5.

PARTIAL DIFFERENTIAL EQUATIONS SPRING 2014

5.1 Introduction
5.2 Characteristics
5.2.1 First-order equations
5.2.2 Second-order wave equation
5.3 Classification of second-order PDEs
5.3.1 Hyperbolic, parabolic and elliptic PDEs
5.3.2 Domain of dependence and boundary conditions

5.1 Introduction

Two numerical strategies have been considered for ordinary differential equations (ODEs):
 forward marching – solution advanced forwards step-by-step from initial conditions;
 all points simultaneously – derivatives are approximated by finite differences, leading
to a set of simultaneous algebraic equations.
For any given ODE, the solution method is determined by the boundary conditions.

It will be shown that essentially the same two types of solution strategy – forward-marching
or all-points-simultaneously – may be employed for partial differential equations (PDEs).
However, for PDEs, the equation itself determines both the required boundary conditions and
the type of solution method.

We will see that certain second-order PDEs (which are very common in science and
engineering) can be categorised as hyperbolic, parabolic or elliptic, and that these have very
different types of solution and boundary conditions.

Hyperbolic and parabolic equations are solved by forward-marching – advancing in the

direction of one independent variable. In practice, this is often the time t, and solution
techniques are then referred to as time-marching methods.

Elliptic equations arise where there is no natural preferred direction of propagation. A

common example is steady-state diffusion and the independent variables are the space
coordinates x and y. In these cases, finite-difference methods are used to solve for all points
simultaneously.

5.2 Characteristics

PDEs are more complex than ODEs because they contain derivatives with respect to more
than one variable. Wouldn’t it be nice if some manipulation could reduce them to ordinary
differential equations? This can be done for some equations: they have characteristics and
are called hyperbolic.

Loose definition: A characteristic is a curve along which a partial differential equation

reduces to an ordinary differential equation.

Computational Mechanics 5-1 David Apsley

5.2.1 First-Order Equations

Example.
First-order wave equation with constant wave speed c.
u u
c  0, u  u 0 ( x) on t  0 (1)
t x

Suppose that one follows some particular path x(t). Then, along that line, u = u(x(t),t) is a
function of t only and its total derivative with respect to t is
du u u dx
  (2)
dt t x dt

Compare this with the original differential equation (1). It is exactly the same as the LHS of
that equation provided that the chosen path satisfies
dx
t
c
dt
or x0 +ct
x  ct  const .

nt
sta
on
=c

Along these lines,
du
 0 (an ordinary differential equation) x
dt x0
Hence, u propagates unchanged along the lines ξ  x  ct  constant . Physically, a wave
pulse propagates unchanged with wave speed c.

(Although we have written a total derivative du/dt for that particular path, u is actually a
function of both t and the particular characteristic ξ. Here, ξ = x0, the x value at t = 0.)

Example.
First-order wave equation with amplitude-dependent wave speed.
u u
u  0, u  u 0 ( x) on t  0
t x

Along any line x(t), the total derivative of u is t

du u u dx
  x 0+ut
dt t x dt characteristics

Comparing this with the original equation, choose

lines such that dx/dt = u. Then
du dx
0 along u. x
dt dt x0
i.e. u propagates unchanged along lines x  x0  ut .

Note that larger-amplitude waves travel faster; in general such waves will steepen (and
eventually break).

Computational Mechanics 5-2 David Apsley

Example.
u u
 3x 2  x y, u ( x,0)  x 2
x y
Find u(3,19).

Solution
y
We require a path y(x) which will take us from the x-axis
where the boundary conditions are known to the point (3,19).
(3,19)
Along any curve y(x), the total derivative of u is
du u u dy characteristics
  .
dx x y dx
By comparing with the given equation, one should choose
the path such that dy/dx  3x 2 . Hence, the characteristics are u known
the set of curves
x
y  x3  c (2,0)
where the constant of integration, c, determines the
characteristic. The particular characteristic passing through
the point (3,19) is
y  x3  8

Along this characteristic,

du
 x y
dx
 x  x3  8
and hence, along this characteristic, u varies as
u  12 x 2  14 x 4  8x  d
where d is another constant of integration.

The curve y  x 3  8 cuts the x axis at x = 2; at this point the boundary conditions give u = 4.
Substituting these values fixes d = 14. Hence, along the characteristic passing through (3,19)
u is given by
u  12 x 2  14 x 4  8x  14
Thus, the value when x = 3 is
59
u (3,19)  [ 12 x 2  14 x 4  8 x  14] x 3 
4

Exercise for the willing: show that the general solution is

u  12 x 2  14 x 4  x( x 3  y)  12 ( x 3  y) 2 / 3  34 ( x 3  y) 4 / 3

Computational Mechanics 5-3 David Apsley

General Method of Characteristics For 1st-Order PDEs

The general quasi-linear 1st-order PDE is

u u
a b c
x y
where a, b, c are functions of x, y, u. The dependent variable u is given on some initial curve.

Write the equation as

u u b c
 
x y a a
Now, for any curve y(x):
du u u dy
 
dx x y dx
dy b
Hence if one chooses characteristic curves y(x) such that  , then u will evolve along
dx a
du c
these curves according to  .
dx a

All first-order partial differential equations of the form

u u
a b c
x y
(where a, b and c may be functions of x, y and u) have a formal solution and characteristics
given by the coupled ordinary differential equations
dy b du c
 , 
dx a dx a

For completely, general a, b and c these are simultaneous ordinary differential equations,
which may, if necessary, be solved by the forward-marching techniques of section 2.

The solution and characteristics are sometimes summarised as

dx dy du
 
a b c

Computational Mechanics 5-4 David Apsley

5.2.2 Second-Order Wave Equation

Under certain circumstances (for “hyperbolic” equations) 2nd-order PDEs may be “factorised”
in such a way as to be soluble by the method of characteristics.

Second-order wave equation:

 2u 1  2u
 0
x 2 c 2 t 2

If c (we shall see later that this is a wave-speed) is constant, this can be factorised as
 1   1 
(  )(  )u  0 (*)
x c t x c t

Choose ξ and η such that

 1  1    1  1  
   ,    
ξ 2  x c t  η 2  x c t 
(The factors of ½ aren’t strictly necessary – they just happen to be convenient for what
follows.) This can readily be achieved by taking
1 1
x  (ξ  η) , t  (η  ξ )
2 2c
which can be inverted to give
ξ  x  ct , η  x  ct

Substituting into (*) reduces the wave equation to the simple form
 2u
0
η ξ
This may now be integrated twice (first with respect to η, along ξ = constant; then with
respect to ξ along η = constant):
u
 f (ξ )
ξ
 t
 u   f (ξ) dξ  G(η) zone of
 influence

(where f(ξ) and G(η) are arbitrary). Writing F (ξ)   f (ξ) dξ , the

general solution of the 2-d second-order wave equation is (x,t)
u  F ( x  ct )  G( x  ct )
x+
t
ns

ct=

for any functions F and G.

co

co
ct=

zone of
nst
x-

dependence
The solution consists of a right-travelling wave ( x  ct  constant )
and a left-travelling wave ( x  ct  constant ). Functions F and G will x
be fixed by appropriate boundary conditions – typically initial initial data
boundary conditions; i.e. u and u/t specified everywhere at t = 0.

The key point is that, for this type of equation, information propagates along characteristics.
This must be reflected in the discretisation scheme.

Computational Mechanics 5-5 David Apsley

5.3 Classification of Second-Order PDEs

5.3.1 Hyperbolic, Parabolic and Elliptic PDEs

An equation of the form

 2  2  2
a 2 b c 2  f
x xy y
where a, b, c and f may be functions of x, y, , /x and /y, but not the second
derivatives, is called quasi-linear.

The nature of the differential equation and its analytical or numerical solution is determined
by the value of the discriminant b 2  4ac .

For constant a, b and c this relates to the ability to factorise it into two first-order
differentials, each of which will have its set of characteristics along which the solution can be
forward-marched. (The general analysis is a bit more complex and won’t be given here, but
the discriminant is still the distinguishing factor.).

The cases are:

b2  4ac  0 hyperbolic two sets of characteristics forward-marching solution
b2  4ac  0 parabolic one set of characteristics forward-marching solution
b2  4ac  0 elliptic no characteristics simultaneous solution for all points

If a, b or c is not constant then it is possible for the governing equations to be of different

types in different parts of the domain.

Examples.
 2  2  2 a = 1, b = 4, c = 3;
(i)  3  4  3 hyperbolic
x 2 y 2 xy (b2 – 4ac = 4)

      2   a = 1, b = 1, c = 1;
    0
x  x y  y 2 y
(ii) elliptic
(b2 – 4ac = –3)

 2  2 a = y, b = 0, c = x; elliptic in 1st/3rd quadrants,

(iii) x 2 y 2 
y x (b2 – 4ac = –4xy) hyperbolic in 2nd/4th quadrants

Computational Mechanics 5-6 David Apsley

5.3.2 Domain of Dependence and Boundary Conditions

The nature of the equation determines the required

t
boundary conditions. hyperbolic
For elliptic equations, values of  or its derivative are x
required on all boundaries of the domain.
t
Forward-marching schemes are like initial-value
problems in ODEs. For a second-order equation one parabolic
requires both  and its time derivative on the initial
boundary. x
y
elliptic

x
Domain of dependence for
different types of PDE

Type Example Equation Application(s)

Transverse vibration of a string: c2 = T/ρ
 = displacement;
T = tension;
 2 1  2 ρ = mass per unit length.
 0
Hyperbolic Wave equation x 2 c 2 t 2
Longitudinal vibration in a rod: c2 = E/ρ
 = displacement;
E = Young’s modulus;
ρ = density.
Heat equation: κ = k/ρcP
 = temperature;
k = thermal conductivity;
ρ = density;
  2
Parabolic Diffusion equation κ 2 cP = specific heat capacity.
t x
Consolidation equation:
 = excess pore pressure.
κ = coefficient of consolidation.
Steady-state distributions such as:
Poisson equation  2  2 temperature distribution;
Elliptic (Laplace equation   f electrostatic potential;
if RHS = 0) x 2 y 2 stress distribution;
velocity potential.

(Note the subtle change between t and y for the second independent variable; the former tends
to occur in time-dependent problems).

Computational Mechanics 5-7 David Apsley