Professional Documents
Culture Documents
But the method requires the fundamental solution (Greens function) to be known in advance.
Also, singularity appears in the integrals involving the fundamental solution and hence provisions
are to be made to avoid the singularity during the evaluation of such integrals. Also, the knowledge
of the fundamental solution is extremely difficult if not impossible except for some simple
structures. In the case of a general anisotropic medium, it is extremely difficult to find the greens
function. Moreover, the matrix resulting from the boundary element method is full and nonsymmetric unlike the finite element matrix, which is sparse and symmetric. In spite of these
disadvantages, the boundary element method has the advantage that the radiation condition at
infinity is exactly satisfied, whereas in the case of finite element method, it is only approximately
satisfied due to the application of the absorbing boundary condition [5] in the unbounded medium.
Hence this makes the boundary element method favorable while dealing with unbounded domains
usually encountered in open boundary problems.
In this connection, it is mentioned that, a novel semi-analytical finite element method called
The Scaled Boundary Finite element method was developed jointly by Chongmin Song and
John. P. Wolf in 1997 [13,14] to successfully solve Elastodynamic and allied problems of Civil
Engineering and Soil structure interaction. The novel method is based entirely on finite elements,
but with a discretization only on the boundary [13,14]. The method combines the advantages of
both the finite and boundary element methods [13,14]. This method doesnt require any
fundamental solution to be known in advance, unlike the boundary element method [13,14]. This
novel method is analytical in its approach in the radial direction with respect to an origin called as
scaling center, and implements the finite element method in the circumferential direction [13,14].
Hence the method is semi-analytical.
The key advantages of this method are as follows: [13,14]
a) Reduction of the spatial dimension by one, reducing the discretization effort.
The scaled boundary finite element method, is based entirely on finite elements.
The scaling applies to each surface finite element. Its discretized surface on the boundary is
denoted as Se (superscript e for element). Continuous scaling of the element yields a pyramid with
volume Ve. The scaling center O is at its apex.
The base of the pyramid is the surface finite element. The sides of the pyramid forming the
boundary Ae follow from connecting the curved edge of the surface finite element to the scaling
center by straight lines. No discretization of Ae occurs. Assembling all the pyramids by connecting
their sides, corresponds to enforcing compatibility and equilibrium conditions. This results in the
total medium with volume V and the closed boundary S. No boundaries Ae passing through the
scaling center remain. Mathematically, the scaling corresponds to a transformation of the
coordinates for each finite element resulting in two curvilinear local coordinates along the
tangential directions and a single dimensionless radial coordinate representing the scaling factor.
This transformation becomes unique due to the choice of the scaling center from which the total
boundary of the geometry is visible [13,14].
The scaled boundary transformation is basically a relation between the derivatives in the
cartesian coordinates and the derivatives expressed in the scaled boundary variables [13,14].
4. The scaled boundary finite element method in electromagnetics: The scaled boundary
transformation equations [13,14] are quite general. It can be applied to differential equations
governing the phenomena in any discipline. This feature of the scaled boundary transformations is
used in the reformulation of the novel method for electromagnetics [17]. However, the actual
formulation of the scaled boundary finite-element equation depends upon the additional constraints
that are specific to the discipline, which are to be satisfied. This approach ensures that, the scaled
boundary finite-element equation takes into account, the specific features of that discipline. Hence,
a closest possible representation of the system, represented by the original differential equations
along with the constraints in the form of boundary conditions, is achieved.
5. The Scaled Boundary Finite Element formulation for Cavity Structures: An ideal metallic
cavity structures represent the total confinement of the electromagnetic field in a volume of space
bounded externally by metallic surfaces. The eigen-modes of the cavity structures are in fact, the
standing wave field patterns corresponding to various eigen values. The tangential component of
the electric field and the normal component of the magnetic field vanish on the boundary
comprising the metallic surfaces. It is assumed that, the metallic surface is a perfect conductor.
In this section, the scaled boundary finite-element equation is developed for a general metallic
cavity structure, and a numerical implementation is illustrated for the case of a spherical metallic
cavity.
5.1. Theory
The first step in the development of the appropriate scaled boundary finite element formulation
is the formulation of the variational functional. This is considered in the following discussion.
The vector helmholtz equation in H formulation [5] is given by
r1 H - k 02 r H = 0 ,
k 02 = 2 0 0
(5.1.1)
where r and r are the local material properties corresponding to the relative permittivity and
relative permeability of the medium respectively, is the angular frequency.
The associated essential boundary condition [5] is given by
H 1n = H t
(5.1.2)
The essential boundary condition implies that the magnetic field is purely tangential to the
bounding metallic surface. H t denotes the tangential magnetic field component, and 1n denotes unit
normal vector to the surface.
The associated functional for Eq.(5.1.1) can be developed by taking the dot product of Eq.(5.1.1) by
an arbitrary test vector W, and integrating over the whole domain ,
W (
1
r
H - k02 r H d = 0
(5.1.3)
P ( H ) d = ( P )( Q ) d (P Q ) 1 dS
n
(5.1.4)
( W ) ( H ) d (W
1
r
H 1n dS - k02 r W H d = 0
1
r
(5.1.5)
W (
1
r
H 1n dS
(5.1.6)
Also,
H = j E
(5.1.7)
( W ) ( H ) d k W H d = 0
1
r
2
0
(5.1.8)
The Eq. (5.1.8) represents the weighted residual form of the vector helmholtz equation in H
formulation.
The associated variational form of Eq.(5.1.1) is obtained by substituting W = H in Eq.(5.1.8) and
by using the identity given in [18] for all the three Cartesian components of the vectors occurring
in the integral form given in (5.1.8). This gives the variational functional F given by,
F=
1
( H ) r1 ( H ) k02 H r H d
(5.1.9)
The expression for F given in (5.1.9) is the variational functional associated with the vector
helmholtz equation (5.1.1) subject to the essential boundary condition given in (5.1.2).
The variational functional given in (5.1.9) is made stationary, with respect to the unknown
coefficients in the expansion of H, which corresponds to the solution of the vector helmholtz
equation.
5.2. Expression of the field components in terms of the scaled boundary coordinates
The first step in the development of the scaled boundary finite element formulation, is
expressing the vector variable H in terms of the scaled boundary variables , , . The vector
variable H is expanded in terms of the ortho-normal vectors in the scaled boundary co-ordinate
system [13,14] as
H ( ,, ) = Hn ( ) Hn (, ) n + Hn ( ) Hn (, ) n + Hn ( ) Hn (, ) n
(5.2.1)
H ( , , ) = H x ( ) H x ( , ) H y ( ) H y ( , ) H z ( ) H z ( , )
i
j
k
(5.2.2)
In Eq.(5.2.2) i ,j, k represent the ortho-normal vectors in the Cartesian coordinate system.
The individual terms in the R.H.S of (5.2.2) are given as follows.
Hx ( ) Hx (, ) = Hn ( ) Hn ( ) Hn ( )
Hn (, ) nx
Hn (, ) n x
Hn (, ) n x
(5.2.3)
10
H y ( ) H y (, ) = Hn ( ) Hn ( ) Hn ( )
Hz ( ) Hz (, ) = Hn ( ) Hn ( ) Hn ( )
Hn (, ) ny
Hn (, ) n y
Hn (, ) n y
(5.2.4)
Hn (, ) nz
Hn (, ) n z
Hn (, ) n z
(5.2.5)
The terms of the form H( ) H( , ) occurring in the matrix multiplication have the form,
m
H ( ) H ( , ) = f ( ) hij hi ( ) h j ( ) n( , )
(5.2.6)
i = o j =0
In the above expression, f ( ) denotes the unknown radial function expressed in terms of the radial
variable and h ij denotes the unknown numerical coefficients occurring in the expansions of Hx,
Hy and Hz respectively. hi ( ) denotes the single variable functions of x, y and z components of H
expressed in terms of the tangential variable . h j ( ) denotes the single variable functions of x, y
and z components of H expressed in terms of the tangential variable . The form of the functions
expressed as h( ) and h( ) are given as [5]
h0 ( r ) = 1 r
(5.2.7)
h1 (r ) = 1 + r
(5.2.8)
hi (r ) = (1 r ) 2 r i 2 for i 2
(5.2.9)
It is shown in detail in [5] that, while employing the vector finite element method, only the
tangential continuity between adjacent elements are sufficient to impose inter-element field
continuity. In (5.2.6), n( , ) represent the scalar components of the orthogonal vectors in the
11
scaled boundary coordinate system. The function f() occurring in the field expansions is expanded
in the form of the power series expansion in as
N
f1 ( ) = ak k
(5.2.10)
k =0
Eqs.(5.2.1-5.2.10) enables the field components to be expressed in terms of the scaled boundary
variables. The next step is the derivation of the solenoidality (divergence) condition of the magnetic
field, in terms of the scaled boundary coordinates. This is detailed in the following section.
5.3. Derivation of the solenoidality condition in terms of the scaled boundary coordinates
The derivation of the solenoidality condition plays a crucial role in the development of the
scaled boundary finite element formulation. As mentioned earlier, this is essential for the
elimination of spurious modes in the eigen spectrum of the boundary value problem.
Since the problem is formulated in terms of the scaled boundary variables ( , , ) [13,14] it is
necessary that the divergence condition should be written in terms of ( , , ) . This is achieved as
follows.
The divergence condition for the magnetic field is given by,
( H ) = 0
(5.3.1)
H = 0
(5.3.2)
Expanding Eq.(5.3.2),
H x H y H z
=0
+
+
y
z
x
(5.3.3)
12
H x H y H z
+
+
y
z
x
(5.3.4)
Rewriting the Eq.(5.3.4) in terms of scaled boundary coordinates by using the three dimensional
scaled boundary transformation [13,14] and using the convention that H x , H y and H z respectively
denote the scalar part of the i , j , and k components given in (5.2.2), we get
g H x 1 g H x g H x g H y 1 g H y g H y
+
+
+
+
+
nx
nx
nx
ny
ny
ny
| J |
| J |
| J |
| J |
| J |
|J|
g H z 1 g H z g H z
=0
nz
nz
nz
+
+
+
|J|
| J |
| J |
(5.3.5)
nx
nx
ny
ny
ny
+
+
+
+
|J|
| J |
| J |
| J |
| J |
| J |
g H z g H z g H z
=0
+
+
+
nz
nz
nz
| J |
| J |
|J|
nx
(5.3.6)
g
d[N1 ]
d[N 2 ]
d[N 3 ] g
d[N1 ]
d[N 2 ]
d[N 3 ] k
[
]
[
]
[
]
n
a
n
b
n
c
+ n y [bk ]
+ nz [ck ]
+ nx [ak ]
+
+
y k
z
k
| J | x k d
d
d | J |
d
d
d
k = 0
m
g
+ k [ak ] nx [N1 ( , )] + [bk ] ny [N 2 ( , )] + [ck ] nz [N 3 ( , )] k = 0
k =0 | J |
(5.3.7)
The Eq.(5.3.7) holds for all values of . This is possible if the coefficient terms of k are zero.
Equating the coefficient of k to zero, gives,
[ak ] g
nx k [N1 ] +
g
g d[N1 ] g d[N 3 ]
g d[N 2 ] g d[N 2 ]
+ [bk ]
nx
nx
n y k [N 2 ] +
ny
ny
+
+
|J|
d
|J|
d
|J|
d
|J|
d
| J|
|J |
g
g d[N 3 ] g d[N 3 ]
=0
nz
nz
+ [ck ] nz k [N 3 ] +
+
|J|
d
|J|
d
| J |
(5.3.8)
13
Eq.(5.3.8) is valid for k 0 and [N1] ,[N2] and [N3] are matrices containing the functions in terms of
the variables and of the double summation series given in (5.2.6). [ak], [bk] and [ck] are the
matrices containing the unknown constants of the radial power series expansion given in (5.11) and
H H in the functional.
Following this procedure, the variational form of the functional given in Eq.(5.1.9) written in terms
of the variables of the scaled boundary coordinates is given by,
F=
[((
)]
1
r1
1
r1
2
2
2
g
T
+
g
T
+
g
T
+
2
1
2
3
2 0 | J |
|J|
) ((
) ((
[((g
) ) ((
k02 r ( f1 ( )N1 ( , )) + ( f 2 ( )N 2 ( , )) + ( f 3 ( )N 3 ( , )) d d d = 0
2
) ) ((
) )]
g T4 + g g T5 + g g T6
(5.4.1)
14
H x
H x
H z H y
H x H x
T1 = ny
+ nx
ny
+ nz
nx
nz
H y H x
H z
H z
H z H y
T2 = ny
+ nx
ny
+ nz
nx
nz
(5.4.2)
H y H x
H z
H z
H z H y
T3 = ny
+ n x
ny
+ nz
nx
nz
(5.4.3)
(5.4.4)
H y H x
H y H z
H z
H x
H z H x
T4 = ny
ny
nz
nx
+ nz
nx
nz
nz
H y
H x
H x H y
ny
+ nx
ny
nx
(5.4.5)
H y H x
H y H z
H z
H z
H z H x
T5 = ny
ny
nz
nx
+ nz
nx
nz
nz
H y
H x
H x H y
ny
+ nx
ny
nx
(5.4.6)
H y H x
H y H z
H z
H z
H z H x
T6 = ny
ny
nz
nx
+ nz
nx
nz
nz
H y
H x
H x H y
ny
+ nx
ny
nx
(5.4.7)
Having expressed the functional in terms of the scaled boundary variables, and the next step is to
implement the solenoidality condition given in Eq.(5.3.8), in the variational functional given in
Eq.(5.4.1). The Lagrange multiplier technique [19] is made use of which results in a modified
functional containing ( k + 1) Lagrange multiplier terms for every value k. These Lagrange
multiplier terms account for the implementation of the solenoidality condition of the magnetic field.
The radial coordinate in the modified functional is independent of the two circumferential
15
coordinates and . The functional is integrated with respect to with its lower and upper limits
being 0 and 1. This renders the functinal entirely in terms of the circumferential variables and .
The resulting modified functional is given by,
F=
1
T1 '+T2 '+ T3 '+ T4 '+T5 '+ T6 ' k02 [T7 '+ T8 '+ T9 '] d d + ( k + 1) Lagrange multiplier terms
2 ,
(5.4.8)
In (5.4.8), the letters with a prime denote the terms after the integration with the radial variable
. An important observation that is to be noted is that, the terms in Eq.(5.4.8) contain only
the surface finite element discretization factor denoted by | J | even for the general 3-D
structures.
This unique feature of the scaled boundary finite element method, in contrast with the
conventional finite element method, where the finite element discretization is to be necessarily
performed in all the three dimensions, for arbitrary three dimensional structures devoid of
uniformity along a particular axis. The forth-coming section deals with the generation of the
After the expression of the functional purely in terms of the tangential variables, as given in
Eq.(5.4.8), the next step is the generation of the scaled boundary finite element matrices.
The expression given in (5.4.8) is evaluated for every surface element characterized by the
circumferential variables (,). Then the variation with respect to each undetermined coefficient is
set to zero. This process leads to a set of linear equations. Imposing only tangential continuity of the
field component between the adjacent elements, the final system equation is of the form
A h + k 02 B h = 0
(5.4.9)
16
The Eq.(5.4.9) is a standard form of the matrix eigen value equation which can be solved
numerically. The solution of Eq.(5.4.9) gives the solution of the vector helmholtz equation given in
Eq.(5.1.1) along with the essential boundary condition given in (5.1.2).
An important feature that is to be noted in the theoretical formulation developed above is
that, there is no specific assumption on the shape of the geometry under consideration. Hence,
the formulation thus developed, holds independent of geometry.
5.7. Numerical Implementation
The theoretical formulation thus developed, is implemented numerically for the case of a
spherical metallic cavity, with air being the dielectric medium. The schematic diagram of the
spherical metallic cavity considered for the numerical implementation is shown in Figure 5.1 in the
following page..
The units of the radius of the spherical metallic cavity denoted as a is taken to be in cms. For
the finite element discretization of the surface, an eight node curvilinear quadrilateral elements were
used with the mesh of one octant consisting of three finite elements. The discretized boundary of
the solid sphere of one octant is shown below in Figure 5.2 in the following page.
17
Figure 5.2 Finite Element mesh of one octant of boundary of solid sphere.
After the discretization of the boundary of the spherical metallic cavity by surface finite
elements, the functional given in Eq.(5.4.8) is evaluated for every finite element followed by the
numerical integration along the radial and tangential coordinates. During this process, only the
inter-element tangential continuity is imposed. The integration which is performed involving the
circumferential variables are done numerically using 5 point gaussian quadrature. Then, the
variation with respect to every undetermined coefficient is taken, and set to zero.
The eigen value equation resulting from the process of assembling the element matrices were
solved by using the standard LAPACK [20] collection of Fortran subroutines. The number of terms
were used in the radial expansion of the fields were five. The resonant frequencies of TM 011 mode
for the spherical cavity were computed by varying the radius of the spherical cavity, and the
numerically obtained results show close agreement to the values obtained by a full theoretical
analysis [21], and is shown below in Table 5.1.
Table 5.1 The close agreement between the theoretical and the numerical values of the resonant frequency of
TM (011) mode.
S.No
4.367
3.277
3.278
2.622
2.6215
2.185
2.1848
a (in cms)
18
By theoretical modal field analysis [21], the resonant frequency fr for a spherical cavity of radius
3 cms for TM mode satisfies the condition TM
(even)
(011) = TM
(even)
Accordingly, the resonant frequency obtained through the scaled boundary formulation for the
spherical cavity of radius 3cms for the above mentioned modes were also constant with a value of
4.3668 x 109 Hz, satisfying the above property, confirming the validity of the new method.
Another important feature of the scaled boundary finite element formulation, is the effect of the
number of terms in the radial expansion of the field variable on the accuracy of the eigen values.
Table 5.2, shows the effect of increase in the number of terms in the radial expansion of the field
variable, on the accuracy of the resonant frequency.
Table 5.2 The effect of increase in the number of terms in the radial expansion of the field variable, on the
resonant frequency of TM (011) mode.
No. of terms in
the radial
expansion of the
field variable
Radius of the
spherical
metallic
cavity.
Resonant frequency
computed for the TM011 mode
by theory.
fr (10 9 Hz)
a (in cms)
3
Resonant frequency
computed by scaled
boundary
Finite element method.
Fr (10 9 Hz)
4.367
4.3654
4.367
4.3660
4.367
4.3664
4.367
4.3670
From Table 5.2, it can be inferred that, the number of terms in the radial expansion of the field
variable, has a marked influence on the accuracy of the resonant frequency. It is observed from the
above table that, the increase in the number of terms in the radial expansion of the field variable is
accompanied by the corresponding increase in the accuracy of the eigen values.
5.8. Conclusion: The scaled boundary finite element method is a novel semi analytical method
based on finite elements, originally developed in the field of civil engineering to study problems
pertaining to elastodynamics and Soil structure interaction. The crucial aspects of the reformulation
19
of the novel method for the full-wave analysis of cavity structures, is reported in this paper. The
closeness of the numerical results obtained from the scaled boundary finite element method to those
obtained from analytical approach validates the methodology developed in this paper, for analyzing
cavity structures.
An important consequence of the method being semi-analytical is that, the accuracy of the
numerical values depend not only on the element size as in the conventional finite element method,
but also on the number of terms that are used in the radial series expansion of the fields. This
enables to get accurate numerical results even by increasing the number of terms in the radial
expansion of the fields without tampering with the element size. Also, the novel formulation
reported in this paper, is independent of the geometry under consideration. The scaled boundary
finite element formulation in electromagnetics is further developed to analyze multi-layered and
multi-conductor micro-strip transmission lines, and VLSI interconnects.
Acknowledgements: The author thanks Prof. John. P. Wolf, Institute of Hydraulics and Civil
Engineering, Swiss Federal Institute of Technology, Lausanne, Switzerland, and Dr. Chongmin
Song, Department of Environmental and Civil Engineering, University of New South Wales,
Australia, for their valuable suggestions and providing their research articles on Scaled Boundary
Finite Element method.
References:
[1]. Monk, P. Finite Element Methods for Maxwell's Equations Oxford University Press, Oxford,
2003.
[2]. J. Jin, The Finite Element Method in Electromagnetics J. Wiley & Sons, New York, 2002
[3]. Salazar-Palma, M., Sarkar, T. K., Garcia-Castillo, L. E., Roy, T. and Djordjevic, A. R. Iterative
and Self-Adaptive Finite-Elements in Electromagnetic Modeling Artech House Publishers, Inc.,
1998.
[4]. Volakis, J. L., Chatterjee, A. and Kempel, L. C., Finite Element Method for Electromagnetics:
Antennas, Microwave Circuits, and Scattering Applications IEEE Press and Oxford University
Press, New York, 1998.
20
[5]. P.P. Silvester, and R.L. Ferrari, Finite Elements for Electrical Engineers,3rd Ed., Cambridge
University Press., 1996.
[6]. P.P. Silvester and G.Pelosi, Finite Elements for Wave Electromagnetics, IEEE Press, 1994.
[7]. Wang, X. H., Finite Element Methods for Nonlinear Optical Waveguides Gordon & Breach,
New York, 1996
[8]. Fernandez, F. A. and Lu, Y., Microwave and Optical Waveguide Analysis by the Finite
Element Method, J. Wiley & Sons, New York, 1996.
[9] Luis valor and Juan Zapata, Efficient Finite Element Analysis of Waveguides with Lossy
Inhomogeneous Anisotropic materials Characterized by Arbitrary Permittivity and
Permeability Tensors, IEEE Trans. Microwave Theory Tech., vol. MTT-43, no.10, pp.2452
2459, Oct.1995.
[10] Luis valor and Juan Zapata, An Efficient Finite Element Formulation to Analyze
Waveguides with Lossy Inhomogeneous Bi-Anisotropic materials, IEEE Trans. Microwave
Theory Tech.,vol. MTT-44, no.2, pp.291 -296, Feb.1996.
[11] Javier Arroyo and Juan Zapata, Subspace Iteration Search Method for Generalized
Eigenvalue problems with Sparse Complex Unsymmetric Matrices in Finite-Element Analysis
of Waveguides, IEEE Trans. Microwave Theory Tech., vol. MTT-46, no.8, pp.1115 -1123,
Feb.1996.
[12] Andrew J. Deeks and John P. Wolf, An h-hierarchical adaptive procedure for the scaled
boundary finite-element method, Int. J. Numer. Meth. Engng, 54, pp.585-605,2002.
[13] Chongmin Song and John. P. Wolf, The Scaled boundary finite-element method- alias
Consistent infinitesimal finite-element cell method for elastodynamics, Computer Methods
in applied mechanics and engineering, (1997), No.147, pp.329-355.
[14] John. P.Wolf and Chongmin Song, The Scaled boundary finite-element method a primer :
derivations, Computers and Structures, (2000), 78, pp.191-210.
[15] Chongmin Song and John P. Wolf, Consistent Infinitesimal Finite-Element Cell Method :
Three-Dimensional Vector Wave Equation, International Journal for Numerical Methods in
Engg, (1996), Vol.39, pp. 2189-2208.
[16]
Andrew J. Deeks, John P. Wolf, An h-hierarchical adaptive procedure for the scaled
boundary finite-element for elastodynamics, Int. J. Numer. Meth. Engng, (2002), 54,
pp.585-605.
[17]
V.S. Prasanna Rajan, The Theory and application of a Novel Scaled Boundary Finite
Element method in computational electromagnetics, Ph. D thesis, University of Hyderabad,
India, Dec.2002.
[18]
J.N.Reddy, An Introduction to the Finite Element Method, McGraw Hill, p.11-138, 1984.
21
[19]. Irving H. Shames, Clive L. Dym, Energy and Finite Element Methods in Structural
Mechanics, New Age International Publishers Ltd., Wiley Eastern Ltd, pp.671674,1995.
[20]
[21] C.A. Balanis, Advanced Engineering Electromagnetics, John Wiley & Sons, New York,
pp.560-562.