John L.

Volakis
Rad. Lab.,EECS Dept.
University of Michigan
Ann Arbor, MI 48109-2122
(313) 647-1 797
(313) 647-2106 (Fax)
volakisQumich.edu(email)

Forward by JLV involves approximating an integral by a summation of integrand

samples. Each sample point has a weighting coefficient associated
It is not unlikely to find that a numerical approach is the most with it. On a triangular domain, the approximation has the form
efficient way to perform integration of complex hnctions, even
when an analytical result is available. In finite-element and moment-
method solutions, some of the analytical expressions for the inte-
grals encountered are, in many cases, lengthy and more complex
than a numerical result. This issue’s article, by Savage and
Peterson, provides a review of quadrature numerical-integration where the hnction is written in local or simplex coordinates,
rules over two- and three-dimensional solutions. The authors give (a,P,y) [l], A is the area of the triangle, and N i s the total number
an introduction, and introduce the reader to the appropriate refer- of integrand sample points. Similarly, on a tetrahedral domain, the
ences for complete listings of higher-order integration rules. approximation has the form

uadrature Rules for

Numerical Integration over Here, Vis the volume of the tetrahedra, and the function is written
in terms of the four three-dimensional simplex coordinates
Triangles and Tetrahedra (a,P,y,s).This method uses a fixed number of integrand sample
points, and yields exact answers as long as the integrand obeys
certain criteria. Many quadrature rules, including those reviewed
J. Scott Savage arid Andrew F, Peterson here, are formulated to exactly integrate polynomial functions of
Georgia Institute o f Technology order I I 5 11, and are said to be accurate to degree p . As the num-
School of Electrical and Computer Engineering ber of sample points is increased, the degree of the rule usually
Atlanta, GA 30332-0250 increases. This implies that general functions, which can be
E-mail peterson@ee gatech edu approximated by polynomials, are integrated with less error.

The derivation of a quadrature rule of a given order involves

1. Introduction
determining the number of sample points, the sample-point loca-
tions, and their corresponding weights. Optionally, symmetry con-
Many computational methods in electromagnetics, including
straints may be placed on the test-point locations to eliminate
the method of moments (MOM) and the finite-element method
variations caused by element vertex reordering. The efficiency of a
(FEM), require integrating a hnction over a given domain. In many
quadrature rule is based on the number of sample points required to
cases, analytical solutions for the integrals are not available, and
give a certain order of accuracy. For example, a fifth-order rule
some type of numerical integration must be used. One-dimensional
using seven points is said to be more efficient than a fifth-order rule
numerical-integration techniques are widespread throughout popu-
using eleven points. Since many computational electromagnetics
lar mathematical and engineering texts. However, two-dimensional
algorithms require a large number of integral evaluations, efficiency
and three-dimensional numerical-integration methods are discussed
is a primary concern.
only in specialized reports and journal articles, and almost never in
the electrical engineering literature This synopsis reviews the avail- The first known paper, which derives efficient, symmetric
able literature on numerical integration over triangular and tetrahe- quadrature rules for triangles and tetrahedra, is by Hammer,
dral domains, in order to point the reader toward usehl references.
Marlowe, and Stroud [2]. In this paper, procedures for determining
the sample-point locations and weights are given, although little is
2. Quadrature rules said about the number of samples required for a given order of
accuracy. Each of the formulas in [2] was derived analytically, and
One specific approach to numerical integration is called quad- although of relatively low degree, they provide insight into deriving
rature (sometimes cubature, in the multi-dimensional case), and higher-order rules. Formulas up to fifth degree, for triangular

100 IEEEAntennas and Propagation Magazine, Vol. 38, No. 3, June 1996

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on September 01,2023 at 07:59:23 UTC from IEEE Xplore. Restrictions apply.
 domains, and up to third degree, for tetrahedral domains, were pre- Table 1. The number of sample points for triangle rules.
sented.

As an example, consider integrating a quadratic hnction over 1 1 1 3 1'

a triangle. In simplex coordinates, a general quadratic fbnction can
2 4 6 3*
be written
3 4 IO 4*
4 9 15 6*

The linear dependence of y on a and p eliminates it from Equa- 6 16 28 12

tion (3). Clearly, one test point is insufficient to capture the behav- 7 16 36 12*
ior of the quadratic, and two sample points would lead to a non- 8 25 45 15:
symmetric rule on a triangle, so a second-order rule with three test 9 25 55 19
points is sought. In simplex coordinates, the three symmetric sam- 10 36 66 22
ple points will have the form 11 36 78 27
12 I 49 91 33
13 I 49 105 36
14 1 64 120 42
Each of these points must have the same weight, MI. The quadrature 15 I 64 136 48
rule must exactly satisfy the following terms: 16 81 153 52
17 81 171 61
jjdA= A , I 18 I 100 I 190 I 70 I
19 100 210 73
20 121 23 1 79
'Known to be the minimum number of samples possible for a given
degree.

Table 2. Number of sample points for tetrahedra rules.

1 1 1 4 1*

I I I I
The other two terms in Equation (3) will be satisfied by symmetry. 3 8 20 5*
By substituting Equations (5)-(8) into (I), the following equations 4 27 35 11
result: 5 27 56 14
6 1 64 84 24
3w=1, (9) 7 1 64 120 31
8 125 165 43
1
w(?? + 25) = - , (10) 9 125 220 53
3
'Known to be the minimum number of samples possible for a given

r I
w 17-+252 = - ) : degree.

quadrature rules on triangles and tetrahedra [3]. This derivation is

based on orthogonal polynomials, and the paper gives a straight-
forward procedure for obtaining high-order rules. However, these
rules are not as efficient as those presented in [2]. Tables 1 and 2
These four nonlinear equations are not linearly independent, but compare the number of sample points for a given degree, based on
have the solution the current literature [4-71. Another alternative is the use of a prod-
uct rule based on one-dimensional Gaussian-quadrature formulas.
The necessary number of sample points for the product rule are also
shown in Tables I and 2.

In this example, an analytical solution is obtained easily. However,

as the desired order of accuracy increases, the number of unknowns 3. Higher-order rules for triangles
and the number of equations grows rapidly. Therefore, deriving
higher-order quadrature rules of this type usually involves a trial Several authors extended the work of [2]. Cowper (41 ana-
and error process to determine the number of sample points, and a lytically determined sixth- and seventh-degree rules for triangles.
numerical solution to the resulting system of nonlinear equations. Rules for every degree up to 20 were presented by Dunavant [5],
who obtained them by solving the nonlinear equations numerically.
As an alternative to solving nonlinear equations, Silvester pre- A recent review of other rules is presented by Cools and
sented a Newton-Cotes type of procedure for deriving symmetric- Rabinowitz [6]. In some cases, the optimal rule requires integrand

IEEE Antennas and Propagation Magazine, Vol. 38, No. 3 , June 1996 101

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on September 01,2023 at 07:59:23 UTC from IEEE Xplore. Restrictions apply.
 samples outside the triangular domain, which may be impractical, 3 . P. Silvester, “Symmetric quadrature formulae for simplexes,”
depending on the application. Mathematics of Coniptatioia, 1970, pp. 95-100.

4. G. R. Cowper, “Gaussian quadrature formulas for triangles,’’ Znt.

4. Higher-order rules for tetrahedra
Joiirrial Nimmer. Meth. Gig., 1973, pp. 405-408.
The most complete collection of rules for tetrahedral domains
5. D. A. Dunavant, “High degree efficient symmetrical Gaussian
was presented by Keast [7]. Alternatives are cited in [6]. The table
quadrature rules for the triangle,” hit. Jozrrnal Nunier. Meth. Eng.,
of coefficients in [7] contains an error in the eighth-degree rule. The
1985, pp. 1129-1 148.
weight associated with the centroid of the tetrahedra should be
negative (-0.039327...). It is noteworthy that some of the other
6. R. Cools and P. Rabinowitz, “Monomial cubature rules since
published reports in this area are clearly incorrect. For example,
‘Stroud:’ a compilation,” J. ConipirtafioimlAppl. Math., 1993, pp.
reference [SI presents rules for tetrahedra that appear to require
309-326.
minimal samples, but are not accurate to the stated degree.
7. P. Keast, “Moderate-degree tetrahedral quadrature formulas,”
5. Summary Computer Meth. Appl. Mechanics Er7g., 1986, pp. 339-348.

This report surveys quadrature rules for triangles and tetrahe- 8. K. S. Sunder and R. A. Cookson, “Integration points for trian-
dra, which are of growing importance for numerical techniques in gles and tetrahedrons obtained from the Gaussian quadrature points
electromagnetics. Any quadrature rule taken from the literature for a line,” C o n i p t e n ar7dStrirctiires, 1985, pp. 881-885
should be tested, to ensure that it is accurate to the degree speci-
fied. It is hoped that this brief summary is of some assistance to the
reader interested in efficient, high-accuracy quadrature formulas.

6. References Correction
1 P. Silvester and R. L. Ferrari, Finite Elenleiits for Electrical
E~gineers.Cambridge, UK, Cambridge University Press, 1983 In “An Efficient Way to Compare ABCs,” which appeared in
the February “EM Programmer’s Notebook” (IEEE Aiiferirias and
2. P. C. Hammer, 0. J. Marlowe, and A. H. Stroud, “Numerical Propagatioii Magazine, 38, 1, pp. 71-75), the plots in Figures 4a
integration over simplexes and cones,” Math. Tab/. Nnt17. Res. and 5a, on page 74, should be switched.
Colin. Wash., 1956, pp. 130-137.

ANTENNA/MICROWAVE FF’BII
ENGINEERS
Wang Electro-Opt0 Corp., located in the northern suburbs
of Atlanta, is a small, rapidly growing company with patented
technology to position it as a leading technology company in
the wireless market. We have immediate openings for
antennaimicrowave engineers to work on major programs to
develop state-of-the-art antenna systems for shipboard,
airborne, and land-mobile applications.
Engineers with hands-on technical, marketing, and/or
customerlsponsor interfacing experience in antenna and
microwave systems are preferred. Familiarity with spiral
antennas, photonics, semiconductors, digital circuits, and
optoelectronics is a plus. BSEE is required; advanced degree
not necessary but preferred.
Send resume and salary history to WEO, 1335 Capital
Circle, Suite G, Marietta, GA 30067. EOE.
Wang Electro-Opt0 Corporation
Air InterfaceTM

102 IEEE Antennas and Propagation Magazine, Vol. 38, No. 3,June 1996

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY MADRAS. Downloaded on September 01,2023 at 07:59:23 UTC from IEEE Xplore. Restrictions apply.