We are IntechOpen,

the world’s leading publisher of

Open Access books
Built by scientists, for scientists

4,400
Open access books available
117,000
International authors and editors
130M Downloads

Our authors are among the

154
Countries delivered to
TOP 1%
most cited scientists
12.2%
Contributors from top 500 universities

Selection of our books indexed in the Book Citation Index

in Web of Science™ Core Collection (BKCI)

Interested in publishing with us?

Contact [email protected]
Numbers displayed above are based on latest data collected.
For more information visit www.intechopen.com
 Chapter 11

Solution of Differential Equations with Applications to

Engineering Problems

Cheng Yung Ming

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/67539

Abstract
Over the last hundred years, many techniques have been developed for the solution of
ordinary differential equations and partial differential equations. While quite a major
portion of the techniques is only useful for academic purposes, there are some which are
important in the solution of real problems arising from science and engineering. In this
chapter, only very limited techniques for solving ordinary differential and partial differ-
ential equations are discussed, as it is impossible to cover all the available techniques
even in a book form. The readers are then suggested to pursue further studies on this
issue if necessary. After that, the readers are introduced to two major numerical methods
commonly used by the engineers for the solution of real engineering problems.

Keywords: differential equations, analytical solution, numerical solution

1. Introduction
1.1. Classification of ordinary and partial equations

To begin with, a differential equation can be classified as an ordinary or partial differential

equation which depends on whether only ordinary derivatives are involved or partial deriva-
tives are involved. The differential equation can also be classified as linear or nonlinear. A
differential equation is termed as linear if it exclusively involves linear terms (that is, terms to
the power 1) of y, y0, y″ or higher order, and all the coefficients depend on only one variable x as
shown in Eq. (1). In Eq. (1), if f(x) is 0, then we term this equation as homogeneous. The general
solution of non-homogeneous ordinary differential equation (ODE) or partial differential
equation (PDE) equals to the sum of the fundamental solution of the corresponding homogenous
equation (i.e. with f(x) = 0) plus the particular solution of the non-homogeneous ODE or PDE.
On the other hand, nonlinear differential equations involve nonlinear terms in any of y, y0, y″,
or higher order term. A nonlinear differential equation is generally more difficult to solve than
linear equations. It is common that nonlinear equation is approximated as linear equation

© 2017 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is properly cited.
234 Dynamical Systems - Analytical and Computational Techniques

(over acceptable solution domain) for many practical problems, either in an analytical or
numerical form. The nonlinear nature of the problem is then approximated as series of linear
differential equation by simple increment or with correction/deviation from the nonlinear
behaviour. This approach is adopted for the solution of many non-linear engineering
problems. Without such procedure, most of the non-linear differential equations cannot be
solved. Differential equation can further be classified by the order of differential. In general,
higher-order differential equations are difficult to solve, and analytical solutions are not avail-
able for many higher differential equations. A linear differential equation is generally
governed by an equation form as Eq. (1).

dn y dn 1 y
þ a 1 ðxÞ þ … þ an ðxÞy ¼ f ðxÞ (1)
dxn dxn 1

“Non-linear” differential equation can generally be further classified as

1. Truly nonlinear in the sense that F is nonlinear in the derivative terms.

∂2 u
 
∂u ∂u
F x1 , x1 , xn , u, , , ¼0 (2)
∂x1 ∂x2 ∂x1 ∂x2

2. Quasi-linear 1st PDE if nonlinearity in F only involves u but not its derivatives

∂u ∂u
A1 ðx1 , x2 , uÞ þ A2 ðx1 , x2 , uÞ ¼ Bðx1 , x2 , uÞ (3)
∂x1 ∂x2

3. Quasi-linear 2nd PDE if nonlinearity in F only involves u and its first derivative but not its
second-order derivatives

∂u ∂u ∂2 u
 2
∂u ∂u ∂2 u
    
∂u ∂u ∂ u
A11 x1 , x2 , u, , þ A12 x1 , x2 , u, , þ A22 x1 , x2 , u, ,
∂x1 ∂x2 ∂x21 ∂x1 ∂x2 ∂x1 ∂x2 ∂x1 ∂x2 ∂x22
 
∂u ∂u
¼ F x1 , x2 , u ,
∂x1 ∂x2
(4)

Examples of differential equations:

dy
1. dx ¼ 3x þ 2; first-order ODE (linear)/nonhomogeneous

2. ðy 2xÞdy 3ydx ¼ 0; first-order ODE (nonlinear)/homogenous

 3
d2 y
3. dt2
þ t2 y dy
dt þ y ¼ 0; second-order ODE (nonlinear)/homogenous

d4 x 2
4. dt4
þ 5 ddt2x þ 7x ¼ sint; fourth-order ODE (linear)/nonhomogeneous
∂z ∂z
5. ∂x þ ∂y ¼ 2z; first-order PDE (linear)/homogeneous
 Solution of Differential Equations with Applications to Engineering Problems 235
http://dx.doi.org/10.5772/67539

∂2 u 2
6. ∂x2
þ ∂∂yu2 þ 4x þ 3y uz ¼ 0; second-order PDE (linear)/nonhomogeneous
2 2
7. x ∂∂xu2 þ 2u ∂∂yu2 þ 3u2 ¼ 0; second-order PDE (linear)/homogeneous

du dv
8. dx dx ¼ 6x; 1st ODE (linear) for two unknowns/nonhomogeneous

1.2. Typical differential equations in engineering problems

Many engineering problems are governed by different types of partial differential equations,
and some of the more important types are given below.

∂2 u ∂2 u y > 0 : elliptic
Tricomi equation: y ∂x2 þ ∂y2 ¼ 0
y < 0 : hyperbolic
2 2
Laplace equation (or variants): ∂∂xϕ2 þ ∂∂yϕ2 ¼ ∇2 ϕ ¼ 0
2 2
Poisson’s equation: ∂∂xϕ2 þ ∂∂yϕ2 ¼ f ðx, yÞ
2 2
Helmholtz equation: ∂∂xϕ2 þ ∂∂yϕ2 þ c2 ϕ ¼ 0

Plate bending: ∇2 ∇2 w ¼ ∇4 w ¼ Dq
 
∂2 u 2 ∂2 u ∂2 u
Wave equation (1D-3D): ∂t2 c ∂x2 þ ∂y2 ¼ 0
2 
Fourier equation: ∂T
∂t ¼ α ∂∂xT2

There are many methods of solutions for different types of differential equations, but most of
these methods are not commonly used for practical problems. In this chapter, the most impor-
tant and basic methods for solving ordinary and partial differential equations will be
discussed, which will then be followed by numerical methods such as finite difference and
finite element methods (FEMs). For other numerical methods such as boundary element
method, they are less commonly adopted by the engineers; hence, these methods will not be
discussed in this chapter.

1.3. Separable differential equations

For equations which can be expressed in separable form as shown below, the solution can be
obtained easily as

dy dy dy
ð ð
¼ Fðx, yÞ ¼ f ðxÞdx ¼ f ðxÞdx þ c (5)
dx ΦðyÞ ΦðyÞ

Mðx, yÞdx þ Nðx, yÞdy ¼ 0 MðxÞdx ¼ NðyÞdy (6)

ð ð
then MðxÞdx ¼ NðyÞdy þ c (7)
236 Dynamical Systems - Analytical and Computational Techniques

Example:

dy dy
¼ x3 ðy2 þ 1Þ ) 2 ¼ x3 dx
dx y þ1
ð ð

dy 3 1 4 1 1 4
¼ x dx þ c ) tan y ¼ x þ C ) y ¼ tan x þ c
y2 þ 1 4 4

Example:
dy 2
dx ¼ 3x2ðyþ4xþ2
1Þ subject to yð0Þ ¼ 1

Since this is a separable function, the problem can be solved as

2ðy 1Þdy ¼ ð3x2 þ 4x þ 2Þdx

y2 2y ¼ x3 þ 2x2 þ 2x þ c

Based on the boundary condition, c = 3, hence y2 2y ¼ x3 þ 2x2 þ 2x þ 3.

This quadratic equation in y2 can be solved with two solutions by the quadratic equation as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
y¼1 x3 þ 2x2 þ 2x þ 4 and y ¼ 1 þ x3 þ 2x2 þ 2x þ 4:

Since the second solution does not satisfy the boundary condition, it will not be accepted;
hence, the solution to this differential equation is obtained.

1.4. Variation of parameters

For the following equation form, it is possible to solve it by variations of parameters.

dy
For ¼ pðxÞy þ QðxÞ (8)
dx
Ð Ð Ð
pðxÞdx dy dcðxÞ pðxÞdx
Put y ¼ cðxÞe . By differentiating, it gives dx ¼ dx e þ cðxÞpðxÞe pðxÞdx . Substitute
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
pðxÞy
Ð
it to the original ODE dcðxÞ
dx ¼ QðxÞe
pðxÞdx
. Comparing the terms, it gives
ð Ð
pðxÞdx
cðxÞ ¼ QðxÞe dx þ c: (9)

Example:

dy
ðx þ 1Þ ny ¼ ex ðx þ 1Þnþ1
dx

This equation is now expressed as

 Solution of Differential Equations with Applications to Engineering Problems 237
http://dx.doi.org/10.5772/67539

dy
¼ pðxÞy þ QðxÞ
dx
dy n
¼ y þ ex ðx þ 1Þn
dx x þ 1 |fflfflfflfflfflffl{zfflfflfflfflfflffl}
QðxÞ

For x 6¼ 1
Solving the homogeneous part of the ODE
dy
dx
n
¼ xþ1 y then dy n
y ¼ xþ1 dx

lnjyj ¼ nlnjx þ 1j þ c1
y ¼ cðx þ 1Þn

Look for solution y ¼ cðxÞðx þ 1Þn , where c(x) is the variation of parameters. Substitute it to the
ODE

dcðxÞ 1 1
ðx þ 1Þn þ ncðxÞðx þ 1Þn ¼ ncðxÞðx þ 1Þn þ ex ðx þ 1Þn
dx
dy n
¼ y þ ex ðx þ 1Þn
dx x þ 1

Comparison gives dcðxÞ

dx ¼ e
x

Integration of this equation gives cðxÞ ¼ ex þ C

General solution is hence given by y ¼ ðx þ 1Þn ðex þ CÞ

The Bernoulli equation is an important equation type which can be solved in a similar way by
variation of parameters. Consider the following form of equation

dy
¼ pðxÞy þ QðxÞyn (10)
dx

Step 1 : Put z ¼ y1 n
(11)

dz n dx
Step 2 : Then ¼ ð1 nÞy
dx dy
(12)
dz
¼ ð1 nÞPðxÞz þ ð1 nÞQðxÞ
dx

The non-linear ODE now becomes linear ODE. It can be solved by formula

Step 3: n = 1, z = y2. Inverting z to get y

dy y x2
¼ þ (13)
dx 2x 2y
238 Dynamical Systems - Analytical and Computational Techniques

dz 1
¼ z þ x2 (14)
dx x
Ð 1
ð 
1
Ð1
dx 2 dx
z¼e x x e x dx þ c ¼ cx þ x3 (15)
2

Back substitution of z ¼ y2

1
y2 ¼ cx þ x3 (16)
2

1.5. Homogeneous equations

For equation of the following type, where all the coefficients are constant, it can be evaluated
according to different conditions.

dy a1 x þ b1 y þ c1
¼ (17)
dx a2 x þ b2 y þ c2

Case 1: c1 ¼ c2 ¼ 0
y
dy a1 x þ b1 y a1 þ b1 x y
¼ ¼ ¼ g (18)
dx a2 x þ b2 y a2 þ b2 yx x

Step 1: Set u ¼ yx, then dy du

dx ¼ x dx þ u

gðuÞ u
Step 2: du
dx ¼ x . The resulting non-linear ODE is hence separable and can be solved
implicitly.
Step 3: Inverting u to get y.

a a
1 2
Case 2: ¼0
b1 b2

a1 b2 a2 b1 ¼ 0 then aa12 ¼ bb12 ¼ k

dy a1 x þ b1 y þ c1 kða2 x þ b2 yÞ þ c1
¼ ¼ ¼ f ða2 x þ b2 yÞ (19)
dx a2 x þ b2 y þ c2 a2 x þ b2 y þ c2

By change of variables as u ¼ a2 x þ b2 y
du
dx ¼ a2 þ b2 dy
dx ¼ a2 þ b2 f ðuÞ, then

du
¼ dx (20)
a2 þ b2 f ðuÞ

The resulting non-linear ODE is now separable and can be solved.

 Solution of Differential Equations with Applications to Engineering Problems 239
http://dx.doi.org/10.5772/67539

a a
1 2
Case 3: 6¼ 0 c1 6¼ 0 and c2 6¼ 0
b1 b2

a1 x þ b1 y þ c1 ¼ 0
Set . Intersecting point of these two lines on xy - plane and (α, β) 6¼ 0
a2 x þ b2 y þ c2 ¼ 0

xy plane and ðα, βÞ ¼

6 ð0, 0Þ (21)

Apply change of variables

X¼x x¼Xþα
 
α
(22)
Y¼y β y¼Yþβ

a1 x þ b1 y þ c1 ¼ a1 ðX þ αÞ þ b1 ðY þ βÞ þ c1 ¼ a1 X þ b1 Y þ ða1 α þ b1 β þ c1 Þ
(23)
a2 x þ b2 y þ c2 ¼ a2 ðX þ αÞ þ b2 ðY þ βÞ þ c2 ¼ a2 X þ b2 Y þ ða2 α þ b2 β þ c2 Þ

dY
The original ODE will now become dX ¼ aa12 Xþb1Y
Xþb2 Y which is homogeneous and separable!

xþy 1
Example: dy
dx ¼ x yþ3

xþy 1¼0
Solve for we have α ¼ 1, β ¼ 2
x yþ3¼0

Change of variables X = x + 1, Y = y 2
Y
xþy 1 1þX
dY
Then, dX ¼ dy XþY
dx ¼ x yþ3 ¼ X Y ¼ 1 Y
X

2 ð1 uÞdu
Use a change of variable u ¼ XY X dX
du
¼ 1þu
1 u 1þu2 ¼ dX
X

1
) tan 1 u lnð1 þ u2 Þ ¼ lnjXj þ c
2
pffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
) tan 1 u ¼ ln½ 1 þ u2 XŠ þ c ¼ ln½ ðX2 þ Y2 ފ þ c


1 y 2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
) tan ¼ ln ðx þ 1Þ2 þ ðy 2Þ2 þ c
xþ1

There are various tricks to solve the differential equations, like integration factors and other
techniques. A very good coverage has been given by Polyanin and Nazaikinskii [29] and will
not be repeated here. The purpose of this section is just for illustration that various tricks have
been developed for the solution of simple differential equations in homogeneous medium, that
is, the coefficients are constants inside a continuous solution domain. The readers are also
suggested to read the works of Greenberg [14], Soare et al. [34], Nagle et al. [28], Polyanin et al.
[30], Bronson and Costa [4], Holzner [18], and many other published books. There are many
elegant tricks that have been developed for the solution of different forms of differential
equations, but only very few techniques are actually used for the solution of real life problems.
240 Dynamical Systems - Analytical and Computational Techniques

1.6. Partial differential equations

In many engineering or science problems, such as heat transfer, elasticity, quantum mechanics,
water flow and others, the problems are governed by partial differential equations. By nature,
this type of problem is much more complicated than the previous ordinary differential equa-
tions. There are several major methods for the solution of PDE, including separation of
variables, method of characteristic, integral transform, superposition principle, change of
variables, Lie group method, semianalytical methods as well as various numerical methods.
Although the existence and uniqueness of solutions for ordinary differential equation is well
established with the Picard-Lindelöf theorem, but that is not the case for many partial differ-
ential equations. In fact, analytical solutions are not available for many partial differential
equations, which is a well-known fact, particularly when the solution domain is nonregular
or homogeneous, or the material properties change with the solution steps.

1.6.1. Classification of second-order PDE

Refer to the following general second-order partial differential equation:

∂2 u ∂2 u ∂2 u ∂u ∂u
A 2
þ B þ C 2
þ D þ E þ Fu þ G ¼ 0 (24)
∂x ∂x∂y ∂y ∂x ∂y

To begin with, let us consider a review of conic curves (ellipse, parabola and hyperbola)

Ax2 þ Bxy þ Cy2 þ Dx þ Ey þ F ¼ 0 (25)

The conic curve can be classified with the following criterion.

8
< > 0 hyperbola
>
B2 4AC ¼ ¼ 0 parabola (26)
>
< 0 ellipse
:

Following the conic curves, the general partial differential is also classified according to similar
criterion as
2
8
< B
> 4AC > 0 : elliptic
2
Classification B 4AC ¼ 0 : parabolic (27)
: 2
>
B 4AC < 0 : hyperbolic

This classification was proposed by Du Bois-Reymond [41] in 1839. In this section, only some
of the more common techniques are discussed, and the readers are suggested to read the
works of Hillen et al. [16], Salsa [33], Polyanin and Zaitsev [31], Bertanz [2], Haberman [15]
and many other published texts.

1.7. Parabolic type: heat conduction/soil consolidation/diffuse equation

The following equation form is commonly found in many engineering applications.
 Solution of Differential Equations with Applications to Engineering Problems 241
http://dx.doi.org/10.5772/67539

∂2 u ∂u
α2 ¼ , 0 < x < L, t > o (28)
∂x2 ∂t

Initial condition: uðx, 0Þ ¼ f ðxÞ, 0 ≤ x ≤ L

Boundary condition: uð0, tÞ ¼ 0, uðL, tÞ ¼ 0, t > 0

α2 is a constant known as the thermal diffusivity or coefficient of consolidation. For soil
consolidation problem, the governing conditions are given by

Initial excess pore pressure

ue ðz, 0Þ ¼ ui ðzÞ, 0 ≤ z ≤ 2d
(29)
ue ð0, tÞ ¼ 0, ue ð2d, tÞ ¼ 0, t > 0

Drained boundary

α2 uxx ¼ ut , 0 < x < L, t > 0

uð0, tÞ ¼ 0, uðL, tÞ ¼ 0, t > 0 (30)
uðx, 0Þ ¼ f ðxÞ, 0 ≤ x ≤ L

Assuming variable u(x, t) can be separated, using separation of variables

uðx, tÞ ¼ XðtÞTðtÞ (31)

α2 X″ T ¼ XT 0
X″ 1 T0
¼ 2
X α T (32)
( ″
″ 0 X þ λX ¼ 0
X 1 T
¼ 2 ¼ λ!
X α T T 0 þ α2 λT ¼ 0

A PDE now becomes two ODE which can be solved readily. Based on the boundary condition
uð0, tÞ ¼ 0, uðL, tÞ ¼ 0, t > 0

uð0, tÞ ¼ Xð0Þ, TðtÞ ¼ 0

Xð0Þ ¼ 0, XðLÞ ¼ 0 (33)
X″ þ λX ¼ 0, Xð0Þ ¼ 0, XðLÞ ¼ 0

This is an eigenvalue problem which has solution only for certain λ. The eigenvalues are given
by

n2 π2
λn ¼ , n ¼ 1, 2, 3, … (34)
L2

Hence the eigenfunctions are expressed as

242 Dynamical Systems - Analytical and Computational Techniques

nπx
Xn ðxÞ ¼ sin , n ¼ 1, 2, 3… (35)
L

For the time-dependent function T,

0
T þ α2 λT ¼ 0 (36)

dT
¼ α2 λdt
T
(37)
α2 n2 π2 t
lnjTj ¼ þC
L2

ðnπα=LÞ2 t
hence T n ¼ kn e , kn constant. The fundamental solutions are then expressed as
2
nπx
uðx, tÞ ¼ e ðnπα=LÞ t sin , n ¼ 1, 2, 3… (38)
L

The Fourier series expansion in x is given by

uð0, tÞ ¼ f ðxÞ, 0 ≤ x ≤ L (39)

∞ ∞ nπx
ðnπα=LÞ2 t
X X
uðx, tÞ ¼ cn un ðx, tÞ ¼ cn e sin (40)
n¼1 n¼1
L

Initial condition is given as

∞
X nπx
uðx, 0Þ ¼ f ðxÞ ¼ cn sin (41)
n¼1
L
ðL mπx ∞ ðL 
X mπx nπx
f ðxÞsin dx ¼ cn sin sin dx
0 L n¼1 0 L L
ðL nπx ðL mπx L
f ðxÞsin dx ¼ cn sin2 dx ¼ cn
0 L 0 L 2

Solution of the soil consolidation equation is hence given by

∞ nπx
ðnπα=LÞ2 t
X
uðx, tÞ ¼ cn e sin (42)
n¼1
L
ðL
2 nπx
cn ¼ f ðxÞsin dx ðEulerFourier formulasÞ (43)
L 0 L

1.8. One-dimensional wave equation

One-dimensional (1D) wave equation appears in many physical and engineering problems.
For example, a vibrating string or pile driving process is given by this type of differential
equation. This problem is also commonly solved by the method of separation of variables
 Solution of Differential Equations with Applications to Engineering Problems 243
http://dx.doi.org/10.5772/67539

a2 uxx ¼ utt , 0 < x < L, t > 0

uð0, tÞ ¼ 0, uðL, tÞ ¼ 0, t ≥ 0 (44)
uðx, 0Þ ¼ f ðxÞ, uðx, 0Þ ¼ 0, 0 ≤ x ≤ L

Consider u(x, t) is given by X(x)T(t). The wave equation will give

X″ 1 T0

X″ þ λx ¼ 0
¼ 2 ¼ λ! (45)
X a T T 0 þ a2 λt ¼ 0

The partial differential equation will then be given by two equivalent ODEs.
0 0
ut ðx, 0Þ ¼ XðxÞT ð0Þ ¼ 0, 0 ≤ x ≤ L ! T ð0Þ ¼ 0
0 (46)
uð0, tÞ ¼ Xð0ÞTðtÞ ¼ 0, uðL, tÞ ¼ XðLÞTðtÞ 0 ≤ x ≤ L ! T ð0Þ ¼ 0

X″ þ λX ¼ 0, Xð0Þ ¼ XðLÞ ¼ 0 (47)

nπx
Xn ðxÞ ¼ sin , n ¼ 1, 2, 3, … (48)
L
n2 π2
λn ¼ , n ¼ 1, 2, 3, … (49)
L2

For the time-dependent function T,

0
T þ a2 λT ¼ 0 (50)
0
T ð0Þ ¼ 0 λn ¼ nπ=L
(51)
Then TðtÞ ¼ k1 cosðnπat=LÞ k2 sinðnπat=LÞ

0
Since T ð0Þ ¼ 0 k2 ¼ 0
Therefore, TðtÞ ¼ k1 cosðnπat=LÞ
Fundamental solution is given by
nπx  
nπat
un ðx, tÞ ¼ sin cos , n ¼ 1, 2, 3… (52)
L L

The general solution is then given by

∞
X X∞ nπx nπat
uðx, tÞ ¼ cn un ðx, tÞ ¼ cn sin cos (53)
n¼1 n¼1
L L

Applying the boundary condition

uðx, 0Þ ¼ f ðxÞ, 0 ≤ x ≤ L
∞ nπx 2 L nπx
ð
uðx, 0Þ ¼ f ðxÞ ¼
X
cn sin ! cn ¼ f ðxÞsin dx (54)
n¼1
L L 0 L
244 Dynamical Systems - Analytical and Computational Techniques

The final solution is then given by

X∞ nπx nπat
uðx, tÞ ¼ cn sin cos (55)
n¼1
L L
ðL
2 nπx
cn ¼ f ðxÞsin dx (56)
L 0 L

1.9. Laplace equation

Laplace equation forms an important governing condition for many types of problems. Some
of the more common forms are given by
three-dimensional Laplace equation uxx þ uyy þ uzz ¼ 0

two-dimensional heat conduction α2 ðuxx þ uyy Þ ¼ ut

two-dimensional seepage problem ðkx uxx þ ky uyy Þ ¼ 0

There are two major types of boundary conditions to this problem:

Dirichlet problem: boundary conditions prescribed as u

Neumann problem: normal derivative ux or uy are usually prescribed on the boundary for many
mathematical problems. This case can be solved by the use of complex analysis or series
method for which many analytical solutions are available in the literature. In many aniso-
tropic seepage problems, however, the normal of a derived quantity at any arbitrary direc-
tion (seepage flow normal to an impermeable surface) is 0 instead of ux or uy being zero. For
such cases, it is very difficult to obtain the analytical solution if the solution domain is
nonhomogeneous, and the use of numerical method such as the finite element method
appears to be indispensable.
Consider the given Laplace equation, using separation of variables for the analysis.

uxx þ uyy ¼ 0, 0 < x < a, 0 < y < b

uðx, 0Þ ¼ 0, uðx, bÞ ¼ 0, 0 < x < a (57)
uð0, yÞ ¼ 0, uða, yÞ ¼ f ðyÞ, 0 < y ≤ b

Using separation of variables, uðx, tÞ ¼ XðxÞYðyÞ

X″ Y þ XY″ ¼ 0
X″ Y″ X″ λX ¼ 0 (58)
¼ ¼λ!
X Y Y 0 þ λY ¼ 0

uxx þ uyy ¼ 0, 0 < x < a, 0 < y < b (59)

uðx, 0Þ ¼ 0, uðx, bÞ ¼ 0, 0 < x < a

(60)
uð0, yÞ ¼ 0, uða, yÞ ¼ f ðyÞ, 0 < y ≤ b
 Solution of Differential Equations with Applications to Engineering Problems 245
http://dx.doi.org/10.5772/67539

uð0, yÞ ¼ Xð0ÞYðyÞ ¼ 0, 0 < y < b ! Xð0Þ ¼ 0,

uðx, 0Þ ¼ XðxÞYð0Þ ¼ 0, 0 < x < a ! Yð0Þ ¼ 0, (61)
uðx, bÞ ¼ XðxÞYðbÞ ¼ 0, 0 < x < a ! YðbÞ ¼ 0,

X″ λX ¼ 0, Xð0Þ ¼ 0
″ (62)
Y þ λY ¼ 0, Yð0Þ ¼ 0, YðbÞ ¼ 0

n2 π2 nπy
λn ¼ , Y n ðyÞ ¼ sin , n ¼ 1, 2, 3, … (63)
b2 b

X″ λX ¼ 0, hence XðxÞ ¼ k1 coshðnπx=bÞ k2 sinðnπx=bÞ

Since X(0) = 0, k1 = 0
nπx
XðxÞ ¼ k2 sinh (64)
b
nπx nπy
un ðx, yÞ ¼ sinh sin n ¼ 1, 2, 3… (65)
b b
uða, yÞ ¼ f ðyÞ, 0 ≤ y ≤ b
∞ ∞ nπx nπy
uðx, yÞ ¼
X
cn un ðx, yÞ ¼
X
cn sin cos (66)
n¼1 n¼1
b b

Based on the Fourier expansion as given by

ðb mπy X∞ nπað b mπy nπy
f ðyÞsin dy ¼ c sinh
n¼1 n
sin sin dy
0 b b 0 b b
ðb nπx ðb
mπa nπx mπa b (67)
f ðxÞsin dx ¼ sinh cn sin2 dx ¼ sinh cn
0 b b 0 b b 2
X∞ nπa nπy
uða, yÞ ¼ f ðyÞ ¼ c sinh
n¼1 n
sin
b b
nπa 2 ð b nπy
cn sinh ¼ f ðyÞsin dy
b b 0 b
2 nπa 1 ð b nπy
cn ¼ sinh f ðyÞsin dy
b b 0 b

1.10. Introduction to numerical methods

In general, analytical solutions are not available for most of the practical differential equations,
as regular solution domain and homogeneous conditions may not be present for practical
problems. Moreover, the solution domain may be indeterminate (free surface seepage flow),
the displacement is large so that the solution may deform under motion, or in an extreme case
part of the material may tear off from the main body with continuous formation and removal
of contacts. Many engineering problems fall into such category by nature, and the use of
numerical methods will be necessary. Currently, there are several major numerical methods
246 Dynamical Systems - Analytical and Computational Techniques

commonly used by the engineers: finite difference method, finite element method, boundary
element method and distinct element. There are also other less common numerical methods
available for practical problems, and many researchers also try to combine two or even more
fundamental numerical methods so as to achieve greater efficiency in the analysis. In general,
the solution domain is discretized into series of subdomains with many degrees of freedom.
The number of variables or degrees of freedom may even exceed millions for large-scale
problems, and sometimes very special material properties are encountered so that the system
is highly sensitive to the method of discretization and the method of solution. Similar to the
ODE and PDE, it is impossible to discuss the details of all the numerical methods and the
author choose to discuss the finite element method due to the wide acceptance of the method
and this method is more suitable for general complicated methods.
Except for some simple problems with regular geometry and loading, it is very difficult to solve
most of the boundary value problems with the yield of analytical solutions. Towards this, the
use of numerical method seems indispensable, and the finite element is one of the most popular
methods used by the engineers [32, 38]. There are two fundamental approaches to FEM, which
are the weighted residual method (WRM) and variational principle, but there are also other
less popular principles which may be more effective under certain special cases. In finite
element analysis of an elastic problem, solution is obtained from the weak form of the equiva-
lent integration for the differential equations by WRM as an approximation. Alternatively,
different approximate approaches (e.g. collocation method, least square method and Galerkin
method) for solving differential equations can be obtained by choosing different weights based
on the WRM and the Galerkin method appears to be the most popular approach in general.
Specifically, in elasticity for instance, the principle of virtual work (including both principle of
virtual displacement and virtual stress) is considered to be the weak form of the equivalent
integration for the governing equilibrium equations. Furthermore, the aforementioned weak
form of equivalent integration on the basis of the Galerkin method can also be evolved to a
variation of a functional if the differential equations have some specific properties such as
linearity and selfadjointness. Principles of minimum potential energy and complementary
energy are two variational approaches equivalent to the fundamental equations of elasticity.
Since displacement is usually the basic unknown quantity in FEM, only the principle of virtual
displacement and minimum potential energy will be introduced in the following section. In
this case, the FEM introduced herein is also called displacement finite element method
(DFEM). There are other ways to form the basis of FEM with advantages in some cases, but
these approaches are less general and will not be discussed here.

1.11. Principle of virtual displacement

The principle of virtual displacement is the weak form of the equivalent integration for
equilibrium equations and force boundary conditions. Given the equilibrium equations and
force boundary conditions in index notation,

σij, j þ f i ¼ 0, ðin domain VÞ (68)

σij nj T i ¼ 0, ðon domain boundary Sσ Þ (69)

 Solution of Differential Equations with Applications to Engineering Problems 247
http://dx.doi.org/10.5772/67539

In WRM, without loss of generality, the variation of true displacement δui and its boundary
value (i.e. δui ) can be selected as the weight functions in the equivalent integration
ð ð
δui ðσij, j þ f i ÞdV δui ðσij nj T i ÞdS ¼ 0 (70)
V Sσ

The weak form of Eq. (70) is given as

ð ð
ð δεij σij þ δui f i ÞdV þ δui T i dS ¼ 0 (71)
V Sσ

It can be seen clearly from Eq. (71) that the first item in the volume integral indicates the work
done by the stresses under the virtual strain (i.e. internal virtual work), while the remaining
items indicate the work done by the body force and surface force under the virtual displace-
ment (i.e. external virtual work). In other words, the summation of the internal and external
virtual works is equal to 0, which is called the principle of virtual displacement. Under this
case, we can conclude that a force system will satisfy the equilibrium equations if the summa-
tion of the work done by it under any virtual displacement and strain is equal to 0.

1.12. Principle of minimum potential energy (PMPE)

Based on Eq. (71), we can deduce that

ð ð
ðδεij Dijkl εkl δui f i ÞdV þ δui T i dS ¼ 0 (72)
V Sσ

Due to the symmetry of the constitutive matrix Dijkl , we can further obtain
 
1
ðδεij ÞDijkl εkl ¼ δ Dijkl εij εkl ¼ δUðεmn Þ (73)
2

where Uðεmn Þ is the unit volume strain energy. Given the assumptions in linear elasticity

δφðui Þ ¼ f i δui , δψðui Þ ¼ T i δui (74)

Eq. (72) is further simplified to

δΠP ¼ 0 (75)

ΠP is the total potential energy of the system, which is equal to the summation of the potential
energy of deformation and external force and can be expressed as
ð  
1
ð
ΠP ¼ ΠP ðui Þ ¼ Dijkl εij εkl f i ui dV T i ui dS (76)
V 2 Sσ

Eq. (75) shows that, among all the potential displacements, the total potential energy of system
will take stationary value at the real displacement, and it can be further verified that this station-
ary value is exactly the minimum value which is the principle of minimum potential energy.
248 Dynamical Systems - Analytical and Computational Techniques

1.13. General expressions and implementation procedure of FEM

The solution of a general continuum problem by FEM always follows an orderly step-by-step
process which is easy to be programmed and used by the engineers. For illustration, a three-
node triangular element for plane problems is taken as an example to illustrate the general
expressions and implementation procedures of FEM.

1.13.1. Discretization of domain

The first step in the finite element method is to divide the structure or solution region into
subdivisions or elements. Hence, the structure is to be modelled with suitable finite elements.
In general, the number, type, size, and arrangement of the elements are critical towards good
performance of the numerical analysis. A typical discretization with three-node triangular
element is shown schematically in Figure 1.
Mesh generation can be a difficult process for a general irregular domain. If only triangular
element is to be generated, this is a relatively simple work, and many commercial programs
can perform well in this respect. There are also some public domain codes (EasyMesh or
Triangle written in C) which are sufficient for normal purposes. For quadrilateral or higher
elements, mesh generation is not that simple, and it is preferable to rely on the use of commer-
cial programs for such purposes.

1.13.2. Interpolation or displacement model

As can be seen from Figure 1(b), the nodal number of a typical three-node triangular element
is coded in anticlockwise order (i.e. in the order of i, j and m), and each node has two degrees of
freedom (DOFs) or two displacement components which is stored in a column vector in index
notation as follows:


ui
ai ¼ ði, j, mÞ (77)
vi

Figure 1. Discretization of a two-dimensional domain with three-node triangular element.

 Solution of Differential Equations with Applications to Engineering Problems 249
http://dx.doi.org/10.5772/67539

Totally, each element has six nodal displacements, i.e. six DOFs. Putting all the displacements
in a column vector, we can obtain the element nodal displacement column matrix as
2 3
ai h
a ¼ aj 5 ¼ ui vi uj vj um vm ŠT
e 4 (78)
am

In FEM, a nodal displacement is chosen as the basic unknowns, so interpolation at any

arbitrary point is based on the three nodal displacements of each element, which is called a
displacement mode. For a three-node triangular element, linear polynomial is utilized, and the
element displacement in both x -direction and y-direction are

u ¼ β1 þ β2 x þ β3 y (79)

v ¼ β4 þ β5 x þ β6 y (80)

Obviously, displacements of all the three nodes should satisfy Eqs. (79) and (80). By substitut-
ing the six nodal displacement components into these equations, it is easy to obtain another
form of displacement mode as

u ¼ Ni ui þ N j uj þ N m um (81)

v ¼ N i vi þ N j vj þ N m vm (82)

where

1
Ni ¼ ðai þ bi x þ ci yÞði, j, mÞ: (83)
2A

In Eq. (81), N i , N j and Nm denote the interpolation function or shape function for the three
nodes, respectively. A is the area of the element, and ai , bi , ci ⋯, cm are constants related to the
coordinates of the three nodes. Similarly, Eqs. (81) and (82) can also be expressed in the form of
matrix as
ui
2 3
v 7



N 0 6 ui 7
6
u Ni 0 j Nm 0 6 j 7
u¼ ¼ 7 ¼ Nae (84)
v 0 Ni 0 N j 0 Nm 6 vj 7
6
4u 5
m
vm

where N is the shape function matrix and ae is the element nodal displacement vector. For the
geometric equations, element strains are
2 3
εx
ε ¼ 4 εy 5 ¼ Lu ¼ LNae ¼ L N i N j Nm ae

γxy (85)
¼ ½ Bi Bj Bm Šae ¼ Bae
250 Dynamical Systems - Analytical and Computational Techniques

where L is the differential operator and B is the element strain displacement matrix which can
be given as

∂ ∂N i
2 3 2 3
6 ∂x 0 7 6 ∂x 0 7
6 7
6 7
6 0 ∂ 7 Ni 0 ∂N i 7
6 7  6
6 0
Bi ¼ LN i ¼ 6 ¼6 7ði, j, mÞ (86)
7
6 ∂y 7 0 Ni
7
6 ∂y 7
6 7 6 7
4 ∂ ∂ 5 4 ∂N i ∂N i 5
∂y ∂x ∂y ∂x

Substitute Eq. (85) by the stress-strain relation,

2 3
σx
σ ¼ 4 σy 5 ¼ Dε ¼ DBae ¼ Sae (87)
τxy

where

S ¼ DB ¼ D½ Bi Bj Bm Š ¼ ½ Si Sj Sm Š (88)

S is called the element stress matrix. It should be noted that both the strain and stress matrices
are constant for each element, because in a three-node triangular element, the displacement
mode is a first-order function, and differentiating this function will give a constant function.

1.13.3. Stiffness equilibrium equation (SEE) of FEM derived from PMPE

For elastic plane problems, the total potential energy ΠP in Eq. (76) can be expressed in matrix
formulation as follows:

1 T
Y ð ð ð
T
¼ ε Dεtdxdy u f tdxdy uT TtdS (89)
P
Ω 2 Ω Sσ

where t, f, and T denote the thickness, body force and surface force, respectively. For an FEM
problem, the total potential energy is the summation of that from all the elements. Therefore,
substituting Eqs. (84) and (85) into Eq. (89) gives
X X
ð 1  X ð X ð
e T T
ΠP ¼ ΠP ¼ aeT
B DBtdxdya e
ðaeT
N f tdxdyÞ ðaeT
N T TtdxdyÞ
e e Ωe 2 e Ωe e Sσ e

(90)

Eq. (90) can be viewed as

ð ð
e T
K ¼ B DBtdxdy, Pef ¼ NT f tdxdy
Ωe Ωe
ð (91)
PeS ¼ T
N Ttdxdy, P ¼ e
Pef þ PeS
Sσ e
 Solution of Differential Equations with Applications to Engineering Problems 251
http://dx.doi.org/10.5772/67539

where Ke and Pe are named as the element stiffness matrix and equivalent element nodal load
matrix, respectively. Substitute Eq. (91) to Eq. (90), the total potential energy of the structure
can be simplified as

1X e X
Π P ¼ aT ðK Þa aT ðPe Þ (92)
2 e e

Given
X X
K¼ Ke , P ¼ Pe (93)
e e

Eq. (92) is further simplified as

1
ΠP ¼ aT Ka aT Pa (93a)
2

where Kand P are global stiffness matrix and global nodal load matrix, respectively.
For PMPE, the variation of ΠP is equal to 0 and the unknown variable is a, thus Eq. (75) gives

∂ΠP
¼0 (94)
∂a

which finally comes to the SEE of FEM as

Ka ¼ P (95)

From Eq. (93), we know that the global stiffness matrix and the global load matrix are the
assemblage of the element stiffness matrices and equivalent element nodal load matrices,
respectively. Specifically, in order to solve Eq. (93), element stiffness matrix, element equivalent
nodal load vector, global stiffness matrices and global nodal load vector are all determined
together with some given displacement boundary conditions. Without the provision of ade-
quate boundary condition, the system is singular as rigid body motion will produce no stress
in the system and such mode will be present in the SEE.

1.13.4. Derivation of element stiffness matrices (ESM)

For a three-node triangular element, the element strain matrix B is constant, thus Eq. (91) gives
2 3
Kii Kij Kim
Ke ¼ BT DBtA ¼ 4 Kji Kjj Kjm 5 (96)
Kmi Kmj Kmm

of which the submatrix

k xx kij xy


Kij ¼ ij yx (97)
kij kij yy
252 Dynamical Systems - Analytical and Computational Techniques

Kij indicates the ith nodal force along the x- and y-directions in the Cartesian coordinate system
when the displacement of the jth node is unit along the x- and y-directions, which can be easily
obtained. Moreover, the element stiffness matrix is symmetric, and the computational memory
required in an FEM program can be reduced by using this property.
It should be noted that for a higher order triangular element (e.g. six-node triangular element)
or quadrilateral element for which higher order terms are involved, the strain matrix B is not
constant any more so that the element stiffness matrix needs to be evaluated by numerical
integration (direct integration is seldom adopted). Towards this, numerical integration
methods such as the Gaussian integration or the Newton-Cotes integration can be utilized.

1.13.5. Assembling of ESMs and ENLMs

For an FEM process, we need to solve Eq. (95) which is the global equilibrium equation. Most
of the elements in the matrix Kare 0 simply because each node is only shared by a few
surrounding elements. In view of that, a rectangular matrix can represent the global stiffness
matrix (which is a square matrix), and the half bandwidth D can be defined as

D ¼ ð1 þ NDIFÞ  NDOF (98)

where NDIF denotes the largest absolute difference between the element node numbers among
all the elements in the finite element mesh.
In conclusion, the properties of the global stiffness matrix can be summarized as: symmetric,
banded distribution, singularity and sparsity. Among all the properties, singularity will vanish
by introducing appropriate boundary conditions to Eq. (95) to eliminate the rigid body motion.
Also, other properties like banded distribution should be fully taken into consideration to
reduce the computational memory and enhance the computation efficiency.

1.13.6. Isoparametric element and numerical integration

Most of the engineering structure is not regular in shape, and some of them even have very
complicated boundary shapes. Although the use of triangular element can always fit a com-
plicated boundary, the accuracy of this element is low in general. To cope with the irregular
boundary shape with a higher accuracy in analysis, one of the most common approaches is the
use of higher-order element, and the isoparametric formulation is the most commonly used at
present. Consider an arbitrary four-node quadrilateral element as an example which is sche-
matically shown in Figure 2. If we can find the transformation from Figure 2(a) to (b), then it
will become easier to carry numerical integration with complicated shapes for an arbitrary
element. In Figure 2(a), we define the Cartesian coordinate system, while in Figure 2(b), we
define the local coordinate system (or natural coordinate system) within a specific domain (i.e.
ξ, η ∈ ð 1, 1Þ). The relation between these two kinds of coordinate system can be described as
   
x ξ
¼f (99)
y η

which can be further modified by the interpolation function at nodes in the local coordinate
system as follows:
 Solution of Differential Equations with Applications to Engineering Problems 253
http://dx.doi.org/10.5772/67539

Figure 2. Isoparametric transition.

m
0
X
x ¼ Σm
i¼1 N i xi , y ¼ N 0 i yi (100)
i¼1

where ðxi , yi Þ are coordinates in the Cartesian coordinate system corresponding to the ith node
0
in local coordinate system, Ni is interpolation function of the ith node in local coordinate
system and m is the number of nodes chosen to transform the coordinates. Therefore, the
regular element in the natural coordinate system can be transformed to the irregular element
in the Cartesian coordinate system. The former element is called the parent element, while the
latter is called the subelement. Specifically, Eq. (101) can be further expanded as
8 9
>
> x1 >
>
y
>
> >
>
>
> 1 >
>
> x2 >
>
> >
>
 
> >
x N1 0 N2 0 N3 0 N4 0 y2
< =
¼ (101)
y 0 N1 0 N2 0 N3 0 N4 > > x3 >
>
y3 >
>
> >
>
> >
>
x
> >
4
>
> >
>
> >
y4
: ;

Using the same interpolation functions, the element displacement model can be written as
u1 >
8 9
>
> v1 >
>
> >
>
> >
u
> >
2
>
> >
>
 
>
< >
u N1 0 N2 0 N3 0 N4 0 v2
=
¼ (102)
v 0 N1 0 N2 0 N3 0 N4 > > u3 >>
v >
> >
> u3 >
>
> >
> >
>
>
: 4 >
>
;
v4
254 Dynamical Systems - Analytical and Computational Techniques

where J denotes the Jacobi matrix while the interpolation functions are given by

1 1
N 1 ¼ ð1 þ ξÞð1 þ ηÞ, N 2 ¼ ð1 ξÞð1 þ ηÞ
4 4 (103)
1 1
N 3 ¼ ð1 þ ξÞð1 þ ηÞ, N 2 ¼ ð1 ξÞð1 þ ηÞ
4 4

As mentioned before, during the derivation of the element stiffness matrix and the equivalent
load vector, the derivative of the shape function and the integration in element surface or
volume in the Cartesian coordinate system are required. Since the shape functions adopted
herein are expressed in natural coordinates, therefore, derivative and integration transforma-
tion relationships are essential when isoparametric element is used.

1.13.7. Derivative and integral transformation

According to the law of partial differential,

∂Ni ∂N i ∂x ∂Ni ∂y
¼ þ ,
∂ξ ∂x ∂ξ ∂y ∂ξ
(104)
∂Ni ∂N i ∂x ∂N i ∂y
¼ þ ,
∂η ∂x ∂η ∂y ∂η

or in matrix form

∂N i > 2 ∂x ∂y 3> ∂N i > ∂Ni >

8 9 8 9 8 9
>
> >
< ∂ξ > 6 ∂ξ ∂ξ 7
> > > >
7 ∂x ∂x
= < = < =
¼6 ¼ J (105)
∂N > 4 ∂x ∂y 5> ∂N i > ∂Ni >
: i>
>
> > > >
> >
; : ; : ;
∂η ∂η ∂η ∂y ∂y

Inverse of Eq. (105) gives

∂Ni > ∂N i >

8 9 8 9
>
> >
< ∂x >= >
< ∂ξ > =
1
¼J (106)
∂N > ∂N >
: i> : i>
>
> >
>
; ;
∂y ∂η

where
2 3
4 4
∂x
2
∂y
3 X ∂N i X ∂N i
6 xi yi 7
6 ∂ξ ∂ξ 7 6
i¼1
∂ξ i¼1
∂ξ 7
J¼6 7¼6
6 7
4 ∂x ∂y 5 6 X4 4
7
∂N i X ∂N i 7
∂η ∂η
4 xi yi 5
∂η ∂η
i¼1 i¼1 (107)
∂N 1 ∂N2 ∂N 3 ∂N 4 x1 y1
2 32 3
6 ∂ξ ∂ξ ∂ξ ∂ξ 776 x2 y2 7
6 7
¼6
6
∂N 4 54 x3 y3 5
76 7
4 ∂N 1 ∂N2 ∂N 3
∂η ∂η ∂η ∂η x4 y4
 Solution of Differential Equations with Applications to Engineering Problems 255
http://dx.doi.org/10.5772/67539

For an infinitely small element, the area under the Cartesian coordinate system and the natural
coordinate system are related by

ds ¼ dxdy ¼ jJjdξdη, (108)

where jJj is the determinant of the Jacobian matrix J. Therefore, element stiffness matrix and
equivalent nodal load matrix in Eq. (91) can be transformed to
ð ð
e T e
K ¼ B DBtjJjdξdη, Pf ¼ N T f jJjdξdη
Ωe Ωe
ð (109)
PeS ¼ T
N TjJjdξdη
Sσ e

For solving the integral equation, usually the Gaussian integration method is employed. In
practice, both two and three integration points along each direction of integration are com-
monly used. Since the discretized system is usually overstiff, it is commonly observed that the
use of two integration points along each direction of integration will slightly reduce the
stiffness of the matrix and give better results as compared with the use of three integration
points. The use of exact integration is possible for some elements, but such approaches are
usually tedious and are seldom adopted. The advantage in using the exact integration is that
the integration is not affected by the shape of the element while the transformation as shown in
Eq. (109) may be affected if the poor shape of the element is poor. The author has developed
many finite element programs for teaching and research purposes which can be obtained at
[email protected]. The programs available include plane stress/strain problem, thin/
thick plate bending problem, consolidation in 1D and 2D (Biot), seepage problem, slope
stability problem, pile foundation problems and others.

1.14. Distinct element method

In practical applications, a limit equilibrium method based on the method of slices or method
of columns and strength reduction method based on the finite element method or finite
difference method are used for many types of stability problems. These two major analysis
methods take the advantage that the in situ stress field which is usually not known with good
accuracy is not required in the analysis. The uncertainties associated with the stress-strain
relation can also be avoided by a simple concept of factor of safety or the determination of the
ultimate limit state. In general, this approach is sufficient for engineering analysis and design.
If the condition of the system after failure has initiated is required to be assessed, these two
methods will not be applicable. Even if the in situ stress field and the stress-strain relation can
be defined, the post-failure collapse is difficult to be assessed using the conventional
continuum-based numerical method, as sliding, rotation and collapse of the slope involve very
large displacement or even separation without the requirement of continuity.
The most commonly used numerical methods for continuous systems are the FDM, the FEM
and the boundary element method (BEM). The basic assumption adopted in these numerical
methods is that the materials concerned are continuous throughout the physical processes.
This assumption of continuity requires that, at all points in a problem domain, the material
256 Dynamical Systems - Analytical and Computational Techniques

cannot be torn open or broken into pieces. All material points originally in the neighbourhood
of a certain point in the problem domain remain in the same neighbourhood throughout the
whole physical process. Some special algorithms have been developed to deal with material
fractures in continuum mechanics-based methods, such as the special joint elements by Good-
man [13] and the displacement discontinuity technique in BEM by Crouch and Starfield [5].
However, these methods can only be applied with limitations [21]:
1. large-scale slip and opening of fracture elements are prevented in order to maintain the
macroscopic material continuity;
2. the amount of fracture elements must be kept to relatively small so that the global stiffness
matrix can be maintained well-posed, without causing severe numerical instabilities; and
3. complete detachment and rotation of elements or groups of elements as a consequence of
deformation are either not allowed or treated with special algorithms.

Before a slope starts to collapse, the factor of safety serves as an important index in both the
LEM and SRM to assess the stability of the slope. The movement and growth after failure have
launched which is also important in many cases that cannot be simulated on the continuum
model, and this should be analyzed by the distinct element method (DEM).

In continuum description of soil material, the well-established macro-constitutive equations

whose parameters can be measured experimentally are used. On the other hand, a discrete
element approach will consider that the material is composed of distinct grains or particles that
interact with each other. The commonly used distinct element method is an explicit method
based on the finite difference principles which is originated in the early 1970s by a landmark
work on the progressive movements of rock masses as 2D rigid block assemblages [6]. Later,
the works by Cundall are developed to the early versions of the UDEC and 3DEC codes
[9, 10, 12]. The method has also been developed for simulating the mechanical behaviour of
granular materials [8], with a typical early code BALL [7], which later evolved into the codes of
the PFC group for 2D and 3D problems of particle systems (Itasca, 1995). Through continuous
developments and extensive applications over the last three decades, there has accumulated a
great body of knowledge and a rich field of literature about the distinct element method. The
main trend in the development and application of the method in rock engineering is
represented by the history and results of the code groups UDEC/3DEC. Currently, there are
many open source (Oval, LIGGGHTS, ESyS, Yade, ppohDEM, Lammps) as well as commercial
DEM programs, but in general, this method is still limited to basic research instead of practical
application as there are many limitations which include: (1) difficult to define and determine
the microparameters; (2) there are still many drawbacks in the use of matching with the macro
response to determine the microparameters; (3) not easy to set up a computer model; (4) not
easy to include structural element or water pressure; (5) extremely time consuming to perform
an analysis; and (6) postprocessing is not easy or trivial. It should also be noted that DEM can
be formulated by an energy-based implicit integration scheme which is the discontinuous
deformation analysis (DDA) method. This method is similar in many respect to the force-
based explicit integration scheme as mentioned previously.
 Solution of Differential Equations with Applications to Engineering Problems 257
http://dx.doi.org/10.5772/67539

In DEM, the packing of granular material can be defined from statistical distributions of grain
size and porosity, and the particles are assigned normal and shear stiffness and friction
coefficients in the contact relation. Two types of bonds can be represented either individually
or simultaneously; these bonds are referred to the contact and parallel bonds, respectively
(Itasca, 1995). Although the individual particles are solid, these particles are only partially
connected at the contact points which will change at different time step. Under low normal
stresses, the strength of the tangential bonds of most granular materials will be weak and the
material may flow like a fluid under very small shear stresses. Therefore, the behaviour of
granular material in motion can be studied as a fluid-mechanical phenomenon of particle flow
where individual particles may be treated as “molecules” of the flowing granular material. In
many particle models for geological materials in practice, the number of particles contained in
a typical domain of interest will be very large, similar to the large numbers of molecules.
One of the primary objectives of the particle model is the establishment of the relations between
microscopic and macroscopic variables/parameters of the particle systems, mainly through
micromechanical constitutive relations at the contacts. Compared with a continuum, particles
have an additional degree of freedom of rotation which enables them to transmit couple
stresses, besides forces through their translational degrees of freedom. At certain moment, the
positions and velocities of the particles can be obtained by translational and rotational move-
ment equations and any special physical phenomenon can be traced back from every single
particle interactions. Therefore, it is possible for DEM to analyze large deformation problems
and a flow process which will occur after slope failure has initiated. The main limitation of
DEM is that there is great difficulty in relating the microscopic and macroscopic variables/
parameters; hence, DEM is mainly tailored towards qualitative instead of quantitative analysis.
DEM runs according to a time-difference scheme in which calculation includes the repeated
application of the law of motion to each particle, a force-displacement law to each contact, and
a contact updating scheme. Generally, there are two types of contact in the program which are
the ball-wall contact and the ball-ball contact. In each cycle, the set of contacts is updated from
the known particles and known wall positions. Force-displacement law is firstly applied on
each contact, and new contact force is then calculated according to the relative motion and
constitutive relation. Law of motion is then applied to each particle to update the velocity, the
direction of travel based on the resultant force, and the moment and contact acting on the
particles. Although every particle is assumed as a rigid material, the behaviour of the contacts
is characterized using a soft contact approach in which finite normal stiffness is taken to
represent the stiffness which exists at the contact. The soft contact approach allows small
overlap between the particles which can be easily observed. Stress on particles is then deter-
mined from this overlapping through the particle interface.

1.15. General formulation of DEM

The PFC runs according to a time-difference scheme in which calculation includes the repeated
application of the law of motion to each particle, a force-displacement law to each contact, and
a contact updating a wall position. Generally, there are two types of contact exist in the
258 Dynamical Systems - Analytical and Computational Techniques

program which are ball-to-wall contact and ball-to-ball contact. In each cycle, the set of
contacts is updated from the known particle and the known wall position. The force-
displacement law is first applied on each contact. New contact force is calculated and replaces
the old contact force. The force calculations are based on preset parameters such as normal
stiffness, density, and friction. Next, a law of motion is applied to each particle to update its
velocity, direction of travel based on the resultant force, moment and contact acting on particle.
The force-displacement law is then applied to continue the circulation.

1.16. The force-displacement law

The force-displacement law is described for both the ball-ball and ball-wall contacts. The
contact arises from contact occurring at a point. For the ball-ball contact, the normal vector is
directed along the line between the ball centres. For the ball-wall contact, the normal vector is
directed along the line defining the shortest distance between the ball centre and the wall. The
contact force vector Fi is composed of normal and shear component in a single plane surface

Fi ¼ Fnij ðtÞ þ Fsij ðt þ ΔtÞ (110)

The force acting on particle i in contact with particle j at time t is given by

 
Fnij ðtÞ ¼ kn ri þ rj lij ðtÞ (111)

where rj and ri stand for particle i and particle j radii, lij(t) is the vector joining both centres of
the particles and kn represents the normal stiffness at the contact. The shear force acting on
particle i during a contact with particle j is determined by

Fsij ðt þ ΔtÞ ¼ minðFsij ðtÞ þ ks Δsij , f jFnij ðt þ ΔtÞjÞ (112)

where f is the particle friction coefficient, ks represents the tangent shear stiffness at the contact.
The new shear contact force is found by summing the old shear force (min Fij(t)) with the shear
elastic force. Δsij stands for the shear contact displacement-increment occurring over a time
step Δt.

Δsij ¼ vsij Δt (113)

where V sij is the shear component of the relative velocity at contact between particles i and j
over the time step Δt.

1.17. Law of motion

The motion of the particle is determined by the resultant force and moment acting on it. The
motion induced by resultant force is called translational motion. The motion induced by
resulting moment is rotational motion. The equations of motion are written in vector form as
follows:
 Solution of Differential Equations with Applications to Engineering Problems 259
http://dx.doi.org/10.5772/67539

- (Translational motion)
X
Fij þ mi g þ Fdi ¼ mi xni (114)
j

- (Rotational motion)
X
ri Fij þ Mdi ¼ I r θni (115)
j

where x″ i and θ″ i stand for the translational acceleration and rotational acceleration of particles
i. Ir stands for moment of inertia. Fdi and Mdi stand for the damping force and damping
moment. Unlike finite element formulation, there are now three degree of freedom for 2D
problem and six degree of freedom for 3D problems. In Cundall and Strack’s explicit integra-
tion distinct element approach, solution of the global system of equation is avoided by consid-
ering the dynamic equilibrium of the individual particles rather than solving the entire system
simultaneously. That means, Newton’s law of motion is applied directly. This approach also
avoids the generation and storage of the large global stiffness matrix that will appear in finite
element analysis. On the other hand, the implicit DDA approach will generate a global stiff-
ness matrix which is even larger than that in finite element analysis, as the rotation is involved
directly in the stiffness matrix.
In a typical DEM simulation, if there is no yield by contact separation or frictional sliding, the
particles will vibrate constantly and the equilibrium is difficult to be achieved. To avoid this
phenomenon which is physically incorrect, numerical or artificial damping is usually adopted
in many DEM codes, and the two most common approaches to damping are the mass
damping and non-viscous damping. For mass damping, the amount of damping that each
particle “feels” is proportional to its mass, and the proportionality constant depends on the
eigenvalues of the stiffness matrix. This damping is usually applied equally to all the nodes. As
this form of damping introduces body forces, which may not be appropriate in flowing
regions, it may influence the mode of failure. Alternatively, Cundall [11] proposed an alterna-
tive method where the damping force at each node is proportional to the magnitude of the out-
of-balance-force, with a sign to ensure that the vibrational modes are damped rather than the
steady motion. This form of damping has the advantage that only accelerating motion is
damped and no erroneous damping forces will arise from steady-state motion. The damping
constant is also non-dimensional and the damping is frequency independent. As suggested by
Itasca [20], an advantage of this approach is that it is similar to the hysteretic damping, as the
energy loss per cycle is independent of the rate at which the cycle is executed. While damping
is one way to overcome the non-physical nature of the contact constitutive models in DEM
simulations, it is quite difficult to select an appropriate and physically meaningful value for the
damping. For many DEM simulations, particles are moving around each other and the dom-
inant form of energy dissipation is for frictional sliding and contact breakages. The choice of
260 Dynamical Systems - Analytical and Computational Techniques

damping may affect the results of computations. Currently, most of the DEM codes allow the
use of automatic damping or manually prescribed the damping if necessary.
To capture the inherent non-linearity behaviour of the problem (with generation and removal
of contacts, non-linear contact response and stress-strain behaviour and others), the displace-
ment and contact forces in a given time step must be small enough so that in a single time step,
the disturbances cannot propagate from a particle further than its nearest neighbours. For most
of the DEM programs, this can be achieved automatically and the default setting is usually
good enough for normal cases. It is, however, sometimes necessary to manually adjust the time
step in some special cases when the input parameters are unreasonably high or low. Most of
the DEM codes use the central difference time integration algorithm which is a second-order
scheme in time step.

1.18. Measuring logic

If the local results in DEM are analyzed, it is found that there will be large fluctuations with
respect to both locations and time. Such results are not surprising, as the results are highly
sensitive to the interaction between particles and hence the time step under which the results
are monitored. It can be viewed that such local results can be meaningless unless the results are
monitored over a long time span or region. A number of quantities in a DEM model are
defined with respect to a specified measurement circle. These quantities include coordinate
number, porosity, sliding fraction, stress and strain rate. The coordination number and stress
are defined as the average number of contacts per particle. Only particles with centroids that
are contained within the measurement circle are considered in computation. In order to
account for the additional area of particles that is being neglected, a corrector factor based on
the porosity is applied to the computed value of stress.
Since measurement circle is used, stress in particle is described as the two in-plane force acting
on each particle per volume of particle. Average stress is defined as the total stress in particle
divided by the volume of measurement circle. Thus, shape of particle is regardless of the
average stress measurement because the reported stress is easily scaled by volume unity. The
reported stress is interpreted as the stress per volume of measurement circle.

1.19. Discussion and conclusion

There are also various publications on the numerical solutions of differential equations, and
the readers are suggested to the works of Lee and Schiesser [24], Jovanoic and Suli [22], Veiga
et al. [37], Sewell [35], Morton and Mayers [27], Logg et al. [25], Holmes [17], Lui [26], Lapids
and Pinder [23] and Iserles [19]. It is impossible for the author to cover every available
analytical or numerical method; hence, the author has chosen some methods that are actually
used for teaching and research. The readers are strongly encouraged to consult the numerous
resources available in various books and publications. There are still new developments
available for the solutions of specific differential equations in large-scale problems, and this is
also the current trend in the development of differential equation solution.
 Solution of Differential Equations with Applications to Engineering Problems 261
http://dx.doi.org/10.5772/67539

Due to the importance of the solution of differential equations, there are other important
numerical methods that are used by different researchers but are not discussed here, which
include the finite difference and boundary element methods (computer codes for learning can
also be obtained from the author). Differential equations rely on the Taylor’s series, and the
derivatives in the differential equation can be replaced with finite difference approximations
on a discretized domain. This will result in a system of algebraic equations that can be solved
implicitly or explicitly. There are various ways to form the derivatives, and the most common
methods are the forward difference, backward difference and the central difference schemes.
While the finite difference methods may be more suitable for different types of differential
equations, this method is less convenient to deal with irregular boundary conditions as com-
pared with the finite element method. For highly irregular domain where it is not easy to form
a nice discretization, the finite element method will also be much easier and natural to deal
with for such condition. In this respect, it is not surprising that many engineering programs are
written by the use of the finite element method than the finite difference method.
The boundary element method (BEM) is another numerical method for solving linear partial
differential equations which can be formulated as integral equations. The boundary element
method uses the given boundary conditions to fit boundary values into the integral equation.
In the post-processing stage, the integral equation will be used to calculate the solution directly
at any given point inside the solution domain numerically. BEM is applied to problems for
which Green’s functions can be calculated, thus this method is initially designed for problems
in linear homogeneous media. The dimension of the problem will then be reduced by one. For
example, two-dimensional problem will be effectively reduced to one-dimensional problem
along the boundary, and this will greatly improve the efficiency of computation. The require-
ment from the boundary element method imposes considerable restrictions on the range and
generality of problems to which the boundary element method can usefully be applied. There
are some new developments to the boundary element method so that it can be used for non-
linear problem or problems with several major materials (problems with random distribution
of material properties are still not applicable). The fundamental solutions are often difficult to
integrate. One important property of boundary element analysis is the solution of a fully
populated matrix as compared with that in the finite element/difference method. For compli-
cated problems, the boundary element will lose its advantage as compared with other numer-
ical methods. Due to the various limitations, there are only limited boundary element
programs available to the researchers. Interested readers can consult the works of Banerjee
[1], Brebbia et al. [3] and Trevelyan [36]. It appears that there are less interest in the use and
development of the boundary element method in the recent years, due to the various limita-
tions of this method in general non-linear non-homogeneous problem.

In history, various techniques have been developed for ordinary differential equations and
partial differential equations under different boundary conditions. While these tricks appear to
be elegant, they are not readily adopted for normal engineering use due to various limitations.
Being an engineer, the author seldom adopted the methods as outlined in this chapter in actual
applications (but do adopt for teaching), except the numerical methods as outlined in this
chapter. At present, there are many proprietary or open source finite elements or distinct
262 Dynamical Systems - Analytical and Computational Techniques

element codes being used for many complicated real problems. The computer codes (usually
in Fortran or C) are usually difficult to be read (if available), and the computer codes for all the
partial differential forms (including some extended formats) that have been discussed in this
chapter can be readily available from the author for learning purposes. There are also very
powerful and general finite element tools or differential equations solver such as FreeFem++,
Comsol, Matlab, Mathematica, Maple and Maxima which are used by many scientists and
engineers [39, 40]. The use of parallel computing is also strongly influenced by the needs to
solve complicated partial differential equations over large solution domain.

Author details

Cheng Yung Ming

Address all correspondence to: [email protected]

Department of Civil and Environmental Engineering, Hong Kong Polytechnic University,

Hong Kong SAR, China

References

[1] Banerjee P.K. (editor). Developments in boundary element methods, Vols. 1,2,3, Applied
Science Publishers, London, 1979.
[2] Bertanz R. Fourier series and numerical methods for partial differential equations, John
Wiley, USA, 2010.
[3] Brebbia C.A., Telles J.C.F. and Wrobel L.C. Boundary element techniques, Springer-
Verlag, Berlin, 1984.
[4] Bronson R. and Costa G.B. Differential equations, McGraw-Hill, USA, 2006.

[5] Crouch S.L. and Starfield A.M. (editors). Boundary element methods in solid mechanics,
George Allen & Unwin, London, 1983.
[6] Cundall P. A. A computer model for simulating progressive, large-scale movements in
blocky rock systems. In: Proceedings of the International Symposium on Rock Mechanics.
Nancy, France, 1971: 129–136.
[7] Cundall P.A. Ball – A computer program to model granular medium using the distinct
element method, Technical note TN-LN-13, Advanced Technology Group, Dames and
Moore, London, 1978:129–163.
[8] Cundall P.A. and Strack O.D.L. A discrete model for granular assemblies, Geotechnique,
1979, 29(1):47–65.
 Solution of Differential Equations with Applications to Engineering Problems 263
http://dx.doi.org/10.5772/67539

[9] Cundall P.A. UDEC-A generalized distinct element program for modelling jointed rock,
Final Tech. Rep. Eur. Res. Office (US Army Contract DAJA37-79-C-0548), Springer, Neth-
erland, 1980.
[10] Cundall P.A. and Hart R.D. Development of generalized 2-D and 3-D distinct element pro-
grams for modeling jointed rock, Misc. Paper SL-85-1, US Army Corps of Engineers, USA,
1985.
[11] Cundall, P.A. Distinct element models of rock and soil structure, in Analytical and
Computational Methods in Engineering Rock Mechanics, Brown, E.T. ed., George Allen
and Unwin, London, 1987:129–163.
[12] Cundall P.A. and Hart D.H. Numerical modelling of discontinua, Engineering with
Computers, 1992, 9:101–11.
[13] Goodman, R. E. Methods of geological engineering in discontinuous rocks. West Publish-
ing Company, San Francisco, CA, USA, 1976.

[14] Greenberg M.D. Solutions manual to accompany ordinary differential equations, John
Wiley, USA, 2012.

[15] Haberman R. Applied partial differential equations, Pearson Education, USA, 2005.
[16] Hillen T., Leonard I.E. and Roessel H.V. Partial differential equations, John Wiley, USA, 2012.
[17] Holmes M.H. Introduction to numerical methods in differential equations, Springer,
USA, 2007.

[18] Holzner S. Differential equations for dummies, John Wiley, USA, 2008.
[19] Iserles A. A first course in the numerical analysis of differential equations, Cambridge
University Press, UK, 2009.
[20] Itasca Consulting Group Inc, a company based in US. PFC3D 3.1, User Guide, 2004.
[21] Jing L.R., Stephansson O. and Erling N. Study of rock joints under cyclic loading condi-
tions, Rock Mechanics and Rock Engineering, 1993, 26, 3:215–232.

[22] Jovanoic B.S. and Suli E. Analysis of finite difference schemes, Springer, USA, 2014.
[23] Lapids L. and Pinder G.F. Numerical solution of partial differential equations in science
and engineering, John Wiley, Italy, 1999.
[24] Lee H.J. and Schiesser W.E. Ordinary and partial differential equation routines in C, C++,
Fortran, Java, Maple, and MATLAB, Chapman and Hall, USA, 2004.
[25] Logg A., Mardal K.A. and Wells G.N. Automated solution of differential equations by the
finite element method, Springer, France, 2012.
[26] Lui S.H. Numerical analysis of partial differential equations, John Wiley, USA, 2011.

[27] Morton K.W. and Mayers D.F. Numerical solution of partial differential equations, Cam-
bridge University Press, UK, 2005.
264 Dynamical Systems - Analytical and Computational Techniques

[28] Nagle R.K., Saff E.B. and Snider A.D. Fundamentals of differential equations and bound-
ary value problems, Addison Wesley, USA, 2012.
[29] Polyanin A.D. and Nazaikinskii V.E. Handbook of linear partial differential equations for
engineers and scientists, 2nd edition, CRC Press, USA, 2016.
[30] Polyallin A.D., Zaitsev V.F. and Moussiaux A. Handbook of first order partial differential
equations, Taylor and Francis, UK, 2002.

[31] Polyanin A.D. and Zaitsev V.F. Handbook of nonlinear Partial Differential Equations, 2nd
edition, Chapman & Hall, USA, 2004.

[32] Rao S.S. The finite element method in engineering, Butterworth Heinemann, London,
2011.
[33] Salsa S. Partial differential equations in action - from modelling to theory, Springer, Italy,
2008.
[34] Soare M.V., Teodorescu P.P. and Toma I. Ordinary differential equations with applications
to mechanics, Springer, Netherland, 2000.
[35] Sewell G. The numerical solution of ordinary and partial differential equations, John
Wiley, USA, 2005.

[36] Trevelyan J. Boundary elements for engineers: theory and applications, Computational
Mechanics Publications, Boston, USA, 1994.

[37] Veiga L.B.D., Lipnikov K. and Manzini G. The mimetic finite difference method for
elliptic problems, Springer, Switzerland, 2014.
[38] Zienkiewicz O.C., Taylor R.L. and Zhu J.Z. The finite element method: Its basis and
fundamentals, 6th edition, Butterworth Heinemann, London, 2011.
[39] Comsol, https://www.comsol.com/
[40] FreeFem++, http://www.freefem.org/

[41] H Weber, Paul du Bois-Reymond, Mathematische Annalen 35 (1890), 457–462