1

Lecture 1: Natural Coordinates

Natural coordinate system is basically a local coordinate system which allows the specification of a
point within the element by a set of dimensionless numbers whose magnitude never exceeds unity.
This coordinate system is found to be very effective in formulating the element properties in finite
element formulation. This system is defined in such that the magnitude at nodal points will have
unity or zero or a convenient set of fractions. It also facilitates the integration to calculate element
stiffness.

3.1.1 One Dimensional Line Elements

The line elements are used to represent spring, truss, beam like members for the finite element
analysis purpose. Such elements are quite useful in analyzing truss, cable and frame structures. Such
structures tend to be well defined in terms of the number and type of elements used. For example, to
represent a truss member, a two node linear element is sufficient to get accurate results. However,
three node line elements will be more suitable in case of analysis of cable structure to capture the
nonlinear effects. The natural coordinate system for one dimensional line element with two nodes is
shown in Fig. 3.1.1. Here, the natural coordinates of any point P can be defined as follows.
x x
N1  1 and N 2  (3.1.1)
l l
Where, x is represented in Cartesian coordinate system. Similarly, x/l can be represented as ξ in
natural coordinate system. Thus the above expression can be rewritten in the form of natural
coordinate system as given below.
N1  1 x and N 2  x (3.1.2)
Now, the relationship between natural and Cartesian coordinates can be expressed from eq. (3.1.1) as
1 1 1  N1 
      (3.1.3)
 x 0 l   N 2 
Here, N1 and N2 is termed as shape function as well. The variation of the magnitude of two linear
shape functions (N1 and N2) over the length of bar element are shown in Fig. 3.1.2. This example
displays the simplest form of interpolation function. The linear interpolation used for field variable φ
can be written as
f x   f1 N1  f2 N 2 (3.1.4)
 2

Fig. 3.1.1 Two node line element

Fig. 3.1.2 Linear interpolation function for two node line element

Similarly, for three node line element, the shape function can be derived with the help of natural
coordinate system which may be expressed as follows:
 3


 3x 2 x 2  

1  2  
 l  
 
 N1 

l  1  3x  2x 2 

    2 
 
 
 (3.1.5)

   2 
    4x 4x
    x  x 
  
2
N N 4 4

 
 
 l l 2

 
 

       
x  2x
2
 N3 
  x 2 x2  


  


 l2 

 l 
The detailed derivation of the interpolation function will be discussed in subsequent lecture. The
variation of the shape functions over the length of the three node element are shown in Fig. 3.1.3

Fig. 3.1.3 Variation of interpolation function for three node line element

Now, if φ is considered to be a function of L1 and L2, the differentiation of φ with respect to x for
two node line element can be expressed by the chain rule formula as
df f L1 f L 2
 .  . (3.1.6)
dx L1 x L 2 x
Thus, eq.(3.1.4) can be written as
L1 1 L2 1
  and  (3.1.7)
x l x l
Therefore,
 4

d 1    
   
dx l  L2 L1 
(3.1.7)

The integration over the length l in natural coordinate system can be expressed by
p !q !
L L2 q dl 
p
l (3.1.9)
l
1
 p  q 1!
Here, p! is the factorial product p(p-1)(p-2)….(1) and 0! is defined as equal to unity.

3.1.2 Two Dimensional Triangular Elements

The natural coordinate system for a triangular element is generally called as triangular coordinate
system. The coordinate of any point P inside the triangle is x,y in Cartesian coordinate system. Here,
three coordinates, L1, L2 and L3 can be used to define the location of the point in terms of natural
coordinate system. The point P can be defined by the following set of area coordinates:
𝐴1 𝐴2 𝐴3
𝐿1 = ; 𝐿2 = ; 𝐿3 = (3.1.10)
𝐴 𝐴 𝐴
Where,
𝐴1 = Area of the triangle P23
𝐴2 = Area of the triangle P13
𝐴3 = Area of the triangle P12
A=Area of the triangle 123
Thus,
𝐴 = 𝐴1 + 𝐴2 + 𝐴3
and
𝐿1 + 𝐿2 + 𝐿3 = 1 (3.1.11)
Therefore, the natural coordinate of three nodes will be: node 1 (1,0,0); node 2 (0,1,0); and node 3
(0,0,1).
 5

Fig. 3.1.4 Triangular coordinate system

The area of the triangles can be written using Cartesian coordinates considering x, y as coordinates
of an arbitrary point P inside or on the boundaries of the element:

1 𝑥1 𝑦1
1
A = �1 𝑥2 𝑦2 �
2
1 𝑥3 𝑦3
1 𝑥 𝑦
1
A1 = �1 𝑥2 𝑦2 �
2
1 𝑥3 𝑦3
1 𝑥 𝑦
1
A2 = � 1 𝑥3 𝑦3 �
2
1 𝑥1 𝑦1
1 𝑥 𝑦
1
A3 = �1 𝑥1 𝑦1 �
2
1 𝑥2 𝑦2

The relation between two coordinate systems to define point P can be established by their nodal
coordinates as
1 1 1 1 𝐿1
� � = � 1 𝑥2
𝑥 𝑥 𝑥3 � �𝐿2 � (3.1.12)
𝑦 𝑦1 𝑦2 𝑦3 𝐿3
Where,
𝑥 = 𝐿1 𝑥1 + 𝐿2 𝑥2 + 𝐿3 𝑥3
 6

𝑦 = 𝐿1 𝑦1 + 𝐿2 𝑦2 + 𝐿3 𝑦3
The inverse between natural and Cartesian coordinates from eq.(3.1.12) may be expressed as
𝐿1 𝑥2 𝑦3 −𝑥3 𝑦2 𝑦2 − 𝑦3 𝑥3 − 𝑥2 1
1
�𝐿2 � = 2𝐴 �𝑥3 𝑦1 −𝑥1 𝑦3 𝑦3 − 𝑦1 𝑥1 − 𝑥3 � �𝑥 � (3.1.13)
𝐿3 𝑥1 𝑦2 −𝑥2 𝑦1 𝑦1 − 𝑦2 𝑥2 − 𝑥1 𝑦
The derivatives with respect to global coordinates are necessary to determine the properties of an
element. The relationship between two coordinate systems may be computed by using the chain rule
of partial differentiation as
  L1  L 2  L 3
 .  .  .
x L1 x L 2 x L3 x
b1  b  b 
 .  2.  3.
2A L1 2A L 2 2A L3 (3.1.14)
3
bi 
 .
i1 2A Li

Where, b1 = y2 – y3; b2 = y3 – y1 and b3 = y1 – y2. Similarly, following relation can be obtained.

 3
c 
 i .
y i1 2A Li (3.1.15)

Where, c1 = x3 – x2; c2 = x1 – x3 and c3 = x2 – x1. The above expressions are looked cumbersome.
However, the main advantage is the ease with which polynomial terms can be integrated using
following area integral expression.

p!q!r!
L L 2 q L3r dA 
p
2A (3.1.16)
A
1
p  q  r  2!

Where 0! is defined as unity.

3.1.3 Shape Function using Area Coordinates

The interpolation functions for the triangular element are algebraically complex if expressed in
Cartesian coordinates. Moreover, the integration required to obtain the element stiffness matrix is
quite cumbersome. This will be discussed in details in next lecture. The interpolation function and
subsequently the required integration can be obtained in a simplified manner by the concept of area
coordinates. Considering a linear displacement variation of a triangular element as shown in Fig.
3.1.5, the displacement at any point can be written in terms of its area coordinates.
 7

u  1L1   2 L 2   3L3
u  f  
T
Or, (3.1.17)

where, f   L1 L3  and    1  2  3 

T T
L2
𝐴1 𝐴2 𝐴3
And 𝐿1 = ; 𝐿2 = ; 𝐿3 = (3.1.18)
𝐴 𝐴 𝐴
Here, A is the total area of the triangle. It is important to note that the area coordinates are dependent
as 𝐿1 + 𝐿2 + 𝐿3 = 1 . It may be seen from figure that at node 1, L1 = 1 while L2 = L3 = 0. Similarly
for other two nodes: at node 2, L2 = 1 while L1 = L3 = 0, and L3 = 1 while L2 = L1 = 0. Now,
substituting the area coordinates for node 1, 2 and 3, the displacement components at nodes can be
written as
 u1  1 0 0
   
u i   u 2   0 1 0   (3. 1.19)
 
u 3  0 0 1
Thus, from the above expression, one can obtain the unknown coefficient  :
1 0 0  u1 
   
   0 1 0 u 2 
 (3. 1.20)
 
 0 0 1 u 3 
  

Fig. 3.1.5 Area coordinates for triangular element

Now, eq.(3.1.17) can be written as:

 8

1 0 0  u1   1 0 0
    
 
T 
u  f 0 1 0 u 2   f 0 1 0 u i 
 T
 (3. 1.21)
 
0 0 1 u 3   0 0 1
     
The above expression can be written in terms of interpolation function as u  N u i 
T

Where,
 1 0 0
 
N   L1 L2 L3  0 1 0   L1 L 2 L3 
T
(3. 1.22)
 
 0 0 1
 
Similarly, the displacement variation v in Y direction can be expressed as follows.

v   N  v i 
T
(3.1.23)
Thus, for two displacement components u and v of any point inside the element can be written as:
 0  ui 
u  N 
T T

d  =   =  T T  
(3.1.24)
v  0  N   vi 
Thus, the shape function of the element will become
 L1 L2 L3 0 0 0
 N     (3.1.25)
0 0 0 L1 L2 L3 
It is important to note that the shape function Ni become unity at node i and zero at other nodes of
the element. The displacement at any point of the element can be expressed in terms of its nodal
displacement and the interpolation function as given below.
u = N1u1 + N 2u2 + N 3u3
(3.1.26)
v = N1v1 + N 2v2 + N 3v3
 9

Lecture 2: Triangular Elements

The triangular element can be used to represent the arbitrary geometry much easily. On the other
hand, rectangular elements, in general, are of limited use as they are not well suited for representing
curved boundaries. However, an assemblage of rectangular and triangular element with triangular
elements near the boundary can be very effective (Fig. 3.2.1). Triangular elements may also be used
in 3-dimensional axi-symmetric problems, plates and shell structures. The shape function for
triangular elements (linear, quadratic and cubic) with various nodes (Fig. 3.2.2) can be formulated.
An internal node will exist for cubic element as seen in Fig. 3.2.2(c).

Fig. 3.2.1 Finite element mesh consisting of triangular and rectangular element

Fig. 3.2.2 Triangular elements

In displacement formulation, it is very important to approximate the variation of displacement in the

element by suitable function. The interpolation function can be derived either using the Cartesian
coordinate system or by the area coordinates.
 10

3.2.1 Shape function using Cartesian coordinates

Polynomials are easiest way of mathematical operation for expressing variation of displacement. For
example, the displacement variation within the element can be represented by the following function
in case of two dimensional plane stress/strain problems.
u = α⍺0 + α⍺1 x + α⍺2 y
(3.2.1)
v = α⍺3 + α⍺4 x + ⍺α5 y
(3.2.2)
where α0, α1, α2 ….. are unknown coefficients. Thus the displacement vectors at any point P, in the
element (Fig.3.2.3) can be expressed with the following relation.
𝛼0
⎧𝛼1 ⎫
⎪ ⎪
𝑢 1 𝑥 𝑦 0 0 0 𝛼2
{𝑑} = � � = � � 𝛼 (3.2.3)
𝑣 0 0 0 1 𝑥 𝑦 ⎨ 3⎬
⎪𝛼4 ⎪
⎩𝛼5 ⎭
Or, {d}=[ 𝜙]{ α } (3.2.4)

Fig. 3.2.3 Triangular element in Cartesian Coordinates

Similarly, for “m” node element having three degrees of freedom at each node, the displacement
function can be expressed as
𝑢 = 𝛼0 + 𝛼1 𝑥 + 𝛼2 𝑦 + 𝛼3 𝑥 2 + 𝛼4 𝑥𝑦 + 𝛼5 𝑦 2 + . . … . . . +𝛼𝑚−1 𝑦 𝑛
𝑣 = 𝛼𝑚 + 𝛼𝑚+1 𝑥 + 𝛼𝑚+2 𝑦 + 𝛼𝑚+3 𝑥 2 + 𝛼𝑚+4 𝑥𝑦+. . . . . +𝛼2𝑚−1 𝑦 𝑛 (3.2.5)
2 𝑛
𝑤 = 𝛼2𝑚 + 𝛼2𝑚+1 𝑥 + 𝛼2𝑚+2 𝑦 + 𝛼2𝑚+3 𝑥 + 𝛼2𝑚+4 𝑥𝑦+. . . . . +𝛼3𝑚−1 𝑦
 11

Hence, in such case,

𝑢 {ϕ}𝑇 0 0
{𝑑} = � 𝑣 � = � 0 𝑇
{ϕ} 0 � {𝛼} (3.2.6)
𝑤 0 0 {ϕ}𝑇
Where, {𝛼}𝑇 = [𝛼0 𝛼1 . . . 𝛼3𝑚−1 ] and, [𝜙]𝑇 = [1 𝑥 𝑦 𝑥 2 𝑥𝑦 . . . . 𝑦 𝑛 ]
Now, for a linear triangular element with 2 degrees of freedom, eq. (3.2.3) can be written in terms of
the nodal displacements as follows.
𝑢1 1 𝑥1 𝑦1 0 0 0 𝛼0
⎧𝑢2 ⎫ ⎡1 𝑥2 𝑦2 0 0 0 ⎤ ⎧𝛼1 ⎫
⎪𝑢 ⎪ ⎢1 𝑥 𝑦 0 0 0 ⎥ ⎪𝛼 ⎪
3 ⎥ 2
{𝑑} = 𝑣 = ⎢ 3 3
(3.2.7)
⎨ 1 ⎬ ⎢0 0 0 1 𝑥1 𝑦1 ⎥ ⎨𝛼3 ⎬
⎪𝑣2 ⎪ ⎢0 0 0 1 𝑥2 𝑦2 ⎥ ⎪𝛼4 ⎪
⎩𝑣3 ⎭ ⎣0 0 0 1 𝑥3 𝑦3 ⎦ ⎩𝛼5 ⎭

Where, {d} is the nodal displacements. To simplify the above expression for finding out the shape
function, the displacements in X direction can be separated out which will be as follows:

𝑢1 1 𝑥1 𝑦1 ⍺∝0
{𝑢𝑖 }=� 2 � = �1 𝑥2 𝑦2 � �⍺∝1 �
𝑢 (3.2.8)
𝑢3 1 𝑥3 𝑦3 ⍺∝2
To obtain the polynomial coefficients, {α} the matrix of the above equation are to be inverted. Thus,
1
 0  1 x1 y1 
 u1   x 2 y3  x 3 y 2 x 3 y1  x1 y3 x1 y 2  x 2 y1  u1 
    1    
   1 x 2 y2  u   y 2  y3 y3  y1 y1  y 2  u 2 
 1     2  2A   
 2  1 x 3 y3  u 3   x3  x 2 x1  x 3 x 2  x1 u 3 
  
 a1 a2 a 3  u1 
1    
 b1 b2 b 3  u 2  (3.2.9)
2A   
 c1 c2 c3 u 3 
Where, A is the area of the triangle and can be obtained as follows.
1 𝑥1 𝑦1
1
A = 2 �1 𝑥2 𝑦2 � (3.2.10)
1 𝑥3 𝑦3
Now, eq. (3.2.1) can be written from the above polynomial coefficients.
 12

1 
u  x 2 y3  x 3 y 2    y 2  y3  x   x 3  x 2  y u1
2A 
1 
  x 3 y1  x1y3    y3  y1  x   x1  x 3  y u 2 (3.2.11)
2A 
1 
  x1y 2  x 2 y1    y1  y 2  x   x 2  x1  y u 2
2A 
Thus, the interpolation function can be obtained from the above as:

 

1 
 x y  x y    y  y  x   x  x  y 


  2 3 3 2 2 3 3 2 

 N1 
   2A 

     
N   N 2   
1  x 3 y1  x1y3    y3  y1  x   x1  x 3  y  
   2A    (3.2.12)

 N 
 
 


   1
3  

              

x
 1 2 y x y
2 1 y1 y 2 x x 2 x 1 y 
 2A
 


Such three node triangular element is commonly known as constant strain triangle (CST) as its strain
is assumed to be constant inside the element. This property may be derived from eq. (3.2.1) and
eq.(3.2.2). For example, in case of 2-D plane stress/strain problem, one can express the strain inside
the triangle with the help of eq.(3.2.1) and eq.(3.2.2):
u  ( 0  1x   2 y)
x    1
x x
v  ( 3   4 x   5 y)
y    5 (3.2.13)
y y
v u
 xy    2  2
x y
CST is the simplest element to develop mathematically. As there is no variation of strain inside the
element, the mesh size of the triangular element should be small enough to get correct results. This
element produces constant temperature gradients ensuring constant heat flow within the element for
heat transfer problems.

3.2.2 Higher Order Triangular Elements

Higher order elements are useful if the boundary of the geometry is curve in nature. For curved case,
higher order triangular element can be suited effectively while generating the finite element mesh.
Moreover, in case of flexural action in the member, higher order elements can produce more
accurate results compare to those using linear elements. Various types of higher order triangular
 13

elements are used in practice. However, most commonly used triangular element is the six node
element for which development of shape functions are explained below.

3.2.2.1 Shape function for six node element

Fig. 3.2.4 shows a triangular element with six nodes. The additional three nodes (4, 5, and 6) are
situated at the midpoints of the sides of the element. A complete polynomial representation of the
field variable can be expressed with the help of Pascal triangle:
φ ( x, y ) =α 0 + α1 x + α 2 y + α 3 x 2 + α 4 xy + α 5 y 2
(3.2.14)

Fig. 3.2.4 (a) Six node triangular element (b) Lines of constant values of the area coordinates

Using the above field variable function, one can reach the following expression using interpolation
function and the nodal values.
6
f x, y   N i  x, yfi (3.2.15)
i1

Here, the every shape function must be such that its value will be unity if evaluated at its related
node and zero if evaluated at any of the other five nodes. Moreover, as the field variable
representation is quadratic, each interpolation function will also become quadratic. Fig. 3.2.4(a)
shows the six node element with node numbering convention along with the area coordinates at three
corners. The six node element with lines of constant values of the area coordinates passing through
the nodes is shown in Fig. 3.2.4(b). Now the interpolation functions can be constructed with the help
of area coordinates from the above diagram. For example, the interpolation function N1 should be
unity at node 1 and zero at all other five nodes. According to the above diagram, the value of L1 is 1
 14

at node 1 and ½ at node 4 and 6. Again, L1 will be 0 at nodes 2, 3 and 5. To satisfy all these
conditions, one can propose following expression:
 1
N1  x, y  N1 L1 ,L 2 ,L3   L1 L1  
 2 (3.2.16)

Evaluating the above expression, the value of N1 is becoming ½ at node 1 though it must become
unity. Therefore, the above expression is slightly modified satisfying all the conditions and will be as
follows:
 1
N1  2L1 L1    L1 2L1 1
 2 (3.2.17)

Eq. (3.2.17) assures the required conditions at all the six nodes and is a quadratic function, as L1 is a
linear function of x and y. The remaining five interpolation functions can also be obtained in similar
fashion applying the required nodal conditions. Thus, the shape function for the six node triangle
element can be written as given below.
N1  L1 2L1 1
N 2  L 2 2L 2 1
N 3  L3 2L3 1 (3.2.18)
N 4  4L1L 2
N 5  4L 2 L3
N 6  4L3L1
Such six node triangular element is commonly known as linear strain triangle (LST) as its strain is
assumed to vary linearly inside the element. In case of 2-D plane stress/strain problem, the element
displacement field for such quadratic triangle may be expressed as
u ( x, y ) =α 0 + α1 x + α 2 y + α 3 x 2 + α 4 xy + α 5 y 2
v ( x, y ) =α 6 + α 7 x + α 8 y + α 9 x 2 + α10 xy + α11 y 2 (3.2.19)

So the element strain can be derived from the above displacement field as follows.
u
x   1  2 3 x   4 y
x
v
y    8  10 x  211y (3.2.20)
y
v  u
 xy     2   4 x  2 5 y   7  2 9 x  10 y
x y
The above expression shows that the strain components are linearly varying inside the element.
Therefore, this six node element is called linear strain triangle. The main advantage of this element is
that it can capture the variation of strains and therefore stresses of the element.
 15

3.2.3 Construction of Shape Function by Degrading Technique

Sometimes, the geometry of the structure or its loading and boundary conditions are such that the
stresses developed in few locations are quite high. On the other hand, variations of stresses are less
in some areas and as a result, refinement of finite element mesh is not necessary. It would be
economical in terms of computation if higher order elements are chosen where stress concentration
is high and lower order elements at area away from the critical area. Fig. 3.2.5 shows graphical
representations where various order of triangular elements are used for generating a finite element
mesh.

Fig. 3.2.5 Triangular elements with different number of nodes

Fig. 3.2.5 contains four types of element. Type 1 has only three nodes, type 2 element has five
nodes, type 3 has four nodes and type 4 has six nodes. The shape function for 3-node and 6-node
triangular elements has already been derived. The shape functions of 6-node element can suitably be
degraded to derive shape functions of other two types of triangular elements.

3.2.3.1 Five node triangular element

Let consider a six node triangular element as shown in Fig. 3.2.6(a) whose shape functions and nodal
displacements are (N1, N2, N3, N4, N5, N6) and (u1, u2, u3, u4, u5, u6) respectively. Similarly, for a five
node triangular element as shown in Fig. 3.2.6(b), the shape functions and nodal displacements are
considered as (N’1, N’2, N’3, N’4, N’5) and (u’1, u’2, u’3, u’4, u’5) respectively. Thus, the displacement
at any point in a six node triangular element will become
 16

u  N1u1  N 2 u 2  N 3u 3  N 4 u 4  N 5u 5  N 6 u 6 (3.2.21)
Where, N1, N2, …, N6 are the shape functions and is given in eq.(3.2.18). If there is no node between
2 and 3, the displacement along line 2-3 is considered to vary linearly. Thus the displacement at an
assumed node 5´ may be written as
u 2  u3
u '5 
2 (3.2.22)
Substituting, the value of u’5 for u5 in eq.(3.2.21) the following expression will be obtained.
u 2  u3
u  N1u1  N 2 u 2  N 3u 3  N 4 u 4  N 5  N6u 5 (3.2.23)
2

Fig. 3.2.6 Degrading for five node element

Thus, the displacement function can be expressed by five nodal displacements as:
 N   N 
u  N1u1   N 2  5  u 2   N 3  5  u 3  N 4 u 4  N 6 u 5
 2  2 (3.2.24)

However, the displacement function for the five node triangular element can be expressed as
u  N1u1  N 2 u 2  N 3 u 3  N 4 u 4  N 5 u 5 (3.2.25)
Comparing eq.(3.2.24) and eq.(3.2.25) and observing node 6 of six node triangle corresponds to
node 5 of five node triangle, we can write
N5 N
N '1  N1, N 2  N 2  , N 3  N 3  5 , N 4  N 4 an dN 5  N 5
2 2 (3.2.26)
Hence, the shape function of a five node triangular element will be
 17

N1  L1 2L1 1

4L 2 L3
N 2  L 2 2L 2 1   L 2 1 2L1 
2
N 4L L
N 3  N 3  5  L3 2L3 1  2 3  L3 1 2L1 
2 2 (3.2.27)
N 4  4L1L 2
N 5  4L3L1
Thus, for a five node triangular element, the above shape function can be used for finite element
analysis.
 18

Lecture 3: Rectangular Elements

Rectangular elements are suitable for modelling regular geometries. Sometimes, it is used along with
triangular elements to represent an arbitrary geometry. The simplest element in the rectangular
family is the four node rectangle with sides parallel to x and y axis. Fig. 3.3.1 shows rectangular
elements with varying nodes representing linear, quadratic and cubic variation of function.

Fig. 3.3.1 Rectangular elements

3.3.1 Shape Function for Four Node Element

Shape functions of a rectangular element can be derived using both Cartesian and natural coordinate
systems. A four term polynomial expression for the field variable will be required for a rectangular
element with four nodes having four degrees of freedom. Since there is no complete four term
polynomial in two dimensions, the incomplete, symmetric expression from the Pascal’s triangle may
be chosen to ensure geometric isotropy.

3.3.1.1 Shape function using Cartesian coordinates

For the derivation of interpolation function, the sides of the rectangular element (Fig. 3.3.2) are
assumed to be parallel to the global Cartesian axes. From the Pascal’s triangle, a linear variation may
be assumed to define filed variable to ensure inter-element continuity.
f  x, y   0  1 x  2 y  3 xy (3.3.1)
 19

Fig. 3.3.2 Rectangular element in Cartesian coordinate

Applying nodal conditions, the above expression may be written in matrix form as
f1  1 x1 y1 x1 y1  0 
     
f2  1 x2 y2 x2 y2  1 
      (3.3.2)
f3  1 x3 y3 x3 y3  2 
    
f4  1 x4 y4 x4 y4  3 
The unknown polynomial coefficients may be obtained from the above equation with the use of
nodal field variables.
1
0 
  1 x1 y1 x1 y1  
 f1 

    
1   x2 y2 

f2  
 
   1 x2 y2
 
   (3.3.3)

2  1 x3 y3 x3 y3  
f3 

   1  
 3
   x4 y4 x4 y4  
f4 
 

Thus, the field variable at any point inside the element can be described in terms of nodal values as
1
0  1 x1 y1 x1 y1  f1 
     
 1 x2 y2  f2 
y xy  1   1 x
x2 y2
f  x, y   1 x y xy     
2 
1 x3 y3 x3 y3  f3 
  1  
3   x4 y4 x4 y4  f4  (3.3.4)
f1 
 
f 
  N1 N2 N3 N 4  2 
f3 
 
f4 
From the above expression, the shape function Ni can be derived and will be as follows.
 20

 x − x2   y − y4 
N1 =   
 x1 − x2   y1 − y4 
 x − x1   y − y3 
N2 =   
 x2 − x1   y2 − y3 
(3.3.5)
 x − x4   y − y2 
N3 =   
 x3 − x4   y3 − y2 
 x − x3   y − y1 
N4 =   
 x4 − x3   y4 − y1 
Now, substituting the nodal coordinates in terms of (x1, y1) as (–a, –b) at node 1; (x2, y2) as (a, –b) at
node 2; (x3, y3) as (a, b) at node 3 and (x4, y4) as (–a, b) at node 4 the above expression can be re-
written as:
1
N1 = ( x − a )( y − b )
4ab
1
N2 = ( x + a )( y − b )
4ab
(3.3.6)
1
N3 = ( x + a )( y + b )
4ab
1
N4 = ( x − a )( y + b )
4ab
Thus, the shape function N can be found from the above expression in Cartesian coordinate system.

3.3.1.2 Shape function using natural coordinates

The derivation of interpolation function in terms of Cartesian coordinate system is algebraically
complex as seen from earlier section. However, the complexity can be reduced by the use of natural
coordinate system, where the natural coordinates will vary from -1 to +1 in place of –a to +a or –b to
+b. The transformation of Cartesian coordinates to Natural coordinates are shown in Fig. 3.3.3.
 21

Fig. 3.3.3 Four node rectangular element

From the figure, the relation between two coordinate systems can be expressed as
x−x y− y
=ξ = and η (3.3.7)
a b
Here, 2a and 2b are the width and height of the rectangle. The coordinate of the center of the
rectangle can be written as follows:
x1 + x2 y1 + y4
=x = and y (3.3.8)
2 2
Thus, from eq. (3.3.7) and eq.(3.3.8), the nodal values in natural coordinate systems can be derived
which is shown in Fig. 3.3.4(b). With the above relations variations of x & h will be from -1 to +1.
Now the interpolation function can be derived in a similar fashion as done in section 3.3.1.1. The
filed variable can be written in natural coordinate system ensuring inter-element continuity as:
f x, h  0  1x  2 h  3xh (3.3.9)
The coordinates of four nodes of the element in two different systems are shown in Table 3.3.1 for
ready reference for the derivation purpose. Applying the nodal values in the above expression one
can get
f1  1 1 1 1  0 
   
  
f2   1 1 1 1 1  (3.3.10)
f3  1 1 1 1  2 
     
f4  1 1 1 1 3 
 22

Table 3.3.1 Cartesian and natural coordinates for four node element

Node Cartesian Coordinate Natural Coordinate

x y ξ η
1 x1 y1 -1 -1
2 x2 y2 1 -1
3 x3 y3 1 1
4 x4 y4 -1 1

Thus, the unknown polynomial coefficients can be found as

  1 1 1 1  1 
0  f1 
 1 1 1 1  f1 

 
     
    

  1 1 1 1 
1    1 1 1
f2  1 1 

f2 
        (3.3.11)

2  1 1 1 1   f3  4 1 1 1 1  
f3 
            

3 
  1 1 1 1  f4 
  1 1 1 1 
f4 
 
The field variable can be written as follows using eq.(3.3.9) and eq.(3.3.11).
0  1 1 1 1  f1 
    
 1 1 1 1 1 f2 
f x, h  1 x h xh 1   1 x h xh   
2  4 1 1 1 1  f3 
   1 1  
3   1 1 f4 
(3.3.12)
f1 
 
f 
  N1 N 2 N 3 N 4  2 
f3 
 
f4 
Where, Ni are the interpolation function of the element in natural coordinate system and can be
found as:


 1 x1 h 

 

 4 

  
 



N 
  1  x 1  h  
 
1
 
N2   
   4 
Ni       (3.3.13)

 N3 
 
 1  x1  h

  
  


N4  
 4 



 1 x1  h

 


 4 

 23

3.3.2 Shape Function for Eight Node Element

The shape function of eight node rectangular element can be derived in similar fashion as done in
case of four node element. The only difference will be on choosing of polynomial as this element is
of quadratic in nature. The derivation will be algebraically complex in case of using Cartesian
coordinate system. However, use of the natural coordinate system will make the process simpler as
the natural coordinates vary from -1 to +1 in the element. The variation of filed variable ϕ can be
expressed in natural coordinate system by the following polynomial.
f x, h  0  1x  2 h  3x 2  4 xh  5h2  6 x 2 h  7 xh2 (3.3.14)
It may be noted that the cubic terms ξ3 and η3 are omitted and geometric invariance is ensured by
choosing the above expression. Fig. 3.3.4 shows the natural nodal coordinates of the eight node
rectangle element in natural coordinate system.

The nodal field variables can be obtained from the above expression after putting the coordinates at
nodes.
f1  1 1 1 1 1 1 1 1 0 
    
f2  1 1 1 1 1 1 1 1  1 
 
f  1 1
 3   1 1 1 1 1 1  2 
f  1 1 1  
1 1 1 1 1 3 
fi    4        Ai  (3.3.15)
f5  1 0 1 0 0 1 0 0  4 
    
f6  1 1 0 1 0 0 0 0  5 
     
f7  1 0 1 0 0 1 0 0  6 
  
f  1 1 0
 8   1 0 0 0 0  7 

Fig. 3.3.4 Natural coordinates of eight node rectangular element

Replacing the unknown coefficient αi in eq.(3.3.14) from eq.(3.3.15), the following relations will be
obtained.
 24

f x, h  1 x h x 2 xh h2 x 2 h xh2   A fi 

1

1 1 1 1 2 2 2 2  f1 
   

0 0 2   f2 

0 0 0 0 2
  
0
 0 0 0 2 0 2 0  

f3



1    f 
 
 1 x h x xh h x h xh  
2 2 2 2 1 1 1 1 2 0 2 0  
 4

4  1 1 1 1 0 0 0 0  
f5 
  
1 1 1 1 
0 2 0 2 f6  
  

1 1 1 1 2 0 2 0   f7

  
 
1 1 1 1 0 2 0 f8 
2  
 

f1 
 
f2 
 
f3 
 
f
  N1 N2 N3 N4 N5 N6 N7 N8  4 
f5 
 
f6 
 
f7 
f 
 8  (3.3.16)
Thus, the interpolation function will become

1 x1 hx  h 1 1  x1 hx  h 1

N1  ; N2 ;
4 4
1  x1  hx  h 1 1 x1  hx  h 1
N3  ; N4  ;
4 4
1  x1 x1 h 1  x1  h1 h
N5  ; N6  ;
2 2
1  x1 x1  h 1 x1  h1 h
N7  ; N8 
2 2 (3.3.17)

The shape functions of rectangular elements with higher nodes can be derived in similar manner
using appropriate polynomial satisfying all necessary criteria. However, difficulty arises due to the
inversion of large size of the matrix because of higher degree of polynomial chosen. In next lecture,
the shape functions of rectangular element with higher nodes will be derived in a much simpler way.
 25

Lecture 4: Lagrange and Serendipity Elements

In last lecture note, the interpolation functions are derived on the basis of assumed polynomial from
Pascal’s triangle for the filed variable. As seen, the inverse of the large matrix is quite cumbersome
if the element is of higher order.

3.4.1 Lagrange Interpolation Function

An alternate and simpler way to derive shape functions is to use Lagrange interpolation polynomials.
This method is suitable to derive shape function for elements having higher order of nodes. The
Lagrange interpolation function at node i is defined by
n
x  x j  x  x1 x  x2 ....x  xi1 x  xi1 ....x  xn 
f i (x )   
j1 x i  x j  xi  x1 xi  x 2 ....xi  xi1 xi  xi1 ....xi  x n 
ji
(3.4.1)
The function fi (ξ) produces the Lagrange interpolation function for i node, and ξj denotes ξ th

coordinate of jth node in the element. In the above equation if we put ξ = ξj, and j ≠ i, the value of the
function fi(ξ) will be equal to zero. Similarly, putting ξ = ξi, the numerator will be equal to
denominator and hence fi(ξ) will have a value of unity. Since, Lagrange interpolation function for ith
node includes product of all terms except jth term; for an element with n nodes, fi(ξ) will have n-1
degrees of freedom. Thus, for one-dimensional elements with n-nodes we can define shape function
as N i (x)  f i (x) .

3.4.1.1 Shape function for two node bar element

Consider the two node bar element discussed as in section 3.1.1. Let us consider the natural
coordinate of the center of the element as 0, and the natural coordinate of the nodes 1 and 2 are -1
and +1 respectively. Therefore, the natural coordinate ξ at any point x can be represented by,

2  x  x1 
x 1
l (3.4.2)

Fig. 3.4.1 Natural coordinates of bar element

 26

The shape function for two node bar element as shown in Fig. 3.4.1 can be derived from eq.(3.4.1) as
follows:

x  x2  x 1 1
N1  f1 x    1  x
x1  x 2  1  (1) 2
x  x1  x  1 1 (3.4.3)
N 2  f1 x    1  x
x 2  x1  1 (1) 2

Graphically, these shape functions are represented in Fig.3.4.2.

Fig. 3.4.2 Shape functions for two node bar element

3.4.1.2 Shape function for three node bar element

For a three node bar element as shown in Fig. 3.4.3, the shape function will be quadratic in nature.
These can be derived in the similar fashion using eq.(3.4.1) which will be as follows:

x  x2 x  x3  (x)x 1 1

N1 x  f1 x    x x 1
x1  x 2 x1  x3  1(2) 2

x  x1 x  x3  x  1x 1

N 2 x   f 2 x     1  x 2  (3.4.4)
x 2  x1 x 2  x3  11

x  x1 x  x 2  x  1(x) 1
N 3 x  f3 x    x x  1
x3  x1 x3  x 2  2(1) 2
 27

Fig. 3.4.3 Quadratic shape functions for three node bar element

3.4.1.3 Shape function for two dimensional elements

We can derive the Lagrange interpolation function for two or three dimensional elements from one
dimensional element as discussed above. Those elements whose shape functions are derived from
the products of one dimensional Lagrange interpolation functions are called Lagrange elements. The
Lagrange interpolation function for a rectangular element can be obtained from the product of
appropriate interpolation functions in the ξ direction [fi(ξ )] and η direction [fi(η)]. Thus,
N i x, h  f i xf i h Where, i = 1,2,3, …., n-node (3.4.5)
The procedure is described in details in following examples.

Four node rectangular element

The shape functions for the four node rectangular element as shown in the Fig. 3.4.4 can be derived
by applying eq.(3.4.3) eq.(3.4.5) which will be as follows.
x  x 2  h  h 2 
N1 x, h  f1 xf1 h 
`
x1  x 2  h1  h2  (3.4.6)
x 1 h 1
1
   1  x1  h
1  (1) 1  (1) 4

Similarly, other interpolation functions can be derived which are given below.
1
N 2 x, h  f 2 xf1 h  1  x1  h
4
1
` N 3 x, h  f 2 xf 2 h  1  x1  h (3.4.7)
4
1
N 4 x, h  f1 xf 2 h  1  x1  h
4
These shape functions are exactly same as eq.(3.3.13) which was derived earlier by choosing
polynomials.
 28

Fig. 3.4.4 Four node rectangular element

Nine node rectangular element

In a similar way, to the derivation of four node rectangular element, we can derive the shape
functions for a nine node rectangular element. In this case, the shape functions can be derived using
eq.(3.4.4) and eq.(3.4.5).
1 1 1
N1 x, h  f1 xf1 h  x x 1 hh 1  xhx 1h 1
2 2 4 (3.4.8)
In a similar way, all the other shape functions of the element can be derived. The shape functions of
nine node rectangular element will be:
1 1
N1  xhx 1h 1, N 2  xhx  1h 1
4 4
1 1
N 3  xhx  1h  1, N 4  xhx 1h  1
4 4
N 5  h 1  x 2 h 1, N 6  x x  11  h 2 
1 1
(3.4.9)
2 2
N 7  h 1  x 2 h  1, N8  x x 11  h 2 
1 1
2 2
N 9  1  x 2 1  h 2 
 29

Fig. 3.4.5 Nine node rectangular element

Thus, it is observed that the two dimensional Lagrange element contains internal nodes (Fig. 3.4.6)
which are not connected to other nodes.

Fig. 3.4.6 Two dimensional Lagrange elements and Pascal triangle

3.4.2 Serendipity Elements

Higher order Lagrange elements contains internal nodes, which do not contribute to the inter-
element connectivity. However, these can be eliminated by condensation procedure which needs
extra computation. The elimination of these internal nodes results in reduction in size of the element
matrices. Alternatively, one can develop shape functions of two dimensional elements which contain
nodes only on the boundaries. These elements are called serendipity elements (Fig. 3.4.7) and their
interpolation functions can be derived by inspection or the procedure described in previous lecture
 30

(Module 3, lecture 3). The interpolation function can be derived by inspection in terms of natural
coordinate system as follows:
(a) Linear element
1
N i x, h  1  xxi 1  hhi  (3.4.10)
4
(b) Quadratic element
(i) For nodes at x  1, h  1
1
N i x, h  1  xxi 1  hhi xxi  hhi 1 (3.4.11a)
4
(ii) For nodes at x  1, h  0

N i x, h  1  xxi 1  h2 

1
2 (3.4.11b)
(iii) For nodes at x  0, h  1

1  x 2 1  hhi 
1
N i x, h  (3.4.11c)
2
(c) Cubic element
(i) For nodes at x  1, h  1

1  xxi 1  hhi  9x2  h2  10

1
N i x, h  (3.4.12a)
32
1
(ii) For nodes at x  1, h  
3

1  xxi 1  h2 1  9hhi 

9
N i x, h  (3.4.12b)
32
And so on for other nodes at the boundaries.
 31

Fig. 3.4.7 Two dimensional serendipity elements and Pascal triangle

Thus, the nodal conditions must be satisfied by each interpolation function to obtain the functions
serendipitously. For example, let us consider an eight node element as shown in Fig. 3.4.8 to derive
its shape function. The interpolation function N1 must become zero at all nodes except node 1, where
its value must be unity. Similarly, at nodes 2, 3, and 6, ξ = 1, so including the term ξ − 1 satisfies the
zero condition at those nodes. Similarly, at nodes 3, 4 and 7, η = 1 so the term η − 1 ensures the zero
condition at these nodes.

Fig. 3.4.8 Two dimensional eight node rectangular element

Again, at node 5, (ξ, η) = (0, −1), and at node 8, (ξ, η) = (−1, 0). Hence, at nodes 5 and 8, the term (ξ
+ η + 1) is zero. Using this reasoning, the equation of lines are expressed in Fig. 3.4.9. Thus, the
 32

interpolation function associated with node 1 is to be of the form N1  y1 h 1x 1x  h  1

where, ψ1 is unknown constant. As the value of N1 is 1 at node 1, the magnitude unknown constant
ψ1 will become -1/4. Therefore, the shape function for node 1 will become
1
N1   1 h1 x x  h  1 .
4
Similarly, ψ2 will become -1/4 considering the value of N2 at node 2 as unity and the shape function
1
for node 2 will be N 2  y2 h 1x  1x  h 1   1  x1 hx  h 1 . In a similar
4
fashion one can find out other interpolation functions from Fig. 3.4.9 by putting the respective values
at various nodes. Thus, the shape function for 8-node rectangular element is given below.

Fig. 3.4.9 Equations of lines for two dimensional eight node element

N 5  1  x 2 1  h,
1 1
N1   1  x1  h1  x  h,
4 2
N 6  1  x1  h2 ,
1 1
N 2   1  x1  h1  x  h,
4 2
(3.4.13)
N 7  1  x 2 1  h,
1 1
N 3   1  x1  h1  x  h,
4 2
N 4   1  x1  h1  x  h and N8  1  x1  h2 
1 1
4 2
 33

It may be observed that the Lagrange elements have a better degree of completeness in polynomial
function compare to serendipity elements. Therefore, Lagrange elements produce comparatively
faster and better accuracy.
 34

Lecture 5: Solid Elements

There are two basic families of three-dimensional elements similar to two-dimensional case.
Extension of triangular elements will produce tetrahedrons in three dimensions. Similarly,
rectangular parallelepipeds are generated on the extension of rectangular elements. Fig. 3.5.1 shows
few commonly used solid elements for finite element analysis.

Fig. 3.5.1 Three-dimensional solid elements

Derivation of shape functions for such three dimensional elements in Cartesian coordinates are
algebraically quite cumbersome. This is observed while developing shape functions in two
dimensions. Therefore, the shape functions for the two basic elements of the tetrahedral and
parallelepipeds families will be derived using natural coordinates.

The polynomial expression of the field variable in three dimensions must be complete or incomplete
but symmetric to satisfy the geometric isotropy requirements. Completeness and symmetry can be
 35

ensured using the Pascal pyramid which is shown in Fig. 3.5.2. It is important to note that each
independent variable must be of equal strength in the polynomial.

Fig. 3.5.2 Pascal pyramid in three dimensions

The following 3-D quadratic polynomial with complete terms can be applied to an element having
10 nodes.
f x, h, z   0  1x  2 h  3z  4 x 2  5h2  6z 2  7 xh  8hz  9zx
(3.5.1)
However, the geometric isotropy is not an absolute requirement for field variable representation to
derive the shape functions.

3.5.1 Tetrahedral Elements

The simplest element of the tetrahedral family is a four node tetrahedron as shown in Fig. 3.5.3. The
node numbering has been followed in sequential manner, i.e, in this case anti-clockwise direction.
Similar to the area coordinates, the concept of volume coordinates has been introduced here. The
coordinates of the nodes are defined both in Cartesian and volume coordinates. Point P(x, y, and z)
as shown in Fig. 3.5.2 is an arbitrary point in the tetrahedron.
 36

Fig. 3.5.3 Four node tetrahedron element

The linear shape function for this element can be expressed as,
{N} = [ L1 L4 ]
T
L2 L3
(3.5.2)
Here, L1 , L2 , L3 , L4 are the set of natural coordinates inside the tetrahedron and are defined as follows
Vi
Li =
V (3.5.3)
Where Vi is the volume of the sub element which is bound by point P and face i and V is the total
volume of the element. For example L1 may be interpreted as the ratio of the volume of the sub
element P234 to the total volume of the element 1234. The volume of the element V is given by the
determinant of the nodal coordinates as follows:
1 1 1 1
1 x1 x2 x3 x4
V=
6 y1 y2 y3 y4 (3.5.4)
z1 z2 z3 z4
The relationship between the Cartesian and natural coordinates of point P may be expressed as

1   1 1 1 1   L1 
x  x x4   L2 
   1 x2 x3
 =  
 y   y1 y 4   L3 
y2 y3 (3.5.5)
 z   z1 z2 z3

z 4   L4 
It may be noted that the identity included in the first row ensure the matrix invertible.
 37

L1 + L2 + L3 + L4 = 1
(3.5.6)
The inverse relation is given by
 L1  V1 a1 b1 c1  1 
L  
 2  1 V2 a2 b2 c2   x 
 =  
 L3  6V V3 a3 b3 c3   y  (3.5.7)
 
 L4  V4 a4 b4 c4   z 

Here, Vi is the volume subtended from face i and terms ai , bi , and ci represent the projected area of
face i on the x, y, z coordinate planes respectively and are given as follows:
ai = ( z j y k − z k y j ) + ( z k y l − z l y k ) + ( z l y j − z j y l )
bi = ( z j xk − z k x j ) + ( z k xl − z l xk ) + ( z l x j − z j xl )
(3.5.8)
ci = ( y j x k − y k x j ) + ( y k xl − y l x k ) + ( y l x j − y j xl )
i, j , k , l will be in cyclic order (i.e., 1 2  3  4  1). The volume coordinates fulfil all nodal
conditions for interpolation functions. Therefore, the field variable can be expressed in terms of
nodal values as
f  x, y, z   L1f1  L2f2  L3f3  L4f4
(3.5.9)
Though the shape functions (i.e., the volume coordinates) in terms of global coordinates is
algebraically complex but they are straightforward. The partial derivatives of the natural coordinates
with respect to the Cartesian coordinates are given by
∂Li ai ∂Li b ∂Li c
= , = i , = i
∂x 6V ∂y 6V ∂z 6V (3.5.10)
Similar to area integral, the general integral taken over the volume of the element is given by,
p!q!r!s!
∫ L L L L dV = ( p + q + r + s + 3)!.6V
p q r s
1 2 3 4
(3.5.11)
v

The four node tetrahedral element is a linear function of the Cartesian coordinates. Hence, all the
first partial derivatives of the field variable will be constant. The tetrahedral element is a constant
strain element as the element exhibits constant gradients of the field variable in the coordinate
directions.
Higher order elements of the tetrahedral family are shown in Fig. 3.5.1. The shape functions for such
higher order three dimensional elements can readily be derived in volume coordinates, as for higher-
order two-dimensional triangular elements. The second element of this family has 10 nodes and a
cubic form for the field variable and interpolation functions.
+++++++++++
3.5.2 Brick Elements
 38

Various orders of elements of the parallelepiped family are shown in Fig. 3.5.1. Fig. 3.5.4 shows the
eight-node brick element with reference to a global Cartesian coordinate system and then with
reference to natural coordinate system. The natural coordinates for the brick element can be relate
Cartesian coordinate system by
x−x y− y z−z
=ξ = , η = and ζ (3.5.12)
a b c
Here, 2a, 2b and 2c are the length, height and width of the element. The coordinate of the center of
the brick element can be written as follows:
x1 + x2 y1 + y4 z1 + z5
=x = , y = and z (3.5.13)
2 2 2
Thus, from eq.(3.5.12) and eq.(3.5.13), the nodal values in natural coordinate systems can be derived
which is shown in Fig. 3.5.4(b). With the above relations variations of x, h & z will be from -1 to +1.
Now the interpolation function can be derived in several procedures as done in case of two
dimensional rectangular elements. For example, the interpolation function can be derived by
inspection in terms of natural coordinate system as follows:
1
N i x, h, z  1  xxi 1  hhi 1  zzi  (3.5.14)
8

Fig. 3.5.4 Eight node brick element

By using field variable the following terms of the polynomial may be used for deriving the shape
function for eight-node brick element.
f x, h, z   0  1x  2 h  3z  4 xh  5hz  6zx  7zhx
(3.5.14)
The above equation is incomplete but symmetric. However, such representations are quite often used
and solution convergence is achieved in the finite element analysis. Again, the shape functions for
 39

three dimensional 8-node or 27-node brick elements can be derived using Lagrange interpolation
function. For this we need to introduce interpolation function in the ζ-direction. Thus, for example,
the Lagrange interpolation function for a three dimensional 8 node brick element can be obtained
from the product of appropriate interpolation functions in the ξ, η and ζ directions. Therefore, the
shape function will become
N i x, h, z  f i xf i hf i z Where, i = 1,2,3, …., n-node (3.5.15)
Thus using the Lagrange interpolation function the shape function at node 1 can be expressed as
x  x 2  h  h 2  z  z 2 
N1 x, h, z  f1 xf1 hf1 z 
x1  x 2  h1  h2  z1  z 2  (3.5.16)
x 1 h 1 z 1 1
    1  x1  h1  z
1  (1) 1  (1) 1  (1) 4
Using any of the above concepts, the interpolation function for 8-node brick element can be found as
follows:

1 1
N1  1  x1  h1  z, N1  1  x1  h1  z,
4 4
1 1
N 3  1  x1  h1  z, N 4  1  x1  h1  z,
4 4 (3.5.17)
1 1
N 5  1  x1  h1  z, N 6  1  x1  h1  z,
4 4
1 1
N 7  1  x1  h1  z, N8  1  x1  h1  z
4 4

The shape functions of rectangular parallelepiped elements with higher nodes can be derived in
similar manner satisfying all necessary criteria.
 40

Lecture 6: Isoparametric Formulation

3.6.1 Necessity of Isoparametric Formulation

The two or three dimensional elements discussed till now are of regular geometry (e.g. triangular
and rectangular element) having straight edge. Hence, for the analysis of any irregular geometry, it is
difficult to use such elements directly. For example, the continuum having curve boundary as shown
in the Fig. 3.6.1(a) has been discretized into a mesh of finite elements in three ways as shown.

(a) The Continuum to be discritized (b) Discritization using Triangular Elements (c)
Discritization using rectangular elements (d) Discritization using a combination of
rectangular and quadrilateral elements
 41

Fig 3.6.1 Discretization of a continuum using different elements

Figure 3.6.1(b) presents a possible mesh using triangular elements. Though, triangular elements can
suitable approximate the circular boundary of the continuum, but the elements close to the center
becomes slender and hence affect the accuracy of finite element solutions. One possible solution to
the problem is to reduce the height of each row of elements as we approach to the center. But,
unnecessary refining of the continuum generates relatively large number of elements and thus
increases computation time. Alternatively, when meshing is done using rectangular elements as
shown in Fig 3.6.1(c), the area of continuum excluded from the finite element model is significantly
adequate to provide incorrect results. In order to improve the accuracy of the result one can generate
mesh using very small elements. But, this will significantly increase the computation time. Another
possible way is to use a combination of both rectangular and triangular elements as discussed in
section 3.2. But such types of combination may not provide the best solution in terms of accuracy,
since different order polynomials are used to represent the field variables for different types of
elements. Also the triangular elements may be slender and thus can affect the accuracy. In Fig.
3.6.1(d), the same continuum is discritized with rectangular elements near center and with four-node
quadrilateral elements near boundary. This four-node quadrilateral element can be derived from
rectangular elements using the concept of mapping. Using the concept of mapping regular triangular,
rectangular or solid elements in natural coordinate system (known as parent element) can be
transformed into global Cartesian coordinate system having arbitrary shapes (with curved edge or
surfaces). Fig. 3.6.2 shows the parent elements in natural coordinate system and the mapped
elements in global Cartesian system.
42
 43

(a) Natural Coordinate System (b) Global Coordinate System

Fig. 3.6.2 Mapping of isoparametric elements in global coordinate system

3.6.2 Coordinate Transformation

The geometry of an element may be expressed in terms of the interpolation functions as follows.
n
x= N1 x1 + N 2 x2 + ... + N n x=
n ∑N x
i =1
i i

n
(3.6.1)
y= N1 y1 + N 2 y2 + ... + N n y=
n ∑N y
i =1
i i

n
z= N1 z1 + N 2 z2 + ... + N n z=
n ∑N z
i =1
i i

Where,
n=No.of Nodes
N i =Interpolation Functions
x i ,yi ,zi =Coordinates of Nodal Points of the Element
One can also express the field variable variation in the element as
 44

n
φ (ξ ,η , ζ ) = ∑ N i (ξ ,η , ζ ) φi (3.6.2)
i =1

As the same shape functions are used for both the field variable and description of element
geometry, the method is known as isoparametric mapping. The element defined by such a method is
known as an isoparametric element. This method can be used to transform the natural coordinates of
a point to the Cartesian coordinate system and vice versa.

Example 3.6.1
Determine the Cartesian coordinate of the point P (ξ= 0.8, η= 0.9) as shown in Fig. 3.6.3.

Fig. 3.6.3 Transformation of Coordinates

Solution:
As described above, the relation between two coordinate systems can be represented through their
interpolation functions. Therefore, the values of the interpolation function at point P will be
(1 − ξ )(1 − n) (1 − 0.8)(1 − 0.9)
=N1 = = 0.005
4 4
(1 + ξ )(1 − n) (1 + 0.8)(1 − 0.9)
=N2 = = 0.045
4 4
(1 + ξ )(1 + n) (1 + 0.8)(1 + 0.9)
=N3 = = 0.855
4 4
(1 − ξ )(1 + n) (1 − 0.8)(1 + 0.9)
=N4 = = 0.095
4 4

Thus the coordinate of point P in Cartesian coordinate system can be calculated as

 45

4
x = ∑ N i xi = 0.005 ×1 + 0.045 × 3 + 0.855 × 3.5 + 0.095 ×1.5 = 3.275
i =1
4
y = ∑ N i yi = 0.005 ×1 + 0.045 ×1.5 + 0.855 × 4.0 + 0.095 × 2.5 = 3.73
i =1

Thus the coordinate of point P (ξ= 0.8, η= 0.9) in Cartesian coordinate system will be 3.275, 3.73.

Solid isoparametric elements can easily be formulated by the extension of the procedure followed for
2-D elements. Regardless of the number of nodes or possible curvature of edges, the solid element is
just like a plane element which is mapped into the space of natural co-ordinates, i.e,
ξ = ±1,η = ±1, ζ = ±1 .

3.6.3 Concept of Jacobian Matrix

A variety of derivatives of the interpolation functions with respect to the global coordinates are
necessary to formulate the element stiffness matrices. As the both element geometry and variation of
the shape functions are represented in terms of the natural coordinates of the parent element, some
additional mathematical obstacle arises. For example, in case of evaluation of the strain vector, the
operator matrix is with respect to x and y, but the interpolation function is with x and h . Therefore,
the operator matrix is to be transformed for taking derivative with x and h . The relationship between
two coordinate systems may be computed by using the chain rule of partial differentiation as
  x   y   x  y
  and   (3.6.3)
x  x x  y x h  x h  y  h
The above equations can be expressed in matrix form as well.

   x y    
 

 
     
 x   x x   
 x   
 x 
 
        J 
   (3.6.4)

    x y     
       
 

   h h  
 y  
 y 
 h 
    
  

 x y 
  n
 x
The matrix [J] is denoted as Jacobian matrix which is: 
x 
 . As we know, x = ∑ N i xi
 x y  i =1
 
 h h 
n
∂ ∑ N i xi
∂x n
∂N i
where, n is the number of nodes in an element. Hence, J=
11 =
∂ξ
i =1

∂ξ
= ∑ ∂ξ
i =1
xi

Similarly one can calculate the other terms J12, J21 and J22 of the Jacobian matrix. Hence,
 46

 n N i n
N i 
 x  x yi 
 i1 x i 
 J    n i 1
 (3.6.5)
  N i xi
n
N i 
 i1 h 
i 1 h
yi 
 

From eq. (3.6.4), one can write


  
 
  
 


 
x  1  x 

   J      (3.6.6)

  
 
  

   
 y 
 
 
 
 x 


 J11* J12* 
Considering  * are the elements of inverted [J] matrix, we may arise into the following
J * 
J 22
 21 
relations.
  
 J11*   J12* 
x x h
(3.6.7)
  
 J 21
*
  J 22
*

y x h

Similarly, for three dimensional case, the following relation exists between the derivative operators
in the global and the natural coordinate system.

 ∂   ∂x ∂y ∂z   ∂  ∂
 ∂ξ   ∂ξ ∂ξ ∂ξ     ∂x 
     ∂x   
 ∂   ∂x ∂y ∂z   ∂  ∂
=  =   [J ]  (3.6.8)
 ∂η   ∂η ∂η ∂η   ∂y   ∂y 
 ∂   ∂x ∂y ∂z   ∂  ∂
      
 ∂ζ   ∂ζ ∂ζ ∂ζ   ∂z   ∂z 

Where,
 47

 ∂x ∂y ∂z 
 ∂ξ ∂ξ ∂ξ 

 ∂x ∂y ∂z 
[J ] =  (3.6.9)
 ∂η ∂η ∂η 
 ∂x ∂y ∂z 
 ∂ζ ∂ζ ∂ζ 

[J] is known as the Jacobian Matrix for three dimensional case. Putting eq. (3.6.1) in eq. (3.6.9) and
after simplifying one can get
 ∂N i ∂N i ∂N i 
 ∂ξ xi ∂ξ
yi zi
∂ξ 

n
 ∂N ∂N i ∂N i 
[ J ] = ∑  i xi (3.6.10)
∂η 
yi zi
i =1
 ∂η ∂η
 ∂N i ∂N i ∂N i 
 ∂ζ xi ∂ζ
yi
∂ζ 
zi


From eq. (3.6.8), one can find the following expression.

∂  ∂ 
 ∂x   
   ∂ξ 
∂ −1  ∂ 
  = [J ]   (3.6.11)
 ∂y   ∂η 
∂  ∂ 
   
 ∂z   ∂ζ 

 J11* J12* J13* 

 * * 
Considering [ J ] =  J 21
−1 *
J 22 J 23  we can arrived at the following relations.
 J 31
* *
J 32 * 
J 33 

   
 J11*   J12*   J13* 
x x h z
   
 J 21
*
  J 22
*
  J 23
*
 (3.6.12)
y x h z
   
 J 31
*
  J 32
*
  J 33
*

z x h z
 48

Lecture 7: Stiffness Matrix of Isoparametric Elements

3.7.1 Evaluation of Stiffness Matrix of 2-D Isoparametric Elements

For two dimensional plane stress/strain formulation, the strain vector can be represented as

 u  
 u u 


 
 
 J11*   J12*  


  
 
  x  h 


 x  
  x
 
 

       

     y    

v v v

*
J 21   J 22 *
  (3.7.1)

 
 
 
y   x h 

 
 xy  
     

  v u 
 
  v  v  u  u 
    J *
  J *
  J *
  J *
 

 
 x y 

  x h x h 
11 12 21 22
  
 


The above expression can be rewritten in matrix form


 u 

 


  x 


 

 J11* J12*   u 
 0 0    
*   h 
    0   (3.7.2)
*
0 J 21 J 22  
 
J*
 21
*
J 22 J11* J12   v 
* 

  x 

 


 v 

 
 h
  

n
For an n node element the displacement u can be represented as, u   N i ui and similarly for v &
i 1

w. Thus,

 u 
   N1 N n 

 
   0  0 
 u1 

 
 

x   x
 x 

 

 u 
   N1 N n   
    0  0   


  h 
  h h  u  

      n (3.7.3)
 v  N1 N n  
 v1 

   0  0    

 x  
 x x   

 
 
 

  
N n  



v 

  0  0
N1
  n
 v 

 h 

  
 h h 

As a result, eq. (3.7.2) can be written using eq. (3.7.3) which will be as follows.
 49

 N N n  u1 
 1  0  0   
 x x  : 
   
 J11  N N n  : 
0   1
(3.7.4)
0   
* *
 J12 0  0 
 u 
   0 0 J*21 J*22   h h
  n 
 J* *  N1 N n  v1 
J*22  0  
*
 21 J11 J12  0 
  x x  : 
  
 N1 N n  : 
 0    
h  v n 
0
 h
 
Or,
   Bd (3.7.5)

Where {d} is the nodal displacement vector and [B] is known as strain displacement relationship
matrix and can be obtained as
 N1 N n 
  0 
0 
 x x 
 
 J11
* *   N N n 

J12 0 0  1 0  0  (3.7.6)
*  h h 
 B   0 0 J 21 J 22   
*

 J* J* J* J*   0 N1 N n 
 21 22 11 12    0  
  x x 
 
 N1 N n 
 0  0  
 h h 
It is necessary to transform integrals from Cartesian to the natural coordinates as well for calculation
of the elemental stiffness matrix in isoparametric formulation. The differential area relationship can
be established from advanced calculus and the elemental area in Cartesian coordinate can be
represented in terms of area in natural coordinates as:
dA  dx dy  J dx dh (3.7.7)

Here J is the determinant of the Jacobian matrix. The stiffness matrix for a two dimensional
element may be expressed as

 k     B  D Bd  t   B  D  Bdxdy (3.7.8)

T T

 A

Here, [B] is the strain-displacement relationship matrix and t is the thickness of the element. The
above expression in Cartesian coordinate system can be changed to the natural coordinate system as
follows to obtain the elemental stiffness matrix
 50

1 1

k   t   B DB J dxdh

T
(3.7.9)
1 1

Though the isoparametric formulation is mathematically straightforward, the algebraic difficulty is

significant.

Example 3.7.1:
Calculate the Jacobian matrix and the strain displacement matrix for four node two dimensional
quadrilateral elements corresponding to the gauss point (0.57735, 0.57735) as shown in Fig. 3.6.4.

Fig. 3.7.1 Two dimensional quadrilateral element

Solution:
The Jacobian matrix for a four node element is given by,
 n N i n 
N i
 x  x yi 
 i1 x i 
 J    n i 1

  N i xi
n
N i 
 i1 h 
i 1 h
yi 
 

For the four node element one can find the following relations.
1 x1 h N1 1  h N1 1 x
N1  ,  , 
4 x 4 h 4
1  x1 h N 2 1  h N 2 1 x
N2  ,  , 
4 x 4 h 4
1  x1  h N 3 1  h N 3 1  x
N3  ,  , 
4 x 4 h 4
 51

1 x1  h N 4 1  h N 4 1  x
N4  ,  , 
4 x 4 h 4
Now, for a four node quadrilateral element, the Jacobian matrix will become
 N1 N 2 y1 
N 3 N 4   x 1
  
 x x y2 x x   x 2
 J     
N N 2 y3 
N 3
 N 4   x 3
 1 
 h h y 4 
h h   x 4
 1 h 1 h 1  h 1  h   x1 y1 

  
 4 4 4 4   x 2 y2 

 
 1 x 1  x 1  x 1  x   x 3 y3 

  
 4 4 4 4   x 4 y 4 

Putting the values of ξ & η as 0.57735 and 0.57735 respectively, one will obtain the following.
N1 N1
 0.10566  0.10566
x h
N 2 N 2
 0.10566  0.39434
x h
N 3 N 3
 0.39434  0.39434
x h
N 4 N 4
 0.39434  0.10566
x h
4
N i
Hence, J11   xi  0.105661  0.105663  0.394343.5  0.394341.5  1.0
i 1 x
Similarly, J12 =0.64632, J21 =0.25462 and J22 =1.14962.
Hence,
1.00000 0.64632
J  
 0.25462 1.14962 
Thus, the inverse of the Jacobian matrix will become:
 * J12*   1.1671 0.6561
 J *    J11  
   J * *   
 21 J 22  0.2585 1.0152 
Hence strain displacement matrix is given by,
 52

 N1 N n 
  0  0 
 x x 
 
 J11
* *   N N n 

J12 0 0  1  0  0 
 
 B   0 0 J*21 J*22   h h

 J* *  N1 N n 
J*22   0 
*
 21 J11 J12  0 
  x x 
 
 N1 N n 
 0  0  
 h h 
 1.1671 0.6561 0 0 

  0 0 0.2585 1.0152  
0.2585 1.0152 1.1671 0.6561
 
0.10566 0.10566 0.39434 0.39434 0 0 0 0 
 
0.10566 0.39434 0.39434 0.10566 0 0 0 0 
 
   
 0 0 0 0 0.10566 0.10566 0.39434 0.39434 
 0.10566 0.39434 0.39434 0.10566 
 0 0 0 0

0.0540 0.3820 0.2015 0.5294 0 0 0 0 

 
 0 0 0 0 0.0800 0.4276 0.2984 0.2092 
 
0.0800 0.4276 0.2984 0.2092 0.0540 0.3820 0.2015 0.5294
 

3.7.2 Evaluation of Stiffness Matrix of 3-D Isoparametric Elements

Stiffness matrix of 3-D solid isoparametric elements can easily be formulated by the extension of the
procedure followed for plane elements. For example, the eight node solid element is analogous to the
four node plane element. The strain vector for solid element can be written in the following form.
 53

 ∂u 
 ∂x 
 
 ∂u 
 ∂y 
 
 ∂u 
 ∂z 
 ε x  1 0 0 0 0 0 0 0 0   ∂v 
 ε  0  
 y  0 0 0 1 0 0 0 0   ∂x 
 ε z  0 0 0 0 0 0 0 0 1   ∂v 
 =  
γ xy  0 1 0 1 0 0 0 0 0   ∂y 
γ yz  0 0 0 0 0 1 0 1 0   ∂v 
    
γ zx  0 0 1 0 0 0 1 0 0   ∂z 
 ∂w 
 ∂x 
 
 ∂w 
 ∂y 
 
 ∂w  (3.7.10)
 ∂z 

The above equation can be expressed as

 ∂u 
 ∂ξ 
 
 ∂u 
 ∂η 
 ∂u   
 ∂x   ∂u 
   ∂ζ 
 ∂v   
 ∂y   J11* J12* J13* 0 0 0 0 0 0   ∂v 
   
 ∂w  0 0 0 *
J 21 *
J 22 *
J 23 0 0 0   ∂ξ 
 
 ∂z  0 * 
 ∂v 
{ε } =
* *
0 0 0 0 0 J 31 J 32 J 33
=   *  
 ∂u + ∂v  0   ∂η 
* *
 J 21 J 22 J 23 J11* J12* J13* 0 0
 ∂y ∂x  0 0 0 *
J 31 *
J 32 *
J 33 *
J 21 *
J 22 * 
J 23 ∂v 
   *   
 ∂v + ∂w   J 31 J11* J12* J13*   ∂ζ 
* *
J 32 J 33 0 0 0
 ∂z ∂y   ∂w 
   
 ∂x + ∂w   ∂ξ 
 ∂z ∂x   ∂w 
 
 ∂η 
 ∂v  (3.7.11)
 
 ∂ζ 
 54

8
For an 8 node brick element u can be represented as, u   N i ui and similarly for v & w.
i 1

∂u 8
∂N ∂u ∂N ∂u ∂N
8 8
= ∑ i ui , = ∑ i ui & = ∑ i ui
∂ξ i =1 ∂ξ ∂η i =1 ∂η ∂ζ i =1 ∂ζ
∂v 8
∂N ∂v 8
∂N ∂v 8
∂N
= ∑ i vi , = ∑ i vi & = ∑ i vi (3.7.12)
∂ξ i =1 ∂ξ ∂η i =1 ∂η ∂ζ i =1 ∂ζ
∂w 8 ∂N i ∂w 8 ∂N i ∂w 8
∂N
=∑ wi , =∑ wi & = ∑ i wi
∂ξ i =1 ∂ξ ∂η i =1 ∂η ∂ζ i =1 ∂ζ

Hence eq. (3.7.11) can be rewritten as

 ∂N i 
 ∂ξ 0 0 
 
 ∂N i 
 J11* J12* J13* 0 0 0 0 0 0   0 ∂η
0 
   
 ∂N i 
* * *
0 0 0 J 21 J 22 J 23 0 0 0 
0 0  ui 
8 
0 0 0 0 0 0 *
J 31 *
J 32 * 
J 33 ∂ζ    (3.7.13)
{ε }  * ×∑   vi 
 J 21
*
J 22 *
J 23 J11* J12* J13* 0 0 0  i =1  ∂N i ∂N i 
 ∂η 0   wi 
0 0 0 J*
J*
J*
J *
J * * 
J 23 ∂ξ
 * 31 32 33 21 22
  
 J 31 J *
J *
0 0 0 J *
J *
J13*   ∂N i ∂N i 
 0 ∂η 
32 33 11 12
∂ζ
 
 ∂N i ∂N i 
 ∂ζ 0
 ∂ξ 
Thu, the strain-displacement relationship matrix [B] for 8 node brick element is
 55

 ∂N i 
 ∂ξ 0 0 
 
 ∂N i 
J *
J*
J*
0 0 0 0 0 0   0 ∂η
0 

11 12 13
  
 ∂N i 
* * *
0 0 0 J21 J22 J23 0 0 0 
0 0
8 
0 0 0 0 0 0 *
J 31 *
J 32 * 
J 33 ∂ζ 
[ B]  * ×∑  (3.7.14)
 J 21
*
J 22 *
J 23 J11* J12* J13* 0 0 0  i =1  ∂N i ∂N i 
 ∂η 0 
0 0 0 *
J 31 *
J 32 *
J 33 *
J 21 *
J 22 * 
J 23 ∂ξ
 *   
 J 31
*
J 32 *
J 33 0 0 0 J11* J12* J13 
*
 ∂N i ∂N i 
 0 ∂ζ ∂η 
 
 ∂N i ∂N i 
 ∂ζ 0
 ∂ξ 

The stiffness matrix may be found by using the following expression in natural coordinate system.
1 1 1

 k     B  D Bd    B  D Bdxdydz     [B]T [D][B]dxdhdz J (3.7.15)

T T

 V 1 1 1
 56

Lecture 8: Numerical Integration: One Dimensional

The integrations, we generally encounter in finite element methods, are quite complicated and it is
not possible to find a closed form solutions to those problems. Exact and explicit evaluation of the
integral associated to the element matrices and the loading vector is not always possible because of
the algebraic complexity of the coefficient of the different equation (i.e., the stiffness influence
coefficients, elasticity matrix, loading functions etc.). In the finite element analysis, we face the
problem of evaluating the following types of integrations in one, two and three dimensional cases
respectively. These are necessary to compute element stiffness and element load vector.

 f x dx;  f x, h dxdh;  f x, h, z dxdhdz; (3.8.1)

Approximate solutions to such problems are possible using certain numerical techniques. Several
numerical techniques are available, in mathematics for solving definite integration problems,
including, mid-point rule, trapezoidal-rule, Simpson’s 1/3rd rule, Simpson’s 3/8th rule and Gauss
Quadrature formula. Among these, Gauss Quadrature technique is most useful one for solving
problems in finite element method and therefore will be discussed in details here.

3.8.1 Gauss Quadrature for One-Dimensional Integrals

The concept of Gauss Quadrature is first illustrated in one dimension in the context of an integral in
1 x2
the form of I   f x dx from  f (x)dx . To transform from an arbitrary interval of x1≤ x ≤ x2
1 x1

to an interval of -1 ≤ ξ ≤ 1, we need to change the integration function from f(x) to ϕ(ξ) accordingly.
Thus, for a linear variation in one dimension, one can write the following relations.
1x 1 x
x x1  x 2  N1x1  N 2 x 2
2 2
1 1 1 1
so for x  1, x  x1  x 2  x1
2 2
x  1, x  x 2
x2 1
I f (x)dx   f x dx
x1 1

Numerical integration based on Gauss Quadrature assumes that the function ϕ(ξ) will be evaluated
over an interval -1 ≤ ξ ≤ 1. Considering an one-dimensional integral, Gauss Quadrature represents
the integral ϕ(ξ) in the form of
1 n
I f x dx   w if xi   w1f x1   w 2f x 2   ..  w hf xh  (3.8.2)
1
i1
 57

Where, the ξ1, ξ2, ξ3, ..., ξn represents n numbers of points known as Gauss Points and the
corresponding coefficients w1, w2, w3, …, wn are known as weights. The location and weight
coefficients of Gauss points are calculated by Legendre polynomials. Hence this method is also
sometimes referred as Gauss-Legendre Quadrature method. The summation of these values at n
sampling points gives the exact solution of a polynomial integrand of an order up to 2n-1. For
example, considering sampling at two Gauss points we can get exact solution for a polynomial of an
order (2×2-1) or 3. The use of more number of Gauss points has no effect on accuracy of results but
takes more computation time.

3.8.2 One- Point Formula

Considering n = 1, eq.(3.8.2) can be written as
1

 1
f(x)dx  w1f(x1 ) (3.8.3)

Since there are two parameters w1 and x1 , we need a first order polynomial for ϕ(ξ) to evaluate the
eq.(3.8.3) exactly. For example, considering, f x  a 0  a1x ,
1
Error   a 0  a1xdx  w1f x1   0
1

 2a 0  w1 a 0  a1x1   0

 a 0 2  w1   w1a1x1  0 (3.8.4)

Thus, the error will be zero if w1  2 and x1  0 . Putting these in eq.(3.8.3), for any general ϕ, we
have
1
I   f x dx  2f 0 (3.8.5)
1

This is exactly similar to the well known midpoint rule.

3.8.3 Two-Point Formula

If we consider n = 2, then the eq.(3.8.2) can be written as
1

 1
f x dx  w1f x1   w 2f x 2  (3.8.6)

This means we have four parameters to evaluate. Hence we need a 3rd order polynomial for ϕ(ξ) to
exactly evaluate eq.(3.8.6).
Considering, f x  a 0  a1x  a 2 x 2  a 3x3
 1 
Error    a 0  a1x  a 2 x 2  a 3x3  dx   w1f x1   w 2f x 2 
  1 
 58

a a 2  w1 a 0  a1x1  a 2 x12  a 3x13   w 2 a 0  a1x 2  a 2 x 2 2  a 3x 23   0

2
2 0
3
2 
 2  w1  w 2  a 0   w1x1  w 2 x 2  a1    w1x12  w 2 x 2 2  a 2   w1x13  w 2 x 23  a 3  0
 3 

Fig 3.8.1 One-point Gauss Quadrature

Requiring zero error yields

w1  w 2  2
w1x1  w 2 x 2  0
2 (3.8.7)
w1x12  w 2 x 22 
3
w1x1  w 2 x 2  0
3 3

These nonlinear equations have the unique solution as

w1  w 2  1 x1  x 2  1 3  0.5773502691 (3.8.8)
From this solution, we can conclude that n-point Gaussian Quadrature will provide an exact solution
if ϕ(ξ) is a polynomial of order (2n-1) or less. Table 3.8.1 gives the values of w1 and x1 for Gauss
Quadrature formulas of orders n = 1 through n = 6. From the table it can be observed that the gauss
 59

points are symmetrically placed with respect to origin and those symmetrical points have the same
weights. For accuracy in the calculation maximum number digits for gauss point and gauss weights
should be taken. The Location and weights given in the Table 3.8.1 must be used when the limits of
integration ranges from -1 to 1. Integration limits other than [-1, 1], should be appropriately changed
to [-1, 1] before applying these values.

Table 3.8.1 Gauss points and corresponding weights

Number of Gauss Point Location, x1 Weight, w1
points, n
1 0.0 2.0
2 ±0.5773502692 (= 1 3) 1.0

3 0.0 0.8888888889 (=8/9)

±0.7745966692 (=  6 ) 0.5555555556 (=5/9)
4 ±0.3399810436 0.6521451549
±0.861363116 0.3478548451
5 0.0 0.5688888889
±0.5384693101 0.4786286705
±0.9061798459 0.2369268851
6 ±0.2386191861 0.4679139346
±0.6612093865 0.3607615730
±0.9324695142 0.1713244924

Example 1:
1 2x 
Evaluate I   e x  2 dx using one, two and three point gauss Quadrature.

0  x  2 

Solution:
Before applying the Gauss Quadrature formula, the existing limits of integration should be changed
from [0, 1] to [-1, +1]. Assuming, x  a  bx , the upper and lower limit can be changed. i.e., at x =
0, ξ = -1 and at x = 1, ξ = +1. Thus, putting these conditions and solving for a & b, we get a = -1 and
b = 2. The relation between two coordinate systems will become x  2x 1 and dx  2dx .
Therefore the initial equation can be written as
 60

  x  1 
  x1   
 2  
2
1
  2 
I   e  dx
1   x   
2

  1  2 

 2  
 

1 1  2 4 x  1 
x1

2 1
Or, I   e  dx
x 1  8 
2

Using one point gauss Quadrature:

w1  2, x1  0 and
I  2f 0
1  4 
Or I  2  e0.5    2.22015
 2  7 
Using two point gauss Quadrature:
w1  w 2  1
x1  0.5773502692
x 2  0.5773502692
Putting these values and calculating, I  2.39831

Using three point gauss Quadrature:

w1 = 0.555555556
ξ1 = −0.774596669
w2 = 0.888888889
ξ 2 = 0.000000000
w3 = 0.555555556
ξ3 = 0.774596669
and I  2.41024
This may be compared with the exact solution as Iexact  2.41193
 61

Lecture 9: Numerical Integration: Two and Three Dimensional

Numerical integrations using Gauss Quadrature method can be extended to two and three
dimensional cases in a similar fashion. Such integrations are necessary to perform for the analysis of
plane stress/strain problem, plate and shell structures and for the three dimensional stress analysis.

3.9.1 Gauss Quadrature for Two-Dimensional Integrals

For two dimensional integration problems the above mentioned method can be extended by first
evaluating the inner integral, keeping η constant, and then evaluating the outer integral. Thus,

 n  n  n 
I    f x, h dxdh    w if xi , h dh   w j   w i f xi , h j 
1 1 1
 
1 1 1    i1 
 i1 i1

Or,
n n
I   w i w jf xi , h j  (3.9.1)
i1 j1

In a matrix form we can rewrite the above expression as

φ (ξ1 ,η1 ) φ (ξ1 ,η2 ) φ (ξ1 ,ηn )   w1 
 
φ (ξ 2 ,η1 ) φ (ξ 2 ,η2 ) φ (ξ 2 ,ηn )   w2 
I ≈ [ w1  wn ]
  
w2 (3.9.2)
 
  
φ ( ξ n , η1 ) φ ( ξ n , η 2 ) φ ( ξ n , η )
n   wn 

Example 1:
yd  4 x  b3

Evaluate the integral: I    1 x  2  y dxdy

2 2

yc4 x a  2

Solution:
Before applying the Gauss Quadrature formula, the above integral should be converted in terms of
x and h and the existing limits of y should be changed from [-4,4] to [-1, 1] and that of x is from
[2,3] to [-1,1].
 62

b  a  b  a  x  5 dx
x x  ; dx 
2 2 2 2
d  c  d  c 
y h  4h; dy  4dh
2 2
h1 x1 h1 x1
 3  x 
2

I  2      h  dxdh    f x, hdxdh

2
 2   2 4
h1 x1 h1 x1

 3  x 
2

f x, h  2    h     x    h 
2 2 2
where
 2   2 4 2 3 1 2

Fig. 3.9.1 Gauss points for two-dimensional integral

 63

1 1 1 1
x1   ; h1   ; x2  ; h2 
3 3 3 3
 1   2 
2 2

f x1 , h1   2 3   1    54.49857

 3   3 
 
2
 3  1 
3   4 
2

f x 2 , h1     2    118.83018
 2   3
 
 
2
 3  1 
3   4 
2

f x 2 , h2     2    0.61254
 2   3
 
 
2
 3  1 
 
3   4 
2

f x1 , h 2   
 2  2    0.28093
 3
 
 
 f x1 , h1  f x1 , h 2    

I  w 1 w 2   w1 

f  x , h  f  x , h   
 2 1 2 2  2
w 

 54.49857 0.28093 1

 1 1  
118.83018 0.61254 1

= 174.22222 agrees with the exact value 174.22222

3.9.2 Gauss Quadrature for Three-Dimensional Integrals

In a similar way one can extend the gauss Quadrature for three dimensional problems also and the
integral can be expressed by.
n n n
f x, h, z dxdhdz   w i w j w k f xi , h j , z k 
1 1 1
I   (3.9.3)
1 1 1
i1 j1 k 1

The above equation will produce exact value for a polynomial integrand if the sampling points are
selected as described earlier sections.
 64

3.9.3 Numerical Integration of Element Stiffness Matrix

As discussed earlier notes, the element stiffness matrix for three dimensional analyses in natural
coordinate system can be written as
1 1 1

 k     B  D Bd    B  D Bdxdydz     [B] [D][B]dxdhdz J

(3.9.4)
T T T

 V 1 1 1

Here, [B] and [D] are the strain displacement relationship matrix and constitutive matrix respectively
and integration is performed over the domain. As the element stiffness matrix will be calculated in
natural coordinate system, the strain displacement matrix [B] and Jacobian matrix [J] are functions
of x, h and z . In case of two dimensional isoparametric element, the stiffness matrix will be
simplified to
1 1

k   t   [B] [D][B]dxdh J
T (3.9.5)
1 1

This is actually an 8×8 matrix containing the integrals of each element. We do not need to integrate
elements below the main diagonal of the stiffness matrix as it is symmetric. Considering,
fx, h  t[B]T [D][B] J , the element stiffness matrix will become after numerical integration as
n n
 k    w i w jfxi , h j  (3.9.6)
i1 j1

Using a 2×2 rule, we get

 k   w12f x1 , h1   w1w 2f x1 , h2   w 2 w1f x2 , h1   w 22f h2 , h2  (3.9.7)

Where w1  w 2  1.0, x1  h1  0.57735...., and x 2  h2  0.57735.... Here, wn is the weight

factor at integration point n. A suitable computer program can be written to calculate the element
stiffness matrix through the numerical integration. The process of obtaining stiffness matrix using
Gauss Quadrature integration will be demonstrated through a numerical example in module 5.

3.10.4 Gauss Quadrature for Triangular Elements

The procedure described for the rectangular element will not be applicable directly. The Gauss
Quadrature is extended to include triangular elements in terms of triangular area coordinates.

I   f L1 , L 2 , L3  d A  w if Li1 , Li2 , Li2 

n
(3.9.8)
A i1

Where, L terms are the triangular area coordinates and the wi terms are the weights associated with
those coordinates. The locations of integration points are shown in Fig. 3.9.2.
 65

Fig. 3.9.2 Gauss points for triangles

The sampling points and their associated weights are described below:
For sampling point =1 (Linear triangle)
1
w1  1 L11  L12  L13  (3.9.9)
3
For sampling points =3 (Quadratic triangle)
1 1
w1  L11  L12  , L13  0
3 2
1 1
w2  L21  0, L22 L23  (3.9.10)
3 2
1 1 1
w3  L31  , L32  0, L33 
3 2 2
For sampling point = 7 (Cubic triangle)
27 1
w1  L11  L12  L13 
60 3
8 1
w2  L21  L22  , L23  0
60 2
8 1
w3  L31  0, L32  L23 
60 2
8 1
w4  L41  L43  , L42  0 (3.9.11)
60 2
3
w5  L51  1, L52  L53  0
60
3
w6  L61  L63  0, L62  1
60
3
w7  L71  L72  0, L73  1
60
 66

3.10.5 Gauss Quadrature for Tetrahedron

The Gauss Quadrature for triangles can be effectively extended to include tetrahedron elements in
terms of tetrahedron volume coordinates.

I   f L1 , L 2 , L3 , L 4  d A  w if Li1 , Li2 , Li3 , Li4 

n

i1
A (3.9.12)
Where, L terms are the volume coordinates and the wi terms are the weights associated with those
coordinates. The locations of Gauss points are shown in Fig. 3.9.3.

Fig. 3.9.3 Gauss points for tetrahedrons

The sampling points and their associated weights are described below:
For sampling point = 1 (Linear tetrahedron)
1
w1  1 L11  L12  L13  L14  (3.9.13)
4
For sampling points = 4 (Quadratic tetrahedron)
1
w1  L11  0.5854102, L12  L13  L14  0.1381966
4
1
w2  L22  0.5854102, L21  L23  L24  0.1381966
4
(3.9.14)
1
w3  L33  0.5854102, L31  L32  L34  0.1381966
4
1
w4  L44  0.5854102, L41  L42  L43  0.1381966
4
For sampling points = 5 (Cubic tetrahedron)
 67

4 1
w1   L11  L12  L13  L14 
5 4
9 1 1
w2  L21  , L22  L23  L24 
20 3 6
9 1 1
w3  L32  , L31  L33  L34  (3.9.15)
20 3 6
9 1 1
w4  L43  , L41  L42  L44 
20 3 6
9 1 1
w5  L54  , L51  L52  L43 
20 3 6
 68

Worked out Examples

Example 3.1 Calculation of displacement using area coordinates
The coordinates of a three node triangular element is given below. Calculate the displacement at
point P if the displacements of nodes 1, 2 and 3 are 11 mm, 14mm and 17mm respectively using the
concepts of area coordinates.

1 𝑥1 𝑦1 1 2 3
1 1 1 8
A = 2� 1 𝑥2 𝑦2 � = �1 5 4� = [(30-12) - (12-9) + (8-15)] = 2 = 4
2 2
1 𝑥3 𝑦3 1 3 6

1 𝑥 𝑦 1 3 4
1 1 1 4
𝐴1 = 2
�1 𝑥2 𝑦2 �= �1 5
2
4� = 2 [(30-12) - (18-12) + (12-20)] = 2 = 2
1 𝑥3 𝑦3 1 3 6

Fig. Ex.3.1 Nodal coordinates of a triangular element

1 𝑥 𝑦 1 3 4
1 1 1 2
𝐴2 = 2
�1 𝑥3 𝑦3 �= �1 3
2
6� = 2 [(9-12) - (9-8) + (18-12)] = 2 = 1
1 𝑥1 𝑦1 1 2 3
 69

1 𝑥 𝑦 1 3 4
1 1 1 2
𝐴3 = 2
�1 𝑥1 𝑦1 �= �1 2
2
3� = 2 [(8-15) - (12-20) + (9-8)] = 2 = 1
1 𝑥2 𝑦2 1 5 4

𝐴1 2
𝑁1 = = 4 = 0.5
𝐴

𝐴2 1
𝑁2 = = 4 = 0.25
𝐴

𝐴3 1
𝑁3 = = 4 = 0.25
𝐴

u = 𝑁1 𝑢1 +𝑁2 𝑢2 +𝑁3 𝑢3
= 0.5 x 11 + 0.25 x 14 + 0.25 x 17 = 13.25 mm
 70

Example 3.2 Derivation of shape function of four node triangular element

Derive the shape function of a four node triangular element.

Fig. Ex.3.2 Degrading for four node element

The procedure for four node triangular element is the same as five node triangular element to derive
its interpolation functions. Here, node 5 and 6 are omitted and therefore displacements in these
nodes can be expressed in terms of the displacements at their corner nodes. Hence,
u 2  u3 u1  u 3
u '5  an d u '6 
2 2 (3.11.1)
Substituting the values of u’5 and u’6 in eq.(3.3.8), the following relations can be obtained.
u 2  u 3   u 3  u1 
u  N1u1  N 2 u 2  N 3 u 3  N 4 u 4  N 5  N6
2 2
(3.11.2)
 N   N   N  N6 
  N1  6  u1   N 2  5  u 2   N 3  5  u 3  N 4 u 4
 2  
 2  
 2 
Now, the displacement at any point inside the four node element can be expressed by its nodal
displacement with help of shape function.
u  N1u1  N 2 u 2  N 3 u 3  N 4 u 4 (3.11.3)
Comparing eq. (3.11.2) and eq. (3.11.3), one can find the following relations.
N6 4L L
N1  N1   L1 2L1 1  3 1  L1 1 2L 2 
2 2
N 4L L
N 2  N 2  5  L 2 2L 2 1  2 3  L 2 1 2L1 
2 2 (3.11.4)
N  N6 4L L  4L3 L1
N 3  N 3  5  L3 2L3 1  2 3  L3
2 2
N 4  N 4  4L1L 2
Thus, the shape functions for the four node triangular element are
 71

N1  L1 1 2L 2 
N 2  L 2 1 2L1 
(3.11.5)
N 3  L3
N 4  4L1L 2

Example 3.3 Numerical integration for two dimensional problems

3

Evaluate the integral: I   x  11x  32 dx using one, two and three point gauss Quadrature.
2

2

Also, find the exact solution for comparison of accuracy.

Solution:
The existing limits of integration should be changed from [-2, +3] to [-1, +1]. Assuming, x  a +bx ,
the upper and lower limit can be changed. i.e., at x1  2, x1  1 and at x2  3, x 2  1 . Thus,
putting these limits and solving for a & b, we get a = -0.2 and b = 0.4. The relation between two
coordinate systems will become:
5x  1
x  0.2  0.4x or x  and dx  2.5dx
2

Thus, the initial equation can be written as

3  5x  12 1
 5x  1 
I    x  11x  32 dx = 2.5 
   
 11 
 32d x
 2   2 
2

1 

2 

(i) Exact Solution:

3

I    x 2  11x  32 dx
2

 x3 11x 2 
3

   32 x 
3 
 2  2
 99   8 
 9   96    22  64
 2   3 
 37.5  83.33333  120.83333
Thus, Iexact = -120.83333
 72

(ii) One Point Formula:

1

I   fx d x  w1fx1 
1

For one point formula in Gauss Quadrature integration, w1  2, x1  0 . Thus,

 5 0  12 
I1  2  2.5    11 5 0  1  32
   2  
 2  
 1 11 
 5    32  131.25
 4 2 
Thus, % of error = (120.83333-131.25)×100/120.83333 = 8.62%

(iii) Two Point Formula:

Here, for two point formula in Gauss Quadrature integration,
1
w1  w 2  1.0 and x1  x 2   . Thus,
3
I 2  w1fx1   w2fx 2 
   5 
 5  1  5    5 
2 2
           
      
1  1 1 

1.0  2.5   3   11  3   32  1.0  2.5   3   11  3   32
 2   2    2   2  
         
         

 0.88996 10.37713  32 2.5  3.77671  21.3771  32 2.5

 48.3333 2.5
= 120.83325
Thus, % of error = (120.83333-120.83325)×100/120.83333 = 6.62×10-05

(iv) Three Point Formula:

Here, for three point formula in Gauss Quadrature integration,
w1  0.8889, x1  0.0
w2  0.5556, x 2  0.7746
w3  0.5556, x3  0.7746
Thus,
I 3  w1fx1   w2fx 2   w3fx3 
 73

 5 0  12 5 0  1 

I 3  0.8889  2.5     11  32
 2 
 2 
 5 0.7746  12 
 0.5556  2.5    11 5 0.7746  1  32
  
 2 2 
 5 0.7746  12 
 0.5556  2.5     11 5 0.7746  1  32
  
 2 2 
I 3  0.8889  2.50.25  5.5  32
 0.5556  2.55.9365  26.8015  32
 0.5556  2.5 2.0635 15.8015  32
 2.523.3336  0.4100  25.4120
 2.5 48.3356  120.839
Thus, % of error = (120.83333-120.839)×100/120.83333 = 4.69×10-03. However, difference of
results will approach to zero, if few more digits after decimal points are taken in calculation.

Example 3.4 Numerical integration for three dimensional problems

1 1 1
Evaluate the integral: I     1 2x 1h 3z  2 dxdhdz
2 2 2

1 1 1

Solution:
Using two point gauss Quadrature formula for the evaluation of three dimensional integration, we
have the following sampling points and weights.
w1  w 2  1
x1  0.5773502692
x 2  0.5773502692
h1  0.5773502692
h 2  0.5773502692
z1  0.5773502692
z 2  0.5773502692

Putting the above values, in f x, h, z  1 2x 1h 3z  2 one can find the following values
2 2 2

in 8 (i.e., 2 × 2 × 2) sampling points.

 74

f x1 , h1 , z1   160.8886
f x1 , h1 , z 2   0.8293
f x1 , h 2 , z1   11.5513
f x1 , h 2 , z 2   0.0595
f x 2 , h1 , z1   0.8293
f x 2 , h1 , z 2   0.0043
f x 2 , h 2 , z1   0.0595
f x 2 , h 2 , z 2   0.0003
2 2 2
Now, I   w i w j w k f xi , h j , z k 
i1 j1 k 1

Thus, I  w1w1w1f x1 , h1 , z1   w1w1w 2f x1 , h1 , z 2     w 2 w 2 w 2f x 2 , h2 , z 2  = 174.222, where as

Iexact = 174.222.