Numerical Integration

(Difference between revisions)
 Revision as of 18:18, 5 November 2009 (view source)JMora (Talk | contribs) (→Numerical Integration for Isoparametric Triangular Domains)← Older edit Latest revision as of 18:41, 11 November 2009 (view source)JMora (Talk | contribs) (→Numerical Integration for Isoparametric Triangular Domains) (8 intermediate revisions by one user not shown) Line 144: Line 144: − A general integral expression form for two dimensional domains can be written in terms of the area coordinates and, therefore, computed by using the Gauss quadrature: + A general integral expression form for two dimensional domains can be written in terms of the [[Two-dimensional_Shape_Functions#Areal_Coordinates | '''area coordinates''']] and, therefore, computed by using the Gauss quadrature: − $\int_0^1 \int_0^{1-L_3} f(L_1,L_2,L_3) dL_2 dL_3 = \sum_{p=1}^{n_p} f(L_{1_p},L_{2_p},L_{3_p}) W_p$ + ::$\int_0^1 \int_0^{1-L_3} f(L_1,L_2,L_3) dL_2 dL_3 = \sum_{p=1}^{n_p} f(L_{1_p},L_{2_p},L_{3_p}) W_p$ with: with: * $n_p \,$   the number of integration points; * $n_p \,$   the number of integration points; − * $L_{1_p}, L_{2_p} & L_{3_p} \,$   the value of the area coordinates; + * $L_{1_p}, L_{2_p}, L_{3_p} \,$   the value of the area coordinates; * $W_p \,$   the weight in the integration point '''''p'''''; * $W_p \,$   the weight in the integration point '''''p'''''; + The following table and picture shows the integration points and weights for triangles obtained from the Gaussian quadrature ('''precision''' means the degree of polynomial for exact integration): − :[[Image:IntegrationPointsTriangularElement.jpg]] + {| border="1" cellpadding="5" cellspacing="0" class="wikitable" style="margin:auto; background:white;" + ! Number of points, ''n'' !! ''precision'' !! Points !! '''L1''' !! '''L2''' !! '''L3''' !! ''W''''i'' + |- align="center" + | 1 || Linear || ''a'' || 1/3 || 1/3 || 1/3 || 1/2 + |- align="center" + | rowspan="3" | 3 || rowspan="3" | Quadratic || ''a'' || 1/2 || 1/2 || 0 || 1/6 + |- align="center" + | ''b'' || 0 || 1/2 || 1/2 || 1/6 + |- align="center" + | ''c'' || 1/2 || 0 || 1/2 || 1/6 + |- align="center" + | rowspan="4" | 4 || rowspan="4" | Cubic || ''a'' || 1/3 || 1/3 || 1/3 || -9/32 + |- align="center" + | ''b'' || 0.6 || 0.2 || 0.2 || 25/96 + |- align="center" + | ''c'' || 0.2 || 0.6 || 0.2 || 25/96 + |- align="center" + | ''d'' || 0.2 || 0.2 || 0.6 || 25/96 + |- align="center" + | rowspan="7" | 7 || rowspan="7" | Quartic || ''a'' || 0 || 0 || 1 || 1/40 + |- align="center" + | ''b'' || 1/2 || 0 || 1/2 || 1/15 + |- align="center" + | ''c'' || 1 || 0 || 0 || 1/40 + |- align="center" + | ''d'' || 1/2 || 1/2 || 0 || 1/15 + |- align="center" + | ''e'' || 0 || 1 || 0 || 1/40 + |- align="center" + | ''f'' || 0 || 1/2 || 1/2 || 1/15 + |- align="center" + | ''g'' || 1/3 || 1/3 || 1/3 || 9/40 + |} + + + ::[[Image:IntegrationPointsTriangularElement.jpg]] + + + Note that the weight values has been normalised in order to sum 1/2 to maintain the exact value for the element area. + + + Therefore: + + :$A^{(e)} = \int \int_{A^{(e)}} dx dy = \int_0^1 \int_0^{1-\beta} |J^{(e)}| d\alpha d\beta = |J^{(e)}| \sum_p W_p = \frac{|J^{(e)}|}{2}$ == References == == References ==

Latest revision as of 18:41, 11 November 2009

Numerical integration refers to all the procedures, algorithms and techniques in the numerical analysis to obtain an approximate solution to a definite integral.

That is, how to obtain a numerical value of:

$\int_{\lambda_a}^{\lambda_b}\! f(\lambda)\, d\lambda.$

where $\lambda \,$ can be a 1D, 2D or 3D domain.

For our interest in the Finite Element Method, the purpose is to describe how the element matrices can be integrated numerically.

Gauss-Legendre Numerical Integration

To fix the most basic concepts on Numerical Integration, we will focus our description on a one dimensional integration using the Gauss-Legendre quadrature, that is, to solve:

$I=\int_{-1}^{+1} f(\xi) d\xi$

The Gauss-Legendre quadrature establish that the definite integral of a function can be approximate by using a weighted sum of function values at specified points within the domain of integration. An p-point Gaussian quadrature rule is constructed to yield an exact result for polynomials of degree 2p − 1 or less by a suitable choice of the points   $\xi_i \,$   and weights   $w_i \,$   for   $i = 1, \cdots p \,$.

$\int_{-1}^{+1} f(\xi)\,d\xi \approx \sum_{i=1}^p w_i f(\xi_i)$

The coordinates and related weights are:

Number of points, p Points, ±ξi Weights, wi
$1\,$ $0.0 \,$ $2.0\,$
$2\,$ $\pm\sqrt{1/3}$ $1.0\,$
$3\,$ $0.0 \,$ $\frac{8}{9}$
$\pm\sqrt{3/5}$ $\frac{5}{9}$
$4\,$ $\pm\sqrt{\Big( 3 - 2\sqrt{6/5} \Big)/7}$ $\tfrac{18+\sqrt{30}}{36}$
$\pm\sqrt{\Big( 3 + 2\sqrt{6/5} \Big)/7}$ $\tfrac{18-\sqrt{30}}{36}$
$5\,$ $0.0 \,$ $\frac{128}{225}$
$\pm\tfrac13\sqrt{5-2\sqrt{10/7}}$ $\tfrac{322+13\sqrt{70}}{900}$
$\pm\tfrac13\sqrt{5+2\sqrt{10/7}}$ $\tfrac{322-13\sqrt{70}}{900}$

or, using numerical values:

Number of points, p Points, ±ξi Weights, wi
1 0.0 2.0
2 0.5773502692 1.0
3 0.0 0.8888888889
0.774596697 0.5555555556
4 0.3399810436 0.6521451549
0.8611363116 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
7 0.0 0.4179591837
0.4058451514 0.3818300505
0.7415311856 0.2797053915
0.9491079123 0.1294849662
8 0.1834346425 0.3626837834
0.5255324099 0.3137066459
0.7966664774 0.2223810345
0.9602898565 0.1012285636

Example of a one dimensional integration

For the function: $f(x)=1+x+x^2+x^3+x^4 \,$, the exact integration in [-1,+1] is:

$I=\int_{-1}^{+1} f(x) dx = \left . \left ( x + \frac{x^2}{2} + \frac{x^3}{3} + \frac{x^4}{4} + \frac{x^5}{5} \right ) \right |_{-1}^{+1} = 2 + 2 \frac{1}{3} + 2 \frac{1}{5} = 3.0666$

Numerically:

$p=1, x_1=0, W_1=2; \qquad I=W_1 f(x_1)=2$

$p=2 \begin{cases} x_1 = - 0.57735, & W_1 = 1 \\ x_2 = + 0.57735, & W_2 = 1 \end{cases} \qquad I=W_1 f(x_1) + W_2 f(x_2) = 0.67464 + 2.21424 = 2.8888$

$p=3 \begin{cases} x_1 = - 0.77459, & W_1 = 0.5555 \\ x_2 = - 0.00000, & W_2 = 0.8888 \\ x_3 = + 0.77459, & W_3 = 0.5555 \end{cases}$
$I=W_1 f(x_1) + W_2 f(x_2) + W_3 f(x_3) = 0.7204·0.5555 + 1.0·0.8888 + 3.19931·0.5555 = 3.0666 \,$

That is the exact value, because for any polynomial function of pth order it is enough to use p-1 integration points.

Two Dimensional Numerical Integration

By using isoparametric formulation we can use natural coordinates to compute any integration. In addition, we can still use the Gauss-Legendre quadrature.

Numerical Integration for Isoparametric Triangular Domains

A general integral expression form for two dimensional domains can be written in terms of the area coordinates and, therefore, computed by using the Gauss quadrature:

$\int_0^1 \int_0^{1-L_3} f(L_1,L_2,L_3) dL_2 dL_3 = \sum_{p=1}^{n_p} f(L_{1_p},L_{2_p},L_{3_p}) W_p$

with:

• $n_p \,$   the number of integration points;
• $L_{1_p}, L_{2_p}, L_{3_p} \,$   the value of the area coordinates;
• $W_p \,$   the weight in the integration point p;

The following table and picture shows the integration points and weights for triangles obtained from the Gaussian quadrature (precision means the degree of polynomial for exact integration):

Number of points, n precision Points L1 L2 L3 Wi
1 Linear a 1/3 1/3 1/3 1/2
3 Quadratic a 1/2 1/2 0 1/6
b 0 1/2 1/2 1/6
c 1/2 0 1/2 1/6
4 Cubic a 1/3 1/3 1/3 -9/32
b 0.6 0.2 0.2 25/96
c 0.2 0.6 0.2 25/96
d 0.2 0.2 0.6 25/96
7 Quartic a 0 0 1 1/40
b 1/2 0 1/2 1/15
c 1 0 0 1/40
d 1/2 1/2 0 1/15
e 0 1 0 1/40
f 0 1/2 1/2 1/15
g 1/3 1/3 1/3 9/40

Note that the weight values has been normalised in order to sum 1/2 to maintain the exact value for the element area.

Therefore:

$A^{(e)} = \int \int_{A^{(e)}} dx dy = \int_0^1 \int_0^{1-\beta} |J^{(e)}| d\alpha d\beta = |J^{(e)}| \sum_p W_p = \frac{|J^{(e)}|}{2}$

References

1. Carlos A. Felippa, "A compendium of FEM integration formulas for symbolic work", Engineering Computations, Vol. 21 No. 8, 2004, pp. 867-890, (c) Emerald Group Publishing Limited [1]
2. Numerical Integration