Introduction to Gauss Quadrature
The whole procedure of performing integration over the domain is called Quadrature (or Numerical Integration)
We know that for a 1-D case, Integration of a function is given as𝐼 𝑓 = න 𝑓 𝑥 ⅆ𝑥
a b
𝑎
1 1
Analytically we know that given 𝑓 𝑥 = 𝑥 The integral is given as 𝐼 𝑓 = 𝑥 𝑓 ⅆ𝑥 = 𝑥 ⅆ𝑥 = 0.5
0 0
𝑏
𝑛
Numerically, the integral is computed as 𝐼(𝑓) = 𝑓 𝑥 ⅆ𝑥 ≈ 𝑤𝑖 𝑓 ξ𝑖
𝑖=0
𝑎
a ξ1 ξ2 ξ3 ξ4 b = 𝑤1 𝑓 ξ1 + 𝑤2 𝑓 ξ2 + 𝑤3 𝑓 ξ3 + 𝑤4 𝑓 ξ4 ……
WEIGHTS SAMPLING POINTS
𝑏
𝑛
Given 𝐼(𝑓) = 𝑓 𝑥 ⅆ𝑥 ≈ 𝑤𝑖 𝑓 ξ𝑖 a ξ1 ξ2 ξ3 ξ4 b
𝑖=0
𝑎
If n=1
𝐼(𝑓) ≈ 𝑤1 𝑓 ξ1
Given that the sample point is a mid point, the above expression will convert to mid point method
𝑎+𝑏
𝑀 𝑓 = (𝑏 − 𝑎) ∗ 𝑓 a ξ b
2
Given that the sample points are end points, we obtain Trapezoidal method
𝑏−𝑎 𝑏−𝑎 𝑏−𝑎
𝑇 𝑓 = ∗ 𝑓 𝑎 +𝑓 𝑏 = ∗𝑓 𝑎 + ∗ 𝑓(𝑏)
2 2 2
a b
𝑤1 = (𝑏 − 𝑎)/2 𝑤2 = (𝑏 − 𝑎)/2 ξ2 = 𝑏 ξ1 = 𝑎
Given that the sample points are both end points as well as middle points, we obtain Simpson’s method
𝑏−𝑎 𝑎+𝑏
S 𝑓 = ∗ 𝑓 𝑎 + 4𝑓 +𝑓 𝑏
6 2 a 𝝃 = (𝒂 + 𝒃)/𝟐 b
𝑤1 = (𝑏 − 𝑎)/6 𝑤2 = 4(𝑏 − 𝑎)/6 𝑤3 = (𝑏 − 𝑎)/6
ξ1 = 𝑎 ξ2 = (𝑎 + 𝑏)/2 ξ3 = 𝑏
So, finally we have
𝑎+𝑏
𝑀 𝑓 = (𝑏 − 𝑎) ∗ 𝑓 + 0 *𝑓 𝑎 + 0 ∗ 𝑓 𝑏 𝑤𝑖 = (𝑏 − 𝑎) = 𝐿𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑡ℎ𝑒 ⅆ𝑜𝑚𝑎𝑖𝑛
2
𝑏−𝑎 𝑏−𝑎 𝑏−𝑎 𝑏+𝑎 𝑤𝑖 = (𝑏 − 𝑎)
𝑇 𝑓 = ∗ 𝑓 𝑎 +𝑓 𝑏 = ∗𝑓 𝑎 + ∗ 𝑓 𝑏 + 0 ∗ 𝑓( )
2 2 2 2
𝑏−𝑎 𝑎+𝑏
S 𝑓 = ∗ 𝑓 𝑎 + 4𝑓 +𝑓 𝑏 𝑤𝑖 = (𝑏 − 𝑎)
6 2
IN ALL THREE CASES INTEGRALS ARE PERFORMED OVER THE LENGTH
In case, if integral is performed over an Area
𝐼(𝑓) = 𝑓 𝑥 ⅆ𝐴 ≈ 𝑤𝑖 𝑓 ξ𝑖 𝑤𝑖 = 𝐴𝑟𝑒𝑎 𝑜𝑓 𝑡ℎ𝑒 ⅆ𝑜𝑚𝑎𝑖𝑛
𝑖=0
𝐴𝑟𝑒𝑎
In case, if integral is performed over a Volume
𝐼(𝑓) = 𝑓 𝑥 ⅆ𝑉 ≈ 𝑤𝑖 𝑓 ξ𝑖 𝑤𝑖 = 𝑉𝑜𝑙𝑢𝑚𝑒 𝑜𝑓 𝑡ℎ𝑒 ⅆ𝑜𝑚𝑎𝑖𝑛
𝑖=0
𝑉𝑜𝑙𝑢𝑚𝑒
QUESTION: WHICH METHOD IS MORE ACCURATE AMONG SIMPSON, TRAPEZOIDAL AND MID POINT ??
LET US UNDERSTAND THAT THROUGH AN EXAMPLE
𝑎+𝑏
Using mid point method 𝑀 𝑓 = (𝑏 − 𝑎) ∗ 𝑓
2
Using Trapezoidal method
𝑏−𝑎 𝑎+𝑏
Using Simpson’s method S 𝑓 = ∗ 𝑓 𝑎 + 4𝑓 +𝑓 𝑏
6 2
Error in TRAPEZOIDAL is twice the error obtained in MID POINT
Error is least in SIMPSON’S METHOD
AMONG THREE, SIMPSON’S METHOD IS THE MOST ACCURATE
IN THREE METHODS, WE FIXED THE SAMPLING POINTS AND VARIED THE WEIGHTS FIXED SAMPLING POINTS
𝑏
𝐼 𝑓 = න 𝑓 𝑥 ⅆ𝑥 = 𝑤1 𝑓 ξ1 + 𝑤2 𝑓 ξ2 + 𝑤3 𝑓 ξ3 + 𝑤4 𝑓 ξ4 ……
𝑎
VARY WEIGHTS
HOW ABOUT VARYING BOTH SAMPLING POINTS AS WELL AS THE WEIGHTS FOR OBTAINING MORE ACCURACY
THE ABOVE IDEA WAS PROPOSED BY THE CARL GAUSS AND INTRODUCED AN INTEGRATION METHOD
CALLED GAUSS QUADRATURE
GAUSS LEGENDRE QUADRATURE
This method has proved most useful in finite element work.
In case of Gauss Quadrature, the integral is considered in the following form (for 1D)
Consider the n -point approximation
GAUSS LEGENDRE SAMPLING POINTS
The idea behind Gauss–Legendre quadrature is to select the n Gauss points and n weights such that the above
equation provides an exact answer for polynomials f(ξ) of as large a degree as possible
+1
We have 𝑛
𝐼(𝑓) = 𝑓 ξ ⅆξ ≈ 𝑤𝑖 𝑓 ξ𝑖
𝑖=0
−1
Let us consider the above formula with n = 1, we get
Assuming and substituting in the above expression, we get
On the LHS of the equation, we have
On the RHS of the equation, we have
Error
MINIMIZING THE ERROR
Solving the above equation, we get
Hence, we have
MID POINT
The above seen to be the familiar midpoint rule
𝑎+𝑏
Mid point method 𝑀 𝑓 = (𝑏 − 𝑎) ∗ 𝑓
2
Consider the formula with n = 2,
We have four parameters to choose: w1, w2, ξ1, and ξ2
We can expect Equation to be exact for a cubic polynomial
Thus, choosing yields
Considering ai ≠ 0, we obtain
With n = 2, We can accurately calculate
Polynomial of order 3
Evaluate
+1
𝑛
𝐼(𝑓) = 𝑓 ξ ⅆξ ≈ 𝑤𝑖 𝑓 ξ𝑖
𝑖=0
−1
This may be compared with the exact solution
We know that with ‘n’ gauss points, we can accurately integrate a polynomial of order ‘2n-1’
2n-1 = 5 n=3
Hence, for the accurate integration of the above polynomial, we need at least 3 Gauss Points
Using two points, we obtain integral value as
Using three points, we obtain integral value as
Using four points, we obtain integral value as
We already discussed about the 1D form of Gauss Quadrature i.e.
w1=01 w2=01
ξ=-1 −1 ξ=0 1 ξ=1
ξ1= ξ2=
3 3
w1=0.55 w3=0.88 w2=0.55
ξ=-1 ξ=1
-0.7745 ξ=0 0.7745
However, we know that the integral in finite element analysis is performed over area as well as volume such as
Therefore, Gauss Quadrature integration is also extended in two and three dimensions
K =
1D form of Gauss Quadrature i.e.
The extension of Gaussian quadrature to two-dimensional integrals of the form
wi is the weight corresponding to the Gauss Point ξi
wj is the weight corresponding to the Gauss Point ηj
Suppose if we take two Gaussian points along the ξ and η direction, then above expression will be expanded as
𝑰 = σ𝒏𝒊=𝟏 σ𝒏𝒋=𝟏 𝒘𝒊𝒘𝒋𝒇 𝝃, 𝜼 = σ𝟐𝒊=𝟏 σ𝟐𝒋=𝟏 𝒘𝒊𝒘𝒋𝒇 𝝃𝒊, 𝜼𝒊 = 𝒘𝟏𝒘𝟐𝒇 𝝃𝟏, 𝜼𝟐 + 𝒘𝟐𝒘𝟏𝒇 𝝃𝟐, 𝜼𝟏 + 𝒘𝟏 𝟐 𝒇 𝝃𝟏, 𝜼𝟏 + 𝒘𝟐 𝟐 𝒇 𝝃𝟐, 𝜼𝟐
𝒇 𝝃, 𝜼 =ξ2η2
I = 1*1* 1Τ
3 ∗ 1Τ3 + 1∗1∗ 1Τ3 ∗ 1Τ3 + 1∗1∗(1Τ3 ∗ 1Τ3)+ 1*1* 1Τ3 ∗ 1Τ3 = 4/9
Considering three sample points in ξ and η direction
w1 = 0.555, for ξ1 = 0.77459
w2 = 0.555, for ξ2 = - 0.77459
−0.77459,0.77459 0,0.77459 0.77459,0.77459
w = 0.3025 w = 0.4937 w = 0.3025 w3 = 0.888, for ξ3 = 0.0
w1 = 0.555, for η1 = 0.77459
0, −0.77459 0,0 0,0.77459 w2 = 0.555, for η2 = - 0.77459
w = 0.4937 w = 0.7899 w = 0.4937
w3 = 0.888, for η3 = 0.0
−0.77459, −0.77459 0, −0.77459 0.77459, −0.77459
w = 0.3025 w = 0.4937 w = 0.3025
