Module 3: Element Properties

Lecture 8: Numerical Integration: One Dimensional

56

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.

ò f (x) dx; ò f (x, h) dxdh; ò 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 = ò f (x) dx from ò 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=ò f (x)dx = ò 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=ò f (x ) dx » å w i f (x i ) » w 1f (x1 ) + w 2f (x 2 ) + .. + w h f (x h ) (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.

3.8.2One- Point Formula

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

ò-1
f(x)dx » w 1f(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 = ò (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 = ò f ( x ) d x = 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

ò f (x) dx » w 1f (x1 ) + w 2f (x 2 ) (3.8.6)

-1

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 2x2 + a 3x3
é 1 ù
Error = ê ò (a 0 + a1x + a 2x2 + a 3x3 ) dxú - éë w1f (x1 ) + w 2f(x2 )ùû
ëê -1 ûú
 58

2
 2a 0 + 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
3
æ2 ö
 (2 - w 1 - w 2 ) a 0 - ( w 1x1 + w 2 x 2 ) a1 + çç - w 1x12 - w 2 x 2 2 ÷÷÷ a 2 - ( w 1x13 + w 2 x 2 3 ) a 3 = 0
çè 3 ø

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 2x 22 =
3
3 3
w1x1 + w 2 x 2 = 0
These nonlinear equations have the unique solution as
w1 = w 2 = 1 x1 = -x2 = -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 points, n Gauss Point Location, xi Weight, wi
1 0.0 2.0
2 ±0.5773502692 (= 1 3) 1.0

3 0.0 0.8888888889 (=8/9)

±0.7745966692 (=  0.6 ) 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:
1æ 2x ö÷
Evaluate I = ò ççe x - 2 ÷dx using one, two and three point gauss Quadrature.
ç
0 è x - 2 ÷ø

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

æ æ x +1ö÷ ö÷
çç æ ö 2 çç ÷
1 ç ççç
x+1 ÷
çç è 2 ÷ø÷ çè 2 ÷÷ø ÷÷÷
I = ò çe - ÷÷dx
-1 ç ÷
2
çç æ x + 1ö
÷ - 2 ÷÷
çç ÷ ÷
çè çè 2 ÷ø ø÷

1 1 æçç 2 4 (x +1) ö÷÷

(x+1)

2 ò-1çè
Or, I = ç e - ÷dx
(x +1) - 8 ÷÷ø
2

Using one point gauss Quadrature:

w1 = 2, x1 = 0 and
I » 2f (0)
æ1 æ 4 öö
Or I » 2 çç ççe0.5 + ÷÷÷÷÷÷ = 2.22015
ç
èç 2 è 7 øø
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
1  0.774596669
w2  0.888888889
2  0.000000000
w3  0.555555556
3  0.774596669
and I = 2.41024
This may be compared with the exact solution as Iexact = 2.41193