CEE570 / CSE 551 Class #31

Previous Class

Isoparametric elements (cont’d)

• Mapping of elements
• 4-node bilinear quadrilateral
• Jacobian matrix
• Understanding of [J] and J
• Examples
• Stiffness matrix
• Q8, T3, T6

This Class
Numerical integration: Gauss quadrature
• Introduction to numerical integration and Gauss quadrature
• Gauss quadrature in one dimension (1D)
• Derivation of Gauss points and weights
(one point, two-point, and n point quadrature)
• FEM example
• Gauss quadrature in 2D
• Exactness of Gauss quadrature in 2D
• Full, reduced and recommended integration in 2D
• Minimum integration order
• Gauss quadrature in 3D
2

1
 Introduction
Numerical integration:
• Evaluating the integrand at specific points
• Multiplying each result by an appropriate weighting factor
• Summing up the results f
x

x2

∫ fdx ≈ W f
x1
1 1 + W2 f 2 + ... + Wn f n
x
x1 x2

The choice of sampling points and weights

defines different quadrature schemes.
3

Gauss quadrature (a numerical integration scheme)

• Formulated to compute exact integration for

polynomials

• Use fewer sampling points compared to other

integration schemes

• Is the most used numerical integration scheme

to obtain element stiffness matrix in FEM

2
 One dimension Gauss quadrature
Transform of integration limit: from x1 and x2 to -1 and 1
so that the formulas will be generalized

x = 12 (1 − ξ ) x1 + 12 (1 + ξ ) x2
x2 1

∫ fdx
x1
∫ φ dξ
−1

φ = f ( x(ξ ) ) dx / d ξ

Transformation Jacobian J = dx / dξ = ( x2 − x1 ) / 2
From now on, we need to work only with -1 to 1 limits
5

One dimension Gauss quadrature

Assume the integrand is a polynomial

φ (ξ ) = a0 + a1ξ + a2ξ 2 + ... + anξ n

Exact integration for each terms:

1 1 1

∫ξ dξ = 2 / 5
4
∫ 1.dξ = 2 ∫ξ dξ = 2 / 3
2

−1 −1 −1
1 1

∫ ξ dξ = 0
−1
∫ξ
3
dξ = 0 ...
−1

Remarks: integration of odd order terms are always zero

3
 Deriving sampling points and weights
Based on the condition that all terms need to be
integrated exactly

Example First order polynomial

f (ξ ) = a0 + a1ξ
1 n

∫ a0 dξ = 2a0 = ∑Wi a0ξi W and ξ

0
i i
−1 i =1 are weights and coordinates of
Gauss points
1 n

∫ a1ξ dξ = 0 = ∑Wiξi
−1 i =1
There are two equations;
We need at least two unknowns for the solution to exist
7 At least 1 Gauss point needed (n=1)

One Gauss point rule (n=1)

Use 1 Gauss point: two unknown W1 and ξ1
1

∫ a d ξ = 2a
−1
0 0 = W1a0 ⇒ W1 = 2
f
1

∫ a ξ dξ = 0 = W ξ
−1
1 1 1 ⇒ ξ1 = 0

Answer : one Gauss point 0

ξ
−1 1
Geometric interpretation

ξ1 = 0 W1 = 2

4
 Two Gauss point rule (n=2)
Example: Third order polynomial

φ (ξ ) = a0 + a1ξ + a2ξ 2 + a3ξ 3

1 n 1 n

∫ a0 dξ = 2a0 = ∑Wi a0
−1 i =1
∫ a1ξ dξ = 0 = ∑Wi a1ξi
−1 i =1
1 n

∫ a3ξ dξ = 0 = ∑Wi a3ξi

1 n 3 3

∫aξ 2
2
dξ = 2 / 3a2 = ∑ W a ξi 2 i
2
−1 i =1
−1 i =1

There are 4 equations;

Need at least 4 unknowns for the solution to exist
So, at least 2 Gauss points needed

Two Gauss point rule (n=2)

Unknown: W1 , ξ1 , W2 , ξ 2
f
W1 + W2 = 2
W1ξ1 + W2ξ 2 = 0
W1ξ12 + W2ξ 22 = 2 / 3 x
−1 −1 0 1 1
W1ξ13 + W2ξ 23 = 0 3 3

W1 = 1
Solving the W2 = 1
equations, we
ξ1 = 1/ 3
have:
ξ 2 = − 1/ 3
10

5
 Three Gauss point rule (n=3)
Fifth order polynomial

φ (ξ ) = a0 + a1ξ + a2ξ 2 + a3ξ 3 + a4ξ 4 + a5ξ 5

1 n 1 n

∫ a0 dξ = 2a0 = ∑Wi a0
i =1
∫ a1ξ dξ = 0 = ∑Wi a1ξi
−1 i =1
−1
1 n 1 n

∫ a2ξ dξ = 2 / 3a2 = ∑Wi a2ξi ∫ a3ξ dξ = 0 = ∑Wi a3ξi

2 2 3 3

−1 i =1 −1 i =1

1 n 1 n

∫ a4ξ dξ = 2 / 5a4 = ∑Wi a4ξi ∫ a5ξ dξ = 0 = ∑Wi a5ξi

4 4 5 5

−1 i =1 −1 i =1

There are 6 equations;

Need at least 6 unknowns for the solution to exist
11 So, at least 3 Gauss points needed

Three Gauss point rule (n=3)

Unknown: W1 , ξ1 , W2 , ξ 2 ,W3 , ξ3

W1 + W2 + W3 = 2 W1ξ13 + W2ξ 23 + W3ξ33 = 0

W1ξ1 + W2ξ 2 + W3ξ3 = 0 W1ξ14 + W2ξ 24 + W3ξ34 = 2 / 5
W1ξ12 + W2ξ 22 + W3ξ32 = 2 / 3 W1ξ15 + W2ξ 25 + W3ξ35 = 0

W1 = 5 / 9 ξ1 = − 3 / 5
Solving the
equations, we W2 = 8 / 9 ξ2 = 0
have: W3 = 5 / 9 ξ3 = 3 / 5

12

6
 Remarks on the derivation
1) Note that the integration for odd order terms are
always zero, e.g.

W1ξ1 + W2ξ 2 = 0 This condition is equivalent to the

symmetry of the Gauss points and
W1ξ13 + W2ξ 23 = 0 weights
...

2) In general, polynomials of order 2n-1 is integrated

exactly by n Gauss point rule

Bonus HW: derive the four point rule

Bonus HW: derive the five point rule

13

FEM example
Beam element (X=ξ l) 2 4
1 3

[K ] = ∫ [ B]T [ D][ B ]dx

2l

length N1

[ B] = l12 [ N1,ξξ N 2,ξξ N 3,ξξ N 4,ξξ ]

N2
[ B] Polynomial order 1
N3
1

[ K ] = ∫ [ B]T [ EI ][ B]ldξ N4
−1
Polynomial order 2 Recall: cubic shape function
(use two point Gauss quadrature)

[ K ] = (1)[ B]T [ EI ][ B]l 1/ 3

+ (1)[ B]T [ EI ][ B]l
− 1/ 3
14

7
 Gauss quadrature in two dimensions
1 1 1
  η
I= ∫ ∫ φ (ξ ,η )dξ dη ≈ ∫ ∑Wiφ (ξi ,η )dη
−1 −1 −1  i 
  1 1
= ∑ W j  ∑ Wiφ (ξi ,η j )  = ∑∑ W jWiφ (ξi ,η j )
1/ 3

j  i  j i
ξ
−1/ 3 1 1
η
−1/ 3 1/ 3

0.6
25
81
40
81
25
81
2 point rule
40 64 40
81 81 81
ξ
− 0.6
25
81
40
81
25
81 m x n rule possible
but not
− 0.6 0.6 recommended

15 3 point rule

Exactness of Gauss quadrature in 2D

1 1
I= ∫ ∫ξ η
l m
dξdη
−1 -1 Gauss rule
for exact
Integration
Constant (l = m = 0) 1
Linear (l + m = 1) ξ η
One point
2
Quadratic (l + m = 2) ξ ξη η2
Cubic (l + m = 3) ξ3 ξ 2η ξη2 η3 2x2
4 3 2 2
Quartic (l + m = 4) ξ ξ η ξ η ξη3 η4
ξ 3 η2 ξ 2 η3
ξ 3 η3

16

8
 Full, reduced and recommended integration in 2D

Table 6.8-1 Cook at al. 2nd Ed. pp226

Element Full Reduced Recommended

Q4 2x2 1 2x2

Q8 3x3 2x2 2x2

Q9 3x3 2x2 2x2 (|J|=const)

3x3 (|J| is not
const)

Q12 4x4 3x3 3x3

17

Minimum integration order

In order to pass the patch test, shape

function should be able to represent
constant strain.
{ε } = const
T
Ue = 1
2 ∫ {ε } {σ } dV
Ve
e = 1
2 {ε }{σ } ∫ dVe = 12 U o ∫ dVe
Ve Ve

U e = 12 U o ∫ J d ξ dη d ς
Ve

Minimum integration order should be

able to compute exactly the volume of
element.
18

9
 Three dimensions

1 1 1
I= ∫ ∫ ∫ φ (ξ ,η , ς )dξ dη dς
−1 −1 −1

≈ ∑∑∑ WW
i jWk φ (ξi ,η j , ς k )
i j k

Computational cost of 3x3x3 Gauss

3x3x3 Gauss quadrature points in 3D
for the brick element:

27 (Gauss points) x 24x24 = 15,552 (function evaluations per element)

19

Next class

Validity of isoparametric element

• Ability to represent rigid body motions
• Generalized Iso-P formulation: GIF
• Graded element and homogeneous element

20

10