Hutton: Fundamentals of 6.

Interpolation Functions Text © The McGraw−Hill

Finite Element Analysis for General Element Companies, 2004
Formulation

176 CHAPTER 6 Interpolation Functions for General Element Formulation

remains that each independent variable must be of equal “strength” in the poly-
nomial. For example, the 3-D quadratic polynomial
P (x , y, z) = a0 + a1 x + a2 y + a3 z + a4 x 2 + a5 y 2 + a6 z 2
+ a7 x y + a8 x z + a9 yz (6.29)
is complete and could be applied to an element having 10 nodes. Similarly, an
incomplete, symmetric form such as
P (x , y, z) = a0 + a1 x + a2 y + a3 z + a4 x 2 + a5 y 2 + a6 z 2 (6.30)
or
P (x , y, z) = a0 + a1 x + a2 y + a3 z + a4 x y + a5 x z + a6 yz (6.31)
could be used for elements having seven nodal degrees of freedom (an unlikely
case, however).
Geometric isotropy is not an absolute requirement for field variable
repesentation [1], hence, interpolation functions. As demonstrated by many
researchers, incomplete representations are quite often used and solution conver-
gence attained. However, in terms of h-refinement, use of geometrically isotropic
representations guarantees satisfaction of the compatibility and completeness
requirements. For the p-refinement method, the reader is reminded that the inter-
polation functions in any finite element analysis solution are approximations to
the power series expansion of the problem solution. As we increase the number
of element nodes, the order of the interpolation functions increases and, in the
limit, as the number of nodes approaches infinity, the polynomial expression of
the field variable approaches the power series expansion of the solution.

6.5 TRIANGULAR ELEMENTS

The interpolation functions for triangular elements are inherently formulated in
two dimensions and a family of such elements exists. Figure 6.7 depicts the first
three elements (linear, quadratic, and cubic) of the family. Note that, in the case
of the cubic element, an internal node exists. The internal node is required to

(a) (b) (c)

Figure 6.7 Triangular elements:

(a) 3-node linear, (b) 6-node quadratic,
(c) 10-node cubic.
Hutton: Fundamentals of 6. Interpolation Functions Text © The McGraw−Hill
Finite Element Analysis for General Element Companies, 2004
Formulation

6.5 Triangular Elements 177

3 (x3, y3)

y
2 (x2, y2)

x 1 (x1, y1)

Figure 6.8 A general three-

node triangular element
referred to global coordinates.

obtain geometric isotropy, as is subsequently discussed. The triangular elements

are not limited to two-dimensional problems. In fact, the triangular elements can
be used in axisymmetric 3-D cases (discussed later in this chapter) as well as in
structural analyses involving out-of-plane bending, as in plate and shell struc-
tures. In the latter cases, the nodal degrees of freedom include first derivatives of
the field variable as well as the field variable itself. While plate and shell prob-
lems are beyond the scope of this book, we allude to those problems again briefly
in Chapter 9.
Figure 6.8 depicts a general, three-node triangular element to which we
attach an element coordinate system that is, for now, assumed to be the same as
the global system. Here, it is assumed that only 1 degree of freedom is associated
with each node. We express the field variable in the polynomial form
␾(x , y) = a0 + a1 x + a2 y (6.32)
Applying the nodal conditions
␾(x 1 , y1 ) = ␾1
␾(x 2 , y2 ) = ␾2 (6.33)
␾(x 3 , y3 ) = ␾3

and following the general procedure previously outlined, we obtain

    
1 x1 y1  a0   ␾1 
 1 x2 y2  a1 = ␾2 (6.34)
   
1 x2 y2 a2 ␾3

To solve for the polynomial coefficients, the matrix of coefficients in Equa-

tion 6.34 must be inverted. Inversion of the matrix is algebraically tedious but
straightforward, and we find
1
a0 = [␾1 (x 2 y3 − x 3 y2 ) + ␾2 (x 3 y1 − x 1 y3 ) + ␾3 (x 1 y2 − x 2 y1 )]
2A
 Hutton: Fundamentals of 6. Interpolation Functions Text © The McGraw−Hill
Finite Element Analysis for General Element Companies, 2004
Formulation

178 CHAPTER 6 Interpolation Functions for General Element Formulation

1
a1 = [␾1 ( y2 − y3 ) + ␾2 ( y3 − y1 ) + ␾3 ( y1 − y2 )]
2A
(6.35)
1
a2 = [␾1 (x 3 − x 2 ) + ␾2 (x 1 − x 3 ) + ␾3 (x 2 − x 1 )]
2A
Substituting the values into Equation 6.32 and collecting coefficients of the nodal
variables, we obtain
 
 [(x2 y3 − x3 y2 ) + ( y2 − y3 )x + (x3 − x2 ) y]␾1 
1
␾(x, y) = + [(x3 y1 − x1 y3 ) + ( y3 − y1 )x + (x1 − x3 ) y]␾2 (6.36)
2A  
+ [(x1 y2 − x2 y1 ) + ( y1 − y2 )x + (x2 − x1 ) y]␾3

Given the form of Equation 6.36, the interpolation functions are observed to be
1
N 1 (x , y) = [(x 2 y3 − x 3 y2 ) + ( y2 − y3 )x + (x 3 − x 2 ) y]
2A
1
N 2 (x , y) = [(x 3 y1 − x 1 y3 ) + ( y3 − y1 )x + (x 1 − x 3 ) y] (6.37)
2A
1
N 3 (x , y) = [(x 1 y2 − x 2 y1 ) + ( y1 − y2 )x + (x 2 − x 1 ) y]
2A

where A is the area of the triangular element. Given the coordinates of the three
vertices of a triangle, it can be shown that the area is given by

1 1 x1 y1
A= 1 x2 y2 (6.38)
2 1 x3 y3

Note that the algebraically complex form of the interpolation functions

arises primarily from the choice of the element coordinate system of Figure 6.8.
As the linear representation of the field variable exhibits geometric isotropy,
location and orientation of the element coordinate axes can be chosen arbitrarily
without affecting the interpolation results. If, for example, the element coordi-
nate system shown in Figure 6.9 is utilized, considerable algebraic simplification
results. In the coordinate system shown, we have x 1 = y1 = y2 = 0, 2 A = x 2 y3 ,

3 (x3, y3)

y Figure 6.9 Three-node

x
2 triangle having an element
(x2, 0) coordinate system attached
1 (0, 0) to the element.
Hutton: Fundamentals of 6. Interpolation Functions Text © The McGraw−Hill
Finite Element Analysis for General Element Companies, 2004
Formulation

6.5 Triangular Elements 179

and the interpolation functions become

1
N 1 (x , y) = [x 2 y3 − y3 x + (x 3 − x 2 ) y]
x 2 y3
1
N 2 (x , y) = [y3 x − x 3 y] (6.39)
x 2 y3
y
N 3 (x , y) =
y3
which are clearly of a simpler form than Equation 6.37. The simplification is
not without cost, however. If the element coordinate system is directly associ-
ated with element orientation, as in Figure 6.9, the element characteristic matri-
ces must be transformed to a common global coordinate system during model
assembly. (Recall the transformation of stiffness matrices demonstrated for bar
and beam elements earlier.) As finite element models usually employ a large
number of elements, the additional computations required for element trans-
formation can be quite time consuming. Consequently, computational efficiency
is improved if each element coordinate system is oriented such that the axes
are parallel to the global axes. The transformation step is then unnecessary
when model assembly takes place. In practice, most commercial finite element
software packages provide for use of either type element coordinate as a user
option [6].
Returning to Equation 6.32, observe that it is possible for the field variable
to take on a constant value, as per the completeness requirement, and that the first
partial derivatives with respect to the independent variables x and y are constants.
The latter shows that the gradients of the field variable are constant in both coor-
dinate directions. For a planar structural element, this results in constant strain
components. In fact, in structural applications, the three-node triangular element
is commonly known as a constant strain triangle (CST, for short). In the case of
heat transfer, the element produces constant temperature gradients, therefore,
constant heat flow within an element.

6.5.1 Area Coordinates

When expressed in Cartesian coordinates, the interpolation functions for the
triangular element are algebraically complex. Further, the integrations required
to obtain element characteristic matrices are cumbersome. Considerable sim-
plification of the interpolation functions as well as the subsequently required
integration is obtained via the use of area coordinates. Figure 6.10 shows a
three-node triangular element divided into three areas defined by the nodes and
an arbitrary interior point P (x , y) . Note: P is not a node. The area coordinates of
P are defined as
A1 A2 A3
L1 = L2 = L3 = (6.40)
A A A
 Hutton: Fundamentals of 6. Interpolation Functions Text © The McGraw−Hill
Finite Element Analysis for General Element Companies, 2004
Formulation

180 CHAPTER 6 Interpolation Functions for General Element Formulation

A2 A1

A3
2
Figure 6.10 Areas used to define area
1 coordinates for a triangular element.

L1 ⫽ 0

3 3
1
L1 ⫽
2

P⬘
L1 ⫽ 1
P
2 2
1 1
(a) (b)

Figure 6.11
(a) Area A1 associated with either P or P  is constant.
(b) Lines of the constant area coordinate L1.

where A is the total area of the triangle. Clearly, the area coordinates are not
independent, since
L1 + L2 + L3 = 1 (6.41)
The dependency is to be expected, since Equation 6.40 expresses the location of
a point in two-dimensions using three coordinates.
The important properties of area coordinates for application to triangular
finite elements are now examined with reference to Figure 6.11. In Figure 6.11a,
a dashed line parallel to the side defined by nodes 2 and 3 is indicated. For any
two points P and P  on this line, the areas of the triangles formed by nodes 2 and
3 and either P or P  are identical. This is because the base and height of any tri-
angle so formed are constants. Further, as the dashed line is moved closer to node
1, area A 1 increases linearly and has value A 1 = A , when evaluated at node 1.
Therefore, area coordinate L 1 is constant on any line parallel to the side of the tri-
angle opposite node 1 and varies linearly from a value of unity at node 1 to value
of zero along the side defined by nodes 2 and 3, as depicted in Figure 6.11b. Sim-
ilar arguments can be made for the behavior of L 2 and L 3 . These observations
Hutton: Fundamentals of 6. Interpolation Functions Text © The McGraw−Hill
Finite Element Analysis for General Element Companies, 2004
Formulation

6.5 Triangular Elements 181

can be used to write the following conditions satisfied by the area coordinates
when evaluated at the nodes:

Node 1: L 1 = 1 L2 = L3 = 0
Node 2: L 2 = 1 L1 = L3 = 0 (6.42)
Node 3: L 3 = 1 L1 = L2 = 0

The conditions expressed by Equation 6.42 are exactly the conditions that
must be satisfied by interpolation functions at the nodes of the triangular ele-
ment. So, we express the field variable as
␾(x , y) = L 1 ␾1 + L 2 ␾2 + L 3 ␾3 (6.43)

in terms of area coordinates. Is this different from the field variable representa-
tion of Equation 6.36? If the area coordinates are expressed explicitly in terms
of the nodal coordinates, the two field variable representations are shown to be
identical. The true advantages of area coordinates are seen more readily in
developing interpolation functions for higher-order elements and performing
integration of various forms of the interpolation functions.

6.5.2 Six-Node Triangular Element

A six-node element is shown in Figure 6.12a. The additional nodes 4, 5, and 6 are
located at the midpoints of the sides of the element. As we have six nodes, a com-
plete polynomial representation of the field variable is
␾(x , y) = a0 + a1 x + a2 y + a3 x 2 + a4 x y + a5 y 2 (6.44)

L1 ⫽ 0 L2 ⫽ 0

3
3 L3 ⫽ 1
1
1 L2 ⫽
L1 ⫽ 2
2
1
5 L3 ⫽
6 6 5 2
L2 ⫽ 1
L1 ⫽ 1
L3 ⫽ 0
2 2
1 4 1 4
(a) (b)

Figure 6.12 Six-node triangular elements: (a) Node numbering convention.

(b) Lines of constant values of the area coordinates.
 Hutton: Fundamentals of 6. Interpolation Functions Text © The McGraw−Hill
Finite Element Analysis for General Element Companies, 2004
Formulation

182 CHAPTER 6 Interpolation Functions for General Element Formulation

which is ultimately to be expressed in terms of interpolation functions and nodal

values as

6
␾(x , y) = N i (x , y)␾i (6.45)
i=1

As usual, each interpolation function must be such that its value is unity when
evaluated at its associated node and zero when evaluated at any of the other five
nodes. Further, each interpolation is a quadratic function, since the field variable
representation is quadratic.
Figure 6.12b shows the six-node element with lines of constant values of the
area coordinates passing through the nodes. Using this figure and a bit of logic,
the interpolation functions are easily “constructed” in area coordinates. For
example, interpolation function N 1 (x , y) = N 1 ( L 1 , L 2 , L 3 ) must have value of
zero at nodes 2, 3, 4, 5, and 6. Noting that L 1 = 1/2 at nodes 4 and 6, inclusion
of the term L 1 − 1/2 ensures a zero value at those two nodes. Similarly, L 1 = 0
at nodes 2, 3, 4, so the term L 1 satisfies the conditions at those three nodes.
Therefore, we propose
 
1
N1 = L 1 L 1 − (6.46)
2
However, evaluation of Equation 6.46 at node 1, where L 1 = 1 , results in
N 1 = 1/2 . As N 1 must be unity at node 1, Equation 6.46 is modified to
 
1
N 1 = 2L 1 L 1 − = L 1 (2L 1 − 1) (6.47a)
2
which satisfies the required conditions at each of the six nodes and is a quadratic
function, since L 1 is a linear function of x and y.
Applying the required nodal conditions to the remaining five interpolation
functions in turn, we obtain
N 2 = L 2 (2L 2 − 1) (6.47b)
N 3 = L 3 (2L 3 − 1) (6.47c)
N 4 = 4L 1 L 2 (6.47d)
N 5 = 4L 2 L 3 (6.47e)
N 6 = 4L 1 L 3 (6.47f)
Using a similarly straightforward procedure, interpolation functions for addi-
tional higher-order triangular elements can be constructed. The 10-node cubic
element is left as an exercise.

6.5.3 Integration in Area Coordinates

As seen in Chapter 5 and encountered again in later chapters, integration of var-
ious forms of the interpolation functions over the domain of an element are
Hutton: Fundamentals of 6. Interpolation Functions Text © The McGraw−Hill
Finite Element Analysis for General Element Companies, 2004
Formulation

6.5 Triangular Elements 183

required in formulating element characteristic matrices and load vectors. When

expressed in area coordinates, integrals of the form

L a1 L b2 L c3 d A (6.48)
A

(where A is the total area of a triangle defined by nodes 1, 2, 3) must often be

evaluated. The relation

a!b!c!
L a1 L b2 L c3 d A = (2 A) (6.49)
(a + b + c + 2)!
A

has been shown [7] to be valid for all exponents a, b, c that are positive integers
(recall 0! = 1 ). Therefore, integration in area coordinates is quite straight-
forward.
EXAMPLE 6.2

As will be shown in Chapter 7, the convection terms of the stiffness matrix for a 2-D heat
transfer element are of the form

ki j = h Ni N j d A
A

where h is the convection coefficient and A is element area. Use the interpolation func-
tions for a six-node triangular element given by Equation 6.47 to compute k 24 .
■ Solution
Using Equation 6.47b and 6.47d, we have
N 2 = L 2 (2L 2 − 1)
N 4 = 4L 1 L 2
so (assuming h is a constant)
 
 
k 24 = h L 2 (2L 2 − 1)4L 1 L 2 d A = h 8L 1 L 32 − 4L 1 L 22 d A
A A

Applying Equation 6.49, we have


(1!)(3!)(0!) 96h A 2h A
h 8L 1 L 32 d A = 8h(2 A) = =
(1 + 3 + 0 + 2)! 720 15
A

(1!)(2!)(0!) 16h A 2h A
h 4L 1 L 22 d A = 4h(2 A) = =
(1 + 2 + 0 + 2)! 120 15
A

Therefore,
2h A 2h A
k 24 = − =0
15 15