AQ. That is,
(qki ai j ql j ) ≡ QT AQ
From linear algebra, we know that for any real symmetric matrix A, there is an associate
orthogonal matrix P such that PT AP = Λ. Here P is called diagonalizing matrix of A and Λ is
a diagonal matrix whose element are the eigenvalues, λi , of A and the columns of P the linearly
independent eigenvectors of A, ei = (e1i , e2i , . . ., eni ). So, we have
P = (ei j ) and Λ = (λi δi j ), i, j = 1, 2, . . ., n
where δi j is the Kronecker delta. Now if the transformation is such that Q is taken to be a
diagonalizing matrix of A, it follows that
λ1
 
 λ2 
Q AQ = Λ = 
T
(44)
 
. . .

 
λn
The numbers λ1 , λ2 , . . . , λn are real, because the eigenvalues of a real symmetric matrix are
always real. It is instructive to note that the previously mentioned transformation (for second-
order PDE with two independent variables) to remove cross derivatives is equivalent to finding
eigenvalues λi of the coefficient matrix A.
We are now in a position to classify the equation (42).
• Equation is called elliptic if all eigenvalues λi of A are non-zero and have the same sign.
• Equation is called hyperbolic if all eigenvalues λi of A are non-zero and have the same
sign except for one of the eigenvalues.
• Equation is called parabolic if any of the eigenvalues λi of A is zero. This means that the
coefficient matrix A is singular.
When more than two independent variables are involved, there are other intermediate classifi-
cations exist which depends on the number of zero eigenvalues and the pattern of signs of the
non-zero eigenvalues. These sub classification has not much practical importance and will not
be discussed here.

Example 7
Classify the three-dimensional Laplace equation
uxx + uyy + uzz = 0

Solution The coefficient matrix is given by

 
1 0 0
A= 0 1 0 
0 0 1
As the coefficient matrix is already in diagonalized form it can be seen immediately that it has
three non-zero eigenvalues which are all positive. Hence, according to the classification rule the
given PDE is elliptic.

Example 8
Classify the two-dimensional wave equation
utt − c2 (uxx + uyy ) = 0

Solution The coefficient matrix is given by

 
1 0 0
A =  0 −c2 0 
0 0 −c2
As the coefficient matrix is already in diagonalized form it can be seen immediately that it
has three non-zero eigenvalues which are all negative except one. Hence, according to the
classification rule the given PDE is hyperbolic.
Example 9
Classify the two-dimensional heat equation

ut − α (uxx + uyy ) = 0

Solution The coefficient matrix is given by

 
0 0 0
A =  0 −α 0 
0 0 −α

As the coefficient matrix is already in diagonalized form it can be seen immediately that it has
a zero eigenvalue. Hence, according to the classification rule the given PDE is parabolic.

Example 10
Classify the two-dimensional equation

∂ 2u ∂ 2u ∂ 2u
− + = f (x, y)
∂ x2 ∂ x∂ y ∂ y2

Solution First of we write the given equation in the following form:

∂ 2u 1 ∂ 2u 1 ∂ 2u ∂ 2u
− − + = f (x, y)
∂ x2 2 ∂ x∂ y 2 ∂ y∂ x ∂ y2
The coefficient matrix is then given by

1 − 21
 
A=
− 12 1

Since the coefficient matrix not in diagonalized form, we solve the eigenvalue problem det(A −
λ I) = 0. That is,
 1 − λ −1 
 
2 
 −1 1 − λ  = 0

2
Expanding the determinant to obtain
1 3
(1 − λ )2 − =0 =⇒ λ 2 − 2λ + =0
4 4
Hence the two eigenvalues are λ1 = 1/2 and λ2 = 3/2. The equation is elliptic.
Example 11
Classify the following small disturbance potential equation for compressible flows:

∂ 2φ ∂ 2φ
1 − M∞2 + =0
∂ x2 ∂ y2

Solution The coefficient matrix is then given by

1 − M∞2 0
 
A=
0 1

As the coefficient matrix is already in diagonalized form it can be seen immediately that its
eigenvalues are given by
λ1 = 1, λ2 = 1 − M∞2
It follows that: If M∞2 < 1 then all eigenvalues are nonzero and of same sign thus small distur-
bance potential equation is elliptic, if M∞2 = 1 then one of the eigenvalues is zero thus small
disturbance potential equation is parabolic, and if M∞2 > 1 then all eigenvalues are nonzero and
are of opposite sign thus small disturbance potential equation is hyperbolic.

Classification of First-Order System of Equations

Consider the semilinear1 first-order system of two equations in two dependent variables (u, v)
and three independent variables (x, y, z) (corresponding to a thee-dimensional space) as given
below:
∂u ∂v ∂u ∂v ∂u ∂v
a11 + a12 + b11 + b12 + c11 + c12 = f1
∂x ∂x ∂y ∂y ∂z ∂z
∂u ∂v ∂u ∂v ∂u ∂v
a21 + a22 + b21 + b22 + c21 + c22 = f2
∂x ∂x ∂y ∂y ∂z ∂z
In matrix form, we have
∂u
 
∂x
 ∂v 
 ∂x 
  ∂u   
a11 a12 b11 b12 c11 c12  ∂y  = f1

∂v

a21 a22 b21 b22 c21 c22 
 ∂y

 f2
 ∂u 
 ∂z 
∂v
∂z

Alternatively, the above system can be rewritten as

a11 a12 ∂ b11 b12 ∂ c11 c12 ∂ u

             
u u f1
+ + =
a21 a22 ∂ x v b21 b22 ∂ y v c21 c22 ∂ z v f2
1 For a first-order PDE to be semilinear, ai j = ai j (x, y, z), bi j = bi j (x, y, z), ci j = ci j (x, y, z) and fi = fi (x, y, z, u, v).
or
∂U ∂U ∂U
A(x, y, z) + B(x, y, z) + C(x, y, z) = F(x, y, z,U )
∂x ∂y ∂z
where    
u f1
U = F =
v f2
Using indicial notations, the system of first-order PDE for two dependent variables and three
independent variables can be written as
2  ∂uj ∂uj ∂uj

∑ ai j ∂ x + bi j ∂ y + ci j ∂ z = fi, i = 1, 2
j=1

The system of first-order PDE of three independent variables can be generalized for n dependent
variables, u j , as follows:
n  ∂uj ∂uj ∂uj

∑ ai j ∂ x + bi j ∂ y + ci j ∂ z = fi, i = 1, 2, . . ., n (45a)
j=1

or in matrix form
∂U ∂U ∂U
A +B +C =F (45b)
∂x ∂y ∂z
where
     
a11 · · · a1n b11 · · · b1n c11 · · · c1n
A =  ... . . . ...  B =  ... . . . ...  C =  ... . . . ... 
     

an1 · · · ann bn1 · · · bnn cn1 · · · cnn

   
u1 f1
 u2   f2 
U = .  F = .
   
.
 .   ..


un fn
Finally, we consider the most general form of system of first-order PDEs. Suppose we have
n dependent variables u j , in an m-dimensional space xk (i.e., m independent variables), we can
group all the variables u j in an (n × 1) column vector U and write the system of first-order
PDEs
m
∂U
∑ Ak ∂ xk = F (46a)
k=1
or
m n ∂uj
∑ ∑ akij ∂ xk = fi, i = 1, 2, . . ., n (46b)
k=1 j=1
As an example let us consider the system n equations (and an equal number of dependent
variables) in two dimensions (x, y):
         
a11 · · · a1n u1 b11 · · · b1n u1 f1
 .. . . .  ∂  .   .. . . .  ∂  ..  =  .. 
 . . ..   ..  +  . . ..   .   . 
∂x ∂y
an1 · · · ann un bn1 · · · bnn un fn
or
∂U ∂U
A(x, y) + B(x, y) = F(x, y,U ) (47)
∂x ∂y
Assuming that A is nonsingular, the system (47) can be written in a more convenient form by
pre-multiplying by A−1 :
∂U ∂U
+ D(x, y) = E(x, y,U ) (48)
∂x ∂y
or          
1 ··· 0 u1 d11 · · · d1n u1 e1
 .. . . ..  ∂  ..   .. . . . ∂  .   .. 
. .   . + . . ..   .. = . 

 .
∂x ∂y
0 ··· 1 un dn1 · · · dnn un en
where
D = A−1 B and E = A−1 F
Using indicial notation, the system (48) can be written as

∂uj n ∂uj
+ ∑ di j = ei , i = 1, 2, . . ., n
∂ x j=1 ∂y

and in component form, the system becomes

∂ u1 ∂ u1 ∂ u2 ∂ un
+ d11 + d12 + · · · + d1n = e1
∂x ∂y ∂y ∂y
∂ u2 ∂ u1 ∂ u2 ∂ un
+ d21 + d22 + · · · + d2n = e2
∂x ∂y ∂y ∂y
.. .. .. .. .. ..
. . . . . .
∂ un ∂ u1 ∂ u2 ∂ un
+ dn1 + dn2 + · · · + dnn = en
∂x ∂y ∂y ∂y
Just as in the case of a single partial differential equation, the important properties of
solutions of the system (48) depend only on its principal part Ux + DUy . Since this principal
part is completely determined by the coefficient matrix D(x, y) = A−1 B, this matrix plays a
fundamental role in the study of (47).

Canonical form and classification

Let us consider the transformation

V = P−1U ⇒ U = PV

where P = (pi j ) is an n × n arbitrary nonsingular matrix. Then

∂U ∂V ∂P ∂U ∂V ∂P
=P + V, =P + V
∂x ∂x ∂x ∂y ∂y ∂y
Substituting these into (48) to obtain
∂V ∂P ∂V ∂P
P + V + DP +D V = E
∂x ∂x ∂y ∂y
Rearranging,
∂V ∂V ∂P ∂P
 
P + DP =E− +D V =G
∂x ∂y ∂x ∂y
We multiply the above equation by P−1 to obtain
∂V ∂V
+ P−1 DP =H (49)
∂x ∂y
where
∂P ∂P
   
−1 −1
H =P G=P E− +D V .
∂x ∂y
Let Λ be the n × n diagonal matrix with diagonal entries the eigenvalues of D, and P is taken
to be the n × n diagonalizing matrix of D with columns the corresponding eigenvectors of D,
i.e.,
λ1 · · · 0
   
p11 · · · p1n
Λ = (λi δi j ) =  ... . . . ...  , P = (pi j ) =  ... . . . ... 
   

0 · · · λn pn1 · · · pnn
where δi j is the Kronecker delta. It follows that

P−1 DP = Λ. (50)

It is instructive to note that the previously mentioned transformations (for second-order PDE
with two independent variables) to remove cross derivatives is equivalent to diagonalizing the
matrix D = A−1 B and then finding eigenvalues λi of the matrix D of the system of PDEs
(47). Thus, we can write the system of equations (49) [and consequently (47)] in the following
canonical form:
∂V ∂V
+ Λ(x, y) = H(x, y,V ) (51a)
∂x ∂y
The simplicity of the canonical form (51a) becomes apparent if we write it in component form,
∂ vi ∂ vi
+ λi (x, y) = hi (x, y, v1 , . . . , vn ), i = 1, 2, . . ., n (51b)
∂x ∂y
or
∂ v1 ∂ v1
+ λ1 = h1
∂x ∂y
∂ v2 ∂ v2
+ λ2 = h2
∂x ∂y
..
. (51c)
∂ vn ∂ vn
+ λn = hn
∂x ∂y
It is clear that the principal part of the ith equation involves only the ith unknown vi .
The classification of the system of first-order PDEs (47) is done based on the nature of the
eigenvalues λi of the matrix P−1 DP, which are exactly the eigenvalues values of D = A−1 B.
Recall that an eigenvalue of D is a root λ of the characteristic equation

|D − λ I| = 0.

The system (47) based on the nature of its eigenvalues is classified as follows:

• If all the n eigenvalues of D are real and distinct the system is called hyperbolic type.
• If all the n eigenvalues of D are complex the system is called elliptic type.
• If some of the n eigenvalues are real and other complex the system is considered as hybrid
of elliptic-hyperbolic type.
• If the rank of matrix D is less than n, i.e., there are less than n real eigenvalues (some of
the eigenvalues are repeated) then the system is said to be parabolic type.

For a system of PDE having only two dependent variables, we can determine the eigenvalues
of matrix D analytically and state the conditions classification in an explicit manner as follows.
Consider the system of two equations (with dependent variables u1 and u2 ) in two dimensions
(x, y):

∂ u1 ∂ u2 ∂ u1 ∂ u2
a11 + a12 + b11 + b12 = f1 (52a)
∂x ∂x ∂y ∂y
∂ u1 ∂ u2 ∂ u1 ∂ u2
a21 + a22 + b21 + b22 = f2 (52b)
∂x ∂x ∂y ∂y
In matrix form,

∂ ∂
         
a11 a12 u1 b11 b12 u1 f1
+ =
a21 a22 ∂x u2 b21 b22 ∂y u2 f2
or
∂ ∂
   
u1 u1
A +B =F
∂x u2 ∂y u2
where      
a11 a12 b11 b12 f1
A= , B= , F =
a21 a22 b21 b22 f2
The inverse of A is  
−1 1 a22 −a12
A =
|A| −a21 a11
where |A| is the determinant of matrix A. We now compute the matrix D as
    
−1 1 a22 −a12 b11 b12 1 a22 b11 − a12 b21 a22 b12 − a12 b22
D=A B= =
|A| −a21 a11 b21 b22 |A| a11 b21 − a21 b11 a11 b22 − a21 b12
so that the system (52) may be written as

∂ u1 1 a22 b11 − a12 b21 a22 b12 − a12 b22 ∂

       
u1 e1
+ =
∂ x u2 |A| a11 b21 − a21 b11 a11 b22 − a21 b12 ∂ y u2 e2

To determine eigenvalues of D, we solve the following eigenvalue problem:

|D − λ I| = 0

Expanding the above determinant to obtain the characteristic equation

|A|λ 2 − |b|λ + |B| = 0 (53)

where the determinants |A|, |B|, and |b| are given by

 
 a a12 
|A| =  11 = a11 a22 − a12 a21
a21 a22 
 
 b b12 
|B| =  11 = b11 b22 − b12 b21
b21 b22 
   
 a b12   b11 a12 
|b| =  11 + = a11 b22 − a21 b12 + a22 b11 − a12 b21
a21 b22   b21 a22 

The two roots of the quadratic equation for λ are given by

p
|b| ± |b|2 − 4|A||B|
λ1,2 = (54)
2|A|
Notice that this expression has the same form as equation (17) except that a, b, and c have now
become determinants. Clearly, the nature of the eigenvalues depends on the sign of discriminant
|b|2 − 4|A||B|. The different possibilities are given below:

• If |b|2 − 4|A||B| > 0, there exists two real and distinct eigenvalues and thus the system
is hyperbolic.
• If |b|2 − 4|A||B| < 0, there exists two complex eigenvalues and thus the system is elliptic.
• If |b|2 − 4|A||B| = 0, there is only one real eigenvalue and thus the system is parabolic.

We mention here that classification of second-order system of equations in general is very

complex. It is difficult to determine the mathematical character of these systems except for
simple cases.

A special case

When A is an identity matrix, the system of equation (47) takes the form
∂U ∂U
+ B(x, y) = F(x, y,U ) (55)
∂x ∂y
so that, we have
D=B and E =F
For a system of PDE having only two dependent variables, (55) becomes
∂ u1 ∂ u1 ∂ u2
+ b11 + b12 = f1 (56a)
∂x ∂y ∂y
∂ u2 ∂ u1 ∂ u2
+ b21 + b22 = f2 (56b)
∂x ∂y ∂y
or, in matrix form
∂ ∂
     
u1 b11 b12 u1
+ =F
∂x u2 b21 b22 ∂y u2
To determine eigenvalues of D, we solve the eigenvalue problem |B − λ I| = 0. The corre-
sponding characteristic equation is
λ 2 − |b|λ + |B| = 0
where the determinants |B| and |b| are given by
 
 b11 b12 
|B| =   = b11 b22 − b12 b21
b21 b22 
   
 b11 0   1 b12 
|b| = 
  +   = b11 + b22
b21 1   0 b22 
The two roots of the quadratic equation for λ are given by
p
|b| ± |b|2 − 4|B|
λ1,2 = (57)
2

Example 12
Classify the single first-order equation
∂u ∂u
a +b = f
∂x ∂y
where a and b are real constants.
Solution In the standard matrix form the above equation may be written as
∂U ∂U
A +B =F
∂x ∂y
where        
A= a , B= b , U = u , F = f
The D matrix, in this case, can be easily found:
D = A−1 B = a−1
    
b = b/a
The matrix D has the eigenvalue, λ = b/a. This is always real and hence, a single first-order
PDE is always hyperbolic in the space (x, y). Note that here we have only a single eigenvalue
and thus a characteristic direction.
Example 13
Classify the following system of first-order equation:

∂φ ∂ψ
a +c = f1
∂x ∂y
∂ψ ∂φ
b +d = f2
∂x ∂y
The matrix form of the system is:

a 0 ∂ φ 0 c ∂ φ
         
f1
+ =
0 b ∂x ψ d 0 ∂y ψ f2

This equation may be written as

∂U ∂U
A +B =F
∂x ∂y
where
φ
       
a 0 0 c f1
A= , B= , U = , F =
0 b d 0 ψ f2
Therefore, we have

a11 = a a12 = 0 a21 = 0 a22 = b

b11 = 0 b12 = c b21 = d b22 = 0

The relevant determinants can be now evaluated as

|A| = a11 a22 − a12 a21 = ab

|B| = b11 b22 − b12 b21 = −cd
|b| = a11 b22 − a21 b12 + a22 b11 − a12 b21 = 0

and the D matrix is given by

     
−1 1 a22 b11 − a12 b21 a22 b12 − a12 b22 1 0 bc 0 c/a
D=A B= = =
|A| a11 b21 − a21 b11 a11 b22 − a21 b12 ab ad 0 d/b 0

so that the system (52) may be written as

∂ φ 0 c/a ∂ φ
       
e1
+ =
∂x ψ d/b 0 ∂y ψ e2

The eigenvalues of D = A−1 B are given by

p r
|b| ± |b|2 − 4|A||B| cd
λ1,2 = =±
2|A| ab
 If cd/ab > 0 then the eigenvalues are real and distinct and the system is hyperbolic in the
space (x, y). For instance, a = b = 1; c = d = 1 with vanishing right-hand side, the system of
equation becomes

∂φ ∂ψ
+ =0
∂x ∂y
∂ψ ∂φ
+ =0
∂x ∂y
By eliminating the variable ψ and replacing y by t, we obtain the well-known wave equation in
φ:
∂ 2φ ∂ 2φ
=
∂ t2 ∂ x2
which is a hyperbolic equation as seen previously.
If cd/ab < 0 then the eigenvalues are complex and the system is elliptic in the space (x, y).
For instance, a = b = 1; c = −d = −1 and vanishing right-hand side, the system of equation
becomes the well-known Cauchy–Riemann equation:

∂φ ∂ψ
− =0
∂x ∂y
∂ψ ∂φ
+ =0
∂x ∂y
By eliminating the variable ψ , we obtain the Laplace equation in φ :

∂ 2φ ∂ 2φ
+ 2 =0
∂ x2 ∂y
which is the standard form of elliptic equations and describes steady-state diffusion phenomena.
Note that we could also obtain the Laplace equation in ψ by eliminating the variable φ .
Finally, if one of the coefficients is equal zero, say c, then there is only one real eigenvalue
and the system is parabolic. For instance, with a = −b = 1, c = 0, d = 1 and f1 = ψ , f2 = 0,
the system of equation becomes

∂φ
=ψ
∂x
∂φ ∂ψ
− =0
∂y ∂x
which on eliminating the variable φ and replacing y by t leads to the standard form for a
parabolic equation:
∂φ ∂ 2φ
=
∂t ∂ x2
This is recognizable by the fact that the equation presents a combination of first and second-
order derivatives.
Example 14
Let us find out the canonical form the system of first-order equations
∂ u1 ∂ u2
+c =0
∂x ∂y
(58)
∂ u2 ∂ u1
+d =0
∂x ∂y

Solution The given system is of the form (55) and can be written as

∂ u1 0 c ∂
       
u1 0
+ =
∂ x u2 d 0 ∂ y u2 0
or
∂U ∂U
+B =F
∂x ∂y
where,      
0 c u1 0
B= , U = , F =
d 0 u2 0
Therefore, we have

b11 = 0 b12 = c b21 = d b22 = 0

The relevant determinants can be now evaluated:

|B| = b11 b22 − b12 b21 = −cd

|b| = b11 + b22 = 0

The eigenvalues of B are given by

√
p
|b| ± |b|2 − 4|B|
λ1,2 = = ± cd
2
√
The eigenvector for λ1 = cd can be found as follows
 √ 
√
      
0 c p11 p11 p11 c
= cd =⇒ = √
d 0 p12 p12 p12 d
√
Similarly, the eigenvector for λ2 = − cd is
 √ 
√
      
0 c p21 p21 p21 √c
= − cd =⇒ =
d 0 p22 p22 p22 − d

The P matrix can now be constructed using eigenvectors of B, that is,

 √ √ 
√ c √c
P=
d − d
The inverse of P is
   √ √   √ √ 
1 p −p −1 − d − c 1 1/ c 1/
P−1 = 22 12
= √ √ √ = √ √d
|P| −p21 p11 2 cd − d c 2 1/ c −1/ d

Therefore,  √ 
cd 0
P −1
BP = Λ = √
0 − cd
The canonical form is then given by
∂V ∂V
+Λ =0
∂x ∂y
or  √
∂ ∂
      
v1 cd 0
√ v1 0
+ =
∂x v2 0 − cd ∂y v2 0
or in component form
∂ v1 √ ∂ v1
+ cd =0
∂x ∂y
(59)
∂ v2 √ ∂ v2
− cd =0
∂x ∂y
The canonical form (59) is particularly simple. Each equation involves only one unknown and
can be easily solved by the methods of characteristics. The general solution of the system (59)
is √
v1 = f (x − cdy)
√ (60)
v2 = g(x + cdy)
where f and g are arbitrary functions of a single variable.
The vectors U and V are related through the following transformation:
   √ √    √ 
u1 c c v1 c(v1 + v2 )
U = PV ⇒ = √ √ = √ (61)
u2 d − d v2 d(v1 − v2 )

Now the general solution of (58) can be obtained using the relation (61) as
√ √ √
u1 = c[ f (x − cdy) + g(x + cdy)]
√ √ √ (62)
u2 = d[ f (x − cdy) − g(x + cdy)]

Example 15
The one-dimensional form of the time-dependent shallow water equations can be written as
∂h ∂h ∂u
+u +h =0
∂t ∂x ∂x
∂u ∂u ∂h
+u +g =0
∂t ∂x ∂x
where h represents the water height, g is the gravity acceleration and u the horizontal velocity.
Solution Since A is a unit matrix, the given system in matrix form
∂ h u h ∂
     
h
+ = 0
∂t u g u ∂x u
Introducing the vector  
h
U =
u
the system is written in the condensed form:
∂U ∂U
+B =0
∂t ∂x
where,      
u h h 0
B= , U = , F =
g u u 0
Therefore, we have

b11 = u b12 = h b21 = g b22 = u

The relevant determinants can be now evaluated:

|B| = b11 b22 − b12 b21 = u2 − gh

|b| = a11 b22 − a21 b12 + a22 b11 − a12 b21 = u + u = 2u

The eigenvalues of B are given by

p p
|b| ± |b|2 − 4|B| 2u ± 4u2 − 4(u2 − gh)
λ1,2 =
p
= = u ± gh
2 2
√
The eigenvector for λ1 = u + gh can be found as follows
        √ 
u h e11 p e11 e11 h
= (u + gh) =⇒ = √
g u e12 e12 e12 g
√
Similarly, the eigenvector for λ2 = u − gh is
        √ 
u h e21 p e21 e21 − h
= (u − gh) =⇒ = √
g u e22 e22 e22 g
The P matrix can now be constructed using eigenvectors of B, that is,
 √ √ 
h − h
P= √ √
g g

Now, the diagonal matrix Λ by definition is P−1 BP. As we have already found the eigenvalues
of B, the matrix Λ can be directly obtained as
 √ 
u + gh 0
Λ= √
0 u − gh
The canonical form is then given by

∂V ∂V
+Λ =0
∂t ∂x
or √
∂ ∂ v1
       
v1 u + gh 0 0
+ √ =
∂t v2 0 u − gh ∂ x v2 0
or in component form
∂ v1 √ ∂ v1
+ (u + gh) =0
∂t ∂x
∂ v2 √ ∂ v2
+ (u − gh) =0
∂t ∂x
Each equation involves only one unknown and can be easily solved by the methods of charac-
teristics. The general solution of the system is
√
v1 = f1 [x − (u + gh)t]
√ (63)
v2 = f2 [x − (u − gh)t]

where f1 and f2 are arbitrary functions of a single variable.

The vectors U and V are related through the following transformation:
   √ √    √ 
h h − h v1 h(v1 − v2 )
U = PV ⇒ = √ √ = √
u g g v2 g(v1 + v2 )

Now the general solution can be obtained using the relation (63) as
√ √ √
h = h[ f1 (x − (u + gh)t) − f2 (x − (u − gh)t)]
√ √ √ (64)
u = g[ f1 (x − (u + gh)t) + f2 (x − (u − gh)t)]

The procedure for classification of semilinear PDEs is equally applicable for quasilinear PDEs.
We illustrate this with an example from fluid dynamics.

Example 16
Classify the Euler equations for unsteady, one-dimensional flow:

∂ρ ∂ρ ∂u
+u +ρ =0
∂t ∂x ∂x
∂u ∂u 1 ∂p c2 ∂ ρ
+u =− =−
∂t ∂x ρ ∂x ρ ∂x
where the speed of sound c is given by the isentropic relation between pressure and density as

∂p
 
2
c =
∂ρ s

35
Solution In matrix form the Euler equation can be written as

1 0 ∂ ρ ρ ∂ ρ
       
u
+ = 0
0 1 ∂t u c2 /ρ u ∂ x u

Introducing the vector

ρ
 
U =
u
the system can be written in the condensed form:

∂U ∂U
A +B =0
∂t ∂x
where,

ρ ρ
       
1 0 u 0
A= , B= , U = , F =
0 1 c /ρ u
2 u 0

Therefore, we have

a11 = 1 a12 = 0 a21 = 0 a22 = 1

b11 = u b12 = ρ b21 = c /ρ2
b22 = u

The relevant determinants can be now evaluated:

|A| = a11 a22 − a12 a21 = 1

|B| = b11 b22 − b12 b21 = u2 − c2
|b| = a11 b22 − a21 b12 + a22 b11 − a12 b21 = u + u = 2u

Since A is a unit matrix, the inverse of A is A itself. We now compute the matrix D:

ρ
 
−1 u
D=A B=B=
c2 /ρ u

The system of Euler equation can now be written as

∂U ∂U
+D =0
∂t ∂x
The eigenvalues of D = A−1 B are given by
p p
|b| ± |b|2 − 4|A||B| 2u ± 4u2 − 4(u2 − c2 )
λ1,2 = = = u±c
2|A| 2

Therefore, the two characteristics of this hyperbolic system are given by

dx
= u±c
dt
The eigenvector for λ1 = u + c can be found as follows

ρ ρ
        
u e11 e11 e11
= (u + c) =⇒ =
c2 /ρ u e12 e12 e12 c

Similarly, the eigenvector for λ2 = u − c is

ρ −ρ
        
u e21 e21 e21
= (u − c) =⇒ =
c2 /ρ u e22 e22 e22 c

The P matrix can now be constructed using eigenvectors of D, that is,

ρ −ρ
 
P=
c c

Therefore,  
u+c 0
P DP = Λ =
−1
0 u−c
The canonical form is then given by

∂V ∂V
+Λ =0
∂t ∂x
or
∂ ∂
       
v1 u+c 0 v1 0
+ =
∂t v2 0 u−c ∂x v2 0
or in component form
∂ v1 ∂ v1
+ (u + c) =0
∂t ∂x
∂ v2 ∂ v2
+ (u − c) =0
∂t ∂x
Each equation involves only one unknown and can be easily solved by the standard methods of
characteristics. The general solution of the system is

v1 = f1 [x − (u + c)t]
(65)
v2 = f2 [x − (u − c)t]

where f1 and f2 are arbitrary functions of a single variable.

The vectors U and V are related through the following transformation:

ρ −ρ ρ (v1 − v2 )
      
h v1
U = PV ⇒ = =
u c c v2 c(v1 + v2 )

Now the general solution can be obtained using the relation (65) as

ρ = ρ [ f1 (x − (u + c)t) − f2 (x − (u − c)t)]
(66)
u = c[ f1 (x − (u + c)t) + f2 (x − (u − c)t)]
Note: Since both the eigenvalues are real, for all values of the velocity u, the system is always
hyperbolic in space and time. This is an extremely important property that the steady isentropic
Euler equations are elliptic in the space (x, y) for subsonic velocities and hyperbolic in the space
(x, y) for supersonic velocities.
Here, in space and time, the inviscid isentropic equations are always hyperbolic independently
of the subsonic or supersonic state of the flow. As a consequence, the same numerical algorithms
can be applied for all flow velocities. On the other hand, dealing with the steady state equations,
the numerical algorithms will have to adapt to the flow regime, as the mathematical nature of
the system of equations is changing when passing from subsonic to supersonic, or inversely.

38