CHAPTER 3

Interpolation Models

to derive the element characteristic matrices and vectors. We know that the integral of a
stepwise continuous function, say f(x), is defined if f(x) remains bounded in the interval of
integration. Thus, for the integral

dr ϕ
Z
dx
dxr

to be defined, ϕ must be continuous to the order (r – 1) to ensure that only finite jump
discontinuities occur in the r-th derivative of ϕ. This is precisely the requirement stated
previously.
The elements whose interpolation polynomials satisfy the requirements (1) and (3) are
called compatible or conforming elements and those satisfying condition (2) are called
complete elements. If r-th derivative of the field variable ϕ is continuous, then ϕ is said to
have Cr continuity. In terms of this notation, the completeness requirement implies that ϕ
must have Cr continuity within an element, whereas the compatibility requirement implies
that ϕ must have Cr–1 continuity at element interfaces.2
In the case of general solid and structural mechanics problems, this requirement implies
that the element must deform without causing openings, overlaps, or discontinuities
between adjacent elements. In the case of beam, plate, and shell elements, the first
derivative of the displacement (slope) across interelement boundaries also must be
continuous.
Although it is desirable to satisfy all the convergence requirements, several interpolation
polynomials that do not meet all the requirements have been used in the finite element
literature. In some cases, acceptable convergence or convergence to an incorrect solution has
been obtained. In particular, the interpolation polynomials that are complete but not
83
conforming have been found to give satisfactory results.
If the interpolation polynomial satisfies all three requirements, the approximate
solution converges to the correct solution when we refine the mesh and use
an increasing number of smaller elements. In order to prove the convergence
mathematically, the mesh refinement has to be made in a regular fashion so as to
satisfy the following conditions:

1. All previous (coarse) meshes must be contained in the refined meshes.

2. The elements must be made smaller in such a way that every point of the solution
region can always be within an element.
3. The form of the interpolation polynomial must remain unchanged during the process of
mesh refinement.

Conditions (1) and (2) are illustrated in Figure 3.4, in which a two-dimensional region
(in the form of a parallelogram) is discretized with an increasing number of triangular
elements. From Figure 3.5, in which the solution region is assumed to have a curved
boundary, it can be seen that conditions (1) and (2) are not satisfied if we use elements
with straight boundaries. In structural problems, interpolation polynomials satisfying
all the convergence requirements always lead to the convergence of the displacement
solution from below while nonconforming elements may converge either from below or
from above.

2
This statement assumes that the functional (I) corresponding to the problem contains derivatives of ϕ up to the
r-th order.
 PART 2
Basic Procedure

A A

H E

D B D B

F
G

C C
(a) Idealization with 2 elements (b) Idealization with 8 elements

H E

B
D

G F

C
(c) Idealization with 32 elements

FIGURE 3.4
All Previous Meshes Contained in Refined Meshes.
84

(a) Idealization with 6 elements (b) Idealization with 12 elements

FIGURE 3.5
Previous Mesh Is Not Contained in the Refined Mesh.

NOTES
1. For any physical problem, the selection of finite elements and interpolation polynomials to achieve
C 0 continuity is not very difficult. However, the difficulty increases rapidly when higher order
continuity is required. In general, the construction of finite elements to achieve specified continuity
of order C0, C1, C2, ... , requires skill, ingenuity, and experience. Fortunately, most of the time, we
would be able to use the elements already developed in an established area such as stress analysis
for solving new problems.
 CHAPTER 3
Interpolation Models

2. The construction of an efficient finite element model involves (a) representing the geometry of
the problem accurately, (b) developing a finite element mesh to reduce the bandwidth,
and (c) choosing a proper interpolation model to obtain the desired accuracy in the solution.
Unfortunately, there is no a priori method of creating a reasonably efficient finite element
model that can ensure a specified degree of accuracy. Several numerical tests are available for
assessing the convergence of a finite element model [3.9, 3.10].
Some adaptive finite element methods have been developed to employ the results from previous
meshes to estimate the magnitude and distribution of solution errors and to adaptively improve the
finite element model [3.11–3.15]. There are four basic approaches to adaptively improve a finite
element model:

a. Subdivide selected elements (called h-method).

b. Increase the order of the polynomial of selected elements (called p-refinement).
c. Move node points in fixed element topology (called r-refinement).
d. Define a new mesh having a better distribution of elements.
Various combinations of these approaches are also possible. Determining which of these approaches
is the best for a particular class of problems is a complex problem that must consider the cost of the
entire solution process.

3.7 LINEAR INTERPOLATION POLYNOMIALS IN TERMS

OF GLOBAL COORDINATES
The linear interpolation polynomials correspond to simplex elements. In this section, we
derive the linear interpolation polynomials for the basic one-, two-, and three-dimensional
elements in terms of the global coordinates that are defined for the entire domain or body.
85
3.7.1 One-Dimensional Simplex Element
Consider a one-dimensional element (line segment) of length l with two nodes, one at each
end, as shown in Figure 3.6. Let the nodes be denoted as i and j and the nodal values of the

φ(x )

φ(x) = α1 + α 2 x

Φj

Φi φ(x )

x
0 i j

xi = (xj − xi )

xj

FIGURE 3.6
One-Dimensional Simplex Element.
 PART 2
Basic Procedure

field variable ϕ as Φi and Φj. The variation of ϕ inside the element is assumed to be linear as
ϕðxÞ = α1 + α2 x (3.21)
where α1 and α2 are the unknown coefficients. By using the nodal conditions
ϕðxÞ = Φi at x = xi
ϕðxÞ = Φj at x = xj
and Eq. (3.21), we obtain
Φi = α1 + α2 xi
Φj = α1 + α2 xj

The solution of these equations gives

Φ i x j − Φ j xi >
9
α1 = =
l (3.22)
Φj − Φi >
α2 = ;
l
where xi and xj denote the global coordinates of nodes i and j, respectively. By substituting
Eq. (3.22) into Eq. (3.21), we obtain
Φi xj − Φj xi Φj − Φi
% & % &
ϕðxÞ = + x (3.23)
l l
This equation can be written, after rearrangement of terms, as
!ðeÞ
ϕðxÞ = Ni ðxÞΦi + Nj ðxÞΦj = ½NðxÞ#Φ (3.24)
where
86 ½NðxÞ# = ½Ni ðxÞ Nj ðxÞ# (3.25)
xj − x >
9
Ni ðxÞ = =
l (3.26)
x − xi >
Nj ðxÞ = ;
l
and
' (
!ðeÞ Φi
Φ = = vector of nodal unknowns of elements e (3.27)
Φj
Note that the superscript e is not used for Φi and Φj for simplicity.
The linear functions of x defined in Eq. (3.26) are called interpolation or shape functions.3
Note that each interpolation function has a subscript to denote the node to which it is
associated. Furthermore, the value of Ni(x) can be seen to be 1 at node i (x = xi) and 0 at
node j (x = xj). Likewise, the value of Nj(x) will be 0 at node i and 1 at node j. These
represent the common characteristics of interpolation functions. They will be equal to 1 at
one node and 0 at each of the other nodes of the element.

The original polynomial type of interpolation model ϕ = ! η !

3 T
α (which is often called the interpolation polynomial
or interpolation model of the element) should not be confused with the interpolation functions Ni associated with
the nodal degrees of freedom. There is a clear difference between the two. The expression !η !
T
α denotes an
interpolation polynomial that applies to the entire element and expresses the variation of the field variable inside
the element in terms of the generalized coordinates αi. The interpolation function Ni corresponds to the i-th nodal
degree of freedom Φi
ðeÞ
and only the sum ∑i Ni ΦiðeÞ represents the variation of the field variable inside the element
ðeÞ
in terms of the nodal degrees of freedom Φi . In fact, the interpolation function corresponding to the i-th nodal
degree of freedom (Ni) assumes a value of 1 at node i, and 0 at all the other nodes of the element.
 CHAPTER 3
Interpolation Models

EXAMPLE 3.2
The nodal temperatures of nodes i and j (same as local nodes 1 and 2) of an element in a one-
dimensional fin are known to be Ti = 120°C and Tj = 80°C with the x-coordinates xi = 30 cm and
xj = 50 cm. Find the following:
a. Shape functions associated with the nodal values Ti and Tj.
b. Interpolation model for the temperature inside the element, T(x).
c. Temperature in the element at x = 45 cm.

Solution
1. The shape functions Ni(x) and Nj(x) are given by Eq. (3.26):
xj − x
Ni ðxÞ = = 50 − x = 2:5 − 0:05 x (E.1)
l 50 − 30
x − xi x − 30
Nj ðxÞ = = = 0:05 x − 1:5 (E.2)
l 50 − 30
2. The interpolation model for the temperature inside the element can be expressed, using
Eq. (3.24), as
TðxÞ = Ni ðxÞTi + Nj ðxÞTj = ð2:5 − 0:05 xÞ120 + ð0:05 x − 1:5Þ 80°C (E.3)

3. The temperature at x = 45 cm can be determined from Eq. (E.3) as

Tðx = 45Þ = ð2:5 − 0:05ð45ÞÞ120 + ð0:05ð45Þ − 1:5Þ 80 = 90°C (E.4)

EXAMPLE 3.3
A one-dimensional tapered fin element has the nodal coordinates xi = 20 mm and xj = 60 mm with the
area of cross section changing linearly from a value of Ai = 20 mm2 at xi to a value of Aj = 10 mm2
at xj as shown in Figure 3.7. (1) Determine the matrix of shape functions and (2) express the area 87
of cross section of the fin element in terms of the shape functions.

Approach: (1) Use the shape functions corresponding to linear variation of the field variable. (2) Express
linear variation of cross-sectional area in terms of shape functions similar to the variation of the field
variable.

Solution
1. The linear variation of the field variable ϕ(x) can be expressed by Eq. (3.21) or, equivalently, by
Eq. (3.24):
!ðeÞ
ϕðxÞ = α1 + α2 x = Ni ðxÞΦi + Nj ðxÞΦj = ½NðxÞ#Φ (E.1)

where the matrix of shape functions [N(x)] is given by Eq. (3.25):

xj − x x − xi
) *
½NðxÞ# = ½Ni ðxÞ Nj ðxÞ# ≡
xj − xi xj − xi
(E.2)
60 − x x − 20 60 − x x − 20
h i h i
= =
60 − 20 60 − 20 40 40

Ai = 20 mm2
A (x ) Aj = 10 mm2

0 x (mm)
x
xj = 60
xi = 20
FIGURE 3.7
A Tapered Fin Element.
(Continued )
 PART 2
Basic Procedure

EXAMPLE 3.3 (Continued )

2. The linear variation of the cross-sectional area of the element can be expressed as
AðxÞ = β1 + β2 x (E.3)
where the values of the constants β1 and β2 can be found by using the known areas of cross section
at the two nodes:
Aðx = xi = 20Þ = Ai = 20 mm2 and Aðx = xj = 60Þ = Aj = 10 mm2

This gives
Ai xj − Aj xi 20ð60Þ − 10ð20Þ Aj − Ai
β1 = = = 25, β2 = = 10 − 20 = −0:25 (E.4)
xj − xi 60 − 20 xj − xi 60 − 20

Using Eq. (E.4), Eq. (E.3) can be expressed in terms of the shape functions as
!ðeÞ
AðxÞ = Ai Ni ðxÞ + Aj Nj ðxÞ ≡ ½NðxÞ# A (E.5)
!ðeÞ
where the matrix of shape functions, [N(x)], is given by Eq. (E.2) and A is the vector of nodal
areas of cross section of the element:

Ai
' ( ' (
!ðeÞ 20
A = = mm2 (E.6)
Aj 10

Note
Equation (E.4) gives β1 = 25 and β2 = −0.25 so that the variation of A(x) can also be expressed as

AðxÞ = 25 − 0:25x (E.7)

88
3.7.2 Two-Dimensional Simplex Element
The two-dimensional simplex element is a straight-sided triangle with three nodes, one at
each corner, as indicated in Figure 3.8. Let the nodes be labeled as i, j, and k by proceeding
counterclockwise from node i, which is arbitrarily specified. Let the global coordinates of
the nodes i, j, and k be given by (xi, yi), (xj, yj), and (xk, yk) and the nodal values of the field
variable ϕ (x, y) by Φi, Φj, and Φk, respectively. The variation of ϕ inside the element is
assumed to be linear as

ϕðx, yÞ = α1 + α2 x + α3 y (3.28)

The nodal conditions

ϕðx, yÞ = Φi at ðx = xi , y = yi Þ
ϕðx, yÞ = Φj at ðx = xj , y = yj Þ
ϕðx, yÞ = Φk at ðx = xk , y = yk Þ

lead to the system of equations

Φi = α1 + α2 xi + α3 yi
Φj = α1 + α2 xj + α3 yj (3.29)
Φk = α1 + α2 xk + α3 yk
The solution of Eqs. (3.29) yields

1
α1 = ðai Φi + aj Φj + ak Φk Þ
2A
1
α2 = ðbi Φi + bj Φj + bk Φk Þ (3.30)
2A
1
α3 = ðci Φi + cj Φj + ck Φk Þ
2A
 CHAPTER 3
Interpolation Models

φ(x, y )

Φj
φ(x, y ) = α1 + α 2x + α 3 y

Φi

Φk

φ(x, y )

j y
(xj, yj )

i
x (x, y ) (xi, yi )

k
(xk, yk)

FIGURE 3.8
Two-Dimensional Simplex Element.

where A is the area of the triangle ijk given by

- 1 xi y i -
- -
1 - - 1 89
A = -- 1 xj yj -- = ðxi yj + xj yk + xk yi − xi yk − xj yi − xk yj Þ (3.31)
2- 2
1 xk yk -
a i = xj y k − xk y j
aj = xk˙ yi − xi yk
ak = xi yj − xj yi
bi = yj − yk
bj = yk − yi (3.32)
bk = yi − yj
c i = xk − xj
c j = xi − xk
c k = xj − xi
Substitution of Eq. (3.30) into Eq. (3.28) and rearrangement yields the equation
!ðeÞ
ϕðx, yÞ = Ni ðx, yÞΦi + Nj ðx, yÞΦj + Nk ðx, yÞΦk = ½Nðx, yÞ#Φ (3.33)
where
½Nðx, yÞ# = ½Ni ðx, yÞ Nj ðx, yÞ Nk ðx, yÞ# (3.34)

1
Ni ðx, yÞ = ðai + bi x + ci yÞ
2A
1
Nj ðx, yÞ = ðaj + bj x + cj yÞ (3.35)
2A
1
Nk ðx, yÞ = ðak + bk x + ck yÞ
2A
and
8 9
< Φi =
!ðeÞ
Φ = Φj = vector of nodal unknowns of element e (3.36)
: ;
Φk
 PART 2
Basic Procedure

NOTES
1. The shape function Ni(x, y) when evaluated at node i (xi, yi) gives
1
Ni ðxi , yi Þ = ðai + bi xi + ci yi Þ
2A
(3.37)
1 1 2
= xj yk − xk yj + xi yj − xi yk + xk yi − xj yi = 1
2A
It can be shown that Ni(x, y) = 0 at nodes j and k, and at all points on the line passing through
these nodes. Similarly, the shape functions Nj and Nk have a value of 1 at nodes j and k,
respectively, and 0 at other nodes.
2. Since the interpolation functions are linear in x and y, the gradient of the field variable in x or y
direction will be a constant. For example,
∂ϕðx, yÞ !ðeÞ
= ∂ ½N ðx, yÞ#Φ = ðbi Φi + bj Φj + bk Φk Þ/2A (3.38)
∂x ∂x
Since Φi, Φj, and Φk are the nodal values of ϕ (independent of x and y), and bi, bj, and bk are
constants whose values are fixed once the nodal coordinates are specified, (∂ø/∂x) will be a constant.
A constant value of the gradient of ϕ within an element means that many small elements have to
be used in locations where rapid changes are expected in the value of ϕ.

EXAMPLE 3.4
The temperatures at the nodes of a triangular element are given by Ti = 210°F, Tj = 270°F, and Tk = 250°F.
If the nodal coordinates are (xi, yi) = (50, 30) in, (xj, yj) = (70, 50) in, and (xk, yk) = (55, 60) in, determine (a)
the shape functions of the element and (b) temperature at the point (x, y) = (60, 40) in inside the element.

Solution
90 1. From the known nodal coordinates, the area of the triangular element and the constants ai, bi, ci, ...
involved in the shape functions can be determined as

A = 1 xi yj + xj yk + xk yi − xi yk − xj yi − xk yj
1 2
2
= 1 ð50 × 50 + 70 × 60 + 55 × 30 − 50 × 60 − 70 × 30 − 55 × 50Þ = 250 in2
2
ai = xj yk − xk yj = 70 × 60 − 55 × 50 = 1450
aj = xk yi − xi yk = 55 × 30 − 50 × 60 = −1350
ak = xi yj − xj yi = 50 × 50 − 70 × 30 = 400
bi = yj − yk = 50 − 60 = −10
bj = yk − yi = 60 − 30 = 30
bk = yi − yj = 30 − 50 = −20
ci = xk − xj = 55 − 70 = −15
cj = xi − xk = 50 − 55 = −5
ck = xj − xi = 70 − 50 = 20

The shape functions can be found as

Ni ðx, yÞ = 1 ðai + bi x + ci yÞ = 1 ð1450 − 10x − 15yÞ = 2:9 − 0:02x − 0:03y
2A 500
1 1 2 1
Nj ðx, yÞ = aj + bj x + cj y = ð−1350 + 30x − 5yÞ = −2:7 + 0:06x − 0:01y
2A 500
1 1
Nk ðx, yÞ = ðak + bk x + ck yÞ = ð400 − 20x + 20yÞ = 0:8 − 0:04x + 0:04y
2A 500
2. The temperature distribution in the element can be expressed as
T ðx, yÞ = Ni ðx, yÞTi + Nj ðx, yÞTj + Nk ðx, yÞTk

= 210ð2:9 − 0:02x − 0:03yÞ + 270ð−2:7 + 0:06x − 0:01yÞ + 250ð0:8 − 0:04x + 0:04yÞ

The temperature at the point (x, y) = (60, 40) in can be found as
T ð60, 40Þ = 210ð2:9 − 1:2 − 1:2Þ + 270ð−2:7 + 3:6 − 0:4Þ + 250ð0:8 − 2:4 + 1:6Þ = 240°F
 CHAPTER 3
Interpolation Models

3.7.3 Three-Dimensional Simplex Element

The three-dimensional simplex element is a flat-faced tetrahedron with four nodes, one at
each corner, as shown in Figure 3.9. Let the nodes be labeled as i, j, k, and l, where i, j, and
k are labeled in a counterclockwise sequence on any face as viewed from the vertex opposite
this face, which is labeled as l. Let the values of the field variable be Φi, Φj, Φk, and Φl and
the global coordinates be (xi, yi, zi), (xj, yj, zj), (xk, yk, zk), and (xl, yl, zl) at nodes i, j, k, and l,
respectively. If the variation of ϕ (x, y, z) is assumed to be linear,
ϕðx, y, zÞ = α1 + α2 x + α3 y + α4 z (3.39)
the nodal conditions ϕ = Φi at (xi, yi, zi), ϕ = Φj at (xj, yj, zj), ϕ = Φk at (xk, yk, zk), and ϕ = Φl
at (xl, yl, zl) produce the system of equations
Φi = α1 + α2 xi + α3 yi + α4 zi
Φj = α1 + α2 xj + α3 yj + α4 zj
(3.40)
Φk = α1 + α2 xk + α3 yk + α4 zk
Φl = α1 + α2 xl + α3 yl + α4 zl
Equation (3.40) can be solved and the coefficients α1, α2, α3, and α4 can be expressed as
1 1 2
α1 = ai Φi + aj Φj + ak Φk + al Φl
6V
1 1 2
α2 = bi Φi + bj Φj + bk Φk + bl Φl
6V (3.41)
α3 = 1 ci Φi + cj Φj + ck Φk + cl Φl
1 2
6V
1 1 2
α4 = di Φi + dj Φj + dk Φk + dl Φl
6V
where V is the volume of the tetrahedron i j k l given by
- -
- 1 xi y i z i -
- - 91
1 - 1 xj yj zj -
- -
V= - (3.42)
6 - 1 xk yk zk -
-
- -
- 1 xl yl zl -

z
(xi, yi, zi)
i

k
(xk, yk, zk)

j
(xj, yj, zj)

0
(x , y , z )

FIGURE 3.9
A Three-Dimensional Simplex Element.
 PART 2
Basic Procedure

- xj yj zj -
- -
- -
ai = -- xk yk zk -- (3.43)
-x yl zl -
l

-1 yj zj -
- -
- -
bi = − -- 1 yk zk -- (3.44)
-1 yl zl -

- xj 1 zj -
- -
- -
ci = − -- xk 1 zk -- (3.45)
-x 1 zl -
l

and
- xj yj 1-
- -
- -
di = − -- xk yk 1 -- (3.46)
-x yl 1-
l

with the other constants defined by cyclic interchange of the subscripts in the order l, i, j,
and k. The signs in front of determinants in Eqs. (3.43) to (3.46) are to be reversed when
generating aj, bj, cj, dj, and al, bl, cl, dl. By substituting Eq. (3.41) into Eq. (3.39), we obtain

ϕðx, y, zÞ = Ni ðx, y, zÞΦi + Nj ðx, y, zÞΦj + Nk ðx, y, zÞΦk + Nl ðx, y, zÞΦl

!ðeÞ (3.47)
= ½N ðx, y, zÞ#Φ
where
3 4
92 ½N ðx, y, zÞ# = Ni ðx, y, zÞ Nj ðx, y, zÞ Nk ðx, y, zÞ Nl ðx, y, zÞ
Ni ðx, y, zÞ = 1 ðai + bi x + ci y + di zÞ
6V
1 1a 2
Nj ðx, y, zÞ = j + b j x + c j y + dj z (3.48)
6V
1
Nk ðx, y, zÞ = ðak + bk x + ck y + dk zÞ
6V
1
Nl ðx, y, zÞ = ðal + bl x + cl y + dl zÞ
6V
and
8 9
>
> Φi >
>
<Φ >
>
!ðeÞ
=
j
Φ = (3.49)
> Φk >
> >
: >
>
Φl
;

PROBLEM 3.5
Consider a tetrahedron element with node numbers i, j, k, and l as shown in Figure 3.9. Noting that
any of the nodes can be considered as the first (local) node, and the next three (local) nodes must
follow a counterclockwise direction as viewed from the first node, enumerate the 12 different ways in
which the node (local) numbers of the element can be assigned.

Solution
If node i is labeled as the first (local) node, the other nodes j, k, and l can be numbered as j, k, l, or k, l,
j or l, j, k to satisfy the counterclockwise requirement (as viewed from node i). Similar considerations
when node j (or k or l) is labeled as the first (local) node lead to the permissible numbering schemes
as indicated in Table 3.1.
 CHAPTER 3
Interpolation Models

TABLE 3.1 Proper Node Numbering Schemes for Tetrahedron Element

Local Node Local Node Local Node Local Node
1 2 3 4
i j k l
i k l j
i l j k
j k i l
j i l k
j l k i
k i j l
k j l i
k l i j
l i k j
l k j i
l j i k

EXAMPLE 3.6
A tetrahedron element with global node numbers 7, 8, 12, and 17 is shown in Figure 3.10.
Determine which of the following (local) numbering sequences satisfy the node numbering
convention.
8, 12, 7, 17; 17, 7, 8, 12; 12, 7, 8, 17

Solution
The numbering scheme 8, 12, 7, 17 (also the scheme 12, 7, 8, 17) does not satisfy the node
numbering convention because the nodes 12, 7, and 17 (7, 8, and 17) correspond to a clockwise
order as seen from node 8 (12). Only the numbering scheme 17, 7, 8, 12 satisfies the node
93
numbering convention because the nodes 7, 8, and 12 correspond to a counterclockwise order as
viewed from node 17.
12

17

FIGURE 3.10
A Tetrahedron Element.

EXAMPLE 3.7
The nodal coordinates and nodal temperatures of a tetrahedron simplex element are given by
Node i: (xi, yi, zi) = (0,0,0) mm, Ti = 100°C
Node j: (xj, yj, zj) = (20,0,0) mm, Tj = 80°C
Node k: (xk, yk, zk) = (0,30,0)mm, Tk = 120°C
Node l: (xl, yl, zl) = (0,0,40) mm, Tl = 50°C
(Continued )