Lecture -5

Numerical Integration

FEM Dr. Riessom 1

 Axially loaded Bar
y A(x) = cross section at x
b(x) = body force
F distribution (force per unit
x
x
length)
x=0 x=L E(x) = Young’s modulus
x1 x2
For each element 1 2

Element stiffness matrix

x2
d1x d2x
kij   Bi EB j Adx
x2
k   B EB Adx
T
x1 x1

FEM Dr. Riessom 2

 Linear Finite Element


x1
x2
B EB Adx 
T 1
 x 2  x1 
2
 1 1 x2
 1 1  x1 AEdx 
 

x1
x2
AEdx  x  x 
2
1
1
2
 1 1
 1 1 
 

Element nodal load vector

x2 x2
f b   N b dx f b i   N i b dx
T
x1 x1

How do we compute these integrals using a computer?

Any integral from x1 to x2 can be transformed to the
following integral on (-1, 1)

FEM Dr. Riessom 3

 Numerical Integration
1
I   f ( ) d
1

Use the following change of variables

1 1 
x x1  x2
2 2
Goal: Obtain a good approximate value of this integral
1. Newton-Cotes Schemes (trapezoidal rule, Simpson’s rule,
etc)
2. Gauss Integration Schemes
NOTE: Integration schemes in 1D are referred to as
“quadrature rules”
FEM Dr. Riessom 4
 Trapezoidal Rule
Approximate the function f(x) by a straight line g(x) that
passes through the end points and integrate the straight
line
g f1

f1
f


-1 1
1 1 
g ( )  f (1)  f (1)
2 2
1 1
I   f ( ) d   g ( ) d  f (1)  f (1)
1 1

FEM Dr. Riessom 5

 Trapezoidal Rule Continued…
•Requires the function f(x) to be evaluated at 2 points
(-1, 1)
• Constants and linear functions are exactly integrated
• Not good for quadratic and higher order polynomials
How can I make this better?

Simpson’s rule
Approximate the function f(x) by a parabola g(x) that
passes through the end points and through f(0) and
integrate the parabola

FEM Dr. Riessom 6

 Simpson’s Rule
g f1
f

f1

f(0

-1 1
 (  1)  (1   )
g ( )  f (1)  (1   )(1   ) f (0)  f (1)
2 2
1 1 1 4 1
I   f ( ) d   g ( ) d  f (1)  f (0)  f (1)
1 1 3 3 3
Requires the function f(x) to be evaluated at 3 points (-1,0, 1)
Constants, linear functions and parabolas are exactly integrated
Not good for cubic and higher order polynomials
FEM Dr. Riessom 7
 Simpson’s Rule Continued…
M
I   f ( ) d  Wi f ( i )
1

1
i 1

Weight Integration point

Trapezoidal rule:
W1  1 1  1 Accurate for polynomial of
M=2 degree at most 1 (=M-1)
W2  1  2  1

Simpson’s rule: W1  1 / 3 1  1 Accurate for polynomial of

M=3 W2  4 / 3  2  0 degree at most 2 (=M-1)
W2  1 / 3  2  1

Notice that both the integration formulas had the general form

FEM Dr. Riessom 8

 Newton-Cotes
Generalization of these two integration rules
Divide the interval (-1,1) into M-1 equal intervals using M
points
Pass a polynomial of degree M-1 through these M points (the
value of this polynomial will be equal to the value of the
function at these M-1 points)
Integrate this polynomial to obtain an approximate value of the
integral With ‘M’ points we may
f1
f integrate a polynomial of
degree ‘M-1’ exactly.
f1 With ‘M’ integration points
g and ‘M’ weights, I should be
able to integrate a polynomial

-1 1 of degree 2M-1 exactly!!
FEM Dr. Riessom 9
 Gaussian Quadrature
To evaluate the integral:

Where y = y(x), we might choose (sample or evaluate) y at

the midpoint y(0) = y1 and multiply by the length of the
interval, as shown below to arrive at I = 2y1, a result that is
exact if the curve happens to be a straight line.
M
I   f ( ) d  Wi f ( i )
1

1
i 1

Weight Integration point

FEM Dr. Riessom 10

Generalization of the formula leads to:

That is, to approximate the integral, we evaluate the

function at several sampling points n, multiply each
value yi by the appropriate weight Wi , and add the terms.

Gauss's method chooses the sampling points so that for a

given number of points, the best possible accuracy is
obtained.

Sampling points are located symmetrically with respect to

the centre of the interval.
FEM Dr. Riessom 11
In general, Gaussian quadrature using n points (Gauss points)
is exact if the integrand is a polynomial of degree 2n - 1 or
less.
In using n points, we effectively replace the given function y =
f(x) by a polynomial of degree 2n- 1.
The accuracy of the numerical integration depends on how
well the polynomial fits the given curve.
If the function f(x) is not a polynomial, Gaussian quadrature
is inexact, but it becomes more accurate as more Gauss
points are used.
Also, it is important to understand that the ratio of two
polynomials is, in general, not a polynomial; therefore,
Gaussian quadrature will not yield exact integration of the
ratio.
FEM Dr. Riessom 12
To illustrate the derivation of a two-point (n = 2) consider:

There are four unknown parameters to determine: W1, W2,

x1, and x2.
Therefore, we assume a cubic function for y as follows:

In general, with four parameters in the two-point formula,

we would expect the Gauss formula to exactly predict the
area under the curve.

FEM Dr. Riessom 13

However, we will assume, based on Gauss's method, that W1 =
W2 and that x1 = x2 as we use two symmetrically located
Gauss points at x = ±a with equal weights.
The area predicted by Gauss's formula is

If the error, e = A - AG, is to vanish for any C0 and C2, we

must have, in the error expression:

Now W = 1 and a = 0.5773 ... are the Wi’s and ai’s (xi’s) for
the two-point Gaussian quadrature as given in the table.
FEM Dr. Riessom 14
 Example
Use three-point Gaussian Quadrature evaluate the integrals:

FEM Dr. Riessom 15

 Gauss Integration
m Gauss Point j Gauss Weight wj Accuracy order n

1 0 2 1

2 -1/3, 1/3 1, 1 3

3 -0.6, 0, 0.6 5/9, 8/9, 5/9 5

4 -0.861136, -0.339981, 0.347855, 0.652145, 7

0.339981, 0.861136 0.652145, 0.347855
5 -0.906180, -0.538469, 0, 0.236927, 0.478629, 9
0.538469, 0.906180 0.568889, 0.478629,
0.236927
6 -0.932470, -0.661209, 0.171324, 0.360762, 11
-0.238619, 0.238619, 0.467914, 0.467914,
0.661209, 0.932470 0.360762, 0.171324

FEM Dr. Riessom 16

In two dimensions, we obtain the quadrature formula by
integrating first with respect to one coordinate and then
with respect to the other as

In three dimensions, we obtain the quadrature formula by

integrating first with respect to one coordinate and then
with respect to the other two as

FEM Dr. Riessom 17

Example: M=1 (Midpoint Qudrature)
1
I   f ( ) d  W1 f (1 )
1

How can we choose W1 and x1 so that we may integrate

a (2M-1=1) linear polynomial exactly?
f ( )  a0  a1
1
1
f ( ) d  2a0

But we want
1

1
f ( ) d  W1 f (1 )  a0W1  a0W1

FEM Dr. Riessom 18

 Hence, we obtain the identity
2a0  a0W1  a1W11

For this to hold for arbitrary a0 and a1 we need to satisfy

2 conditions
Condition 1 : W1  2
Condition 2 : W11  0
i.e.,
W1  2; 1  0

FEM Dr. Riessom 19

For M=1 1
I   f ( ) d  2 f (0)
1

f
f0

g


-1 1
Midpoint quadrature rule:
• Only one evaluation of f(x) is required at the midpoint
of the interval.
• Scheme is accurate for constants and linear
polynomials (compare with Trapezoidal rule)
FEM Dr. Riessom 20
 Example: M=2
1
I   f ( ) d  W1 f (1 )  W2 f ( 2 )
1

How can we choose W1 ,W2 x1 and x2 so that we may

integrate a polynomial of degree (2M-1=4-1=3)
exactly?
f ( )  a0  a1  a2 2  a3 3
1 2
1 f ( ) d  2a0  3 a2
But we want
1

1
f ( ) d  W1 f (1 )  W2 f ( 2 )

 2 2
 
 a0 W1  W2   a1 W11  W2 2   a2 W11  W2 2  a3 W11  W2 2
3 3

FEM Dr. Riessom 21
 Hence, we obtain the 4 conditions to determine the 4
unknowns (W1 ,W2 x1 and x2 )

Condition 1 : W1  W2  2
Condition 2 : W11  W2 2  0
2
Condition 3 : W11  W2 2 
2 2

3
Condition 4 : W11  W2 2 0
3 3

Check that the following is the solution

W1  W2  1
1 1
1   ; 2 
3 3
FEM Dr. Riessom 22
 For M=2
1 1 1
I   f ( ) d  f ( ) f( )
1
3 3
1 1
f ( ) f( )
3 3
f


-1
* * 1
Only two evaluations of f(x) is required.
Scheme is accurate for polynomials of degree at most 3
(compare with Simpson’s rule)

FEM Dr. Riessom 23

 Stiffness Matrix by Gaussian
Quadrature
For the two-dimensional element, we have shown in previous
chapters that

Where, in general, the integrand is a function of x and y and

nodal coordinate values.
For a quadrilateral element can be evaluated in terms of a local
set of coordinates s-t, with limits from -1 to 1within the
element.

Each coefficient of the integrand [B]T [D] [B] |[J]|

evaluated by numerical integration in the same manner
as f(s, t) was integrated.
FEM Dr. Riessom 24
 In FEM we always use Gauss
quadrature 2
1
Linear Element
Stiffness matrix   1  1
 1

 1
T 1  1 1 1
k   B EB Ad   
4  1 1  1
AEd   1
1  1 1
1 AEd 4 1 1 
Nodal load vector
1
1
f b   N b d
T f b i   N i b d
1
1

Usually a 2-point Gauss integration is used. Note that if

A, E and b are complex functions of x, they will not be
accurately integrated
FEM Dr. Riessom 25
 Quadratic Element
1 2 3
Nodal shape functions
   1  0  1
N1 ( )    1
2
N 2 ( )  1   2 
You should be able to derive these!

N 3 ( )    1
2
Stiffness matrix
1 1
k   B EB Ad  AE  B B d
T T
1 1
Assuming E and A are constants

d N  dN1 dN 2 dN3   1
 2  1
1
B      2  1  2
d  d d d   2 2 
FEM Dr. Riessom 26
 Quadratic Element Continued…
1 1
k   B EB Ad  AE  B B d
T T
1 1

1 
   1/ 2 2 2   1/ 2   
 2  1/ 4 

 AE   2   1/ 2  4 2
2   1/ 2  d
1  
 
   1/ 4
2
 2   1/ 2    1/ 2  
2

Need to exactly integrate quadratic terms. Hence we

need a 2-point Gauss quadrature scheme.. Why?

FEM Dr. Riessom 27

 Example
f


f  1/ 3  
f 1/ 3 


-1  1 / 3 1/ 3 1

A 2-point Gauss quadrature rule

1 1 1

1
f ( ) d  f (
3
)  f (
3
)

is exact for a polynomial of degree 3 or less

FEM Dr. Riessom 28

 1 1 
2D
t
Square Domain 1 1 
 
  
1 1 ,
,  3 3 
 3 3 

1
1 1
s I   f (s, t ) dsdt
1 1
1

 1 1   1 1 
  ,   , 
 3 3  3 3
1 1
I   f ( s, t ) dsdt
1 1

M 
    W j f ( s, t j )  ds Using 1D Gauss rule to integrate along ‘t’
1

1
 j 1 
M M
  WiW j f ( si , t j ) Using 1D Gauss rule to integrate along ‘s’
i 1 j 1
M M
  Wij f ( si , t j ) Where Wij =Wi Wj
i 1 j 1
FEM Dr. Riessom 29
 The Rule
M M
f ( s, t ) dsdt  Wij f ( si , t j )
1 1
I  
1 1
i 1 j 1

Uses M2 integration points on a non-uniform grid

inside the parent element and is exact for a polynomial
of degree (2M-1) i.e.,
exact M M

Wij si t j
1 1
 
   
s t dsdt  for     2M  1
1 1
i 1 j 1

A M2 –point rule is exact for a complete polynomial of

degree (2M-1)

FEM Dr. Riessom 30

 CASE II: M=2 (2x2 GQ rule)
t  1 1 
 1 1  1 1  , 
  ,   3 3
 3 3

1
s
1

 1 1   1 1 
  ,   , 
 3 3   3 3
2 2
I  Wij f ( si , t j )
i 1 j 1

1 1 1 1 1 1 1 1
 f( , )  f ( , )  f ( , ) f( , )
3 3 3 3 3 3 3 3
is exact for a product of two cubic polynomials
FEM Dr. Riessom 31
 CASE III: M=3 (3x3 GQ rule)
t
1 1
64
3 6 2 W1  ,
1 3
81
7 1 9 5 25
3
s W2  W3  W4  W5 
81
1 4 8 5 5
40
W6  W7  W8  W9 
3 3 81
5 5
3 3
f ( s, t ) dsdt  Wij f ( si , t j )
1 1
I  
1 1
i 1 j 1

is exact for a product of two 1D polynomials of degree 5

FEM Dr. Riessom 32
 Examples
If f(s,t)=1
1 1
I   f ( s, t ) dsdt  4
1 1

A 1-point GQ scheme is sufficient

If f(s,t)=s
1 1
I   f ( s, t ) dsdt  0
1 1

A 1-point GQ scheme is sufficient

If f(s,t)=s2t2
1 1 4
I    f ( s, t ) dsdt 
1 1 9
A 3x3 GQ scheme is sufficient
FEM Dr. Riessom 33
 2D Gauss quadrature for
Triangular Domains
Remember that the parent element is a right angled
triangle with unit sides
t The type of integral encountered
1 1t
I   f ( s, t ) dsdt
1 t 0 s 0

t 1 1t
s=1-t t s
I   f ( s, t ) dsdt
t 0 s 0
M
  WIP f IP
1

IP 1

FEM Dr. Riessom 34