0% found this document useful (0 votes)

146 views

Introduction To Nonlinear Finite Element Analysis

Uploaded by

Kevin Arias

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

146 views

Introduction To Nonlinear Finite Element Analysis

Uploaded by

Kevin Arias

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 48

2

The Finite Element Method:

A Review

2.1 Introduction
The main ideas of the Þnite element method were presented in Chapter 1. To
summarize, the Þnite element method has the following three basic features:
1. Divide the whole (i.e. domain) into parts, called Þnite elements.
2. Over each representative element, develop the relations among the
secondary and primary variables (e.g. “forces” and “displacements”,
“heats” and “temperatures”, and so on).
3. Assemble the elements (i.e. combine the relations of all elements) to obtain
the relations between the secondary and primary variables of the whole
system.
In the present chapter, we review the basic steps of the Þnite element model
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

development as applied to one- and two-dimensional problems described by

typical second-order diﬀerential equations. The main objective is to familiarize
the reader with the speciÞc steps involved in the Þnite element formulation and
its applications. Readers who are already familiar with the author’s approach
to Þnite element modeling may skip this chapter.

2.2 One-Dimensional Problems

2.2.1 Governing Diﬀerential Equation
Consider the diﬀerential equation
µ ¶
d du
− a + cu = f for 0<x<L (2.2.1)
dx dx
where a = a(x), c = c(x), and f = f (x) are the data (i.e. known quantities) of
the problem, and u(x) is the solution to be determined. The data depends on
the material properties, geometry, and “loads” or source. The equation arises
in a number of Þelds, and it must be solved subject to appropriate boundary
conditions. Table 2.2.1 contains a list of Þelds, by no means exhaustive, in
which Eq. (2.2.1) arises.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
14 NONLINEAR FINITE ELEMENT ANALYSIS

2.2.2 Finite Element Approximation

The domain (0, L) of the problem consists of all points between x = 0 and
x = L. The points x = 0 and x = L are the boundary points of the total
domain. In the Þnite element method, the domain (0, L) is divided into a set of
intervals. A typical interval, called a Þnite element, is of length he and located
between points x = xa and x = xb , where xa and xb denote the coordinates of
the end points of the Þnite element with respect to the coordinate x.

Table 2.2.1 List of Þelds in which the model equation (2.2.1) arises, with
meaning of various parameters and variables (see the bottom
of the table for the meaning of some parameters*).

Field of Primary Problem data Secondary

study variable variable
u a c f Q
Heat Tempe- Thermal Surface Heat Heat
transfer rature conductance convection generation
T − T∞ kA Apβ f Q

Flow Fluid- Permea- InÞl- Point

through head bility tration source
porous
medium φ µ 0 f Q

Flow Pressure Pipe Point

through resistance source
pipes P 1/R 0 0 Q
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

Flow of Velocity Viscosity Pressure Shear

viscous gradient stress
ßuids vz µ 0 −dP/dx σxz
Elastic Displa- Tension Transverse Point
cables cement force force
u T 0 f P
Elastic Displa- Axial Axial Point
bars cement stiﬀness force force
u EA 0 f P

Torsion Angle of Shear Torque

of twist stiﬀness
bars θ GJ 0 0 T
Electro- Electrical Dielectric Charge Electric
statics potential constant density ßux
φ ε 0 ρ E

*k = thermal conductance; β = convective Þlm conductance; p = perimeter; P = pressure or

force; T∞ = ambient temperature of the surrounding ßuid medium; R = 128µh/(πd4 ) with
µ being the viscosity; h, the length and d the diameter of the pipe; E = Young’s modulus;
A = area of cross-section; J = polar moment of inertia.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 15

In the Þnite element method, we seek an approximate solution to Eq.

(2.2.1) over each Þnite element; a typical Þnite element is shown in Figure
2.2.1. The Þnite element approximation ueh (x) is sought in the form
u(x) ≈ ueh (x) = ce1 ϕe1 (x) + ce2 ϕe2 (x) + . . . + cen ϕen (x)
n
X
= cej ϕej (x) (2.2.2)
j=1

where ϕej (x) are functions to be selected and cej are constants to be determined
such that Eq. (2.2.2) satisÞes the diﬀerential equation (2.2.1) and appropriate
end conditions over the element. Since there are n unknown parameters, we
need n relations to determine them. Substituting the approximate solution
(2.2.2) into the left-hand side of Eq. (2.2.1), we obtain an expression that, in
general, will not be equal to the right-hand side of the equation, f(x). The
diﬀerence between the two sides of the equation is called the residual
µ ¶
d due
− a h + cueh (x) − f (x) ≡ Re (x, ce1 , ce2 , . . . , cen ) 6= 0 (2.2.3)
dx dx
We wish to determine cej (j = 1, 2, · · · , n) such that the residual is zero, in
some sense, over the element.
One way of making the residual zero is in weighted-integral sense
Z xb
wie (x)Re (x, c1 , c2 , . . .) dx = 0, i = 1, 2, . . . , n (2.2.4)
xa

where wie (i = 1, 2, . . . , n) are weight functions. Equation (2.2.4) provides a

set of n algebraic relations among the parameters cej (j = 1, 2, . . . , n). The set
{w1e (x), w2e (x), . . . , wne (x)} must be linearly independent so that the algebraic
equations (2.2.4) are also linearly independent and invertible.
There are other choices of wie that may be used. In the present study
we take wie (x) to be the same as the approximation functions ϕei (x). This
particular choice is known as the Galerkin method. Different choice of the
weight functions will result in a different set of algebraic equations or different
Þnite element models of the same differential equation.

f(x)
⎛ du ⎞ 1 2 ⎛ du ⎞
⎜− a ⎟ = Qa ⎜a ⎟ = Qb
⎝ dx ⎠ x = x ⎝ dx ⎠ x = xb
a u(xa) = u1 u(xb) = u2

Figure 2.2.1 A typical Þnite element in one dimension.

Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
16 NONLINEAR FINITE ELEMENT ANALYSIS

2.2.3 Derivation of the Weak Form

To weaken the continuity required of ueh (x), we must trade some of the
differentiation in (2.2.4) from ueh to wie such that both ueh and wie are
differentiated equally, once each in the present case. The resulting integral
form is termed the weak form of Eq. (2.2.1) because it allows approximation
functions with less (or weaker) continuity (or differentiability).
A three-step procedure of constructing the weak form of Eq. (2.2.1) is
presented next.
Step 1. The Þrst step is to write the weighted-residual statement as in Eq.
(2.2.4) Z xb ∙ µ ¶ ¸
d due
0= wie − a h + cueh − f dx (2.2.5)
xa dx dx

Step 2. The second step is to trade diﬀerentiation from ueh to wie , using
integration by parts. We obtain
Z xb µ ¶ ∙ ¸xb
dwie dueh dueh
0= a + cwie ueh − wie f dx − wie · a (2.2.6)
xa dx dx dx xa

Step 3. Examining the boundary term appearing in the weak form (2.2.6),
namely, the expression ∙ ¸
e du xb
wi · a
dx xa
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

The coeﬃcient of the weight function wie in the boundary expression, a(du/dx),
is called the secondary variable, and its speciÞcation constitutes the natural
or Neumann boundary condition. The primary variable is the dependent
unknown of the diﬀerential equation, u, in the same form as the weight
function in the boundary expression (i.e. replace wie with u). The speciÞcation
of a primary variable on the boundary constitutes the essential or Dirichlet
boundary condition. For the model equation at hand, the primary and
secondary variables are

du
Primary variable: u Secondary variable: a ≡Q (2.2.7)
dx
In writing the Þnal weak form, we denote the secondary variables at the
ends of the element as
µ ¶ µ ¶
du du
Qea = − a , Qeb = a (2.2.8)
dx xa dx xb

The primary and secondary variables at the nodes are shown on the typical
element in Figure 2.2.1. Students of engineering who have taken a course in
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 17

mechanics of deformable bodies (or strength of materials) recognize that this

Þgure shows the free-body diagram of a typical but arbitrary portion of a bar,
with Qea and Qeb denoting the axial forces; Qea is a compressive force while Qeb
is a tensile force (algebraically, both are in the positive x direction, as shown
in Figure 2.2.1). For heat conduction problems, Qea denotes the heat input at
the left end and Qeb the heat output from the right end of the element. With
the notation in (2.2.8), the Þnal expression for the weak form is given by
Z xb µ ¶
dwie dueh
0= a + cwie ueh − wie f dx − wie (xa )Qea − wie (xb )Qeb (2.2.9)
xa dx dx

This completes the three-step procedure of constructing the weak form.

Remarks
1. The weak form in (2.2.9) contains two types of expressions: those
containing both wie and ueh , and those containing only wie . The expression
containing both wie and ueh is called the bilinear form (i.e. linear in wie and
linear in ueh ):
Z xb µ ¶
dwie dueh
B(wie , ueh ) ≡ a + cwie ueh dx (2.2.10)
xa dx dx

Similarly, the expression containing only wie (but not ueh ) is called the linear
form: Z xb
`(wie ) = wie f dx + wie (xa )Qea + wie (xb )Qeb
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

(2.2.11)
xa

More formally, we say that B(u, v) is linear in both u and v, and `(v) is
linear in v, if and only if the following conditions hold:

B(c1 u1 + c2 u2 , v) = c1 B(u1 , v) + c2 B(u2 , v)

(2.2.12a)
B(u, c1 v1 + c2 v2 ) = c1 B(u, v1 ) + c2 B(u, v2 )

`(c1 v1 + c2 v2 ) = c1 `(v1 ) + c2 `(v2 ) (2.2.12b)

where c1 and c2 are constants and u, v, u1 , u2 , v1 , and v2 are dependent
variables.
2. In view of the above deÞnitions, the weak form (2.2.9) of Eq. (2.2.1) can
now be expressed as
B(wie , ueh ) = `(wie ) (2.2.13)
which is called the variational problem associated with Eq. (2.2.1). As will
be seen later, the bilinear form B(wie , ueh ) results directly in the element
coeﬃcient matrix, and the linear form `(wie ) leads to the right-hand side
column vector of the Þnite element equations.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
18 NONLINEAR FINITE ELEMENT ANALYSIS

3. Those who have a background in applied mathematics or solid and

structural mechanics will appreciate the fact that the weak form (2.2.9)
or the variational problem (2.2.13) is nothing but the statement of the
principle of the minimum total potential energy Π(ueh ) (applied to the bar
element):
0 = δΠ = B(δueh , ueh ) − `(δueh )
where δ is the variational symbol and Π(ueh ) is the quadratic functional
deÞned by
1
Π(ueh ) = B(ueh , ueh ) − `(ueh ) (2.2.14)
2 " #
Z xb µ ¶
a duei 2 c e 2
= + (uh ) − ueh f dx − ueh (xa )Qea − ueh (xb )Qeb
xa 2 dx 2
(2.2.15)

Equation (2.2.14) holds only when the bilinear form B(w, u) is symmetric
in u and w,
B(w, u) = B(u, w) (2.2.16)
and `(u) is linear in u. The expression 12 B(ueh , ueh ) represents the elastic
strain energy stored in the bar Þnite element and `(ueh ) represents the work
done by applied distributed force f (x) and point loads Qea and Qeb .

2.2.4 Interpolation Functions

Recall that the weak form over an element is equivalent to the diﬀerential
equation, and it contains the end conditions on the “forces” Qei (see Figure
2.2.1). Therefore, the approximate solution ueh (x) should be selected such that
the diﬀerentiability (or continuity) conditions implied by the weak form are
met and the end conditions on the primary variables u(xi ) = uei are satisÞed.
Since the weak form contains the Þrst-order derivative of ueh , any function
with a non-zero Þrst derivative would be a candidate for ueh . Thus, the Þnite
element approximation ueh of u(x) can be an interpolant, that is, must be equal
to uea at xa and ueb at xb . Thus, a linear polynomial (see Figure 2.2.2)

ueh (x) = ce1 + ce2 x (2.2.17)

is admissible if we can select ce1 and ce2 such that

ueh (xa ) = ce1 + ce2 xa = uea , ueh (xb ) = ce1 + ce2 xb = ueb

or
∙ ¸½ e ¾ ½ e¾
1 xa c u ueb xb − uea xa ueb − uea
1 = a → ce1 = , ce2 = (2.2.18)
1 xb ce2 ueb xb − xa xb − xa
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 19

Substitution of (2.2.18) for cei into (2.2.17) yields

2
X
ueh (x) = Le1 (x)ue1 + Le2 (x)ue2 = Lej (x)uej (2.2.19)
j=1

where
xb − x x − xa
Le1 (x) = , Le2 (x) = (2.2.20)
xb − xa xb − xa
are the linear Lagrange interpolation functions, and

ue1 = uea , ue2 = ueb (2.2.21)

are the nodal values of ueh (x) at x = xa and x = xb , respectively. Note that
Lei (x) satisfy the interpolation property
½
1, if i = j
Lei (xej ) = (2.2.22)
0, if i = j

where xe1 = xa and xe2 = xb (see Figure 2.2.2). In addition, the Lagrange
interpolation functions satisfy the property, known as the “partition of unity”:
n
X
Lej (x) = 1 (2.2.23)
j=1
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

u ( x) ≈ c1 + c 2 x
u(x) True solution
= u1e L1e ( x) + u 2e Le2 ( x )
u1e L1e ( x)

u 2e Le2 ( x )
x = xa = x e
1 u e
2

x = xb = x2e
u1e
1 2
x
x he

Figure 2.2.2 Linear Þnite element approximation over an element.

If we wish to approximate u(x) with a quadratic polynomial, we write

ueh (x) = ce1 + ce2 x + ce3 x2 (2.2.24)

Since there are three parameters ce1 , ce2 , and ce3 , we must identify one more
nodal point in the element to express all three c, s in terms of the values of ueh
at three nodes. Of course, we can also carry the nodal values uea and ueb (so
that they can be used to join adjacent elements) and the parameter ce3 as the
unknowns of the approximation. Identifying the third node at the center of
the element [see Figure 2.2.3(a)], we can write

ueh (xe1 ) ≡ ue1 = ce1 + ce2 xe1 + ce3 (xe1 )2

ueh (xe2 ) ≡ ue2 = ce1 + ce2 xe2 + ce3 (xe2 )2 (2.2.25)
ueh (xe3 ) ≡ ue3 = ce1 + ce2 xe3 + ce3 (xe3 )2

where
he
xe1 = xa , xe2 = xa + , xe3 = xa + he = xb (2.2.26)
2
Solving Eqs. (2.2.33) for cei in terms of uei , we obtain
3
X
ueh (x) = Le1 (x)ue1 + Le2 (x)ue2 + Le3 (x)ue3 = Lej (x)uej (2.2.27)
j=1

where Lei (x) are the quadratic Lagrange interpolation functions [see Figure
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

2.2.3(b)]
µ ¶µ ¶
x − xe2 x − xe3
Le1 (x) =
xe − xe2 xe1 − xe3
µ 1 ¶ µ ¶
x − xe1 x − xe3
Le2 (x) = (2.2.28)
xe − xe1 xe2 − xe3
µ 2 ¶ µ ¶
e x − xe1 x − xe2
L3 (x) =
xe3 − xe1 xe3 − xe2
Higher-order Lagrange interpolation of u(x) can be developed along the
similar lines. Thus, an (n − 1)st degree Lagrange interpolation of u(x) can be
written as
n
X
ueh (x) = Le1 (x)ue1 + Le2 (x)ue2 + . . . + Len (x)uen = Lej (x)uej (2.2.29)
j=1

where the Lagrange interpolation functions of degree n − 1 are given by

n
Ã !
Y x − xei
Lej (x) = (2.2.30)
i=1,i6=j
xej − xei
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 21

e x3e
x 2
e
x 1
x 1 x 3
he 2
(a)

Lei (x )

L1e ( x ) Le2 ( x )
Le3 ( x )

1.0 1.0 1.0

x
1 2 3
(b)

Figure 2.2.3 (a) Quadratic Þnite element. (b) Quadratic Lagrange

interpolation functions.

The Þnite element solution ueh (x) must fulÞll certain requirements in order
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

that it be convergent to the actual solution u(x) as the number of elements

(h reÞnement) or the degree of the polynomials (p reÞnement) is increased.
These are:
1. The approximate solution should be continuous and diﬀerentiable as
required by the weak form.
2. It should be a complete polynomial, that is, include all lower-order terms
up to the highest order term used.
3. It should be an interpolant of the primary variables at the nodes of the
Þnite element (at least interpolate the solution at the end points).
The reason for the Þrst requirement is obvious; it ensures that every term
of the governing equation has a non-zero contribution to the coeﬃcient matrix.
The second requirement is necessary in order to capture all possible states, that
is, constant, linear and so on, of the actual solution. For example, if a linear
polynomial without the constant term is used to represent the temperature
distribution in a one-dimensional system, the approximate solution can never
be able to represent a uniform state of temperature Þeld in the element. The
third requirement is necessary in order to enforce continuity of the primary
variables at the end points where the element is connected to other elements.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
22 NONLINEAR FINITE ELEMENT ANALYSIS

2.2.5 Finite Element Model

Substitution of (2.2.29) into (2.2.9) will give the necessary algebraic equations
among the nodal values uei and Qei of the element. In order to formulate
the Þnite element model based on the weak form (2.2.9), it is not necessary
to decide a priori the degree of approximation of ueh (x). The model can be
developed using an arbitrary degree of interpolation. For n > 2, the weak
form in Eq. (2.2.9) must be modiÞed to include non-zero secondary variables,
if any, at interior nodes:
Z xb µ ¶ Z xb n
X
dw dueh
0= a + cwueh dx − wf dx − w(xei ) Qei (2.2.31)
xa dx dx xa i=1

where xei is the global coordinate of the ith node of element Ωe . If nodes 1 and
n denote the end points of the element, then Qe1 and Qen represent the unknown
point sources, and all other Qei are the point sources at nodes 2, 3, . . . , n, which
are always known.
Substituting Eq. (2.2.29) for ueh and w1e = Le1 , w2e = Le2 , . . . , wie =
Li , . . . , wne = Len into the weak form (2.2.31), we obtain n algebraic equations.
e

The ith algebraic equation can be written as

⎡ ⎛ ⎞ ⎛ ⎞ ⎤
Z xb n n
dLe X dLej X
0= ⎣a i ⎝ uej ⎠ + cLe ⎝
i uej Lej ⎠ − Lei f ⎦ dx
xa dx j=1 dx j=1
n
X
− Lei (xej )Qej
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

j=1
" Ã ! #
X Z xb
n
dLe dLej
Z xb
= a i + cLei Lej dx uej − Lei f dx − Qei
j=1 xa dx dx xa

Xn
e e
0= Kij uj − fie − Qei (2.2.32)
j=1

for i = 1, 2, . . . , n, where
Z xb Ã ! Z xb
e dLei dLej
Kij = a + cLei Lej dx = B(Lei , Lej ), fie = f Lei dx
xa dx dx xa
(2.2.33)
Note that the interpolation property (2.2.22) is used to write
n
X
Lej (xei )Qej = Qei (2.2.34)
j=1

In matrix notation, these algebraic equations can be written as

[K e ]{ue } = {f e } + {Qe } (2.2.35)

The matrix [K e ] is called the coeﬃcient matrix, or stiﬀness matrix in structural

mechanics applications. The column vector {f e } is the source vector, or force
vector in structural mechanics problems.
Equation (2.2.35) has 2n unknowns (ue1 , ue2 , . . . , uen ) and (Qe1 , Qe2 , . . . , Qen ),
called primary and secondary element nodal degrees of freedom; hence, it
cannot be solved without having an additional n conditions. Some of these
equations are provided by the boundary conditions of the problem and the
remaining by the balance of the secondary variables Qei at nodes common
to elements. The balance of equations can be implemented by putting the
elements together (i.e. assembling the element equations) to form the total
system. Upon assembly and imposition of boundary conditions, we shall
obtain exactly the same number of algebraic equations as the total number of
unknown primary (uei ) and secondary (Qei ) degrees of freedom.
The coeﬃcient matrix [K e ], which is symmetric, and source vector {f e }
can be evaluated for a given interpolation and data (a, c, and f ). When a, c,
and f are functions of x, it may be necessary to evaluate [K e ] and {f e } using
numerical integration. We will discuss the numerical integration concepts in
the sequel. Here we give the exact values of [K e ] and {f e } for linear as well
as quadratic interpolations for element-wise constant values of a, c, and f.
Suppose that ae , ce , and fe denote the element-wise constant values of a(x),
c(x), and f (x). Then the following matrices can be derived by evaluating the
integrals exactly.
Linear element
x̄ x̄
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

L1 (x̄) = 1 − , L2 (x̄) =
he he
µ ∙ ¸ ∙ ¸¶ ½ e ¾ ½ ¾ ½ ¾
ae 1 −1 ce he 2 1 u 1 fe he 1 Qe1
+ = + (2.2.36)
he −1 1 6 1 2 ue2 2 1 Qe2

Quadratic element
µ ¶µ ¶ µ ¶ µ ¶
2x̄ x̄ 4x̄ x̄ x̄ 2x̄
L1 (x̄) = 1 − 1− , L2 (x̄) = 1− , L3 (x̄) = − 1−
he he he he he he

Ã ⎡ ⎤ ⎡ ⎤ ⎧ ⎫
7 −8 1 4 2 −1 ! ⎨ ue1 ⎬
ae ⎣ ce he
−8 16 −8 ⎦ + ⎣ 2 16 2⎦ ue
3he 30 ⎩ 2e ⎭
1 −8 7 −1 2 4 u3
⎧ ⎫ ⎧ e⎫
1 Q
fe he ⎨ ⎬ ⎨ 1e ⎬
= 4 + Q2 (2.2.37)
6 ⎩ ⎭ ⎩ e⎭
1 Q3

Note that the contribution of uniform source to the nodes in a quadratic

element is non-uniform, that is, fie 6= fe he /3.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
24 NONLINEAR FINITE ELEMENT ANALYSIS

2.3 Two-Dimensional Problems

2.3.1 Governing Differential Equation
Consider the problem of Þnding u(x, y) such that the following partial
differential equation is satisÞed
∙ µ ¶ µ ¶¸
∂ ∂u ∂ ∂u
− axx + ayy = f (x, y) in Ω (2.3.1)
∂x ∂x ∂y ∂y
where Ω is a two-dimensional domain with boundary Γ. Here axx and ayy are
material coefficients in the x and y directions, respectively, and f (x, y) is the
known source. For example, in a heat transfer problem, u denotes temperature
T , axx and ayy denote the conductivities, kxx and kyy , and f is the internal
heat generation. For an isotropic medium, we set kxx = kyy = k. Similarly, for
a ground water ßow problem u denotes the water head (i.e. velocity potential),
axx and ayy are the permeabilities in the x and y directions, respectively, and
f(x, y) is distributed water source. Equation (2.3.1) also arises in other Þelds
of science and engineering, and some of them are listed in Table 2.3.1.
Equation (2.3.1) must be solved in conjunction with speciÞed boundary
conditions of the problem. The following two types of boundary conditions
are assumed:
u = û(s) on Γu (2.3.2)
µ ¶
∂u ∂u
qn = axx nx + ayy ny + qc = q̂(s) on Γq (2.3.3)
∂x ∂y
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

where Γu and Γq are disjoint portions of the boundary Γ such that Γ = Γu ∪Γq ,
qc refers to the convective component of ßux (e.g. in heat transfer problems)

qc = hc (u − uc ) (2.3.4)

and (nx , ny ) denote the direction cosines of the unit normal vector on the
boundary. In Eq. (2.3.4), hc denotes the convective heat transfer coeﬃcient.
The radiative heat transfer boundary condition (which is a nonlinear function
of u) is not considered here. However, radiation boundary condition will be
considered in the nonlinear analysis.

2.3.2 Finite Element Approximation

In the Þnite element method, the domain Ω̄ = Ω ∪ Γ is divided into a set
of subdomains Ω̄e = Ωe ∪ Γe , called Þnite elements (see Figure 2.3.1). Any
geometric shape for which the approximation functions can be derived uniquely
qualiÞes as an element. We shall discuss simple geometric shapes and orders
of approximation shortly. To keep the formulative steps very general (i.e. not
conÞne the formulation to a speciÞc geometric shape or order of the element),
we have denoted the domain of a typical element by Ωe and its boundary by Γe .
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 25

The element Ω̄e can be a triangle or quadrilateral in shape, and the degree of
interpolation over it can be linear, quadratic, and so on. The non-overlapping
sum of all elements Ω̄e is denoted by Ωh , and it is called the Þnite element
mesh of the domain Ω. In general, Ωh may not equal Ω̄ when the boundary Γ
is curved. Of course, for polygonal domains, the Þnite element mesh exactly
represents the actual domain.

Table 2.3.1 List of Þelds in which the model equation (2.3.1) arises, with
meaning of various parameters and variables (see the bottom of
the table for the meaning of some parameters*).

Field Variable Data Data Data Variable

u axx ayy f qn

Heat Tempe- Thermal Thermal Heat Heat

transfer rature conductance conductance generation ßux
T kxx kyy f qn

Flow Fluid- Permea- Permea- InÞl- Flux

through head bility bility tration
porous
medium φ µxx µyy f qn

∂φ
Torsion of Warping 1 1 0 ∂n
cylindrical function
members φ
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

∂φ
Torsion of Stress 1 1 2Gθ ∂n
cylindrical function
members ψ

Deßection Displa- Tension Tension Transverse qn

of cement force
∂u
membranes u T T f ∂n

Flows of Velocity 1 1 0 qn
inviscid potential
∂φ
ßows φ ∂n

Flows of Stream 1 1 0 qn
inviscid function
∂ψ
ßows ψ ∂n

Electro- Electrical Dielectric Dielectric Charge Electric

statics potential constant constant density ßux
∂ψ
φ ε ε ρ ∂n
* k = thermal conductance; β = convective Þlm conductance; T∞ = ambient temperature
of the surrounding ßuid medium; G = shear modulus; θ = angle of twist.

Γe

e
y Ω

Figure 2.3.1 Finite element discretization of a domain.

Suppose that the dependent unknown u is approximated over a typical

Þnite element Ω̄e by the expression
n
X
u(x, y) ≈ ueh (x, y) = uej Lej (x, y) (2.3.5)
j=1

where ueh (x, y) represents an approximation of u(x, y) over the element Ω̄e ,
parameters uej denote the values of the function ueh (x, y) at a selected number
of points, called element nodes, in the element Ω̄e , and Lej are the Lagrange
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

interpolation functions associated with the element. As we shall see shortly,

the interpolation functions depend not only on the number of nodes in the
element, but also on the shape of the element. The shape of the element must
be such that its geometry is uniquely deÞned by a set of nodes. A triangle
(n = 3) is the simplest two-dimensional geometric shape in two dimensions.

2.3.3 Weak Formulation

The n parameters (or nodal values) uej in Eq. (2.3.5) must be determined
such that the approximate solution ue (x, y) satisÞes the governing Eq. (2.3.1)
and boundary conditions of the problem. As in the case of a variational
and weighted-residual method, we seek to satisfy the governing diﬀerential
equation in a weighted-integral sense, as described in Section 2.2. The type of
Þnite element model depends on the weighted-integral form used to generate
the algebraic equations. Thus, if one uses the weak form, the resulting model
will be diﬀerent from those obtained with a weighted-residual statement in
which the weight function can be any one of several choices. In the remainder
of this chapter, we shall be primarily concerned with the weak form Þnite
element models.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 27

The weak form of a diﬀerential equation is a weighted-integral statement

that is equivalent to both the governing differential equation as well as the
associated natural boundary conditions. We shall develop the weak form
of Eqs. (2.3.1) and (2.3.3) over the typical element Ωe using the three-step
procedure. The Þrst step is to take all non-zero expressions in Eq. (2.3.1) to
one side of the equality, multiply the resulting equation with a weight function
w, and integrate the equation over the element domain Ωe :
Z ∙ µ ¶ µ ¶ ¸
∂ ∂ue ∂ ∂ue
0= w − axx h − ayy h − f (x, y) dx dy (2.3.6)
Ωe ∂x ∂x ∂y ∂y
The expression in the square brackets of the above equation represents a
residual of the approximation of the differential equation (2.3.1), because
ueh (x, y) is only an approximation of u(x, y). For n independent choices of
w, we obtain a set of n linearly independent algebraic equations.
In the second step, we distribute the differentiation among u and w equally,
so that both u and w are required to be differentiable only once with respect
to x and y. To achieve this we use the component form of the gradient (or
divergence) theorem,
Z I
∂
(wF1 ) dx dy = (wF1 )nx ds (2.3.7a)
Ωe ∂x Γe
Z I
∂
(wF2 ) dx dy = (wF2 )ny ds (2.3.7b)
Ωe ∂y Γe
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

where nx and ny are the components (i.e. the direction cosines) of the unit
normal vector
n̂ = nx êx + ny êy = cos α êx + sin α êy (2.3.8)
on the boundary Γe , and ds is the arc length of an inÞnitesimal line element
along the boundary. We obtain
Z ∙ ¸ I ∙ ¸
∂w ∂u ∂w ∂u ∂u ∂u
0= axx + ayy − wf dx dy− w axx nx + ayy ny ds
Ω e ∂x ∂x ∂y ∂y Γe ∂x ∂y
(2.3.9)
From an inspection of the boundary term in Eq. (2.3.9), we note that u is
the primary variable, and speciÞcation of u constitutes the essential boundary
condition. The coeﬃcient of the weight function in the boundary expression,
namely
∂u ∂u
qn = axx nx + ayy ny (2.3.10)
∂x ∂y
is the secondary variable. Its speciÞcation constitutes the natural boundary
condition. By deÞnition qn is positive outward from the surface as we move
counterclockwise along the boundary Γe . The secondary variable qn denotes
the ßux normal to the boundary of the element.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
28 NONLINEAR FINITE ELEMENT ANALYSIS

The third and last step of the formulation is to use the deÞnition (2.3.10)
in Eq. (2.3.9) and write it as
Z µ ¶ I
∂w ∂u ∂w ∂u
0= axx + ayy − wf dx dy − w qn ds (2.3.11)
Ωe ∂x ∂x ∂y ∂y Γe

or
0 = B(w, u) − `(w) (2.3.12)

where the bilinear form B(·, ·) and linear form `(·) are deÞned by
Z µ ¶
∂w ∂u ∂w ∂u
B(w, u) = axx + ayy dx dy (2.3.13a)
Ωe ∂x ∂x ∂y ∂y
Z I
`(w) = w f dx dy + wqn ds (2.3.13b)
Ωe Γe

Note that the bilinear form is symmetric in its arguments (w, u)

B(w, u) = B(u, w)

and `(w) is linear in w. Therefore, it is possible to construct the associated

quadratic functional from the formula

1
I(u) = B(u, u) − `(u) (2.3.14)
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

2.3.4 Finite Element Model

The weak form in Eq. (2.3.11) requires that the approximation chosen for u
should be at least linear in both x and y so that there are no terms in (2.3.11)
that become identically zero. Suppose that u is approximated over a typical
Þnite element Ωe by the expression of the form (2.3.5). Substituting the Þnite
element approximation (2.3.5) for u into the weak form (2.3.11), we obtain

n
(Z " Ã ! Ã ! # ) I
X ∂w ∂Lej ∂w ∂Lej
0= axx + ayy − w f dx dy uej − w qn ds
j=1 Ωe ∂x ∂x ∂y ∂y Γe
(2.3.15)
This equation must hold for any weight function w. Since we need n
independent algebraic equations to solve for the n unknowns, ue1 , ue2 , ..., uen ,
we choose n independent functions for w: w = Le1 , Le2 , ..., Len . This particular
choice of weight functions is a natural one when the weight function Pn
is viewed
as a virtual variation of the dependent unknown (i.e. w = δu = i=1 δuei Lei ).
e

For each choice of w we obtain an algebraic relation among (ue1 , ue2 , ..., uen ).
The ith algebraic equation is obtained by substituting w = Lei into Eq.
(2.3.15):
n
X
e e
Kij uj = Qei + qie (2.3.16)
j=1
e , Qe , and q e are deÞned by
where the coeﬃcients Kij i i

Z Ã !
e ∂Le ∂Lej ∂Le ∂Lej
Kij = axx i + ayy i dx dy (2.3.17a)
Ωe ∂x ∂x ∂y ∂y
Z I
Qei = f Lei dx dy, qie = qn Lei ds (2.3.17b)
Ωe Γe

We note that Kij e = K e (i.e. [K e ] is symmetric). The symmetry of the

ji
coeﬃcient matrix is due to the symmetry of the bilinear form, which in turn
is due to the weak form development. In matrix notation, Eq. (2.3.16) takes
the form
[K e ]{ue } = {Qe } + {q e } ≡ {F e } (2.3.18)
This completes the Þnite element model development. Before we discuss
assembly of elements, it is informative to determine the interpolation functions
Lei for certain basic two-dimensional Þnite elements.

2.3.5 Interpolation Functions

The Þnite element approximation ue (x, y) of u(x, y) over an element Ωe

must satisfy the following conditions in order for the approximate solution
to converge to the true solution:
1. ue (x, y) must be continuous as required in the weak form of the problem
(i.e. all terms in the weak form are represented as non-zero values).
2. The polynomials used to represent ue (x, y) must be complete (i.e. all
terms, beginning with a constant term up to the highest order used in
the polynomial should be included in the expression of ue (x, y)).
3. All terms in the polynomial should be linearly independent.
The number of linearly independent terms in the representation of ue dictates
the shape and number of degrees of freedom of the element. Here we review
the interpolation functions of linear triangular and rectangular elements.
An examination of the variational form (2.3.11) and the Þnite element
matrices in Eq. (2.3.17a) shows that the Lei should be at least linear functions
of x and y. The polynomial

ue (x, y) = ce1 + ce2 x + ce3 y (2.3.19)

is the lowest-order polynomial that meets the requirements. It contains three

linearly independent terms, and it is linear in both x and y. The polynomial is
complete because the lower-order term, namely, the constant term, is included.
To write the three constants (ce1 , ce2 , ce3 ) in terms of the nodal values of ue , we
must identify three points or nodes in the element Ωe . The three nodes must
be such that they uniquely deÞne the geometry of the element and allow the
imposition of inter-element continuity of the variable ue (x, y). Obviously, the
geometric shape deÞned by three points in a two-dimensional domain is a
triangle. Thus the polynomial in Eq. (2.3.19) is associated with a triangular
element and the three nodes are identiÞed as the vertices of the triangle.
On the other hand, the polynomial

ue (x, y) = ce1 + ce2 x + ce3 y + ce4 xy (2.3.20)

contains four linearly independent terms, and is linear in x and y, with a

bilinear term in x and y. This polynomial requires an element with four
nodes. It is a rectangle with nodes at the four corners of the rectangle.
The interpolation functions for linear triangular and rectangular elements
are given below. Higher-order two-dimensional elements (i.e. element with
higher-order interpolation polynomials) will be discussed in Section 2.4.
Linear triangular element
The linear interpolation functions for the three-node triangle [see Figure
2.3.2(a)] are (see Reddy [1, pp. 304—307])
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

1
Lei (x, y) = (αe + βie x + γie y), (i = 1, 2, 3) (2.3.21)
2Ae i

where Ae is the area of the triangle, and αie , βie , and γie are geometric constants
known in terms of the nodal coordinates (xi , yi ):

αie = xj yk − xk yj ; βie = yj − yk ; γie = −(xj − xk ) (2.3.22)

for i 6= j 6= k, and i, j, and k permute in a natural order. Note that (x, y)

are the global coordinates used in the governing equation (2.3.1) over the
domain Ω. The interpolation functions Lei (i = 1, 2, . . . , n) satisfy the following
interpolation properties:

3
X
(i) Lei (xj , yj ) = δij , (i, j = 1, 2, 3); (ii) Lei (x, y) = 1 (2.3.23)
i=1

and they are called the Lagrange interpolation functions. Note that use of the
linear interpolation functions Lei of a triangle will result in thePapproximation
of the curved surface u(x, y) by a planar function ueh (x, y) = 3i=1 uei Lei (x, y).
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 31

y y _
y
3
4 3
Γe
Ωe b Ωe Γe
1 _
1 2 x
2 a
x
x

Figure 2.3.2 The linear (a) triangular and (b) rectangular Þnite elements.

e
The integrals in the deÞnition of Kij and Qei can be evaluated for given
data: axx , ayy , and f . For example, for element-wise constant values of the
data, that is, axx = aexx , ayy = aeyy , and f = f e , we have (see Reddy [1, pp.
311—313]) the following results:

e 1 fe Ae
Kij = (ae β e β e + aeyy γie γje ); Qei = (2.3.24)
4Ae xx i j 3
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

where Ae is the area of the triangular element, and βie and γie are known in
terms of the global nodal coordinates of the element nodes, as given in Eq.
(2.3.22). For a right-angled triangular element with base a and height b, and
node 1 at the right angle (nodes are numbered counterclockwise), [K e ] takes
the form (see Reddy [1, p. 387])
⎡ ⎤ ⎡ ⎤
α −α 0 aeyy β 0 −β
ae
[K e ] = xx ⎣ −α α 0⎦ + ⎣ 0 0 0⎦ (2.3.25)
2 0 0 0 2 −β 0 β

where α = b/a and β = a/b. Of course, for cases in which the conductivities
are functions of (x, y), numerical integration can be used to evaluate the
coeﬃcients (see Section 2.5.3).
The evaluation of boundary integrals of the type
I
qie = qne Lei (s) ds (2.3.26)
Γe

where qne is a known function of the distance s along the boundary Γe , involves
evaluation of line integrals. It is necessary to compute such integrals only when
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
32 NONLINEAR FINITE ELEMENT ANALYSIS

Γe , or a portion of it, coincides with the boundary Γq of the total domain Ω

on which the ßux is speciÞed. On portions of Γe that are in the interior of
the domain Ω, qne on side (i, j) of element Ωe cancels with qnf on side (p, q) of
element Ωf when sides (i, j) of element Ωe and (p, q) of element Ωf are the
same (i.e. at the interface of elements Ωe and Ωf ). This can be viewed as the
balance of the internal ßux. When Γe falls on the boundary Γu of the domain
Ω, qne is not known there and can be determined in the post-computation.
Note that the primary variable u is speciÞed on Γu . For additional details, see
Reddy [1, pp. 313—318].
Linear rectangular element
For a linear rectangular element, we have
4
X
u(x̄, ȳ) = uei Lei (x̄, ȳ) (2.3.27)
i=1

where Li are the Lagrange interpolation functions expressed in terms of the

element coordinates (x̄, ȳ) (see Reddy [1, pp. 308—311])
x̄ ȳ x̄ ȳ
Le1 = (1 − )(1 − ), Le2 = (1 − )
a b a b
e x̄ ȳ e x̄ ȳ
L3 = , L4 = (1 − ) (2.3.28)
ab a b
and (x̄, ȳ) denote the local coordinates with origin located at node 1 of
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

the element, and (a, b) denote the horizontal and vertical dimensions of the
rectangle [see Figure 2.3.2(b)].
The integrals in the deÞnition of Kije and Qe can be easily evaluated over a
i
rectangular element of sides a and b. For example, for element-wise constant
values of the data, that is, axx = aexx , ayy = aeyy , and f = f e , we have (see
Reddy [1, p. 313; p. 387]) the following results:

fe ab
[K e ] = aexx [S 11 ] + aeyy [S 22 ], Qei = (2.3.29)
4
where
⎡ ⎤
2α −2α −α α
1 ⎢ −2α 2α α −α ⎥
[S 11 ] = ⎢⎣
⎥
⎦ (2.3.30a)
6 −α α 2α −2α
α −α −2α 2α
⎡ ⎤
2β β −β −2β
1 ⎢ β 2β −2β −β ⎥
[S 22 ] = ⎢ ⎥ (2.3.30b)
6 ⎣ −β −2β 2β β⎦
−2β −β β 2β
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 33

and α = b/a and β = a/b. Again, for cases in which the conductivities are
functions of (x, y), numerical integration is used to evaluate the coeﬃcients,
as discussed in Section 2.5.4. When the element is non-rectangular, that is,
a quadrilateral, we use coordinate transformations to represent the integrals
over a square geometry and then use numerical integration to evaluate them.

2.3.6 Assembly of Elements

The assembly of Þnite elements to obtain the equations of the entire domain
is based on the following two rules:
1. Continuity of the primary variable (i.e. temperature)
2. Balance of secondary variables (i.e. heat ßux)
We illustrate the assembly procedure by considering a Þnite element mesh
consisting of a triangular element and a quadrilateral element (see Figure
2.3.3).
Let Kij1
(i, j = 1, 2, 3) denote the coeﬃcient matrix corresponding to the
triangular element, and let Kij 2 (i, j = 1, 2, 3, 4) denote the coeﬃcient matrix

corresponding to the quadrilateral element. The nodes of the Þnite element

mesh are called global nodes. From the mesh shown in Figure 2.3.3, it is
clear that the following correspondence between global and element nodes
exists: nodes 1, 2, and 3 of element 1 correspond to global nodes 1, 2, and
3, respectively. Nodes 1, 2, 3, and 4 of element 2 correspond to global nodes
2, 4, 5, and 3, respectively. Hence, the correspondence between the local and
global nodal values of temperature is
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

u11 = u1 , u12 = u21 = u2 , u13 = u24 = u3 , u22 = u4 , u23 = u5 (2.3.31)

which amounts to imposing the continuity of the primary variables at the
nodes common to elements 1 and 2. Note that the continuity of the primary
variables at the inter-element nodes guarantees the continuity of the primary
variable along the entire inter-element boundary.

Global node numbers

5
3
3 Element (local) node numbers
4
3
2 2
4
1
1
1
2 1
2

Figure 2.3.3 Global—local correspondence of nodes for assembly of elements.

Next, we consider the balance of secondary variables at the interelement

boundaries. At the interface between the two elements, the ßux from the
two elements should be equal in magnitude and opposite in sign. For the
two elements shown in Figure 2.3.3, the interface is along the side connecting
global nodes 2 and 3. Hence, the internal ßux qn1 on side 2—3 of element 1
should balance the ßux qn2 on side 4—1 of element 2 (recall the sign convention
on qne ):
(qn1 )2−3 = (qn2 )4−1 or (qn1 )2−3 = (−qn2 )1−4 (2.3.32)

In the Þnite element method, the above relation is imposed in a weighted-

integral sense:
Z Z Z Z
qn1 L12 ds = − qn2 L21 ds, qn1 L13 ds = − qn2 L24 ds (2.3.33)
h123 h214 h123 h214

where hepq denotes length of the side connecting node p to node q of element
Ωe .
Now we are ready to assemble the element equations for the two-element
mesh. The element equations of the two elements are written separately Þrst.
For the triangular element, the element equations are of the form

1 1 1 1 1 1
K11 u1 + K12 u2 + K13 u3 = Q11 + q11
1 1 1 1 1 1
K21 u1 + K22 u2 + K23 u3 = Q12 + q21 (2.3.34a)
1 1 1 1 1 1
K31 u1 + K32 u2 + K33 u3 = Q13 + q31
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

For the rectangular element the element equations are given by

2 2 2 2 2 2 2 2
K11 u1 + K12 u2 + K13 u3 + K14 u4 = Q21 + q12
2 2 2 2 2 2 2 2
K21 u1 + K22 u2 + K23 u3 + K24 u4 = Q22 + q22
2 2 2 2 2 2 2 2
K31 u1 + K32 u2 + K33 u3 + K34 u4 = Q23 + q32 (2.3.34b)
2 2 2 2 2 2 2 2
K41 u1 + K42 u2 + K43 u3 + K44 u4 = Q24 + q42

In order to impose the balance condition in (2.3.32), it is necessary to add the

second equation of element 1 to the Þrst equation of element 2, and also add
the third equation of element 1 to the fourth equation of element 2:

1 1 1 1 1 1 2 2 2 2 2 2 2 2
(K21 u1 + K22 u2 + K23 u3 ) + (K11 u1 + K12 u2 + K13 u3 + K14 u4 )
1 1 2 2
= (Q2 + q2 ) + (Q1 + q1 )

1 1 1 1 1 1 2 2 2 2 2 2 2 2
(K31 u1 + K32 u2 + K33 u3 ) + (K41 u1 + K42 u2 + K43 u3 + K44 u4 )
1 1 2 2
= (Q3 + q3 ) + (Q4 + q4 )
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 35

Using the local-global nodal variable correspondence in Eq. (2.3.1), we can

rewrite the above equations as
1 1 2 1 2 2 2
K21 u1 + (K22 + K11 )u2 + (K23 + K14 )u3 + K12 u4 + K13 u5
1 2 1 2
= Q2 + Q1 + (q2 + q1 )

1 1 2 1 2 2 2
K31 u1 + (K32 + K41 )u2 + (K33 + K44 )u3 + K42 u4 + K43 u5
1 2 1 2
= Q3 + Q4 + (q3 + q4 )
Now we can impose the conditions in Eq. (2.3.33) by setting appropriate
portions of the expressions in parenthesis on the right-hand side of the
above equations to zero (or a speciÞed non-zero value). In general, when
several elements are connected, the assembly of the elements is carried out by
putting element coeﬃcients Kij e , Qe , and q e into proper locations of the global
i i
coeﬃcient matrix and right-hand column vectors. This is done by means of
the connectivity relations, that is, correspondence of the local node number
to the global node number.
The assembly procedure described above can be used to assemble elements
of any shape and type. The procedure can be implemented in a computer with
the help of the local-global nodal correspondence.
For heat conduction problems that involve convection heat transfer at
the boundary, that is, when heat is transferred from one medium to the
surrounding medium (often, a ßuid) by convection, the Þnite element model
developed earlier requires some modiÞcation. For a convection boundary, the
natural boundary condition is a balance of energy transfer across the boundary
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

due to conduction and/or convection (i.e. Newton’s law of cooling):

∂u ∂u
(axx nx + ayy ny ) + hc (u − uc ) = qn (2.3.35)
∂x ∂y
where hc is the convective conductance (or the convective heat transfer
coeﬃcient), uc is the (ambient) temperature of the surrounding ßuid medium,
and qn is the speciÞed heat ßux. The Þrst term accounts for heat transfer by
conduction, the second by convection, and the third accounts for the speciÞed
heat ßux, if any. It is the presence of the term hc (u − uc ) that requires some
modiÞcation of the weak form in Eq. (2.3.11). To include the convective
boundary condition (2.3.35), the boundary integral in Eq. (2.3.9) should be
modiÞed. Instead of replacing the coeﬃcient of w in the boundary integral
with qn , we use Eq. (2.3.35):
Z ∙ ¸ I ∙ ¸
∂w ∂u ∂w ∂u ∂u ∂u
0= axx + ayy − wf dx dy − w axx nx + ayy ny ds
Ωe ∂x ∂x ∂y ∂y Γe ∂x ∂y
Z µ ¶ I
∂w ∂u ∂w ∂u
= axx + ayy − wf dx dy − w[qn − hc (u − uc )] ds
Ωe ∂x ∂x ∂y ∂y Γe
(2.3.36)
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
36 NONLINEAR FINITE ELEMENT ANALYSIS

or
0 = B(w, u) − `(w) (2.3.37)
where w is the weight function, and B(·, ·) and `(·) are the bilinear and linear
forms
Z µ ¶ I
∂w ∂u ∂w ∂u
B(w, u) = axx + ayy dx dy + hc wu ds (2.3.38a)
Ωe ∂x ∂x ∂y ∂y Γe
Z I I
`(w) = f w dx dy + hc uc w ds + qn w ds (2.3.38b)
Ωe Γe Γe

Note that the unknown surface temperature in the convective boundary

condition has been made part of the bilinear form B(·, ·) while all the known
quantities remain part of the linear form `(·). The Þnite element model of Eq.
(2.3.36) is (see Reddy [1], pp. 341—346) given by

[K e ]{ue } = {F e } (2.3.39)

where
Z µ ¶ I
e ∂Li ∂Lj ∂Li ∂Lj
Kij = axx + ayy dx dy + hc Li Lj ds (2.3.40a)
Ωe ∂x ∂x ∂y ∂y Γe
Z I I
Fi = f Li dx dy + hc uc Li ds + qn Li ds (2.3.40b)
Ωe Γe Γe
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

The Þnite element model (2.3.39) is valid for both conductive and
convective heat transfer boundary conditions. For problems with no convective
boundary conditions, the convective contributions to the element coeﬃcients
are omitted. Indeed, these contributions have to be included only for those
elements whose sides fall on the boundary with speciÞed convection heat
transfer. The contribution due to convective boundaries to the element
coeﬃcient matrix and source vector can be computed by evaluating line
integrals, as discussed in Reddy [1], pp. 342—345.

2.4 Library of Two-Dimensional Finite Elements

2.4.1 Introduction
The objective of this section is to present a library of two-dimensional
triangular and rectangular elements of the Lagrange family, that is, elements
over which only the function — not its derivatives — are interpolated. Once
we have elements of diﬀerent shapes and order at our disposal, we can
choose appropriate elements and associated interpolation functions for a given
problem. The interpolation functions are developed here for regularly shaped
elements, called master elements. These elements can be used for numerical
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 37

evaluation of integrals deÞned on irregularly shaped elements. This requires a

transformation of the geometry from the actual element shape to its associated
master element. We will discuss the numerical evaluation of integrals in
Section 2.6.

2.4.2 Triangular Elements

The three-node triangular element was developed in Section 2.3.5. Higher-
order triangular elements (i.e. triangular elements with interpolation functions
of higher degree) can be systematically developed with the help of the so-called
area coordinates. For triangular elements, it is possible to construct three non-
dimensional coordinates L̂i (i = 1, 2, 3), which vary in a direction normal to
the sides directly opposite each node (see Figure 2.4.1). The coordinates are
deÞned such that
X3
Ai
L̂i = , A= Ai (2.4.1)
A i=1
where Ai is the area of the triangle formed by nodes j and k and an arbitrary
point P in the element, and A is the total area of the element. For example, A1
is the area of the shaded triangle which is formed by nodes 2 and 3 and point
P . The point P is at a perpendicular distance of s from the side connecting
nodes 2 and 3. We have A1 = bs/2 and A = bh/2. Hence
A1 s
L̂1 = = (2.4.2)
A h
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

Clearly, L̂1 is zero on side 2—3 (hence, zero at nodes 2 and 3) and has a value
of unity at node 1. Thus, L̂1 is the interpolation function associated with node
1. Similarly, L̂2 and L̂3 are the interpolation functions associated with nodes
2 and 3, respectively. In summary, we have
Li = L̂i (2.4.3)

3 L1= 0 5
b 4
A1 2 L1 3
h A2 L 1 = 0.5
P P
s 6
A3
2
L1= 1

1 1

Figure 2.4.1 DeÞnition of area coordinates L̂i for triangular elements.

The area coordinates L̂i can be used to construct interpolation functions

for higher-order triangular elements. For example, a higher-order element with
k nodes per side (equally spaced on each side) has a total of n nodes

k−1
X k
n= (k − i) = k + (k − 1) + · · · + 1 = (k + 1) (2.4.4)
i=0
2

and its degree is equal to k−1. The explicit forms of the interpolation functions
for the linear and quadratic elements are recorded below:
⎧ ⎫
⎪
⎪ L̂1 (2L̂1 − 1) ⎪
⎪
⎪
⎪ ⎪
⎪
⎧ ⎫ ⎪
⎪ L̂2 (2L̂2 − 1) ⎪
⎪
⎨ L̂1 ⎬ ⎪
⎨ ⎪
⎬
{Le } = ; {Le } = L̂3 (2L̂3 − 1) (2.4.5)
L̂
⎩ 2⎭ ⎪
⎪ 4L̂1 L̂2 ⎪ ⎪
L̂3 ⎪
⎪ ⎪
⎪
⎪
⎪ 4L̂2 L̂3 ⎪ ⎪
⎪
⎩ ⎪
⎭
4L̂3 L̂1

Note that the order of the interpolation functions in the above arrays
corresponds to the node numbers shown in Figure 2.4.2(a). Thus, the Þrst
three rows of the vectors in Eq. (2.4.5) correspond to the Þrst three nodes of
the linear and quadratic elements, which correspond to the three vertices of
the triangular element. The last three rows of the second vector in Eq. (2.4.5)
associated with the quadratic element correspond to the mid-side nodes of the
triangular element. A similar node-numbering scheme is used for rectangular
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

elements, which are discussed next.

2.4.3 Rectangular Elements

The Lagrange interpolation functions associated with rectangular elements
can be obtained from the tensor product of corresponding one-dimensional
Lagrange interpolation functions. We take a local coordinate system (ξ, η)
such that −1 ≤ (ξ, η) ≤ 1. This choice of local coordinate system is dictated
by the Gauss quadrature rule used in the numerical evaluation of integrals
over the element (see Section 2.5).
The linear and quadratic interpolation functions are given in Eqs. (2.4.6)
and (2.4.7), respectively (see Figure 2.4.2(b) for the node numbers).
⎧ ⎫
⎪
⎪ (1 − ξ)(1 − η) ⎪
⎪
1 ⎨ ⎬
(1 + ξ)(1 − η)
{Le } = (2.4.6)
4⎪
⎪
⎩
(1 + ξ)(1 + η) ⎪
⎪
⎭
(1 − ξ)(1 + η)

3 3

6
5

1 1
4
2 2
(a)
η η
4 3 4 7 3

6
ξ 8 9 ξ

1 2 1 5 2
(b)

Figure 2.4.2 Linear and quadratic (a) triangular and (b) rectangular
elements.
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

⎧ ⎫
⎪
⎪ (1 − ξ)(1 − η)(−ξ − η − 1) + (1 − ξ 2 )(1 − η 2 ) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪ (1 + ξ)(1 − η)(ξ − η − 1) + (1 − ξ 2 )(1 − η2 ) ⎪⎪
⎪
⎪
⎪ (1 + ξ)(1 + η)(ξ + η − 1) + (1 − ξ 2 )(1 − η 2 ) ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪ 2 2
⎨ (1 − ξ)(1 + η)(−ξ + η − 1) + (1 − ξ )(1 − η ) ⎪ ⎬
1
{Le } = 2(1 − ξ 2 )(1 − η) − (1 − ξ 2 )(1 − η 2 ) (2.4.7)
4⎪⎪ ⎪
⎪
⎪
⎪
⎪ 2(1 + ξ)(1 − η 2 ) − (1 − ξ 2 )(1 − η 2 ) ⎪
⎪
⎪
⎪
⎪ 2(1 − ξ 2 )(1 + η) − (1 − ξ 2 )(1 − η 2 ) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 2(1 − ξ)(1 − η 2 ) − (1 − ξ 2 )(1 − η 2 ) ⎪
⎪
⎩ ⎭
4(1 − ξ 2 )(1 − η 2 )

The serendipity elements are those rectangular elements that have no

interior nodes. These elements have fewer nodes compared with the higher-
order Lagrange elements. The interpolation functions of the serendipity
elements are not complete, and they cannot be obtained using tensor products
of one-dimensional Lagrange interpolation functions. Instead, an alternative
procedure must be employed, as discussed in Reddy [1]. The interpolation
functions for the two-dimensional quadratic serendipity element are given by
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
40 NONLINEAR FINITE ELEMENT ANALYSIS

(see Figure 2.4.3)

⎧ ⎫
⎪
⎪ (1 − ξ)(1 − η)(−ξ − η − 1) ⎪
⎪
⎪
⎪ (1 + ξ)(1 − η)(ξ − η − 1) ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ (1 + ξ)(1 + η)(ξ + η − 1) ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎨ ⎪
⎬
1 (1 − ξ)(1 + η)(−ξ + η − 1)
{Le } = (2.4.8)
4⎪
⎪ 2(1 − ξ 2 )(1 − η) ⎪
⎪
⎪ ⎪ 2 ⎪
⎪
⎪
⎪ 2(1 + ξ)(1 − η ) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 2(1 − ξ 2 )(1 + η) ⎪
⎪
⎪
⎩ ⎪
⎭
2(1 − ξ)(1 − η ) 2

η
4 7 3

6
8 ξ

1 5 2

Figure 2.4.3 Quadratic rectangular serendipity element.

2.5 Numerical Integration

2.5.1 Preliminary Comments
An accurate representation of irregular domains (i.e. domains with curved
boundaries) can be accomplished by the use of reÞned meshes and/or
irregularly shaped curvilinear elements. For example, a non-rectangular
region cannot be represented using rectangular elements; however, it can
be represented by quadrilateral elements. Since the interpolation functions
are easily derivable for a rectangular element and it is easier to evaluate
integrals over rectangular geometries, we transform the Þnite element integral
statements deÞned over quadrilaterals to a rectangle. The transformation
results in complicated expressions for the integrands in terms of the
coordinates used for the rectangular element. Therefore, numerical integration
is used to evaluate such complicated integrals. The numerical integration
schemes, such as the Gauss—Legendre numerical integration scheme, require
the integral to be evaluated on a speciÞc domain or with respect to a speciÞc
coordinate system.

2.5.2 Coordinate Transformations

Gauss quadrature requires the integral to be expressed over a square region
Ω̂ of dimension 2 × 2 with respect to the coordinate system, (ξ, η) to be such
that −1 ≤ (ξ, η) ≤ 1. The transformation of the geometry and the variable
coeﬃcients of the diﬀerential equation from the problem coordinates (x, y) to
the local coordinates (ξ, η) results in algebraically complex expressions, and
they preclude analytical (i.e. exact) evaluation of the integrals. Thus, the
transformation of a given integral expression, deÞned over element Ωe , to one
on the domain Ω̂ facilitates the numerical integration. Each element of the
Þnite element mesh is transformed to Ω̂, only for the purpose of numerically
evaluating the integrals (see Figure 2.5.1). The element Ω̂ is called a master
element. For example, every quadrilateral element can be transformed to a
square element with a side of length 2 and −1 ≤ (ξ, η) ≤ 1 that facilitates
the use of Gauss—Legendre quadrature to evaluate integrals deÞned over the
quadrilateral element.
The transformation between a typical element Ωe in the mesh and the
master element Ω̂ [or equivalently, between (x, y) and (ξ, η)] is accomplished
by a coordinate transformation of the form
m
X m
X
x= xej φej (ξ, η) , y= yje φej (ξ, η) (2.5.1)
j=1 j=1

where φj denote the Þnite element interpolation functions of the master

element Ω̂. The coordinates in the master element are chosen to be the
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

natural coordinates (ξ, η) such that −1 ≤ (ξ, η) ≤ 1. This choice is dictated

by the limits of integration in the Gauss quadrature rule used to evaluate
the integrals. For this case, the φej denote the interpolation functions of the
four-node rectangular element shown in Figure 2.4.2(b) (i.e. m = 4).

Ωe

η=1 η

ξ=1 y
ξ
Ω̂
Ω3
1 Ω2
Ω

Figure 2.5.1 Transformation of quadrilateral elements to the master

rectangular element for numerical evaluation of integrals.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
42 NONLINEAR FINITE ELEMENT ANALYSIS

The transformation (2.5.1) maps, for example, the line ξ = 1 in Ω̂ to the line
deÞned parametrically by x = x(1, η) and y = y(1, η) in the xy-plane. In other
words, the master element Ω̂ is transformed, under the linear transformation,
into a quadrilateral element (i.e. a four-sided element whose sides are not
parallel) in the xy-plane. Conversely, every quadrilateral element of a mesh
can be transformed to the same four-noded square (master) element Ω̂ in the
(ξ, η)-plane.
In general, the dependent variable(s) of the problem are approximated by
expressions of the form

n
X
u(x, y) = uej Lej (x, y) (2.5.2)
j=1

The interpolation functions Lej used for the approximation of the dependent
variable, in general, are diﬀerent from φej used in the approximation of the
geometry. Depending on the relative degree of approximations used for the
geometry [see Eq. (2.5.1)] and the dependent variable(s) [see Eq. (2.5.2)], the
Þnite element formulations are classiÞed into three categories:
1. Superparametric (m > n). The approximation used for the geometry is
higher order than that used for the dependent variable.
2. Isoparametric (m = n). Equal degree of approximation is used for both
geometry and dependent variables.
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

3. Subparametric (m < n). Higher-order approximation of the dependent

variable is used.
It should be noted that the transformation of a quadrilateral element of
a mesh to the master element Ω̂ is solely for the purpose of numerically
evaluating the integrals (see Figure 2.5.1). No transformation of the physical
domain or elements is involved in the Þnite element analysis. The resulting
algebraic equations of the Þnite element formulation are always in terms of the
nodal values of the physical domain. Different elements of the Þnite element
mesh can be generated from the same master element by assigning appropriate
global coordinates to each of the elements. Master elements of a different
order deÞne different transformations and hence different collections of Þnite
elements within the mesh. For example, a quadratic rectangular master
element can be used to generate a mesh of quadratic curvilinear quadrilateral
elements. The transformations of a master element should be such that no
spurious gaps exist between elements, and no element overlaps occur. For
example, consider the element coefficients
Z " #
e ∂Lei ∂Lej ∂Lei ∂Lej
Kij = axx (x, y) + ayy (x, y) dx dy (2.5.3)
Ωe ∂x ∂x ∂y ∂y
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 43

The integrand (i.e. the expression in the square brackets under the integral) is
a function of the global coordinates x and y. We must rewrite it in terms of ξ
and η using the transformation (2.5.1). Note that the integrand contains not
only functions but also derivatives with respect to the global coordinates (x, y).
∂Le ∂Le ∂Le ∂Le
Therefore, we must relate ( ∂xi , ∂yi ) to ( ∂ξi , ∂ηi ) using the transformation
(2.5.1).
The functions Lei (x, y) can be expressed in terms of the local coordinates
(ξ, η) by means of the transformation (2.5.1). Hence, by the chain rule of
partial diﬀerentiation, we have
∂Lei ∂Lei ∂x ∂Lei ∂y ∂Lei ∂Lei ∂x ∂Lei ∂y
= + ; = +
∂ξ ∂x ∂ξ ∂y ∂ξ ∂η ∂x ∂η ∂y ∂η
or, in matrix notation,
( ∂Le ) " ∂x ∂y # ( ∂Le )
i i
∂ξ ∂ξ ∂ξ ∂x
∂Lei = ∂x ∂y ∂Lei (2.5.4)
∂η ∂η ∂η ∂y

which gives the relation between the derivatives of Lei with respect to the
global and local coordinates. The matrix in Eq. (2.5.4) is called the Jacobian
matrix of the transformation (2.5.1):
" ∂x ∂y #
e ∂ξ ∂ξ
[J ] = ∂x ∂y (2.5.5)
∂η ∂η
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

e in Eq. (2.5.3) that we must relate

Note from the expression given for Kij
∂Le ∂Le ∂Le ∂Le
( ∂xi , ∂yi )
to ( ∂ξi , ∂ηi ), whereas Eq. (2.5.4) provides the inverse relations.
Therefore, Eq. (2.5.4) must be inverted. We have
( ∂Le ) ( ∂Le )
i i
∂x e −1 ∂ξ
∂Lei = [J ] ∂Lei (2.5.6)
∂y ∂η

This requires that the Jacobian matrix [J e ] be non-singular.

Using the transformation (2.5.1), we can write
m m
∂x X ∂φej ∂y X ∂φej
= xj , = yj (2.5.7a)
∂ξ j=1
∂ξ ∂ξ j=1
∂ξ

m m
∂x X ∂φej ∂y X ∂φej
= xj , = yj (2.5.7b)
∂η j=1 ∂η ∂η j=1 ∂η

and by means of Eq. (2.5.5) one can compute the Jacobian matrix and then
its inverse. Thus, given the global coordinates (xj , yj ) of element nodes and
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
44 NONLINEAR FINITE ELEMENT ANALYSIS

the interpolation functions φej used for geometry, the Jacobian matrix can be
evaluated using Eq. (2.5.5). A necessary and suﬃcient condition for [J]−1
to exist is that the determinant J, called the Jacobian, be non-zero at every
point (ξ, η) in Ω̂ :
∂x ∂y ∂x ∂y
Je ≡ det[J e ] = − 6= 0. (2.5.8)
∂ξ ∂η ∂η ∂ξ
From Eq. (2.5.8) it is clear that the functions ξ(x, y) and η(x, y) must be
continuous, diﬀerentiable, and invertible. Moreover, the transformation should
be algebraically simple so that the Jacobian matrix can be easily evaluated.
Transformations of the form in Eq. (2.5.1) satisfy these requirements and the
requirement that no spurious gaps between elements or overlapping of elements
occur.
Returning to numerical evaluation of integrals, we have from Eq. (2.5.6),
( ∂Le ) ( ∂Le ) ( ∂Le )
i i i
∂x −1 ∂ξ ∗ ∂ξ
∂Lei = [J] ∂Lei ≡ [J ] ∂Lei (2.5.9)
∂y ∂η ∂η

where Jij∗ is the element in position (i, j) of the inverse of the Jacobian matrix
[J e ]. The element area dA = dx dy in element Ωe is transformed to
dA = |[J e ]| dξ dη (2.5.10)
in the master element Ω̂.
Equations (2.5.7)-(2.5.10) provide the necessary relations to transform
integral expressions on any element Ωe to an associated master element Ω̂.
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

For instance, consider the integral expression in Eq. (2.5.3), where axx and
ayy are functions of x and y. Suppose that the Þnite element Ωe can be
generated by the master element Ω̂e . Under the transformation (2.5.1) we can
write
Z " #
e ∂Le ∂Lej ∂Le ∂Lej
Kij = axx (x, y) i + ayy (x, y) i dx dy
Ωe ∂x ∂x ∂y ∂y
Z
= Fij (ξ, η) dξ dη (2.5.11)
Ω̂e
The discussion presented above is valid for master elements of both
rectangular and triangular geometry.

2.5.3 Integration Over a Master Rectangular Element

Integrals deÞned over a rectangular master element Ω̂R can be numerically
evaluated using the Gauss—Legendre quadrature formulas
Z Z 1Z 1 M X
X N
F (ξ, η) dξ dη = F (ξ, η) dξ dη ≈ F (ξI , ηJ ) WI WJ (2.5.12)
Ω̂R −1 −1 I=1J=1
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 45

where M and N denote the number of Gauss quadrature points, (ξI , ηJ ) denote
the Gauss point coordinates, and WI and WJ denote the corresponding Gauss
weights as shown in Table 2.5.1 (from Table 7.2 in Reddy [1]).

Table 2.5.1 Gauss quadrature points and weights for rectangular elements.
R1 PN
−1
F (ξ) dξ = I =1
F (ξI ) WI

N Points, ξI Weights, WI

1 0.0000000000 2.0000000000
2 ± 0.5773502692 1.0000000000
3 0.0000000000 0.8888888889
± 0.7745966692 0.5555555555
4 ± 0.3399810435 0.6521451548
± 0.8611363116 0.3478548451
5 0.0000000000 0.5688888889
± 0.5384693101 0.4786286705
± 0.9061798459 0.2369268850
6 ± 0.2386191861 0.4679139346
± 0.6612093865 0.3607615730
± 0.9324695142 0.1713244924

The selection of the number of Gauss points is based on the formula

N = int[(p+1)/2]+1, where p is the polynomial degree to which the integrand

is approximated. In most cases, the interpolation functions are of the same
degree in both ξ and η, and therefore one has M = N . When the integrand is
of a different degree in ξ and η, we use max(M, N ). The minimum allowable
quadrature rule is one that yields the area or volume of the element exactly.
The maximum degree of the polynomial refers to the degree of the highest
polynomial in ξ or η that is present in the integrands of the element matrices
of the type in Eq. (2.5.3). Note that the polynomial degree of coefficients as
well as Jije should be accounted for in determining the total polynomial degree
of the integrand. Of course, the coefficients axx , ayy , and Jije , in general,
may not be polynomials. In those cases, their functional variations must be
approximated by a suitable polynomial (e.g. by a binomial series) in order to
determine the polynomial degree of the integrand.

2.5.4 Integration Over a Master Triangular Element

In the preceding section, we discussed numerical integration on quadrilateral

elements which can be used to represent very general geometries as well as
Þeld variables in a variety of problems. Here we discuss numerical integration
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
46 NONLINEAR FINITE ELEMENT ANALYSIS

on triangular elements. Since quadrilateral elements can be geometrically

distorted, it is possible to distort a quadrilateral element to obtain a required
triangular element by moving the position of the corner nodes, and the fourth
corner in the quadrilateral is merged with one of the neighboring nodes. In
actual computation, this is achieved by assigning the same global node number
to two corner nodes of the quadrilateral element. Thus, master triangular
elements can be obtained in a natural way from associated master rectangular
elements. Here we discuss the transformations from a master triangular
element to an arbitrary triangular element.
We choose the unit right isosceles triangle (see Table 2.5.2) as the master
element. An arbitrary triangular element Ωe can be generated from the master
triangular element Ω̂T by transformation of the form (2.6.1). The derivatives
of Lei with respect to the global coordinates can be computed from Eq. (2.5.6),
which take the form
⎧ e⎫ ⎧ ∂Le ⎫ ⎡ ⎤
⎪ ∂L ∂x ∂y
⎨ ∂xi ⎪
⎬ ⎪
⎨ ∂ L̂i ⎪
⎬ ∂ L̂1 ∂ L̂1
1 ⎢ ⎥
= [J e ]−1 , [J e ] = ⎣ ⎦ (2.5.13)
⎪
⎩ ∂Lei ⎪
⎭ ⎪
⎩ ∂Li ⎪
e
⎭ ∂x ∂y
∂y ∂ L̂2 ∂ L̂2 ∂ L̂2

Note that only L̂1 and L̂2 are treated as linearly independent coordinates
because L̂3 = 1 − L̂1 − L̂2 .
After transformation, integrals on Ω̂T have the form
Z Z
G(ξ, η) dξ dη = G(L̂1 , L̂2 , L̂3 ) dL̂1 dL̂2 (2.5.14)
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

Ωe Ωe

which can be approximated by the quadrature formula

Z N
X
G(L̂1 , L̂2 , L̂3 ) dL̂1 dL̂2 ≈ G(SI )WI (2.5.15)
Ωe I=1

where WI and SI denote the weights and integration points of the quadrature
rule. Table 2.5.2 contains the location of integration points and weights for
one-, three-, and four-point quadrature rules over triangular elements.

2.6 Computer Implementation

2.6.1 General Comments
In this section, computer implementation of Þnite element calculations is
presented to illustrate the ease with which theoretical ideas can be transformed
into practice. The material presented here is based on a more detailed account
presented by Reddy [1] and Reddy and Gartling [4]. We begin with some
general comments on a typical Þnite element program.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 47

Table 2.5.2 Quadrature weights and points for triangular elements.

Number Degree of
of polynomial Location of integration points
integration Order of the
points residual L̂1 L̂2 L̂3 W Geometric locations
1
1 O(h2 ) 1/3 1/3 1/3 1 a
a

2 1/2 0 1/2 a
3 O(h3 ) 1/2 1/2 0 1/3 b
a
0 1/2 1/2 c b
c

3 1/3 1/3 1/3 −27/48 a

A typical computer program consists of three basic parts:

1. Preprocessor
2. Processor
3. Postprocessor
In the preprocessor part of a program, the input data of the problem
are read in and/or generated. This includes the geometry of the domain,
analysis option (e.g. static, eigenvalue, or transient analysis), the data of
the problem (e.g. deÞnition of the coeﬃcients appearing in the diﬀerential
equation), boundary conditions, Þnite element analysis information (e.g.
element type, number of elements, geometric information required to generate
the Þnite element mesh and element connectivity), and indicators for various
postprocessing options (e.g. print, no print, types of quantities to be
calculated, and so on). In the postprocessor part of the program, the solution is
computed by interpolation at points other than the nodes, secondary variables
that are derivable from the solution are also computed, and the output data are
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
48 NONLINEAR FINITE ELEMENT ANALYSIS

processed in a desired format for printout and/or plotting. The preprocessor

and postprocessor computer modules may contain a few Fortran statements to
read and print pertinent information, simple subroutines (e.g. subroutines to
generate mesh and compute the gradient of the solution), or complex programs
linked to other units via disk and tape Þles.
The processor module, where typically large amounts of computing time
are spent, may consist of several subroutines, each having a special purpose.
The main modules include:
1. Generation of the element matrices using numerical integration.
2. Assembly of element equations.
3. Imposition of the boundary conditions.
4. Solution of the algebraic equations for the nodal values of the primary
variables (see Appendix 1).
The degree of sophistication and the complexity of a Þnite element program
depend on the general class of problems being programmed, the generality of
the data in the equation, and the intended user of the program. Figure 2.6.1
contains the ßow chart of a typical Þnite element computer program. A typical
but simple Þnite element programs are included in the book by Reddy [1] and
interested reader may wish to study Chapters 7 and 13 of [1], where details
of the implementation of one- and two-dimensional problems of heat transfer,
ßuid mechanics, and solid mechanics are discussed.
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

2.6.2 One-Dimensional Problems

Here we discuss the main ideas behind the calculation of element coeﬃcient
matrices, [K e ], [M e ], and {f e } for the model problem discussed in Sections
2.2 and 2.3 (see Chapter 8 for transient problems). Recall that the coeﬃcient
matrices are of the form
Z xb " #
e dLe dLej
Kij = a(x) i + c(x)Lei Lej dx
xa dx dx
Z xb Z xb
Mije = ct (x)Lei Lej dx, fie = f (x)Lei (x) dx (2.6.1)
xa xa

We use the coordinate transformation of the form [see Eq. (2.5.1)]

M
X
x= xej φej (ξ) (2.6.2)
j=1

to express the integrals posed over a typical element Ωe = (xa , xb ) as those

over the interval, −1 ≤ ξ ≤ 1, so that we can use the Gauss quadrature to
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 49

Processor
Initialize global
Preprocessor
[K], [M], {F}

DO e = 1 to n

Transfer global information

(material properties, geometry and solution)
to element

CALL ELKMF to calculate Kij(e)

Mij(e), and fi(e), and assemble
to form global [K], [M], and {F}

Impose boundary conditions

Postprocessor Solve the equations

Figure 2.6.1 The ßow chart of a typical Þnite element program.

evaluate them numerically. In Eq. (2.6.2), xej denote the global coordinates
of node j of element ω e = (xa , xb ), and φej are the approximation functions
used to approximate the geometry. For example, if we use linear interpolation
functions φej (ξ) = Lej (ξ) to represent the geometry of the element, we have
xe1 = xa , xe2 = xb and Eq. (2.6.2) becomes
xa + xb xb − xa he
x = xa Le1 (ξ) + xb Le2 (ξ) = + ξ = xa + ξ (2.6.3)
2 2 2
The Jacobian of transformation is given by
dx he
Je ≡ = (2.6.4)
dξ 2
The above transformation is exact for all (straight) line elements.
The derivatives of Lei (x) with respect to the global coordinate x is given by
dLei dLei dξ dLei −1
= = J
dx dξ dx dξ e
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
50 NONLINEAR FINITE ELEMENT ANALYSIS

Then the integrals in Eq. (2.6.1) become

Z xb " #
e dLe dLej
Kij = a(x) i + c(x)Lei Lej dx
xa dx dx
Z 1 " µ ¶Ã ! #
dLei −1 dLej −1
= â(ξ) Je J + ĉ(ξ)Lei Lej Je dξ
−1 dξ dξ e
Z xb Z 1 ³ ´
Mije = ct (x)Lei Lej dx = ĉt (ξ)Lei Lej Je dξ
xa −1
Z xb Z 1
fie = f (x)Lei (x) dx = fˆ(ξ)Lei Lej Je dξ (2.6.5)
xa −1

where â(ξ) = a(x(ξ)) and so on. Each of the integral expressions above can
be evaluated using the Gauss quadrature
Z 1 NX
GP
F (ξ)Je dξ = F (ξNI ) Je WN I (2.6.6)
−1 NI=1

where N GP is the number of Gauss points, ξNI is the N Ith Gauss point, and
WNI is the N Ith Gauss weight.
To implement the above development into a computer subroutine for
arbitrary degree of Lei , we must Þrst create a subroutine of all interpolations
functions and their derivatives with respect to ξ that we intend to use in our
analysis. For the present discussion, we limit them to the linear and quadratic
Lagrange family of functions. These functions and their derivatives are given
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

below.
Linear
1 dL1 1 dL2
L1 (ξ) = (1 − ξ) , = −0.5; L2 (ξ) = (1 + ξ) , = 0.5 (2.6.7)
2 dξ 2 dξ

Quadratic
1 dL1
L1 (ξ) = − ξ (1 − ξ) , = 0.5 (2ξ − 1)
2 dξ
³ ´ dL2
L2 (ξ) = 1 − ξ 2 , = −2ξ
dξ
1 dL3
L3 (ξ) = ξ (1 + ξ) , = 0.5 (1 + 2ξ) (2.6.8)
2 dξ
The following variables names are used in the subroutine (see INTRPL1D
in Box 2.6.1):
dLi dLi
SFL(i) = Li , DSFL(i) = , GDSFL(i) =
dξ dx
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 51

Box 2.6.1 Listing of subroutine INTRPL1D.

SUBROUTINE INTRPL1D(ELX,GJ,IEL,NPE,XI)
C __________________________________________________________________
C
C The subroutine computes shape functions and their derivatives for
C Hermite cubic and Lagrange linear, quadratic, and cubic elements
C
C X......……... Global (i.e. problem) coordinate
C XI ....…….. Local (i.e. element) coordinate
C H......…….. Element length
C {SFL}...…. Interpolation (or shape) functions
C {DSFL}..... First derivative of SF with respect to XI
C {GDSFL}.. First derivative of SF with respect to X
C GJ.....……. Jacobian of the transformation
C __________________________________________________________________
C
IMPLICIT REAL*8(A-H,O-Z)
COMMON /SHP/SFL(4),GDSFL(4)
DIMENSION DSFL(4),ELX(3)
C
C SHAPE FUNCTIONS AND THEIR DERIVATIVES
C
C Linear functions
C
IF(IEL.EQ.1)THEN
SFL(1)=0.5*(1.0-XI)
SFL(2)=0.5*(1+XI)
DSFL(1)=-0.5
DSFL(2)=0.5
ENDIF
C
C Quadratic functions
C
IF(IEL.EQ.2)THEN
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

SFL(1)=0.5*(XI-1.0)*XI
SFL(2)=1.0-XI*XI
SFL(3)=0.5*(XI+1.0)*XI
DSFL(1)=XI-0.5
DSFL(2)=-2.0*XI
DSFL(3)=XI+0.5
ENDIF
GJ=0.0
DO 10 I=1,NPE
10 GJ=GJ+DSFL(I)*ELX(I)
DO 30 I=1,NPE
GDSFL(I)=DSFL(I)/GJ
30 CONTINUE
C
RETURN
END

The notation should be transparent to the reader: SFL = Shape Functions

of the Lagrange family; DSF L = Derivative of SF L with respect to ξ; and
GDSF L = Global Derivative (i.e. derivative with respect to x) of SF L. All
of them are n × 1 arrays, where n is the Nodes Per Element, NPE (i.e. n =
NPE). In addition, GJ is used for the Jacobian J and XI for ξ.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
52 NONLINEAR FINITE ELEMENT ANALYSIS

Next, we implement the steps to evaluate the matrix coeﬃcients Kij , Mij ,
and fi (i = 1 to NPE) into a subroutine called ELEKMF1D. To this end,
we must assume some form of the data, a(x), c(x), ct (x) and f (x). Although
we can use any integrable functions, we restrict in the present discussion our
choice of these coeﬃcients to linear polynomials for the entire domain of the
problem:

a(x) = a0 + a1 x, c(x) = c0 + c1 x, ct (x) = ct0 + ct1 x,

f (x) = f0 + f1 x
(2.6.9)
Obviously, the coeﬃcients of the polynomials must be read in the preprocessor
and transferred to the subroutine ELEKMF1D through a common block (or
the argument list).
Since the evaluation of the matrix coeﬃcients involves summation on the
number of Gauss points, the arrays used for [K], [M ], and {f } must be
initialized outside the do-loop on NGP, the Number of Gauss Points. The
following notation is used:
e
ELK(i,j) = Kij , ELM(i,j) = Mije ,
ELF(i) = fie , i = 1, 2, . . . , N P E
(2.6.11)
The Gauss points are arranged in a matrix form so that the Jth column
corresponds to the Jth order Guass rule. The same notation is used for the
Gauss weights: GAU SP T (N I, N J) = ξNI , N Ith Gauss point of the N J-
point Gauss rule; GAU SW T (N I, N J) = WNI , N Ith Gauss weight of the
N J weight Gauss rule. Inside the do-loop on N I = 1 to N GP , we must
call Subroutine INTRPL1D to compute SF L(i) and GDSF L(i) at the N Ith
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

Gauss point and then compute all necessary quantities. Box 2.6.2 contains a
listing of Subroutine ELEKMF1D.

2.6.3 Two-Dimensional Problems

The ideas presented in Section 2.6.2 for one-dimensional problems extend in a
straightforward way to two-dimensional problems. The main diﬀerences are:
(a) numerical integration over two-dimensional elements; (b) the Jacobian Je
is the determinant of the Jacobian matrix [J e ] deÞned in Eq. (2.5.5); and
(c) global derivatives of the interpolation functions are determined using Eq.
(2.5.9).
The Fortran statements of generating element coeﬃcient matrices [K e ],
[M ], and {Qe } are provided in the form of subroutine ELEKMF2D in Box
e

2.6.3, along with the subroutine INTRPL2D for interpolation functions

of rectangular elements and subroutine TEMPORAL for generating the
coeﬃcient matrices [K̂ e ] and {F̂ e } deÞned in Eq. (8.2.10) for transient
problems.

Box 2.6.2 Listing of subroutine ELEKMF1D.

SUBROUTINE ELEKMF1D(MODEL,NDF,NGP,NPE)
C __________________________________________________________________
C
C IEL....…….. Element TYPE (1, Linear; 2, Quadratic)
C H......……… Element length
C X......……… Global (i.e. problem) coordinate
C XI ....……… Local (i.e. element) coordinate
C [GAUSPT].. 4×4 matrix of Gauss points: Nth column corresponds
C to the N-point Guass rule
C [GAUSWT]. 4×4 matrix of Gauss weights (see the comment above)
C [ELK]..…… Element coefficient matrix [K]
C {ELF}..….. Element source vector {f}
C {ELX}..….. Vector of the global coordinates of element nodes
C
IMPLICIT REAL*8(A-H,O-Z)
COMMON/STF1/ELK(9,9),ELF(9),ELX(4)
COMMON/STF2/AX0,AX1,CX0,CX1,FX0,FX1
COMMON/SHP/SF(4),GDSF(4)
DIMENSION GAUSPT(5,5),GAUSWT(5,5)
C
DATA GAUSPT/5*0.0D0,−0.57735027D0,0.57735027D0,3*0.0D0,−0.77459667D0,
* 0.0D0,0.77459667D0,2*0.0D0,−0.86113631D0,−0.33998104D0,0.33998104D0,
* 0.86113631D0,0.0D0,−0.906180D0,−0.538469D0,0.0D0,0.538469D0,0.906180D0/
DATA GAUSWT/2.0D0,4*0.0D0,2*1.0D0,3*0.0D0,0.55555555D0,0.88888888D0,
* 0.55555555D0,2*0.0D0,0.34785485D0,2*0.65214515D0,0.34785485D0,0.0D0,
* 0.236927D0,0.478629D0,0.568889D0,0.478629D0,0.236927D0/
C
H = ELX(NPE) - ELX(1)
IEL = NPE-1
C Initialize the arrays DO 10 J=1,NPE
ELF(J) = 0.0
DO 10 I=1,NPE
10 ELK(I,J)=0.0
C DO-LOOP on number of Gauss points begins here
DO 40 NI=1,NGP
XI=GAUSS(NI,NGP)
CALL INTRPL1D(ELX,GJ,IEL,NPE,XI)
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

CNST=GJ*WT(NI,NGP)
X=ELX(1)+0.5*(1.0+XI)*H
AX=AX0+AX1*X
CX=CX0+CX1*X
FX=FX0+FX1*X
C Calculate element coefficients
DO 20 I=1,NPE
ELF(I)=ELF(I)+FX*SFL(I)*CNST
DO 20 J=1,NPE
S00=SFL(I)*SFL(J)*CNST
S11=GDSFL(I)*GDSFL(J)*CNST
ELK(I,J)=ELK(I,J)+AX*S11+CX*S00
20 CONTINUE
40 CONTINUE
RETURN
END

2.7 Closure
The present chapter was devoted to a study of (1) the Þnite element
models of one- and two-dimensional problems involving Poisson’s equation,
(2) a derivation of interpolation functions for basic one- and two-
dimensional elements, (3) numerical evaluation of integrals, and (4) computer
implementation ideas. An understanding of the topics presented in this
chapter is a prerequisite for the subsequent chapters of this book.

Box 2.6.3 Listings of subroutines ELEKMF2D, INTRPL2D, and

TEMPORAL.
SUBROUTINE ELEKMF2D(NPE,NN,ITEM)
C ________________________________________________________________
C
C Element calculations based on linear and quadratic rectangular elements
C and isoparametric formulation are carried out for the heat transfer and
C penalty model of viscous incompressible fluid flow.
C ________________________________________________________________
C
IMPLICIT REAL*8(A-H,O-Z)
COMMON/STF/ELF(18),ELK(18,18),ELM(18,18),ELXY(9,2),ELU(18),A1,A2
COMMON/PST/A10,A1X,A1Y,A20,A2X,A2Y,C0,CX,CY,F0,FX,FY
COMMON/SHP/SF(9),GDSF(2,9),SFH(16)
COMMON/PNT/IPDF,IPDR,NIPF,NIPR
DIMENSION GAUSPT(5,5),GAUSWT(5,5)
COMMON/IO/IN,ITT
C
DATA GAUSPT/5*0.0D0, −0.57735027D0, 0.57735027D0, 3*0.0D0,
2 −0.77459667D0, 0.0D0, 0.77459667D0, 2*0.0D0, −0.86113631D0,
3 −0.33998104D0, 0.33998104D0, 0.86113631D0, 0.0D0, −0.90617984D0,
4 −0.53846931D0,0.0D0,0.53846931D0,0.90617984D0/
C
DATA GAUSWT/2.0D0, 4*0.0D0, 2*1.0D0, 3*0.0D0, 0.55555555D0,
2 0.88888888D0, 0.55555555D0, 2*0.0D0, 0.34785485D0,
3 2*0.65214515D0, 0.34785485D0, 0.0D0, 0.23692688D0,
4 0.47862867D0, 0.56888888D0, 0.47862867D0, 0.23692688D0/
C
NDF = NN/NPE
NET=NPE
C
C Initialize the arrays
C
DO 120 I = 1,NN
ELF(I) = 0.0
DO 120 J = 1,NN
IF(ITEM.NE.0) THEN
ELM(I,J)= 0.0
ENDIF
120 ELK(I,J)= 0.0
C
C Do-loops on numerical (Gauss) integration begin here. Subroutine
C INTRPL2D is called here
C
DO 200 NI = 1,IPDF
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

DO 200 NJ = 1,IPDF
XI = GAUSPT(NI,IPDF)
ETA = GAUSPT(NJ,IPDF)
CALL INTRPL2D (NPE,XI,ETA,DET,ELXY,NDF)
CNST = DET*GAUSWT(NI,IPDF)*GAUSWT(NJ,IPDF)
X=0.0
Y=0.0
DO 140 I=1,NPE
X=X+ELXY(I,1)*SF(I)
140 Y=Y+ELXY(I,2)*SF(I)
C
SOURCE=F0+FX*X+FY*Y
IF(ITEM.NE.0) THEN
CT=C0+CX*X+CY*Y
ENDIF
A11=A10+A1X*X+A1Y*Y
A22=A20+A2X*X+A2Y*Y
C
II=1
DO 180 I=1,NET
JJ=1
DO 160 J=1,NET
S00=SF(I)*SF(J)*CNST
S11=GDSF(1,I)*GDSF(1,J)*CNST
S22=GDSF(2,I)*GDSF(2,J)*CNST
C
C Heat transfer and like problems (i.e. single DOF problems):_______
C
ELK(I,J) = ELK(I,J) + A11*S11 + A22*S22 + A00*S00
IF(ITEM.NE.0) THEN
ELM(I,J) = ELM(I,J) + CT*S00
ENDIF
160 JJ = NDF*J+1
C Source of the form fx = F0 + FX*X + FY*Y is assumed
L=(I-1)*NDF+1
ELF(L) = ELF(L)+CNST*SF(I)*SOURCE
180 II = NDF*I+1
200 CONTINUE
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 55

IF(ITEM.NE.0) THEN
C
C Compute the coefficient matrices of the final algebraic equations
C (i.e. after time approximation) in the transient analysis:_______
C
CALL TEMPORAL(NN)
ENDIF
RETURN
END

SUBROUTINE INTRPL2D(NPE,XI,ETA,DET,ELXY,NDF)
C ________________________________________________________________
C
C The subroutine evaluates the interpolation functions (SF(I)) and
C their derivatives with respect to global coordinates (GDSF(I,J))
C for Lagrange linear & quadratic rectangular elements, using the
C isoparametric formulation. The subroutine also evaluates Hermite
C interpolation functions and their global derivatives using the
C subparametric formulation.
C
C SF(I).....……..Interpolation function for node I of the element
C DSF(J,I)..……Derivative of SF(I) with respect to XI if J=1 and
C and ETA if J=2
C GDSF(J,I)…...Derivative of SF(I) with respect to X if J=1 and
C and Y if J=2
C XNODE(I,J)...J-TH (J=1,2) Coordinate of node I of the element
C NP(I).....…….Array of element nodes (used to define SF and DSF)
C GJ(I,J)...…….Jacobian matrix
C GJINV(I,J)….Inverse of the jacobian matrix
C ________________________________________________________________
C
IMPLICIT REAL*8 (A-H,O-Z)
DIMENSION ELXY(9,2),XNODE(9,2),NP(9),DSF(2,9),GJ(2,2),GJINV(2,2)
COMMON/SHP/SF(9),GDSF(2,9)
COMMON/IO/IN,ITT
DATA XNODE/-1.0D0, 2*1.0D0, -1.0D0, 0.0D0, 1.0D0, 0.0D0, -1.0D0,
* 0.0D0, 2*-1.0D0, 2*1.0D0, -1.0D0, 0.0D0, 1.0D0, 2*0.0D0/
DATA NP/1,2,3,4,5,7,6,8,9/
C
FNC(A,B) = A*B
IF(NPE.EQ.4) THEN
C
C LINEAR Lagrange interpolation functions for FOUR-NODE element
C
DO 10 I = 1, NPE
XP = XNODE(I,1)
YP = XNODE(I,2)
XI0 = 1.0+XI*XP
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

ETA0=1.0+ETA*YP
SF(I) = 0.25*FNC(XI0,ETA0)
DSF(1,I)= 0.25*FNC(XP,ETA0)
10 DSF(2,I)= 0.25*FNC(YP,XI0)
ELSE
IF(NPE.EQ.8) THEN
C
C QUADRATIC Lagrange interpolation functions for EIGHT-NODE element
C
DO 20 I = 1, NPE
NI = NP(I)
XP = XNODE(NI,1)
YP = XNODE(NI,2)
XI0 = 1.0+XI*XP
ETA0 = 1.0+ETA*YP
XI1 = 1.0-XI*XI
ETA1 = 1.0-ETA*ETA
IF(I.LE.4) THEN
SF(NI) = 0.25*FNC(XI0,ETA0)*(XI*XP+ETA*YP-1.0)
DSF(1,NI) = 0.25*FNC(ETA0,XP)*(2.0*XI*XP+ETA*YP)
DSF(2,NI) = 0.25*FNC(XI0,YP)*(2.0*ETA*YP+XI*XP)
ELSE
IF(I.LE.6) THEN
SF(NI) = 0.5*FNC(XI1,ETA0)
DSF(1,NI) = -FNC(XI,ETA0)
DSF(2,NI) = 0.5*FNC(YP,XI1)
ELSE
SF(NI) = 0.5*FNC(ETA1,XI0)
DSF(1,NI) = 0.5*FNC(XP,ETA1)
DSF(2,NI) = -FNC(ETA,XI0)
ENDIF
ENDIF
20 CONTINUE
ELSE

C
C QUADRATIC Lagrange interpolation functions for NINE-NODE element
C
DO 30 I=1,NPE
NI = NP(I)
XP = XNODE(NI,1)
YP = XNODE(NI,2)
XI0 = 1.0+XI*XP
ETA0 = 1.0+ETA*YP
XI1 = 1.0-XI*XI
ETA1 = 1.0-ETA*ETA
XI2 = XP*XI
ETA2 = YP*ETA
IF(I .LE. 4) THEN
SF(NI) = 0.25*FNC(XI0,ETA0)*XI2*ETA2
DSF(1,NI)= 0.25*XP*FNC(ETA2,ETA0)*(1.0+2.0*XI2)
DSF(2,NI)= 0.25*YP*FNC(XI2,XI0)*(1.0+2.0*ETA2)
ELSE
IF(I .LE. 6) THEN
SF(NI) = 0.5*FNC(XI1,ETA0)*ETA2
DSF(1,NI) = -XI*FNC(ETA2,ETA0)
DSF(2,NI) = 0.5*FNC(XI1,YP)*(1.0+2.0*ETA2)
ELSE
IF(I .LE. 8) THEN
SF(NI) = 0.5*FNC(ETA1,XI0)*XI2
DSF(2,NI) = -ETA*FNC(XI2,XI0)
DSF(1,NI) = 0.5*FNC(ETA1,XP)*(1.0+2.0*XI2)
ELSE
SF(NI) = FNC(XI1,ETA1)
DSF(1,NI) = -2.0*XI*ETA1
DSF(2,NI) = -2.0*ETA*XI1
ENDIF
ENDIF
ENDIF
30 CONTINUE
ENDIF
ENDIF
C
C Compute the Jacobian matrix [GJ] and its inverse [GJINV], and [GDSF]
C
DO 40 I = 1,2
DO 40 J = 1,2
GJ(I,J) = 0.0
DO 40 K = 1,NPE
40 GJ(I,J) = GJ(I,J) + DSF(I,K)*ELXY(K,J)
C
DET = GJ(1,1)*GJ(2,2)-GJ(1,2)*GJ(2,1)
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

GJINV(1,1) = GJ(2,2)/DET
GJINV(2,2) = GJ(1,1)/DET
GJINV(1,2) = -GJ(1,2)/DET
GJINV(2,1) = -GJ(2,1)/DET
DO 50 I = 1,2
DO 50 J = 1,NPE
GDSF(I,J) = 0.0
DO 50 K = 1, 2
50 GDSF(I,J) = GDSF(I,J) + GJINV(I,K)*DSF(K,J)
RETURN
END
SUBROUTINE TEMPORAL(NN)
C ____________________________________________________________________________________
C
C The subroutine computes the algebraic equations associated with the parabolic differential equations
C by using the α-family of approximations. A constant source is assumed.
C ____________________________________________________________________________________
C
IMPLICIT REAL*8(A-H,O-Z)
COMMON/STF/ELF(18),ELK(18,18),ELM(18,18),ELXY(9,2),ELU(18),A1,A2
C
C The α-family of time approximation for parabolic equations
C
DO 20 I=1,NN
SUM=0.0
DO 10 J=1,NN
SUM=SUM+(ELM(I,J)-A2*ELK(I,J))*ELU(J)
10 ELK(I,J)=ELM(I,J)+A1*ELK(I,J)
20 ELF(I)=(A1+A2)*ELF(I)+SUM
RETURN
END

Problems
2.1 (Least-Squares Method). In the least-squares method, the residual Re [see Eq. (2.2.3)]
is minimized in the following sense:
Z xb Z xb
e 2 ∂
δ [R ] dx = 0 or [Re (x, c1 , c2 , · · · , cn )]2 dx = 0, i = 1, 2, · · · , n
xa
∂ci xa

(a) Identify the weight function wie if the least-squares method is to be deduced from
Eq. (2.2.4), (b) develop the least-squares Þnite element model, and (c) discuss the
type of Þnite element approximation of u that may be used.
2.2 Consider the diﬀerential equation
³ ´ µ ¶
d du d2 d2 u
− a + 2 b + cu = f
dx dx dx dx2

where a, b, c, and f are known functions of position x. (a) Develop the weak form
over a typical element Ωe = (xa , xb ) such that the bilinear form is symmetric, (b)
identify the bilinear and linear forms and construct the quadratic functional, (c)
develop the Þnite element model of the equation, and (d) discuss the type of Þnite
element approximation of u that may be used.
2.3 Derive the Lagrange cubic interpolation functions for a four-node (one-dimensional)
element (with equally spaced nodes) using the alternative procedure based on
interpolation properties (2.2.22). Use the local coordinate x̄ for simplicity.
2.4 Derive the Þnite element model of the diﬀerential equation
³ ´
d du
− a(x) = f (x) for 0<x<L
dx dx
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

for the boundary conditions

h³ du
´ i
u(0) = u0 , a(x) + ku =P
dx x=L

2.5 The following diﬀerential equation arises in connection with heat transfer in an
insulated rod: ³ dT ´
d
− k = q for 0 < x < L
dx dx
h i¯ ¯
dT
T (0) = T0 , k + β(T − T∞ ) + q̂ ¯¯ =0
dx x=L

where T is the temperature, k the thermal conductivity, and q the heat generation.
Take the following values for the data: q = 0, q̂ = 0, L = 0.1 m, k = 0.01 W m−1 ◦ C−1 ,
β = 25 W m−2◦ C−1 . T0 = 50◦ C, and T∞ = 5◦ C. Solve the problem using two linear
Þnite elements for temperature values at x = 12 L and L. Answer: U2 = 27.59◦ C,
U3 = 5.179◦ C, Q(1)
1 = 4.482 W m
−2 = −Q(2) .
2
2.6 An insulating wall is constructed of three homogeneous layers with conductivities
k1 , k2 , and k3 in intimate contact (see Figure P2.6). Under steady-state conditions,
the temperatures at the boundaries of the layers are characterized by the external
surface temperatures T1 and T4 and the interface temperatures T2 and T3 . Formulate
the problem to determine the temperatures Ti (i = 1, . . . , 4) when the ambient
temperatures T0 and T5 and the (surface) Þlm coeﬃcients β0 and β5 are known.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
58 NONLINEAR FINITE ELEMENT ANALYSIS

Assume that there is no internal heat generation and that the heat ßow is one-
dimensional (∂T /∂y = 0). Answer: U1 = 84.489◦ C, U2 = 68.977◦ C, U3 = 50.881◦ C,
U4 = 45.341◦ C, (Q11 )def = 217.16 W-m−2 , (Q32 )def = −155.11 W-m−2 .

h1 h2 h3

k1 = 90 W/(m ºC)
k2 = 75 W/(m ºC)
k3 = 50 W/(m ºC) L
h1 = 0.03 m
h2 = 0.04 m 1 2 3 4
h3 = 0.05 m
1 2 3
β = 500 W/(m2 ºC)
T∞ = 20ºC
x

Figure P2.6
2.7 Consider the steady laminar ßow of a viscous ßuid through a long circular cylindrical
tube. The governing equation is

where w is the axial (i.e. z) component of velocity, µ is the viscosity, and f0 is

the gradient of pressure (which includes the combined eﬀect of static pressure and
gravitational force). The boundary conditions are

³ dw ´ ¯¯
r ¯ = 0, w(R0 ) = 0
dr ¯ r=0

Using the symmetry and (a) two linear elements, (b) one quadratic element, determine
the velocity Þeld and compare with the exact solution at the nodes:
∙ ³ r ´2 ¸
f0 R02
we (r) = 1−
4µ R0

2.8. In the problem of the ßow of a viscous ßuid through a circular cylinder (see Problem
2.7), assume that the ßuid slips at the cylinder wall; that is, instead of assuming that
w = 0 at r = R0 , use the boundary condition that

dw
kw = −µ at r = R0
dr

in which k is the “coeﬃcient of sliding friction.” Solve the problem with two linear
elements.
Reddy, J. N.. Introduction to Nonlinear Finite Element Analysis, Oxford University Press, Incorporated, 2004. ProQuest Ebook
Central, http://ebookcentral.proquest.com/lib/UNICAF/detail.action?docID=472350.
Created from UNICAF on 2024-05-16 23:18:23.
THE FINITE ELEMENT METHOD: A REVIEW 59

2.9 The two members in Figure P2.9 are fastened together and to rigid walls. If
the members are stress free before they are loaded, what will be the stresses and
deformations in each after the two 50,000 lbs. loads are applied? Use Es = 30 × 106
psi and Ea = 107 psi; the aluminum rod is 2 in. in diameter and the steel rod is 1.5 in.
in diameter. Answer: U2 = 0.0134 in., P2(1) = −21, 052.6 lb, and σ (1) = 6701.25 psi.

Aluminum
Steel
P = 50 k

A C B
P = 50 k
20 in. 16 in.

Figure P2.9
2.10 Evaluate the coeﬃcients e
and Fie of Eqs. (2.3.40a,b) for a linear triangular element.
Kij
2.11 Repeat Problem 2.10 for a linear rectangular element.
2.12 Consider the partial diﬀerential equation governing heat transfer in an axisymmetric
geometry ³ ´ ∂ ³ ∂T ´
1 ∂ ∂T
− rkrr − kzz = f (r, z) (1)
r ∂r ∂r ∂z ∂z
where (krr , kzz ) and f are the conductivities and internal heat generation per unit
volume, respectively. In developing the weak form, we integrate over the elemental
volume of the axisymmetric geometry: r dr dθ dz. Develop the weak form and
associated Þnite element model over an element.
Copyright © 2004. Oxford University Press, Incorporated. All rights reserved.

References
1. Reddy, J. N., An Introduction to the Finite Element Method, 2nd edn, McGraw—Hill,
New York (1993).
2. Reddy, J. N. and Gartling, D. K., The Finite Element Method in Heat Transfer and
Fluid Dynamics, 2nd edn, CRC Press, Boca Raton, FL (2001).

Numerical Methods in Science and Engineering Theories With MATLAB, Mathematica, Fortran, C and Python Programs (P. Dechaumphai, N. Wansophark)
No ratings yet
Numerical Methods in Science and Engineering Theories With MATLAB, Mathematica, Fortran, C and Python Programs (P. Dechaumphai, N. Wansophark)
389 pages
Sow Add Maths F5 2023 (Premier) - Penuh PDF
No ratings yet
Sow Add Maths F5 2023 (Premier) - Penuh PDF
24 pages
Full
No ratings yet
Full
93 pages
BENG 221 Mathematical Methods in Bioengineering
No ratings yet
BENG 221 Mathematical Methods in Bioengineering
19 pages
applied-maths-2
No ratings yet
applied-maths-2
2 pages
Math 221E Module 11st Edition With Erratum Protected
No ratings yet
Math 221E Module 11st Edition With Erratum Protected
20 pages
Solid Mechanics - Week 2
No ratings yet
Solid Mechanics - Week 2
4 pages
Research paper 4
No ratings yet
Research paper 4
12 pages
Research Notes: Boundary Element Method and Wave Equation
No ratings yet
Research Notes: Boundary Element Method and Wave Equation
4 pages
A Transformational Approach To Teaching Matrix Structural Analysis
No ratings yet
A Transformational Approach To Teaching Matrix Structural Analysis
17 pages
Chapter2 PCA
No ratings yet
Chapter2 PCA
65 pages
Unit5 1
No ratings yet
Unit5 1
98 pages
QPE QM-Applications
No ratings yet
QPE QM-Applications
44 pages
Hamilton and Symmetry Project Report PDF
No ratings yet
Hamilton and Symmetry Project Report PDF
11 pages
Illustrative Example: A Blending Process: An Unsteady-State Mass Balance For The Blending System
No ratings yet
Illustrative Example: A Blending Process: An Unsteady-State Mass Balance For The Blending System
23 pages
CE6109 - Chapter1
No ratings yet
CE6109 - Chapter1
11 pages
5 Matemetika-2
No ratings yet
5 Matemetika-2
147 pages
CH 03
No ratings yet
CH 03
8 pages
Ordinary DifferentialEquations 20MAT21C3
No ratings yet
Ordinary DifferentialEquations 20MAT21C3
116 pages
Mathematics Calculus of Variations
No ratings yet
Mathematics Calculus of Variations
46 pages
FEM Introduction
No ratings yet
FEM Introduction
127 pages
Analysis Manual For MIDAS Meshfree
No ratings yet
Analysis Manual For MIDAS Meshfree
16 pages
qq
No ratings yet
qq
12 pages
Kaleay Hagos 2018
No ratings yet
Kaleay Hagos 2018
34 pages
DEVC_QUESTION_BANK
No ratings yet
DEVC_QUESTION_BANK
39 pages
matecconf_simpro21_01015
No ratings yet
matecconf_simpro21_01015
7 pages
Elective Maths Syllabus (1) 11 18
No ratings yet
Elective Maths Syllabus (1) 11 18
8 pages
Mathematics (Elective)
No ratings yet
Mathematics (Elective)
55 pages
Boundary Value Problems For Differential
No ratings yet
Boundary Value Problems For Differential
12 pages
9 Final Exam
No ratings yet
9 Final Exam
3 pages
Differential Equation Hons - Dr. Prasanta Das
No ratings yet
Differential Equation Hons - Dr. Prasanta Das
28 pages
Mini Project Report Structural Dynamics - R1
No ratings yet
Mini Project Report Structural Dynamics - R1
4 pages
Cost Function - Coursera
No ratings yet
Cost Function - Coursera
1 page
Full
No ratings yet
Full
92 pages
Object Oriented Analysis
No ratings yet
Object Oriented Analysis
19 pages
Scaling Monosemanticity - Extracting Interpretable Features From Claude 3 Sonnet
No ratings yet
Scaling Monosemanticity - Extracting Interpretable Features From Claude 3 Sonnet
75 pages
SME1308
No ratings yet
SME1308
224 pages
Calculus 2: Engr. Luisito Lolong Lacatan, Pcpe, PHD
No ratings yet
Calculus 2: Engr. Luisito Lolong Lacatan, Pcpe, PHD
46 pages
Final Exam 23- 12-2021
No ratings yet
Final Exam 23- 12-2021
3 pages
Lecture 1 02-01-2022
No ratings yet
Lecture 1 02-01-2022
49 pages
Problems and Solutions in Quantum Mechanics 1st Edition Kyriakos Tamvakis instant download
100% (1)
Problems and Solutions in Quantum Mechanics 1st Edition Kyriakos Tamvakis instant download
59 pages
Dynamic Neural Networks - Structures and Training Methods
No ratings yet
Dynamic Neural Networks - Structures and Training Methods
57 pages
B.Sc. Allied Mathematics: Pondi Cherry Uni Versi Ty
No ratings yet
B.Sc. Allied Mathematics: Pondi Cherry Uni Versi Ty
8 pages
Lecture1 PDF
No ratings yet
Lecture1 PDF
25 pages
Himanshu Pandey M.sc.
No ratings yet
Himanshu Pandey M.sc.
73 pages
2.1 Communication Skills - Ii
No ratings yet
2.1 Communication Skills - Ii
23 pages
Lecture 3
No ratings yet
Lecture 3
16 pages
COS2633 - Lesson 2 - 3 - 2020
No ratings yet
COS2633 - Lesson 2 - 3 - 2020
14 pages
Maximum Likelihood Estimates of Linear Dynamic Systems (1965)
No ratings yet
Maximum Likelihood Estimates of Linear Dynamic Systems (1965)
14 pages
Ba(Math)s1p1u1&2e
No ratings yet
Ba(Math)s1p1u1&2e
90 pages
Introduction
No ratings yet
Introduction
38 pages
ENST, Signal Dept. & CNRS, URA 820 46, Rue Barrault, 75 634 Paris Cedex 13, Matignon@
No ratings yet
ENST, Signal Dept. & CNRS, URA 820 46, Rue Barrault, 75 634 Paris Cedex 13, Matignon@
6 pages
Database Management System Notes
No ratings yet
Database Management System Notes
28 pages
Learning Guide 2 - Rational Functions
No ratings yet
Learning Guide 2 - Rational Functions
26 pages
Calculus 1A - Syllabus (MAT1060)
No ratings yet
Calculus 1A - Syllabus (MAT1060)
8 pages
11959-Article Text-34037-1-10-20150416 PDF
No ratings yet
11959-Article Text-34037-1-10-20150416 PDF
8 pages
MathPhys2 M1
No ratings yet
MathPhys2 M1
22 pages
Mathematical Modeling of Chemical Processes
No ratings yet
Mathematical Modeling of Chemical Processes
43 pages
Grade 11 Physics Anual Plan
100% (1)
Grade 11 Physics Anual Plan
7 pages
Geometric functions in computer aided geometric design
From Everand
Geometric functions in computer aided geometric design
Oscar Ruiz
No ratings yet
Complex Variables
From Everand
Complex Variables
Francis J. Flanigan
No ratings yet
Math Ass
No ratings yet
Math Ass
13 pages
Math 250 A
No ratings yet
Math 250 A
13 pages
POLYNOMIALS ASSERTION AND REASON (1)
No ratings yet
POLYNOMIALS ASSERTION AND REASON (1)
4 pages
CHE 536 Engineering Optimization: Course Policies and Outline
No ratings yet
CHE 536 Engineering Optimization: Course Policies and Outline
33 pages
OT - Transportation and Assignment
No ratings yet
OT - Transportation and Assignment
74 pages
Load Flow Gauss-Seidel Method EE 452: Computer Methods in Power Systems
No ratings yet
Load Flow Gauss-Seidel Method EE 452: Computer Methods in Power Systems
9 pages
Tugas Aljabar Linear
No ratings yet
Tugas Aljabar Linear
6 pages
Dual Simplex
No ratings yet
Dual Simplex
32 pages
A Level Math Paper 1 Polynomials PDF
No ratings yet
A Level Math Paper 1 Polynomials PDF
7 pages
Practical Methods of Optimization 2nd Edition: Book Information Sheet Book Information Sheet
No ratings yet
Practical Methods of Optimization 2nd Edition: Book Information Sheet Book Information Sheet
1 page
Maths (041) Xii PB 1 QP Set A
No ratings yet
Maths (041) Xii PB 1 QP Set A
7 pages
Dokumen - Pub - Numerical Approximation of Partial Differential Equations 1st Ed 3319323539 978 3 319 32353 4 978 3 319 32354 1 3319323547
100% (1)
Dokumen - Pub - Numerical Approximation of Partial Differential Equations 1st Ed 3319323539 978 3 319 32353 4 978 3 319 32354 1 3319323547
541 pages
Newton
No ratings yet
Newton
437 pages
A Iserles
No ratings yet
A Iserles
70 pages
Lecture 4 - Simplex Method
No ratings yet
Lecture 4 - Simplex Method
19 pages
Secant Method
No ratings yet
Secant Method
8 pages
Binomial Quiz
No ratings yet
Binomial Quiz
1 page
D1, L9 Solving Linear Programming Problems
No ratings yet
D1, L9 Solving Linear Programming Problems
16 pages
Assignment - 2 Numerical Solutions of Ordinary and Partial Differential Equations
No ratings yet
Assignment - 2 Numerical Solutions of Ordinary and Partial Differential Equations
1 page
Graphical Method
No ratings yet
Graphical Method
11 pages
Strassen Algorithm
No ratings yet
Strassen Algorithm
2 pages
Lec - 3 Bisection Method
No ratings yet
Lec - 3 Bisection Method
31 pages
Lab 3 Report
No ratings yet
Lab 3 Report
8 pages
OSM M2Drill 02 e
No ratings yet
OSM M2Drill 02 e
20 pages
ASSIGNMENT: Numerical Analysis Submitted To: Miss Sidra Ayub Submitted by
No ratings yet
ASSIGNMENT: Numerical Analysis Submitted To: Miss Sidra Ayub Submitted by
7 pages
Akar Persamaan 1
No ratings yet
Akar Persamaan 1
15 pages
Chapter 4 Numerical Differentiation and Integration
No ratings yet
Chapter 4 Numerical Differentiation and Integration
3 pages
Adaptive Quad Proof
No ratings yet
Adaptive Quad Proof
6 pages