Professional Documents
Culture Documents
FEM & BEM Course Notes (2003) PDF
FEM & BEM Course Notes (2003) PDF
FEM & BEM Course Notes (2003) PDF
c Copyright 1997-2003
Department of Engineering Science
The University of Auckland
Contents
4 Linear Elasticity 75
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 Truss Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3 Beam Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4 Plane Stress Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.4.1 Notes on calculating nodal loads . . . . . . . . . . . . . . . . . . . . . . . 83
4.5 Three-Dimensional Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.5.1 Weighted Residual Integral Equation . . . . . . . . . . . . . . . . . . . . 85
4.5.2 The Principle of Virtual Work . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5.3 The Finite Element Approximation . . . . . . . . . . . . . . . . . . . . . 87
4.6 Linear Elasticity with Boundary Elements . . . . . . . . . . . . . . . . . . . . . . 89
4.7 Fundamental Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.8 Boundary Integral Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.9 Body Forces (and Domain Integrals in General) . . . . . . . . . . . . . . . . . . . 96
4.10 CMISS Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Bibliography 143
Index 147
Chapter 1
+ +
+ + + +
+ ++ + + ++ +
+ + + +
+ + + +
+ + + +
+ +
+ +
(a) (b)
F IGURE 1.1: (a) Temperature distribution
along a bar. The points are the measured
temperatures. (b) A least-squares polynomial fit to the data, showing the unacceptable oscillation
between data points.
+ +
+ +
+ +
+ + + + + + + + + +
+ + + + + +
+ + + + + +
+ +
+ +
& &
(a) (b)
F IGURE 1.2: (a) Temperature measurements replotted against arclength parameter . (b) The ' '
domain is divided into three subdomains, elements, and linear polynomials are independently fitted
to the data in each subdomain.
+*,-.0/213*,4)()5*6
where *879:*9;/< is a normalized measure of distance along the curve.
We define
= >( *,-?/@13*
= *,-A*
such that
= ()B*, = B *,
/ 1/
* *
7 / 7 /
F IGURE 1.3: Linear basis functions Z (
[P]\:^-_`[ and Z +
[U>\a[ .
interpolation
element
nodes: )( cb cb cb cb cb cb cb f > ( fg gf gf gf gf gf gf )( ed ed ed ed ed ed ed
* * *
0 1 0 1 0 1
element / element j element i
F IGURE 1.4: The relationship between global nodes and element nodes.
with )(kMH and ;H , since the parameter H is shared between the first and second elements
"
4 F INITE E LEMENT BASIS F UNCTIONS
the temperature field is implicitly continuous. Similarly, in the third element is interpolated by
with )(noH and pH]l , with the parameter H being shared between the second and third
" "
elements. Figure 1.6 shows the temperature field defined by the three interpolations (1.1)–(1.3).
node /
+ node i
+ node j
+ +
+
node h
+ + +
+ + +
+ + +
+
+
element / element j element i
&
F IGURE 1.5: Temperature measurements fitted with nodal parameters and linear basis functions.
The fitted temperature field is now continuous across element boundaries.
at *Q7 87!qrs/2157D4)()t7u v
)(
which is the value of at the left hand end of the element and has no dependence on
w / w
at *Q
/ 2
/ 1 / )() / i )(> /
h h)x h>x h h h
which depends on >( and , but is weighted more towards )( than
w w
at *Q
/ / /21 / )() / / )(> /
j jx jyx j j j
which depends equally on )( and
w
i i w i i / )(> i
at *Q 2
/ 1 )
)
(
h h)x h>x h h h
1.3 BASIS F UNCTIONS AS W EIGHTING F UNCTIONS 5
|}(
(a)
&
|
(b)
&
| "
(c)
&
|l
(d)
&
F IGURE 1.6: (a) ~ ~~ (d) The weighting functions E associated with the global nodes \:^)~~~ ,
respectively. Notice the linear fall off in the elements adjacent to a node. Outside the immediately
adjacent elements, the weighting functions are defined to be zero.
For example, | weights the global parameter H and the influence of H falls off linearly in
the elements on either side of node 2.
We now have a continuous piecewise parametric description of the temperature field +*, but
in order to define + we need to define the relationship between and * for each element. A
convenient way to do this is to define as an interpolation of the nodal values of .
For example, in element 1
and similarly for the other two elements. The dependence of temperature on , + , is therefore
6 F INITE E LEMENT BASIS F UNCTIONS
+*,k = E2B*D4FE
E
OB*,k = E2B*DCFE
E
where summation is taken over all element nodes (in this case only j ) and the parameter * (the
“element coordinate”) links temperature to physical position . *, provides the mapping
between the mathematical space 7`9M*9
/ and the physical space )(}9M39M , as illustrated in
Figure 1.7.
)(
*
7 /
*Q7#j + at Q* 7#j
)(
*
7
)(
)(
7 / *
*Q7#j
F IGURE 1.7: Illustrating how and are related through the normalized element coordinate . [
The values of and
[U }
[P
are obtained from a linear interpolation of the nodal variables and
then plotted as . The points at
+ are emphasized. [@\3!~
1.4 Q UADRATIC BASIS F UNCTIONS 7
that since = ()B*D must be zero at *M7# (node j ), = ()B*, must have a factor *157#! and since it
The quadratic basis functions are shown, with their mathematical expressions, in Figure 1.8. Notice
is also zero at *$/ (node i ), another factor is B*1A/P . Finally, since = ()B*D is / at *Q7 (node / )
we have = ()B*D-Mj2*1A/<)B*157#! . Similarly for the other two basis functions.
U
P
(a)
J m P (b)
`¡¢ Q
P
(c)
m
U
F IGURE 1.8: One-dimensional quadratic basis functions.
Let
+*u(N¤* - = ()B*u(N¤* C>() = B *u(N¤* C = " B*u(N¤* " = lm*u(N* Fl
where
= )( B*U(N¤* qrs/213*u(s)s/213*
= B*U(N¤* qA*u()s/21a*
= B*U(N¤* qrs/213*u(sC* (1.6)
"=
lJB*U(N¤* qA*u(z*
Note that = ()B*U(N¤* = = ()+*u( = ()B* where = ()B*U( and = ()B* are the one-dimensional linear
basis functions. Similarly, = B*u(N¤* = = B*u(s = ()B* q#¥#%# etc.
These four bilinear basis functions are illustrated in Figure 1.9.
= ( =
"
node i
* *
node / node h
7
node j *u( / *U(
= =l
*
*
7 7
/ *U( / *U(
F IGURE 1.9: Two-dimensional bilinear basis functions.
Notice that = E2*u(N* is / at node G and zero at the other three nodes. This ensures that the
temperature *U(N¤* receives a contribution from each nodal parameter FE weighted by = E2B*u(N¤*
and that when +*u(N* is evaluated at node G it takes on the value E .
As before the geometry of the element is defined in terms of the node positions FEN¤¦uE6 , Ga
1.5 T WO - AND T HREE -D IMENSIONAL E LEMENTS 9
/!N%#%#¥#)NYh by
/ j i *U(
7
7 7 # /
³
F IGURE 1.10: A -node quadratic-linear element (node numbers circled).
Three-dimensional basis functions are formed similarly, e.g., a trilinear element basis has eight
nodes (see Figure 1.11) with basis functions
= >( *u(N* N* k.0/213*u()0/213* >z/213* = B *U(N¤* N¤* -{*u(ys/213* )s/213*
= *u(N* N* " k.0/213*u(C* s/213 * " = lJ B*U(N¤* N¤* " -{*u(z* s/213 * "
(1.10)
=]"± *u(N* N* " k.0/213*u()0/213* 4* " = B*U(N¤* N¤* " -{*u(ys /213* C" *
(1.11)
=O´ *u(N* N* " k.0/213*u(C* * " =Oµ¬ B*U(N¤* N¤* " -{*u(z* * "
(1.12)
" " " " (1.13)
10 F INITE E LEMENT BASIS F UNCTIONS
*" ·
² ¶
*
i
/ h
j
*u(
¸
F IGURE 1.11: An -node trilinear element.
87!- {>(
0/<-¿m®-¿ {
, À7!- {Á(
!*
, s/<-mtj6-ti6 { Á
!*
These four equations in the four unknowns , , and are solved to give
)(
 ½(
2iu 15iu>(O1¿juF½( 1°F½
F½( ¿F½ ®jU>(O15jU
Substituting , , and back into the original cubic then gives
where the four cubic Hermite basis functions are drawn in Figure 1.12.
w step is required to make cubic Hermite basis functions useful in practice. The
One further
D
derivative
!*Ox E defined at node G is dependent upon the element * -coordinate in the two ad-
w D
jacent elements. It is much more useful to define a global node derivative
C& x E where & is
arclength and then use
w , w D w C&
!*Jx E C&)x OI RSEUT VXWÄ !*)x E (1.15)
w C&
where
!*)x E is an element scale factor which scales the arclength derivative of global node
K to the * -coordinate derivative of element node G . Thus DC& is constrained to be continuous
D . A two- dimensional bicubic Hermite basis requires four
across element boundaries rather than
D*
derivatives per node
à ¼ B *D-v*!P8i15jU*,
/
à ( *,-A*!P*1A/P
7 *
/
7 * slope $/
/
F IGURE 1.12: Cubic Hermite basis functions.
The need for the second-order cross-derivative term can be explained as follows; If is cubic in *U(
and cubic in * , then Å *u ( is quadratic in *U( and cubic in * , and Å * is cubic in *u( and quadratic
in * . Now consider the Å side 1–3 w in Figure 1.13.w The Å of with * is specified by
cubic variation
the four nodal parameters )( , Å
*Å x ( , " and Å Å * x " . But since Å Å *u( (the normal derivative) is
also cubic in * along that side and is entirely independent of these four parameters, w four additional
w
parameters are required to specify this cubic. Two of these are specified by Å and Å
w w Å U
* u
( x ( u
*
Å " u
( x ,
*
w node i node h
Å *u(ux
Å "
w
Å *u(ux *u(
Å ( node / node j
where
à ¼( B *D 2/ 15iu* ®jU* "
à (( B*D *@*1:/P
à ¼ B*D * Bi1¿jU*, (1.17)
à ( B*D * +*1A/P
are the one-dimensional cubic Hermite basis functions (see Figure 1.12).
As in the one-dimensional case above, to preserve derivative continuity in physical x-coordinate
space as well as in * -coordinate space the global node derivatives need to be specified with respect
to physical arclength. There are now two arclengths to consider: &u( , measuring arclength along the
*u( -coordinate, and & , measuring arclength along the * -coordinate. Thus
w w w &u(
Å *u(ux Å &u(ux Å *u(6x
w Å E w Å O
I Ç
R u
E T È
V
W Ä w Å& E
Å* x Å& x Å*x
w Å E w Å O
I Ç
R u
E T È
V W Ä Å w E w C&
(1.18)
C& u (
*Å uÅ ( Å * x E Å &uÅ ( Å & x IORÇEuT VXW Ä !*u(ux E Ä !* x E
w C&u( w 4&
where
!*u(ux E and !* x E are element scale factors which scale the arclength derivatives of
global node K to the * -coordinate derivatives of element node G .
The bicubic Hermite basis is a powerful shape descriptor for curvilinear surfaces. Figure 1.14
shows a four element bicubic Hermite surface in 3D space where each node has the following
twelve parameters
ªN Å &6 ( N Å & N u& Å ( & YN ¦N Å &6¦ ( N Å & ¦ N u& Å ( ¦ & N ÉN Å &uÉ ( N Å & É and Å&6( É &
Å Å Å Å Å Å Å Å Å Å Å Å
1.7 Triangular Elements
Triangular elements cannot use the *u( and * coordinates defined above for tensor product elements
(i.e., two- and three- dimensional elements whose basis functions are formed as the product of one-
dimensional basis functions). The natural coordinates for triangles are based on area ratios and are
called Area Coordinates . Consider the ratio of the area formed from the points j , i and Ê:ªNY¦
in Figure 1.15 to the total area of the triangle
/ ¦ Î
Ë (- Area ÌAÊjuiÍ
Area ÌM/Pj6iÍ
j ¹¹ / ¦ ¹¹ K.¨À(>t¥(z¿<(z¦ Î 8j6Kn
/
¹¹ / " ¦ " ¹¹
¹¹ ¹¹
1.7 T RIANGULAR E LEMENTS 15
É
12 parameters per node
*
¦ *u(
F IGURE 1.14: A surface formed by four bicubic Hermite elements.
P( ,¦ )
jÐ YN ¦
/ B)(NY¦,(
Ë (k7
Ë (- (
Ë (- "
Ë (qr/ "
( / )(Ò¦C(
where KÑ
¹¹¹ // ¦¦ ¹¹¹ is the area of the triangle with vertices /Pjui , and (q ¦ " 13 " ¦ N¥(Ó
¦ 13¦ " NY<(-v ¹¹ " Ë 1a " . " ¹¹
Notice that¹ ( is linear ¹ in and ¦ . Similarly, area coordinates for the other two triangles
containing Ê and two of the element vertices are
/
= -( Ë (ªÀj Ë O( 1A/P
= Ë Ë 1A/P ²
= Ë Àj Ë 1A/P
= l" vh Ë " ( ÀË j " h
=O± vh Ë Ë
= vh Ë Ë "
(
"
¬ j i
F IGURE 1.16: Basis functions for a six node quadratic triangular element.
ample, can be modelled geometrically using one element with cylindrical polar 8×PNØ! -coordinates,
e.g., the annular plate in Figure 1.17a has two global nodes, the first with ׫Ñ×u( and the second
with ×v× .
¦ Ø *
jUÙ j i
j i
/ j
/ h × / h *u(
7 ×u( ×
(a) (b) (c)
F IGURE 1.17: Defining a circular annulus with one cylindrical polar element. Notice that element
^
vertices and in
Ú¥Û0Üu
-space or
[ ( Ûz[
-space, as shown in (b) and (c), respectively, map onto the
single global node in ^
£ÛzÝ,
-space in (a). Similarly, element vertices and map onto global Þ
node .
Global nodes / and j , shown in BªNY¦ -space in Figure 1.17a, each map to two element vertices
in 8 ×PNØ! -space, as shown in Figure 1.17b, and in B*U(N¤* -space, as shown in Figure 1.17c. The
B×PNØ6 coordinates at any B*U(N¤* point are given by a bilinear interpolation of the nodal coordinates
ץE and إE as
× = EÂB*U(N¤* ×¥E
Ø = EÂB*U(N¤* Ä Ø¥E
Ä
=
where the basis functions E@*u(N* are given by (1.6).
Three orthogonal curvilinear coordinate systems are defined here for use in later sections.
Cylindrical polar B×PNØ4NÉ! :
×Oߥà!áØ
¦×Oá¤âäã¾Ø (1.19)
ÉÉ
Spherical polar 8×PNØCNåy :
{×Jߥà6áØOߥà!áå
¦{×Já¤âäã@Ømߥà!á£å (1.20)
É{×Já¤âäã¾å
18 F INITE E LEMENT BASIS F UNCTIONS
`Óߢà!á¤è¾æ¾ß¢à!áç
¦Óáâ
ãè@æ@á¤â
ãÂçߢà!áFØ (1.21)
ÉQÓáâ
ãè@æ@á¤â
ãÂçáâ
ãÂØ
y
Ø
r
z ç
x
F IGURE 1.18: Prolate spheroidal coordinates.
The prolate spheroidal coordinates rae illustrated in Figure 1.18 and a single prolate spheroidal
element is shown in Figure 1.19. The coordinates Èæ>NYçmNØ6 are all trilinear in B*U(N¤* N¤* . Only four
"
global nodes are required provided the four global nodes map to eight element nodes as shown in
Figure 1.19.
1.8 C URVILINEAR C OORDINATE S YSTEMS 19
(a) (b)
j h ¦
É *
*U(
1 *"
3
(c) ç (d)
*
é7o j h j h
j *u( h
j h / i
7 æ *"
/ i / i
jUÙ
/ i
Ø
F IGURE 1.19: A single prolate spheroidal element, shown (a) in
+ÛzÝ4Û0ê6
-coordinates, (c) in
8ëÛzì>Û0Üu
-coordinates and (d) in
+[ ( Ûz[ Ûz[ "
-coordinates, (b) shows the orientation of the
[í
-coordinates on the prolate spheroid.
20 F INITE E LEMENT BASIS F UNCTIONS
6
4 5
1 2
3
1 2
7. To define a bilinear mesh in cylindrical polar coordinates run CMISS example /Pj6j .
1.9 CMISS E XAMPLES 21
1 3
7 9
2
1 3 2 4
8 5
4 10
6
1 11 3
7 13 9
2
1 5 3 6 2 4
5
4 12 8 14 10
6
3
4
1 2
or
w D ¿ï8ªN¤-7
!
1 î ! x
where is temperature, ï28ªN¤ the heat sink and î the thermal conductivity (ð3ñUò¤òYá
ÎuóôÎ6õö ).
Consider the case where ï
w D
1 ! î ! x ¿v7 7Ìt§Ì/ (2.1)
3. Introduce the finite element approximation for the temperature field with nodal parameters
and element basis functions.
4. Integrate over the elements to calculate the element stiffness matrices and RHS vectors.
û( w D ý ýÕÿ
1 ! î ! x ¿ !«7 (2.5)
¼ þ
ý of as forcing the residual or error to
This formulation of the governing equation can be thought
be zero in a spatially averaged sense. More precisely, is chosen such that the residual is kept
orthogonal to the space of functions used in the approximation of (see step 3 below).
formula
û( ( û(
! ! £# ¼ 1
! !
¼ ¼
gives
û( ý w ý w ( û( w ý
! 1î ,DÓx ! 1}î D!q
x 1 1î D, D@x !
¼ ¼ ¼
and Equation (2.5) becomes
û ( w , ý ý
D ý (
î D D ¿ x ! î ! (2.6)
¼ ¼
2.1.3 Finite element approximation
We divide the domain 7aÌÑÌ / into 3 equal length elements and replace the continuous field
variable within each element by the parametric finite element approximation
û( û û û(
!« D ! !
¼ Ä ¼ Ä
Ä
Ä
1
Boris G. Galerkin (1871-1945). Galerkin was a Russian engineer who published his first technical paper on the
buckling of bars while imprisoned in 1906 by the Tzar in pre-revolutionary Russia. In many Russian texts the Galerkin
finite element method is known as the Bubnov-Galerkin method. He published a paper using this idea in 1915. The
method was also attributed to I.G. Bubnov in 1913.
26 S TEADY-S TATE H EAT C ONDUCTION
ûø û(
!` D*
Ä ¼ Ä
ø
where ¹ !D*¾¹ is the Jacobian of the transformation
from coordinates to * coordinates.
¹¹ ¹¹
¹ ¹
2.1.4 Element integrals
The element integrals arising from the LHS of Equation (2.6) have the form
û ( w D ý ý
î ! ! 5 x !* (2.7)
ý ¼
where ` = EuFE and = . Since = E and = are both functions of * the derivatives with respect
to need to be converted to derivatives with respect to * . Thus Equation (2.7) becomes
û ( w = D* = E !*
E î !* ! !* ! == E x D* (2.8)
¼
Notice that FE has been taken outside the integral because it is not a function of * . The term
D* is
/ !
evaluated by substituting the finite element approximation *,Ö = E4#nE . In this case i * or
!* ;i and the Jacobian is ! ( . The term multiplying the nodal parameters FE is called
D !* "
the element stiffness matrix, E
û ( w = !* = E D* û ( w = = E
EQ î !* ! !* ! =Ö= E x !*Õ î D* i D* i¾ =Ö= E x i/ D*
¼ ¼
where the indices and G are / or j . To evaluate E we substitute the basis functions
= ()B*D-$/213* = (
or
!*= r 1Q/
= B *D-{* /
or
!* r
2 .1 O NE -D IMENSIONAL S TEADY-S TATE H EAT C ONDUCTION 27
Node 1 2 3 4
Node 1
X X 0 0 U( X
Node 2
X X X 0 U
X
=
Node 3
0 X X X U
" X
Node 4
0 0 X X Ul X
F IGURE 2.1: The rows of the global stiffness matrix are generated from the global weight
functions. The bar is shown at the top divided into three elements.
Thus,
û/ ( é w = ( / û( é / wé /
(z(k î
= s
( D*Q
î 0
Q
1 P
/ s/213*, ú!*Q
î
i ¼ D*x i ¼ ÷ i i)x
and, similarly, ²
/ éw /
( (-
w i 1 Õî x
/ é /
z i î i)x
( é î ( ú ( 1 é îÕ ( ú
EQ
( " ÷1 é îÕ " ( ú " ( ÷ é î ( ú
" ÷ " ÷ "¬
¬
Notice that the element stiffness matrix is symmetric. Notice also that the stiffness matrix, in this
particular case, is the same for all elements. For simplicity we put înr/ in the following steps.
2.1.5 Assembly
The three element stiffness matrices (with î / ) are assembled into one global stiffness matrix.
This process is illustrated in Figure 2.1 where rows /6N%#ä# NYh of the global stiffness matrix (shown here
multiplied by the vector of global unknowns) are generalised from the weight function associated
with nodes /!N%# # NYh .
Note how each element stiffness matrix (the smaller square brackets in Figure 2.1) overlaps
28 S TEADY-S TATE H EAT C ONDUCTION
with its neighbour because they share a common global node. The assembly process gives
µ ±
1µ ( µ" µ
!#" #! "
± 7± 7 " Hk( "
1 ( "µ ±
µ 1 ( µ" µ 7 ± µ $ H $
7 1 ( µ"
± 1 µ( " H "
7 7 1 ( µ" H]l
Notice that the first row (generating heat flux at node / ) has zeros multiplying H and HOl since
"
nodes i and h have no direct connection through the basis functions to node / . Finite element
matrices are always sparse matrices - containing many zeros - since the basis functions are local
to elements.
The RHS of Equation (2.6) is
D ý ø ( w D ý w D ý
î ! î ! x ¹ 1 î ! x ¹ (2.9)
¼ ø ( ¼
¹¹ ø ý ¹¹ ø
ý ¹ ¹
To evaluate these expressions consider the weighting function corresponding
ý ý = to each global node
(see Fig.1.6). For node / ( is obtained from the basis function ( associated with the first node
ý element / and therefore &( % ¼ / . Also, since ( is identically zero outside element / ,
('% ø ( 7 . Thus Equation (2.9)ø for node / reduces to
of
D ý ø ( w D
î ! ( r1 î !qx ¹ = flux entering node / .
ø ¼ ¹¹ ø ¼
Similarly, ¹
D ý ø (
î ! E 7 (nodes j and i )
ø ¼
and
,ý ø ( w D
î D l î !qx ¹ = flux entering node h .
ø ¼ ¹¹ ø (
¹
Note: î has been left in these expressions to emphasise that they are heat fluxes.
Putting these global equations together we get
w D !"
µ ± !#" !#"
1 î !qx ¹ """
± 1µ ( µ" µ 7± 7 " Hk( " ¼ "
7 ¹¹¹ ø "
1 ( µ" ± µ 1 ( "µ µ 7 ± H
7 1 ( "µ ± 1 µ ( " µ $ H "$ w D 7
(2.10)
7 7 1 ( µ" ]H l î !qx ¹
$
¹¹ ø (
or (*) ¹
,+
2 .1 O NE -D IMENSIONAL S TEADY-S TATE H EAT C ONDUCTION 29
( )
where is the global “stiffness” matrix, the vector of unknowns and + the global “load” vector.
Note that if the governing differential equation had included a distributed source term that was
independent of , this term would appear - via its weighted integral - on the RHS of Equation (2.10)
rather than on the LHS as here. Moreover, if the source term was a function of , the contribution
from each element would be different - as shown in the next section.
2.1.8 ² Fluxes
The fluxes at nodes / and h are evaluated by substituting the nodal solutions Hk(q7 , H 74#Sj 6¶ ¶ ,
H v7# 7 é!¶ and HOlr/ into Equation (2.10)
"
²
w D
flux entering node /¾r1 î !Óxô¹ $1¾7# ¶ h é ( înr/ ; exact solution 7#
¶ 67 é )
w D ¹¹ ø ¼
h î ! x ¹ ¹ $/!#Çi/P ·
flux entering node Q ( înr/ ; exact solution /!#Çi/<i/ )
¹¹ ø (
¹
These fluxes are shown in Figure 2.2 as heat entering node h and leaving node / , consistent with
heat flow down the temperature gradient.
30 S TEADY-S TATE H EAT C ONDUCTION
/ /!#Çi/P ·
²
7# 7 é6¶
² 7#Sj !¶ ¶
74# ¶ h é (
7 " " /
F IGURE 2.2: Finite element solution of one-dimensional heat equation.
û ( w D ý ý
D ý ( û( ý
î ! ! ¿ x D` î ! ! (2.11)
¼ ¼ ¼
where the -dependent source term appears on the RHS because it is not dependent on . Replacing
the domain integral for this source term by the sum of three element integrals
û( ý û ý û ý û( ý
D` ! ! !
¼ ¼
ý
where
(
is chosen to be the appropriate basis function within each element. For example, the first
û( /
é @* 0/21a*,Â!*Q u/ h
¼
and
û( /
é * !*Q j / ·
¼
± (l
Thus, the contribution to the element / RHS vector from the source term is (´ .
Similarly, for element j ,
û( / j· û( /
´
é 0 /Ó5*,ys/213*,!*Q j and é 0 Ö
/ °
,
* C
-
* D* Q uh gives ± ± l
¼ ¼
and for element i ,
û( / · û( /
± ´ l±
é Àj25*,ys/213*,!*Q U h and é À j@°*,C*-D*Q u h gives ±l
¼ ¼
Assembling these into the global RHS vector, Equation (2.10) becomes
w D !"
±
µ ± µ !#" !#"
1 î "
"
± (l !#"
1 ± ( " 7 ± 7 Hk(" " !qx ¹¹ ø ¼ "" "
±± ( ´
1 ( "µ ± 1 ± ( µ" 7 ± H 7 ¹¹ " ´ ´
¬
7 1 ( µ" ± 1 µ ( "µ $ H " $ w , 7 l ± ± l $
7 ¬
7 1 (" µ H]l $ ±l
î Dqx ¹
¹¹ ø (
¹
2.3 The Galerkin Weight Function Revisited
A key idea in the Galerkin finite element method is the choice of weighting functions which are
orthogonal to the equation residual) (thought of here as the error or amount by which the ) equation
)
/102/
fails to 02
be/ exactly zero). This idea is illustrated in Figure 2.3.
In Figure 2.3a an exact vector V (lying in 3D space) )4 is3 approximated by a vector )
where is a basis vector along the first coordinate axis (representing one degree of freedom
in the system). The difference between the exact vector and the approximate vector is the
32 S TEADY-S TATE H EAT C ONDUCTION
=
(a) (b)
" (c)
"
ü = ü
V
)
)( = ( ) )
permitted in the approximation >( = (k{ = ) v = with the result that the residual (now
and hence to . Finally, in Figure 2.3c, a third ) degree
)=3 of freedom (a third axis in Figure 2.3c) is
where î NîA@ and îCB are the thermal diffusivities along the , ¦ and É axes respectively. If the
ø
material is assumed to be isotropic, î î2
@ îÂB î , and the above equation can be written as
ø
1ED ÀîFDn-7 (2.13)
Ä
and, if î is spatially constant (in the case of a homogeneous material), this reduces to Laplace’s
equation îFD ¨7 . Here we consider the solution of Equation (2.13) over the region G , subject
to boundary conditions on H (see Figure 2.4).
Solution region: G
[ [
Ý ^ _
¸ ^ ¸
I " I l
] ]
_
¸ [( ³^ [(
^ ^
I l
I
" ^
]
³
[ [
I
^ ] ^ ]
³
I ( Þ (
I I
^
^ [( Þ
^ [(
^
F IGURE 2.5: Mapping each I to the [ ( Ûz[ plane in a 2` n element plane.
For each element, the basis functions and their derivatives are:
=
= (k¨0/Â13*u(¢s/21a*
Å *u( ( $ 1s/213* (2.19)
Å= (
Å * $1s/213*u( (2.20)
Å (2.21)
= v =
*U(¢s/Â13* Å *u( $/213* (2.22)
Å= (
Å * $12*u( (2.23)
Å (2.24)
= ¨ =
" 0/Â13*u(z* Å *u( " $12* (2.25)
Å=
Å * " $/213*u( (2.26)
Å (2.27)
=
= lÂv*U(z*
Å *u( l {* (2.28)
Å= l
Å * {*u( (2.29)
Å
36 S TEADY-S TATE H EAT C ONDUCTION
2.6 Integration
The equation is
ûK ý ûR ý
îSDn D G: -î Å G H (2.30)
Ä Å
w ý ý
i.e.,
ûK ûR ý
î Å Å Å ¦
Å ¦ x G: H (2.31) î ÅG
Å Å Å
Å ý Å
u has already been approximated by = E6FE
andý is a weight function but what should this be
chosen to be? For a Galerkin formulation choose = i.e., weight function is one of the basis
functions used to approximate the dependent variable.
This gives
ûK w = E = = E = ûR
FE î Å Å Å ¦ Å ¦ x G : î Å G = H (2.32)
í Å Å Å Å Å
where the stiffness matrix is E where r/!N%#%¥# #¢NYh and G$/6N%#%#%#>NYh and a is the (element)
load vector.
The names originated from earlier finite element applications and extension of spring systems,
i.e., M
a îC where î is the stiffness of spring and a is the force/load.
This yields the system of equations E!FE ba . e.g., heat flow in a unit square (see Fig-
ure 2.6).
¦ B *
/
7 / *u(
F IGURE 2.6: Considering heat flow in a unit square.
2 .7 A SSEMBLE G LOBAL E QUATIONS 37
û(û(
(z(qî
0/213¦ s/21a !£D¦
¼ ¼
ij î
and similarly for the other components of the matrix.
Note that if the element was not the unit square we would need to transform from BªNY¦ to
B*U(N¤* coordinates. In this case we would have to include the Jacobian of the transformation and
=í = = = =
also use the chain rule to calculate Å
. e.g., Å E Å u* (E Å *u ( Å * E Å * Å * íE Å * í .
Å
The system of E!FEÕca becomes
LV
Å Å Å Å Å Å Å
1 ( ( (
!" !
" ( ( 1 ( ( 1 " ( (
"
"
)( """
1 " ¬ ( " 1 1 ¬" 1 1( 1 "( "
"
"
"
¬( 1 ( " ¬ ¬(1 " ¬( 1 ( (
"
"
"
" "
1
1( ¬" ( " ( " ( 1 1 1
¬ ( ( ( " ( ( """1 Fl " üedgf
1 ( 1 1 ( 1 "( 1 1 (
" " ¬( " ¬( " 1 1 1 ¬" 1 1 ( 1 "( " ± "
"
¬" "
¬1 " ¬ 1 ¬ ( ¬ 1 1( " ¬ " ¬ ¬ 1 "( ¬ 1 " "
¬ 1 ( ¬1 ¬ 1 ¬ ¬µ´
1 ¬" 1 ( 1 "( ( 1 "(( 1 " ( " ¬( " 1 ( $ $
¬1 " ¬ 1 ¬ 1 " ¬
¬ ¬ (2.34)
38 S TEADY-S TATE H EAT C ONDUCTION
¦
· ¶ é
h
/ j
/ j i
F IGURE 2.7: Assembling unit sized elements into a global stiffness matrix.
there are only i unknown values of u at nodes (2,5 and 8), therefore there is a i
ji matrix to solve.
The RHS is known at these three nodes (see below). We can then solve the iQj i matrix and then
¶
multiply out the original matrix to find the unknown RHS values.
The RHS is 7 at nodes j and because ûR it isý insulated. To find out what the RHS is at node
we need to examine the RHS expression Å G H $7 at node . This is zero as flux is always
Å
7 at internal nodes. This can be explained in two ways.
¾(
G G
n n
Correct way: H does not pass through node and each basis function that is not zero at is zero
on H
B*D
....
o p *
*U( .... * *
F IGURE 2.9: Gaussian quadrature. q
[U is sampled at r
Y
Gauss points [ ( Ûz[ ~~~0[ ~
Y
û(
Let B*D2!*Q (s*u(> *
¼ k k
û( û( û( û( û(
*,Â!*Q !*¾t *k!*@t * !*@¿ * " !* (2.35)
¼ ¼ ¼ ¼ ¼
Since , , and are arbitrary coefficients, each integral on the RHS of 2.35 must be integrated
exactly. Thus,
û(
D*Q?/ (# /Ó #/ (2.36)
¼ k k
û(
*-D*Q j/ (#»*u() #»* (2.37)
¼ k k
û(
* D*Q i/ (#»* ( #»* (2.38)
¼ k k
û(
* " D*Q h / (#»* (" #»* " (2.39)
¼ k k
These four equations yield the solution for the two Gauss points and weights as follows:
2 .8 G AUSSIAN Q UADRATURE 41
(k /
k k j#
Then, from (2.37),
* ?/@13*U(
and, substituting in (2.38),
* ( s/Â13*u(s ij
jU* ( 1¿jU*U() i/
7N
giving
*u(- ju/ t j v / i #
Equation (2.39) is satisfied identically. Thus, the two Gauss points are given by
*U(k j/ 1 j v / iN
* j/ v / N (2.40)
j i
(q j/
k k
A similar calculation for a th
degree polynomial using three Gauss points gives
*U(k j/ 1 xj/ w i N
(k / ¶
k
* j/ N hé (2.41)
k
* " j/ jl/ w iN
" /¶
k
2 For two- or three-dimensional Gaussian quadrature the Gauss point positions are simply the
values given above along each * í -coordinate with the weights scaled to sum to / e.g., for j x j Gauss
/
quadrature the h weights are all . The number of Gauss points chosen for each * í -direction is
h
governed by the complexity of the integrand in the element integral (2.8). In general two- and three-
dimensional problems the integral is not polynomial (owing to the Å L* Ví terms which come from the
Å
42 S TEADY-S TATE H EAT C ONDUCTION
inverse of the matrix Å * VLí ) and no attempt is made to achieve exact integration. The quadrature
error must be balanced Å against the discretization error. For example, if the two-dimensional basis
is cubic in the *u( -direction and linear in the * -direction, three Gauss points would be used in the
*u( -direction and two in the * -direction.
2. To solve for the steady state temperature distribution inside an annulus run CMISS example
i/Pj
3. To investigate the convergence of the steady state temperature distribution with mesh refine-
ment run CMISS examples i/%h/ , i/%h,j , i/%hDi and i/%h6h .
Chapter 3
3.1 Introduction
Having developed the basic ideas behind the finite element method, we now develop the basic ideas
of the boundary element method. There are several key differences between these two methods,
one of which involves the choice of weighting function (recall the Galerkin finite element method
used as a weighting function one of the basis functions used to approximate the solution variable).
Before launching into the boundary element method we must briefly develop some ideas that are
central to the weighting function used in the boundary element method.
{ z
{
( {
{
z (
F IGURE 3.1: Illustrations of unit force distributions E.
As G gets larger we can easily see that the area of application of the force becomes smaller
and smaller, the magnitude of the force increases but the total force applied remains unity. If we
imagine letting G}| ~ we obtain an idealised “point” force of unit strength, given the symbol
B , acting at = 0. Thus, in a nonrigorous sense we have
Bk E
â ó y |E@B the Dirac Delta“function”.
This
is not a function that we are used to dealing with because we have 7 if 7
and “ 87D
~ ” i.e., the “function” is zero everywhere except at the origin, where it is infinite.
ûy ûy
However, we have BÂ!$/ since each |E2@!?/ .
y y
ù ù
The Dirac delta “function” is not a function in the usual sense, and it is more correctly referred
to as the Dirac delta distribution. It also has the property that for any continuous function B
ûy
B +2D`Q À7D (3.1)
y
ù
3 .2 T HE D IRAC -D ELTA F UNCTION AND F UNDAMENTAL S OLUTIONS 45
ù w / /
E&
â ó y G j B*D G j by the Mean Value Theorem, where *O 1 G N ÖG x
w / /
87D
since *O 1 G N Gx and as G| ~N)*| 7
The above result (Equation (3.1)) is often used as the defining property of the Dirac delta in
more rigorous derivations. One does not usually talk about the values of the Dirac delta at a
particular point, but rather its integral behaviour. Some properties of the Dirac delta are listed
below
ûy
B*13Q BÂ!Q *D (3.2)
y
ù
(Note: +*13 is the Dirac delta distribution centred at {* instead of `7 )
d
B*1¯q ½ B*1 (3.3)
d
B*1 =
7 if *ÌP
(i.e., the Dirac Delta function is the slope of the Heaviside2
where
/ if *ÍP
step function.)
*1aªNW1°¦- B*13 1°¦ (3.4)
(i.e., the two dimensional Dirac delta is just a product of two one-dimensional Dirac deltas.)
have well-known fundamental solutions (see Appendix 3.16). We briefly illustrate here how to find
a simple fundamental solution.
Consider solving the Laplace Equation Å Å ¦ 7 in some domain G}@ .
The fundamental solution for this equation Å (analogousÅ to a particular solution in ODE work) is
a solution of
ý ý
Å Å ¦ B*1aªWN 1°¦-v7 (3.5)
Å Å
in 2 (i.e., we solve the above without reference to the ý original domain G or original boundary
conditions). The method is to try and find solution to Dn M7 in 2 which contains a singularity
at the point *WN . This is not
as difficult as it sounds. We expect the solution to be symmetric
about the point *WN since *1aªWN 1°¦ is symmetric about this point. So we adopt a local
polar coordinate system about the singular point B*WN .
Let
For ×ÍA74N *13ªNW1°¦m7 and owing to symmetry, Å
is zero. Thus Equation (3.6) becomes
w ý ÅØ
/ Å × Å 7
× Å× Å× x
This can be solved by straight (one-dimensional) integration. The solution is
ý
cgäà-×27 (3.7)
B*Ns
radius ÍA7 centred at ×Õv7 (Figure 3.2) then from the Green-Gauss theorem
û ý û ý f
D L ÅG Å is the surface of the disk
Å
û ý
f
Å since is a disk centred at ×7 so G and × are in the same direction
Å ×
jUÙ from Equation (3.7), and the fact that is a disc of radius
jU Ù
Therefore, from Equation (3.8)
$1 jU/ Ù #
So we have
ý
r1 jU/ Ù
àk×27
remains arbitrary but usually put equal to zero, so that the fundamental solution for the two-
dimensional Laplace Equation is
ý w
r1 Uj / Ù
à-× jU/ Ù äà )×/ x (3.9)
48 T HE B OUNDARY E LEMENT M ETHOD
This equation contains only boundary integrals (and no domain integrals as in Finite Elements)
and is referred to as a boundary integral equation. It relates the value of at some point inside
the solution domain to integral expressions involving and Å over the boundary of the solution
G
domain. Rather than having an expression relating the value Å of at some point inside the domain
to boundary integrals, a more useful expression would be one relating the value of at some point
on the boundary to boundary integrals. We derive such an expression below.
The previous equation (Equation (3.13)) holds if B*NW 4gG (i.e., the singularity of Dirac Delta
function is inside the domain). If +*Ns
is outside G then
ûK ý ûK
¢D LG®r1 B *13ªNs13¦2 Gt7
since the integrand of the second integral is zero at every point except *NW and this point is
outside the region of integration. The case which needs special consideration is when the singular
point *NW
is on the boundary of the domain G . This case also happens to be the most important
for numerical work as we shall see. The integral expression we will ultimately obtain is simply
+*Ns / *NW .
Equation (3.13) with replaced by
j We can see this in a non-rigorous way as
follows. When B*NW was inside the domain, we integrated around the entire singularity of the
Dirac Delta to get +*NW in Equation (3.13). When B*NW is on the boundary we only have half of
the singularity contained inside the domain, so we integrate around one-half of the singularity to
/ +*NW . Rigorous details of where this coefficient /
get
j j
comes from are given below.
ûK ý
Let Ê denote the point B*NW¤¥G . In order to be able to evaluate ¢D G in this case we
enlarge G to include a disk of radius about Ê (Figure 3.3). We call this enlarged region G ½ and
let H ½ H
ù ¦§ H ¦ .
Now, since Ê is inside the enlarged region G ½ , Equation (3.13) holds for this enlarged domain
ý
i.e.,
RA¨&©«û ª&R © RA¨&¬© û ª'R ©
ý
BÊQ) Å H Å H
ÅG ÅG
(3.14)
H
G ½
¦
H
ù ¦
Ê
G
F IGURE 3.3: Illustration of enlarged domain when singular point is on the boundary.
Firstly consider
û
R ©
ý û
R © w ý
ÅG H Å G 1 Uj / Ù äàk× x H by definition of
Å Å
û
R © w /
Å × 1 jUÙ
àk× x H since Å G¯® Å × on H
¦
Å Å Å
Rû ©
$1 jU/ Ù H
×
$1 jU/ Ù / R û © H since × on H
¦
| 1 jU/ Ù /
BÊ,Ù
ó
â R û © ý Å ó w /
¼ H cäâ ¼ 1 Uj Ù Å G ÀÊQCÙ
à x 7
ÅG
(3.16)
¦M ¦M Å
since
â
ó ¼ as
â ó ¦M ¼ .
¦±°³²«´A¦M
3 .3 T HE T WO -D IMENSIONAL B OUNDARY E LEMENT M ETHOD 51
ó û
RA¨©
äâ ¼µ¶ (nice integrand) · ¸ ¹ºb»¼ ½ (nice integrand) · ¸
¦
¼½ Î · ¸h»bÚ Ù ¼½ Î Ï
· ¸
ÏÆÕÖ× Ø Ï Ñ ÏuÕÖÛ×£Ø
Î Ð
ÎÐ Ñ
or
ÚÙ ¼½ Î · ¸h» ¼½ Î Ï
U ·L¸
ÏÆÕÖ× Ø Ï Ñ
Î Ð
ÎÐ4Ñ
æ ¼½ Î · ¸g» ¼½ Î Ï
· ¸ (3.19)
ÕMÖ×XÏuÕÖÛ×£Ø Ï Ñ
Î Ð
U ÎUÐ Ñ
where
» ù ãÙ
ê
Ñ
ê2»ìë çî çò ç6û
ÕÜ ×ï Ø ÕÞ
× ï ØÕú ×ï
if
ö Ùø Öüßgà
æ » ôõ if ¸ and ¸ smooth at
ÕÖ× inner solid angle Öüß Ö
ù ïã if ¸ and ¸ not smooth at
õ÷ Öüß Ö
Equation (3.19) involves only the surface distributions of and ÎÏ and the value of at a
Ï Ï
ÎUÐ
point . Once the surface distributions of and ÎÏ are known, the value of at any point
Ö Ï Ï Ö
ÎÐ
inside can be found since all surface integrals in Equation (3.19) are then known. The procedure
à
is thus to use Equation (3.19) to find the surface distributions of and ÎÏ and then (if required)
Ï
use Equation (3.19) to find the solution at any point . Thus we solve ÎÐ for the boundary data
Ößýà
first, and find the volume data as a separate step.
Since Equation (3.19) only involves surface integrals, as opposed to volume integrals in a finite
element formulation, the overall size of the problem has been reduced by one dimension (from
volumes to surfaces). This can result in huge savings for problems with large volume to surface
ratios (i.e., problems with large domains). Also the effort required to produce a volume mesh of a
complex three-dimensional object is far greater than that required to produce a mesh of the surface.
Thus the boundary element method offers some distinct advantages over the finite element method
in certain situations. It also has some disadvantages when compared to the finite element method
and these will be discussed in Section 3.6. We now turn our attention to solving the boundary
integral equation given in Equation (3.19).
3þ .4 N UMERICAL S OLUTION P ROCEDURES FOR THE B OUNDARY I NTEGRAL E QUATION 53
¸ÿ» ¸ (3.20)
ø
(a) (b)
F IGURE 3.5: Schematic illustration of a boundary element mesh (a) and a finite element mesh (b).
æ ¼½ Î · ¸g» ¼½ Î Ï
U ·L¸ (3.21)
ÕÖÛ×XÏÆÕÖ× Ø ø Ï Ñ ø
Î Ð
ÎÐ Ñ
This gives
æ
¼½
Î ·L¸h»
¼½
· ¸ (3.23)
ÕMÖ×XÏÆÕÖ×NØ ø
Ï Ñ ø
Î Ð
U Ñ
This equation holds for any point on the surface ¸ . We now generate one equation per node by
Ö
putting the point to be at each node in turn. If is at node , say, then we have
Ö Ö
æ
¼½
Î · ¸g»
¼½
· ¸ (3.24)
Ï Ø ø
Ï Ñ ø
Î Ð
Ñ
where is the fundamental solution with the singularity at node (recall is çÒÚäÙ ã ÂÄèélê , where
ê is theÑ distance from the singularity point). We can write Equation (3.24)Ñ in a more abbreviated
form as
æ
»
(3.25)
Ï Ø ø
Ï ø
where
» ¼½
Î · ¸ and » ¼½
· ¸ (3.26)
Ñ
Î Ð
U Ñ
Equation (3.25) is for node and if we have nodes, then we can generate equations.
)
We can assemble these equations into the matrix system
») ) (3.27)
(compare to the global finite element equations » ) where the vectors and are the vectors
of nodal values of and
. Note that the component of the matrix in general is not and
th
Ï
similarly for .
At each node, we must specify either a value of or
(or some combination of these) to have a
Ï
well-defined problem. We therefore have equations (the number of nodes) and have unknowns
to find. We need to rearrange the above system of equations to get
!#"
» (3.28)
"
where is the vector of unknowns. This can be solved using standard linear equation solvers,
although specialist solvers are required if the problem is large (refer [todo : Section ???]).
!
The matrices and (and hence ) are fully populated and not symmetric (compare to the
finite element formulation where the global stiffness matrix is sparse and symmetric). The
size of the and matrices are dependent on the number of surface nodes, while the matrix
is dependent on the number of finite element nodes (which include nodes in the domain). As
3þ .5 N UMERICAL E VALUATION OF C OEFFICIENT I NTEGRALS 55
mentioned earlier, it depends on the surface to volume ratio as to which method will generate the
smallest and quickest solution.
The use of the fundamental solution as a weight function ensures that the
and matrices
are generally well conditioned (see Section 3.5 for more on this). In fact the matrix is diagonally
!
)
dominant (at least for Laplace’s equation). The matrix is therefore also well conditioned and
Equation (3.28) can be solved) reasonably easily.
"
The vector contains the unknown values of and on the boundary. Once this has been
found, all boundary values of and are known. If a solution is then required at a point inside the
domain, then we can use Equation (3.25) with the singular point located at the required solution
Ö
point i.e.,
$
$
»
ç
% (3.29)
ÏÆÕÖ× ø
ø
Ï
The right hand side of Equation (3.29) contains no unknowns and only involves evaluating the
surface integrals using the fundamental solution with the singular point located at .
Ö
We consider in detail here how one evaluates the and integrals for two-dimensional problems.
These integrals typically must be evaluated numerically, and require far more work and effort than
the analogous finite element integrals.
Recall that
»¼½
Î ·L¸ and » ¼½
·L¸
Ñ
Î Ð
Ñ
where
»ç ÚäÙ ã ÂÄèAé;ê
Ñ
ê » distance measured from node
· ¸ ·.- ·î · ò
» » » / ï / ï (3.32)
(QÕÜC× · · · Ø ·
Ü Ü Ü10 Ü'0
·Cî ·Cò
where - represents the arclength and and can be found by straight differentiation of the
· ·
interpolation expression for î and ò Ü . Ü
ÕíÜ × ÕíÜ ×
The fundamental solution is
»ç Ú'Ù ã ÂÄèé ê
Õ ÕÜ ×É×
Ñ
ê » ë î çî ò çò
ÕÜC× Õ ÕÜC× ×
ïxØcÕ ÕÜ × ×
ï
where î ò are the coordinates of node .
Õ Ý ·C× ê
To find we note that
·
Ð
·Cê 5478 6
»32 ê (3.33)
·
Ð
where 8 6 is a unit outward normal vector. ã To find a unit normal vector, we simply rotate the tangent
vector (given by î:9 ò.9 ) by Ú in the appropriate direction and then normalise.
Õ ÕíÜ ×¢Ý ÕíÜ ×É×
Thus every expression in the integrands of the and integrals can be found at any value of
, and the integrals can therefore be evaluated numerically using some suitable quadrature schemes.
Ü
If node is well removed from element ¸ then standard Gaussian quadrature can be used to
evaluate these integrals. However, if node is in ¸ (or close to it) we see that ê approaches 0
and the fundamental solution tends to ; . The integral still exists, but the integrand becomes
singular. In such cases specialÑ care must be taken - either by using special quadrature schemes,
large numbers of Gauss points or other special treatment.
The integrals for which node lies in element ¸ are in general the largest in magnitude and
lead to the diagonally dominant matrix equation. It is therefore important to ensure that these
integrals are calculated as accurately as possible since these terms will have most influence on the
solution. This is one of the disadvantages of the BEM - the fact that singular integrands must be
accurately integrated.
A relatively straightforward way to evaluate all the integrals is simply to use Gaussian quadra-
ture with varying number of quadrature points, depending on how close or far the singular point is
from the current element. This is not very elegant or efficient, but has the benefit that it is relatively
easy to implement. For the case when node is contained in the current element one can use special
quadrature schemes which are designed to integrate log-type functions. These are to be preferred
when one is dealing with Laplace’s equation. However, these special log-type schemes cannot be
so readily used on other types of fundamental solution so for a general purpose implementation,
Gaussian quadrature is still the norm. It is possible to incorporate adaptive integration schemes
that keep adding more quadrature points until some error estimate is small enough, or also to sub-
divide the current element into two or more smaller elements and evaluate the integral over each
3þ .6 T HE T HREE -D IMENSIONAL B OUNDARY E LEMENT M ETHOD 57
¸ ¸
node
ê
ê
node
(a) (b)
F IGURE 3.6: Illustration of the decrease in < as node = approaches element > .
subelement. It is also possible to evaluate the “worst” integrals by using simple solutions to the
governing equation, and this technique is the norm for elasticity problems (Section 4.8). Details
on each of these methods is given in Section 3.8. It should be noted that research still continues in
an attempt to find more efficient ways of evaluating the boundary element integrals.
expressions given for and apply equally well to the three-dimensional case. The only real
difference between the two procedures is how to numerically evaluate the terms in each integrand
of these coefficient integrals.
As in Section 3.5 we illustrate how to evaluate each of the terms in the integrand of and .
58 T HE B OUNDARY E LEMENT M ETHOD
» ¼½
Î · ¸
Ñ
Î Ð
ø ø
ø ø ·Cê ø
» ¼ ¼
Î · ø · (3.34)
ÕÜ ÝsÜ × Ñ ê íÕ Ü ÝðÜ × · &?(QÕÜ ÝsÜ @× & Ü Ü
È È ï Î ï Ð ï ï
» ¼½
· ¸
Ñ
ø ø
» ¼ ¼
ø ø ø · ø · (3.35)
ÕÜ ÝsÜ × ÕÜ sÝ Ü A × &?(QÕ Ü sÝ Ü ×,& Ü Ü
È È ï Ñ ï ï ï
the BEM - construction of meshes for complicated objects, particularly in 3D, is a very time
consuming exercise.
¼ ÂÄèAé · will be required. It is relatively common to use logarithmic schemes for this.
ÕíÜ × ÕíÜ × Ü
È H
These are obtained by approximating the integral as
ø
¼ Âèé ·
ÕÜ × ÕÜ × ÜNJ ø ÕÜ ×OL
È H H
àVU
àVY
¸
U
¸
¸XW Y
» FEM region
àVY
» BEM region
àVU
¸ » FEM boundary
Y
¸ » BEM boundary
U
¸ W=»
X interface boundary
where is a vector of the nodal values of and is a vector of the nodal values of Î Ï
Ï
The FEM matrices for can be written as) ÎUÐ
à Z
»[ (3.37)
3þ .11 C OUPLING THE FE AND BE TECHNIQUES 63
If we recall what the elements of in Equation (3.37) contained, then we can convert in
Equation (3.38) to an equivalent load vector by weighting the nodal values of by the appropriate
basis functions, producing a matrix ] i.e., » ]B
\
U
Therefore Equation (3.38) becomes )
¾ ø a`
] ^_ » ]B [
b »
U
i.e., )
»[
U U ¾ ø
where »3]c
U
an equivalent stiffness matrix obtain from BEM.
Therefore we can assemble this together with original FEM matrix to produce an FEM-type
system for the entire region .
à U
V
Notes:
1. is in general not symmetric and not sparse. This means that different matrix equation
U
solvers must be used for solving the new combined FEM-type system (most solvers in FEM
codes assume sparse and symmetric). Attempts have been made to “symmetricise” the
` U
matrix - of doubtful quality. (e.g., replace by Ú Ù ^d çegf - often yields inaccurate
U U U
results). )
1. This approach does not require any matrix inversion and is hence easier (cheaper) to imple-
ment
2. Existing BEM solvers will not assume symmetric or sparse matrices therefore no new matrix
solvers to be implemented
64 T HE B OUNDARY E LEMENT M ETHOD
2. Must find a (complete) set of functions (If you just use usual approximations for matrix
Ï
system is not diagonally dominant so not so good)
3. Method is not so popular - Green’s functions more widely available that complete systems
æ ¼ Î · ¸g» ¼ Î Ï
· ¸
ÕMÖ×XÏuÕÖÛ×£Ø jlk Ï Ñ j k
Î Ð
U ÎUÐ Ñ
with » çbÚäÙ ã ÂÄèélê
Ñ
The usual procedure is to put point at each solution variable node - creating an equation for each
Ö
node. This leads to singular integrands.
Another possibility is to put point outside of the domain - this yields
Ö à
¼ Î · ¸h» ¼ Î Ï
· ¸
j k Ï Ñ m
n j k
ÎÐ ÎÐ
m
3þ .13 S YMMETRY 65
¼½
Î · ¸g»
¼½
· ¸
ø
Ï Ñ m
n ø
ÎUÐ Ñnm
1. This method does not involve singular integrands, so that integrals are inexpensive to calcu-
late.
2. There is considerable choice for the location of the point . Often the set of Equations
Ö
generated are ill-conditioned unless chosen carefully. In practise is chosen along the unit
Ö Ö
outward normal of the surface at each solution variable node. The distance along each node
is often found by experimentation - various research papers suggesting “ideal” distances
(Patterson & Shiekh).
4. The idea of placing the singularity point away from the solution variable node is often of
Ö
use in other situations e.g., Exterior Acoustic Problems. For an acoustic problem (governed
by Helmholtz Equation 2 » S ) in an unbounded region the system of Equations pro-
ï Ïp
Ø o ï Ï
duced by the usual (singular) BEM approach is singular for certain “fictitious” frequencies
(i.e., certain values of ). To overcome this further equations are generated (by placing the
o
singular point at various locations outside ). The system of equations are then overde-
Ö à
termined and are solved in a least squares sense.
3.13 Symmetry
Consider the problem given in Figure 3.8 (the domain is outside the circle). Both the boundary
conditions and the governing Equation exhibit symmetry about the vertical axis. i.e., putting î to
ç
î makes no difference to the problem formulation. Thus the solution q î û has the property
Õ Ý ×
that q î û r » q ç¤î û î . This behaviour can be found in many problems and we can make
Õ Ý × Õ Ý .
× s Ú % t
use of this as follows. The Boundary Element Equation is (with Mb» (i.e., M is even) constant
elements)
¼½
ÚÙ Î · ¸ÿ»
¼½ · ¸ » M (3.39)
Ï Ø ø Ï Ñ ø Ù Ý*u*uvuNÝ
Î Ð
Ñ
66 T HE B OUNDARY E LEMENT M ETHOD
û
¾yx{z
q »w
2 qb»\- q
ï ï
î
We have M Equations and M unknowns (allowing for the boundary conditions). From symmetry
we know that (refer to Figure 3.9).
ø¾ t
» ñ» (3.40)
Ï Ï:|,} Ù Ý*uvu*uNÝ
So we can write
ö ö
ÚÙ ôõ ¼½ Î · ¸ ½@O%¼ Ì Î · ¸ õ »
ôõ ¼½ · ¸ ½,O%¼ Ì · ¸ õ (3.41)
Ï Ø ~ ø Ï Ñ Ø Ñ
~ ø Ø
Î Ð
Î Ð
Ñ Ñ
õ÷ õ õ÷ õ
t t
for nodes » . (The Equations for nodes \» M are the same as the Equations
Ù Ý*uvu*uNÝ t t Ø Ù Ý*t u*u*u Ý
for nodes ñ» ). The above Equations have only unknowns.
Ù Ý*uvu*u Ý
If we define
» ¼½ Î · ¸ ½@O%¼ Ì Î · ¸ (3.42)
Ñ Ø Ñ
Î Ð
Î Ð
» ¼½
· ¸ ½@y¼ Ì ·L¸ (3.43)
Ø
Ñ Ñ
t
ÚÙ » \» (3.44)
Ï Ø ~ ø Ï ~ ø Ù Ý*u*u*u£Ý
and solve as before. (This procedure has halved the number of unknowns.)
3þ .14 A XISYMMETRIC P ROBLEMS 67
¸ ø¾
¸
}
¸ ¸ ø
t ø to ¸
Note: Since 4» this means that the integrals over the elements ¸ will never
Ù Ý*u*u*u Ý }
contain a singularity arising from the fundamental solution, except possibly on the axis of symme-
try if linear or higher order elements are used. ~
An alternative approach to the method above arises from the implied no flux across the û axis.
This approach ignores the negative î axis and considers the half plane problem shown.
However now the surface to be discretised extends to infinity in the positive and negative û
directions and the resulting systems of equations produced is much larger.
Further examples of how symmetry can be used (e.g., radial symmetry) are given in the next
section.
æ ¼½ ¼ ï Î · ¹º ê · ¸h» ¼½
¼ ï · ¹º ê · ¸ (3.45)
ÕMÖ×XÏÆÕÖ×NØ Ïg Ñ m
n * *
È ÎÐ È Ñnm
where ê û and ê û are the polar coordinates of and respectively, and ¸ is the
Õ Ý Ý × Õ ±Ý*1Ý × Ö
intersection
m m of ¸ m and »3S semi-plane (Refer Figure 3.10). (n.b. is a point on the surface being
integrated over.)
68 T HE B OUNDARY E LEMENT M ETHOD
û ¸
¸
à
ê
ê
Ö ï
ê ø
ê
m ê
ê
F IGURE 3.11: The distance from the source point ( ) to the point of interest ( ) in terms of
cylindrical polar coordinates.
» ù ãÙ
ê
Ñnm
where ê is the distance from to . From Figure 3.11
Ö
m Ú
ê ø c » ê ê ç ê ê è% ç
ï ï Ø ï : Õ *W×
m ø m m
ê » ë ê ê
ï ï Ø ï
ï Ú
êe» ë ê ê ç ê ê èT ç û ç6û
ï Ø ï Õ * × ØÕ W×Éï
m m m m
» ë ç è% ç û ç6û (3.46)
Õ *× ØÕ × ï
m m
3þ .15 I NFINITE R EGIONS 69
We define
Ú
» ù Ù ã ¼ ï ·
ã
Õ × where » (3.47)
*
ÑXm È n Ñ m Ø Ø
» S
b (3.49) 2
ï Ø
Ñ m $
where is the dirac delta centered at the point and êEçóê is the dirac delta centered on the
Ö Õ ×
ring ê2»c m ê .
Unlikem the two- and three-dimensional cases, the axisymmetric fundamental solution cannot be
written as simply a function of the distance between two points and , but it also depends upon
Ö
the distance of these points to the axis of revolution.
We also define
» ù Ù ã ¼ ï Î · Î (3.50)
Ñ m
n * Ñ m
1
m È ÎÐ ÎUÐ
For Laplace’s equation Equation (3.50) becomes
æ ¼½ Î · ¸h» ¼½
· ¸ (3.52)
ÕÖÛ×XÏÆÕÖ× Ø Ï Ñ m
X
ÎÐ Ñ'm
and the solution procedure for this Equation follows the same lines as the solution procedure given
previously for the two-dimensional version of boundary element method.
¸
§
àV¦
îUÈ
¸
valid for infinite regular (i.e., nice) regions provided the solution and its normal derivative behave
appropriately as ¸gÀ ; .
Consider the problem of solving 2 » S outside some surface ¸ .
ï Ï
¸ is the centre of a circle (or sphere in three dimensions) of radius centred at some point îUÈ on
¸ and surrounding ¸ (see Figure 3.12). The boundary integral equations for the bounded domain
can be written as
àV¦
$ $
æ $ $
¼½ Î · ¸ ¼½ Î · ¸g» ¼½
·L¸ ¼½
· ¸ (3.53)
ÕÖ×XÏÆÕÖ× Ø Ï Ñ Ø Ï Ñ Ø
Î Ð
Î Ð
Ñ Ñ
§
If we let the radius À ; Equation (3.53) will only be valid for the points on ¸ if
$
$
ÂÄÊ©
ÃÅ ¨ ¼ ½ / Î çª
· ¸h»3S (3.54)
¦
Ï Ñ
Î Ð
Ñ 0
If this is satisfied, the boundary integral Equation for will be as expected i.e.,
à
$
æ $
¼½ Î · ¸g» ¼½
· ¸ (3.55)
ÕMÖ×XÏuÕÖ×NØ Ï Ñ
Î Ð
Ñ
3þ .15 I NFINITE R EGIONS 71
¼»
Navier’s Equation Î:º »S for a point load in direction ¿ .
î Øe¾½
Solution Î » ¹ w
¹ e
¹ »ç nÀ ã Ù
çÂÁ ê
ï
ê Ú ÕÚ Ù ï ×
êÆÅ ç
Î / ç Á êÆÅ êÆÅ Ç ç Á êÆÅ w
ÕÙ × ØÄ ØÕ Ù × ÕÐ Ð ×
for
ΠaЪtraction
Ã
in direction where Á is Poisson’s ratio. 0
o
Wave Equation æ
/ Î ï Ï Î ï Ï Î ï Ï ç Î ï Ï »S
ï î ø Ø î Ø î Ø Î
Î æ is
ï theÎ wave Î È ï 0 Î:´ ï
where ï ï speed.
ê
ç : æ Ñ
Solution » ÐÏv´ ù ã
Ï ê
»
Navier’s Equation Î:º »ÓS for a isotropic homogenenous Kelvin
î ØÒ¾½
Î
solution for a point load in direction ¿ .
Solution » » » w
ù
Ï Ï ê ½ ½
ç Á ê
» »
» Î ø ã'Õ Ù / Ä Î
Ï ½ çeÁ ê ¾½ Ø î î
Ù*Ô ÕÙ × direction whereÎ Á Î is Poisson’s
0
for a displacement
Õ
in
o
ï
ratio and is the shear modulus.
Linear Elasticity
4.1 Introduction
To analyse the stress in various elastic bodies we calculate the strain energy of the body in terms of
nodal displacements and then minimize the strain energy with respect to these parameters - a tech-
nique known as the Rayleigh-Ritz. In fact, as we will show later, this leads to the same algebraic
equations as would be obtained by the Galerkin method (now equivalent to virtual work) but the
physical assumptions made (in neglecting certain strain energy terms) are exposed more clearly in
the Rayleigh-Ritz method. We will first consider one-dimensional truss and beam elements, then
two-dimensional plane stress and plane strain elements, and finally three-dimensional elasticity.
In all cases the steps are:
2. Evaluate the components of stress from strain using the elastic material constants,
3. Evaluate the strain energy for each element by integrating the products of stress and strain
components over the element volume,
4. Evaluate the potential energy from the sum of total strain energy for all elements together
with the work done by applied boundary forces,
6. Minimize the potential energy with respect to the unconstrained nodal displacements,
7. Solve the resulting system of equations for the unconstrained nodal displacements,
8. Evaluate the stresses and strains using the nodal displacements and element basis functions,
9. Evaluate the boundary reaction forces (or moments) at the nodes where displacement is
constrained.
76 L INEAR E LASTICITY
Ö
Õ×
Ø6ÏÝÙÔØ ×
Ö
ÕØ×hÝÙu× Ï
¿
î
F IGURE 4.1: A truss of initial length Ú is stretched to a new length Û . Displacements of the right
hand end relative to the left hand end are Ü and Ý in the Þ - and ß - directions, respectively.
¿ ë Ö
w?» ç » Õµ×
Ø7
Ï¢× ï ØÕÙ*Ø ×ï ç
Ù Ù
Ú × ï ØeÙ ï
® Ö Ö
» ï Ø ØÕ ×gÏuØeÙ × Ø7Ï ï Ø ï ç
Ù
Ú Ö Ö
»Óà è% Ï sÃâá Ñ Ï ï Ø ï ç
Ù Ø Ï .u Ø yu Ø Ù
ï
using × » è% and Ù »hsÃâá , where is defined to be positive in the anticlockwise direction.
Neglecting second order terms in the binomial expansion ® » ÚÙ , the strain
Õ Ù Øäã× Ù Ø ã¤Ø Õã ï ×
for small displacements and Ö is
Ï
Ö
wå» èT Ï ðà á (4.1)
yu Ø yu
æ
4 .2 T RUSS E LEMENTS 77
SE » ÚÙ ¼ w·.ç » Ù è ¼
Ún é wñ·î » Ù ¼ é
Úê è w ·î » Ù è
Ú5 w (4.2)
º º ï ï
È È
where » w is the stress in the truss (of cross-sectional area è ), linearly related to the strain w
º
via Young’s modulus . We now substitute for w from Equation (4.1) into Equation (4.2) and put
ø and Ö ë
» ç » Ö çeÖ ø , where ø Ö ø and Ö are the nodal displacements of the two
Ï Ï Ï ÕÏ Ý × ÕÏ Ý ×
ends ofï the truss ï ï ï
ç ø Ö çäÖ ø
SE » ÚÙ è / è% Ï Ï ðà á ï (4.3)
yu ï Ø .u ï
0
The potential energy is the combined strain energy from all trusses in the structure minus the
work done on the structure by external forces. The Rayleigh-Ritz approach is to minimize this
potential energy with respect to the nodal displacements once all displacement boundary conditions
have been applied.
For example, consider the system of three trusses shown in Figure 4.2. A force of S%S¯ì.í
Ú Ù
is applied in the î -direction at node . Node is a sliding joint and has zero displacement in the
Ù
y-direction only. Node is a pivot and therefore has zero displacement in both î - and ò - directions.
Ä
The problem is to find all nodal displacements and the stress in the three trusses.
node
Ù S%SDì.í
Ù
Ù
Ú
S%î
Ä
S î
Ä Ú
node node
Ä
Ä
Ú
The strain in truss (joining nodes and ) is
Ä Ä
Ï è% ç S » Ú Ä Ï
ï Õ Ä × ï
Since a force of S%SDì.í acts at node in the î -direction, the potential energy is
Ù Ù
è
PE » ÚÙ è w ç S%S ø » ÚÙ Ú Ä ø ÚÙ Ö ø¼õ ï Ú Ä õ ï Ö ø ç S%S ø
ï Ù Ï ñòôó Ï Ø Ø ó Ï ØcÕ × ïö÷ Ù Ï
trusses ï
[Note that if the force was applied in the negative î -direction, the final term would be S%S ø ]
Ø Ù Ï
Minimizing the potential energy with respect to the three unknowns ø , Ö ø and gives
Ï Ï
ï
PE è
Î ø » Ú Ä ø ÚÙ Ö ø õ Ú Ä ç S%S »bS (4.4)
ó Ï Ø Ù
ÎÏ
PE è ø
Î ø » Ú Ä ÚÙ Ö ¼ø õ ÚÙ Ö dø ù »bS (4.5)
Ö ø:ó Ï Ø Ø
Î
PE è õ
Î » Ú Ä Ú Ä »3S (4.6)
ó Ï
ÎUÏ ï
ï
G ¾ G
If we choose è »[R S È Å , » Sûúýüþ and c
» Å (e.g., S%S ÅQÅ RTSÅ Å timber
è ¾ Ù ï Ù Ù ¾ ø Ù
truss) then ÿ » R G S È Å G S1ìyü1þ Å»ÿR G SCì.íÔÅ .
Ù ï Ù Ù
Equation (4.6) gives
»3S
Ï
ï
Equation (4.4) gives
ø
ø » ù G
G
`
Ö S ^dR S
ÄAÏ Ø Ä Ù ï Ù
Equation (4.5) gives for two dimensions
Ö ø »ç Ä ø
R Ï
ù
Solving these last two equations gives ø » Å Å and Ö ø » ç R Å Å . Thus the strain in truss
Ï ÄOu?Ä Ù uÙ
ù G ¾ Ú Ú Ú
is Ú Ä ç ÚÙ R S È » S , in truss is çËS R and in truss is zero.
Ù Õ ÄOu?Ä Ù uÙ × Ù u Ä ¾ u ÙÙ Ú Ú G ¾ Ä
The tension in truss is è » è wE\» R G S È Å SXì.üþ G S S » ì.í (tensile),
Ú Ù º Ù ï Ù u Ä Ù ï ÙÙ,Ô
in truss is Ë ç R RDì.í (compressive) and in truss is zero. The nodal reaction forces are shown in
u Ä
Figure 4.3.
æ
4 .3 B EAM E LEMENTS 79
STSDì.í
Ù
S%Sðì.í
Ù
R ðì.í
u
R ðì.í
u
F IGURE 4.3: Reaction forces for the truss system of Figure 4.2.
t
» ¼ &ûÁ· è » § ¼ û · è »
§ (4.8)
º ï
t
where » ¼ û · è is the second moment of area of the beam cross-section. Thus, § » and
ï
Equation (4.7) becomes
t
û
E» (4.9)
º
·
The slope of the beam is L » and the rate of change of slope is the curvature
·î
· ·
» » ï L » §Ù (4.10)
·î ·î
ï
80 L INEAR E LASTICITY
t ·
» ï L » 99 (4.11)
·î
L
ï
and a force balance gives the shear force
t
· ·
ç »ç »ç 99 (4.12)
·î · î Õ
L ×
and the normal force (per unit length of beam)
·.ç ·
¹ » » ç ï 99 (4.13)
·î ·î Õ
L ×
ï
This last equation is the equilibrium equation for the beam, balancing the loading forces ¹ with the
axial stresses associated with beam flexure
· ·
ç ï / ï L »e¹ (4.14)
·Cî ·î
ï ï 0
The elastic strain energy stored in a bent beam is the sum of flexural strain energy and shear
strain energy, but this latter is ignored in the simple beam theory considered here. Thus, the
(flexural) strain energy is
û
» ÚÙ ¼ é ¼ § Ñ ï · è ·î »ÒÚ Ù ¼ é 99 ·î
Ï ÕL
×ï
È È
where î is taken along the beam and è is the cross-sectional area of the beam.
The external work associated with forces ¹ acting normal to the beam and moving through a
The finite element approximation for the transverse displacement must be able to represent
L
the second derivative 9 9 . A linear basis function has a zero second derivative and therefore cannot
L
represent the flexural strain. The natural choice of basis function for beam deflection is in fact cubic
Hermite because the inter-element slope continuity of this basis ensures transmission of bending
moment as well as shear force across element boundaries.
ù
The boundary conditions associated with the th order equilibrium Equation (4.14) or the equa-
æ
4 .4 P LANE S TRESS E LEMENTS 81
Ú
tions arising from minimum potential energy Equation (4.15) (which contain the square of nd
derivative terms) are more complex than the simple temperature or flux boundary conditions for
the (second order) heat equation. Three possible combinations of boundary condition with their
associated reactions are
Boundary conditions Reactions
)
4.4 Plane Stress Elements
For two-dimensional problems, we define the displacement vector » Ï , strain vector
}»
Ö ¥
w
º
w and stress vector *» . The stress-strain relation for two-dimensional plane stress:
ñò ö÷ ñò º ö÷
w
º
Ô»
Á S
Ù
where » Á S . The strain components are given in terms of displacement
ç Á
e ñò Ù ö÷
Ù ï S S çÂÁ
gradients by Ù
w » ÎÏ
î
Î Ö
w » Î (4.17)
ò
Î Ö
w » ÚÙ / ÎÏ Î
ò Ø î
Î Î 0
82 L INEAR E LASTICITY
Ú
» ÚÙ ¼
f
·.ç » ÚÙ ¼ w w ÁOw@w çÂÁ w ·.ç
ç Á
e ï Ø ï Ø ØÕ Ù × ï
Ù ï
The potential energy is
)
w?»
ÎÏ » Î |
î î Ï:|
Î Ö Î
w
»
Î » Î | Ö (4.19)
ò ò |
Î Î
Ö
w
» ÚÙ / ÎÏ Î » ÚÙ / Î | Î | Ö
ò Ø î ò Ï:|
Ø î |
Î Î 0 Î Î 0
or
Î | S )
w î
ñ Î ö
Q» w » S Î | Ï:| »3 (4.20)
ñò ö÷ ò Ö ¥
w Î |
ò ÚÙ Î | ÚÙ Î | ÷
ò î
Î Î
From Equation (4.18) the potential energy is therefore
) ) )
PE » ÚÙ ¼ f ·.çç ¼ f · è
Õ × Õ × )
) )
»bÚ Ù f ¼ ü·.ç
f a ç ¼ fF· è
) u
) )
» ÚÙ f ç ¼ f S· è
æ
4 .4 P LANE S TRESS E LEMENTS 83
where
» ¼ · è is a vector of nodal forces.
where is the normalized element coordinate along the side of length loaded by the constant
Ü ¾ ø
stress ¹pì.í*Å . If the element side has a linear basis, Equation (4.22) gives
ø ø
ø »Â¹5 ¼ ø · »e¹5 ¼ ç · »bÚ Ù ¹5
Ü ÕÙ ÜC× Ü
H È È
ø ø
as shown in Figure 4.4b. If the element side has a quadratic basis, Equation (4.22) gives
ø ø
ø e ø · e Ú
» ¹5 ¼ » ¹: ¼ / ÚÙ ç ç · »bÙ ¹5
Ü Ü ÕÙ Ü × Ü
H È È 0 Ô
ø ø
Ú
ù
» ¹5 ¼
e · e » ¹: ¼ ç · » ¹5
Ü Ü
Õ Ù ÜC× Ü
H ï È ï È Ä
ø ø
Ú
» ¹5 ¼
e · e » ¹: ¼ / ç ÚÙ · » Ù ¹5
È È Ü Ü Ü Ü
H È È 0 Ô
as shown in Figure 4.4c. A node common to two elements will receive contributions from both
elements, as shown in Figure 4.4d.
84 L INEAR E LASTICITY
¾ ø
¹pì.í*Å ø ø ø `
¹5 ¹5 ¹ ^
ï éï `
¹ ^
ø ïÈ ` éï
¹5 ¹¯^
ïÈ È éï `
¹ ^
ø ø ø ïÈ ` éï
¹5 ¹5 ¹ ^
ï éï
(a) (b) (c) (d)
F IGURE ø 4.4: A uniform
ø boundary stress applied to the element
ø side in (a) isø equivalent to nodal
loads of Ú and Ú for the linear basis used in (b) and to Ú , Ú and Ú for the quadratic
basis usedï in (c). Two ïÈ
ï adjacent quadratic elements both contribute to a common node in (d), where
the element length is now .
éï
Å Ú
) w » ÚÙ )Å Û» (4.24)
ÕÏ Ø7Ï × Ý Ù Ý ÝÄ
where is the displacement vector (i.e., is the difference between the final and initial positions
of a material point in question). Note: we are assuming here that the displacement gradients are
small compared to unity, which is appropriate for many materials in solid mechanics. However, for
soft materials, such as rubber or biological tissue, then we need to use the exact finite strain tensor.
The object of solving an elasticity problem is to find the distributions of stress and displacement
in an elastic body, subject to a known set of body forces and prescribed stresses or displacements
at the boundaries. In the general three-dimensional case, this means finding stress components
Ô
which arises from the conservation of angular momentum) and 3 displacements
(» each
º º Ï
as a function of position in the body. Currently we have R unknowns ( stresses, strains and
Ù Ô Ô Ä
displacements), but only ï equations ( equilibrium equations and strain-displacement relations).
Ä Ô
To progress, we require an equation of state, i.e., a stress-strain relation or constitutive law. For
a linear elastic material we may propose that the components of stress depend linearly on w .
º
æ
4 .5 T HREE -D IMENSIONAL E LASTICITY 85
i.e.,
» æ » w »
º ½ ½
À ù
where æ » are the components of a th order tensor, although symmetry of the strain and stress
½ Ù Ú
tensors reduces the number of independent components to .
Ù
If the material is assumed to be isotropic (i.e., the material response is independent of orienta-
tion of the material element), then we end up with the generalized Hooke’s Law.
Ú'&
» w »» w (4.25)
º ° Ø
or inversely
w » ÚÙ & ç Ú& ° Ú& »»
º º
ÕÄ%°ÆØ ×
&
where , are Lamés constants.
° &
Note: , are related to Young’s modules and Poisson’s ratio Á by
°
& 'Ú &
» Õ{Ä%°ÛØ & ×
°ÆØ
ÁQ» Ú ° &
Õ °ÆØ ×
Providing that the displacements are continuous functions of position, then Equation (4.23),
Equation (4.24) and Equation (4.25) are sufficient to determine the R unknown quantities. This
Ù
can often work with some smaller grouping or simplification of these equations, e.g., if all bound-
ary conditions are expressed in terms of displacements, substituting Equation (4.24) into Equa-
tion (4.25) then into Equation (4.23) yields Navier’s equation of motion.
& & » Å» Ú
Å» » b
» S »
Ï ØÕ Û
° Ø ×XÏ Ø Ý o Ù Ý ÝÄ
These equations can be solved for the unknown displacements. Then Equation (4.24) can be used
Ä
to determine the strains and Equation (4.25) to calculate the stresses.
¼ Å · »bS (4.26)
Õº Ø ×XÏ à
) k
where » is a (vector) weighting field. The are usually interpreted as a consistent set of
ÕÏ × Ï
virtual displacements
(hence we use the notation instead of ).
Ï L
86 L INEAR E LASTICITY
By the chain-rule
Å » Å Å
Õº Ï × º Ï Øeº Ï
Therefore, the first term in the integrand of Equation (4.26) can be re-written
¼ Å · » ¼ Å · ç ¼ Å ·
k º Ï à k Õº Ï × à k º Ï à
4
» ¼ 2 · ç ¼ Å ·
k Õº Ï × à k º Ï à
»¼ · ¸ ç6¼ Å · (4.27)
j k º Ï Ð k º Ï à
4
where the domain integral involving “ 2 » Î ” has been transformed into a surface integral
î
using the divergence theorem Î
4 4 8 Å
¼ 2 $· »¼ $ · ¸ or ¼ · »¼ · ¸
k à jlk k ( à j k ( Ð
¼ Å · »¼ · ¼ · ¸
k º Ï à k Ï à¥Ø jlk º Ð Ï
» ¼ · ¼ · ¸ (4.28)
k Ï à¥Ø jlk ´ Ï
E
where are the components of the internal stress vector ( ) and are related to the components of
´
the stress tensor ( ) by Cauchy’s formula
º
E )
» (4.29)
º Ð
To arrive at this point, we have used weighted residuals to tie in with Chapter 2, however
Equation (4.28) is more usually derived using the principle of virtual work (below). Note that the
weighted integral Equation (4.28) is independent of the constitutive law of the material.
) E
is equal to the work done by the stress vector » ) in moving through a compatible set of virtual
´
displacements . In mathematical terms, the principle of virtual work can be written
¼ - · ¸h» ¼ ·L¸ÿ» ¼ · ¸ (4.30)
j k Ï j k ´ Ï j k º Ð Ï
¼ -
· ¸g» ¼ ^
Å ` · (4.31)
Å
j k Ï k º Ï Øº Ï à
Substituting the equilibrium relation (Equation (4.23)) into the first integrand on the right hand
side, yields the virtual work equation
¼ Å · » ¼ · ¼ - · ¸ (4.32)
k º Ï à k Ï àØ j k Ï
where the internal work done due to the stress field is equated to the work due to internal body
forces and external surface forces. Note that Equation (4.32) is equivalent to Equation (4.28) via
Equation (4.30). In practice, Equation (4.32) is in a more useful form than Equation (4.28), because
the right hand side integrals can be expressed in terms of the known body forces and the applied
boundary conditions (surface traction forces or stresses).
» 5 Å 5 Å
¼ Å» ÎÜ · ¹º ±/ - Ñ » ¼ · ¼ - · ¸º¹ ± / - Ñ
î à 3 àØ 3
k 7
8 º 2/ Ï ³ k 7
8 2/ j k87 2/ Ï ³
- Î -
88 L INEAR E LASTICITY
»
¼ Å» ÎFÜ · » ¼ · ¼ - · ¸ º¹ (4.34)
î à à¥Ø
k 7 º 0/ k 7 2/ j k97 2/
- Î -
The next step is to express the stress components in terms of the virtual displacements
º
and their finite element approximation by substituting Equation (4.33) into Equation (4.24) (the
strain-displacement relation) and in turn into Equation (4.25) (the generalized Hooke’s law).
We first introduce the finite element approximation for the displacement field » | which
Ï n
| Ï
gives
`
w » ÚÙ / Î | Î ^
| » ÚÙ / Î | ÎFÜ* ½ | Î | ÎFÜ*½ | (4.35)
î Õ |nÏ × Ø î |nÏ î Ï Ø î Ï
Î Î 0 ÎÜv½ Î ÎÜv½ Î 0
and
Ú'&
» Î | ÎFÜ*»½ |» / ÚÙ Î | ÎFÜ* ½ | ÚÙ Î | ÎÜv½ |
º ° î Ï Ø î Ï Ø î Ï
Î Ü*½ Î
F ÎFÜ*½ Î ÎFÜv½ Î 0
which, due to symmetry of the stress tensor, simplifies to
Ú&
» Î | ÎFÜ*» ½ |» Î | ÎFÜ*½ |
º ° î Ï Ø î Ï
ÎFÜ*½ Î ÎFÜv½ Î
Ú&
» ÎFÜ* ½ Å ÎFÜ*½ | (4.36) / Å
î Ø | ½ î Ï °5 ± | ½
Î Î 0 ³
where the summation index has been replaced with , but the parenthesis in implies that
o ±
there is no sum with respect to that particular index. ³
Substituting this expression into Equation (4.34) and simplifying, we get for each element
» Ú'& »
| ¼ / Å ÎFÜ* ½ Å» ÎFÜ Å ÎFÜ*½ Å» ÎFÜ · » (4.37)
Ï î 0 î Ø î 2 î à
k 7
8 ° | ½ / | ½ / /
Î Î Î Î 0 H
where denotes the right hand side terms in Equation (4.34). (Note that there has been some
/
carefulH manipulation of summation indices with the substitution of Equation (4.36) to arrive at
Equation (4.37).)
So for each element
| »
/ |nÏ /
H
æ
4 .6 L INEAR E LASTICITY WITH B OUNDARY E LEMENTS 89
where
ø
» Ú& »
» ¼Á¼Á¼ / ÎFÜ* ½ F
Î Ü ÎFÜ*½ Î Ü Å Å» ø · ø· ·
/ | ° î î Ø î î | ½2 / ( ÕÜ
; ÝsÜ sÝ Ü È × Ü Ü Ü È
È Î Î Î Î 0 ï ï
ø ø (4.38)
» ¼Á¼Á¼ ø · ø· · ¼Á¼ -
: ø · ø·
/ 2/ (;ÕÜ ÝðÜ ðÝ Ü È × Ü Ü Ü È Ø 2/ ( ÕÜ sÝ Ü × Ü Ü
H È ï ï È ï ï ï
where the Jacobians ø and : ø have been used to transform volume and sur-
(lÕíÜ ÝðÜ ÝðÜ È × ( ÕíÜ ÝsÜ ×
face integrals so that they can ï be can beï calculated ï using -coordinates. (Note: without loss of
Ü
generality, the above definition of
assumes that ø are defined to lie in the surface ¸ .)
So in summary, the finite element H
/ approximationÕíÜ leads
ÝðÜ ×
ï to element stiffness matrix components
that can be calculated from the known material parameters, the chosen interpolation functions, and
the geometry of the material (note that the element stiffness components are independent of the
unknown displacement parameters). Element stiffness components are then assembled into the
global stiffness matrix in the usual manner (as described previously). Note that this is implicitly a
Galerkin formulation, since the unknown displacement fields are interpolated using the same basis
functions as those used to weight the integral equations.
w » ÚÙ
Å ÚÙ Å
º º Ï Ø º Ï
Å
» ÚÙ Å ÚÙ
º Ï Ø º Ï
» ÚÙ Å ÚÙ Å
º Ï Ø º Ï
»
Å
º Ï
where w are the virtual strains corresponding to the virtual displacements.
90 L INEAR E LASTICITY
Using the constitutive law for linearly elastic materials (Equation (4.25)) we have
¼ Å · » ¼ w ·
k º Ï à k º à
Ú'&
» ¼ w »» w · ¼ w w ·
° k àØ k à
Ú&
» ¼ w »» w »» · ¼ w w ·
° k àØ k à
»c¼ w ·
k º à
due to symmetry.
Thus from the virtual work statement, Equation (4.28) and the above symmetry we have
¼ · ¼ ·L¸ÿ»¼
· ¼
· ¸ (4.39)
k Ï à¥Ø j k ´ Ï k Ï à¥Ø jlk ´ Ï
This is known as Betti’s second reciprical work theorem or the Maxwell-Betti reciprocity relation-
ship between two different elastic problems (the starred and unstarred variables) established on the
same domain.
Note that »ç Å (i.e., Å 3
» S ). Therefore Equation (4.39) can be written as
º º Ø
`
¼ ^
Å · ¼ · » ¼
· ¸ ç ¼ · ¸ (4.40)
k º Ï àØ k Ï à j k ´ Ï j k ´ Ï
(
w represents the equilibrium state corresponding to the virtual displacements ).
ºÝ Ý´ Ï
Note:
What we have essentially done is use integration of parts to get Equation (4.28), then use
it again to get Equation (4.39) above (after noting the reciprocity between and w ).
º
Since the body forces, , are known functions, the second domain integral on the left hand
side of Equation (4.40) does not introduce any unknowns into the problem (more about this later).
The first domain integral contains unknown displacements in and it is this integral we wish to
à
remove.
We choose the virtual displacements such that
Å w »S (4.41)
º Ø
(or equivalently ç w »ëS ), where w is the th component of a unit vector in the th direction
" Ø
and w »3w ç . We can interpret this as the body force components which correspond to a
Õ ÖÛ×
positive unit point load applied at a point in each of the three orthogonal directions.
Öüßhà
Therefore
æ
4 .7 F UNDAMENTAL S OLUTIONS 91
"
¼ Å · » ç ¼ ç w · »ç w
k º Ï à k =Õ Ö× Ï à Ï ÕMÖ×
i.e., the volume integral is replaced with a point value (as for Laplace’s equation).
Therefore, Equation (4.40) becomes
w » ¼ · ¸ ç ¼ · ¸ ¼ · (4.42)
Ï ÕMÖ× j k ´ Ï j k ´ Ï Ø k Ï à Öüßgà
»c¼ î î · ¸ î ç6¼ î î · ¸ î
Ï ÕÖÛ× j k Ï ÕÖËÝ .
× ´ Õ × Õ × j k ´ ÕÖ¤Ý S× Ï Õ × Õ ×
¼ î î · î (4.45)
Ø k Ï ÕÖËÝ × Õ × à¥Õ ×
Õ
where = shear Modulus.
Thus satisfy
Ï
Õ
Õ "
Å» » Ú » Å» ç w »S (4.47)
Ï Ø ç Á Ï Øe Õ ÖÛ×
Ù
The solutions to the above equation in either two or three dimensions are known as Kelvin 2 ’s
fundamental solutions and are given by
" ù
» ã Ù Õ ç Á
êÆÅ êÆÅ >= (4.48)
Ï ÕMÖËÝ × çÂÁ ê<; Õ{Ä × Ø
Ù Ô ÕÙ
, ×
for three-dimensions and for two-dimensional plane strain problems,
" ç ù
» Ànã Ù Õ ç Á
ÂÄèélêEçóêÆÅ êÆÅ >= (4.49)
Ï ÕÖ¤Ý × çeÁ ; Õ{Ä ×
ÕÙ ×
and
" ç ê Ú Ú
» ù ã Î ç ç Á ê¢Å çóêÆÅ
Ù (4.50)
ç Á êÆÅ êÆÅ
´ ÕÖËÝ × çÂÁ ê ÕsÕ Ù × Ø@? × ÕÙ × Õ Ð Ð ×,¡
â ÕÙ × Î Ð
U
Ú A
B Ú
where » » for two-dimensional plane strain and three-dimensional problems respec-
â Ù Ý ? ÝÄ
tively.
" " "
Here ê ê , the distance between load point ( ) and field point ( ), ê »Ôî ç î
ê M
Õ Ë
Ö Ý × ê Ö Õ × ÕÖÛ×
and Æê Å » Î " » .
î ê "
Î Õ ×
In addition the strains at an point due to a unit point load applied at in the th direction are
Ö
given by
" ç Ú
w ¼» » À ã Ù Õ
ç Á êÆÅ »
êÆÅ »
çóêÆÅ » êÆÅ ê¢Å ê¢Å »'=
ÕMÖËÝ × çeÁ ê ; ÕÙ ×£Õ Ø × Ø@?
â ÕÙ
×
and the stresses are given by
" ç Ú
» » ù ã Ù
ç Á êÆÅ »
êÆÅ » çóêÆÅ ¼» ê¢Å êÆÅ êÆÅ »D=
º ÕÖ¤Ý × çeÁ ê ; ÕÙ × Õ Ø × ØC?
â ÕÙ ×
where and are defined above.
â ? Á
The plane strain expressions are valid for plane stress if Á is replaced by Á » (This is a
Á
Ù Ø differences.
mathematical equivalence of plane stress and plane strain - there are obviously physical
What the mathematical equivalence allows us to do is to use one program to solve both types of
problems - all we have to do is modify the values of the elastic constants).
Note that in three dimensions
3
» /ËÙ \
» Ó/Ù
Ï ê ´ ê
0 ï 0
2
Lord Kelvin (1824-1907) Scottish physicist who made great contributions to the science of thermodynamics
æ
4 .8 B OUNDARY I NTEGRAL E QUATION 93
3
» ÂÄèAé;ê \
» / Ù
Ï Õ × ´ ê u
0
Somigliana’s identity (Equation (4.45)) is a continuous representation of displacements at any
point . Consequently, one can find the stress at any firstly by combining derivatives
Öìߥà Öߥà
of (4.45) to produce the strains and then substituting into Hooke’s law. Details can be found in
Brebbia, Telles & Wrobel (1984b) pp 190–191, 255–258.
This yields
Note: One can find internal stress via numerical differentiation as in FE/FD but these are not
as accurate as the above expressions.
Expressions for the new tensors » and ¼» are on page 191 in (Brebbia et al. 1984b).
Ï ´
¸N¿
9
à
ã ¸\¾ ¿
F IGURE 4.5: Illustration of enlarged domain when singular point is on the boundary.
94 L INEAR E LASTICITY
Ç
We need to look at each integral in turn as S (i.e., À S from above). The only integral that
ã ã
presents a problem is the second integral. This can be written as
Let
æ " "
» ¿MÃÄÇÉÅ È ¼½ Í · ¸ (4.54)
ÕMÖ× Ø ´ ÕMÖËÝ × Õ ×
Ç "
" "
As S , ¸x¾ ¿4À ¸ and we write the second integral of Equation (4.52) as ¼ ½ ·L¸
ã ´ ÕMÖËÝ ×XÏ Õ × Õ ×
where we interpret this in the Cauchy Principal Value3 sense.
3
What is a Cauchy Principle Value?
Consider LNMPO%QRS on TUV0RXWZY2[]\^Y`_aQb<M_]\4[dc
O
æ
4 .8 B OUNDARY I NTEGRAL E QUATION 95
Ç
Thus as S we get the boundary integral equation
ã
(or, in brief (no body force), æ ¼½ ·L¸h» ¼½ ·L¸ ) where the integral on the left hand
Ï Ø ´ Ï Ï ´
side is interpreted in the Cauchy Principal sense. In practical applications æ and the principal value
integral can be found indirectly from using Equation (4.55) to represent rigid-body movements.
The numerical implementation of Equation (4.55) is similar to the numerical implementation
of an elliptic equation (e.g., Laplace’s Equation). However, whereas with Laplace’s equation the
unknowns were and ÎÏ (scalar quantities) here the unknowns are vector quantities. Thus it is
Ï
ÎÐ with matrices instead of indicial notation.
more convenient to work
i.e., use
)
ø ø
Ï E ´
» »
ñò Ï ö÷hÝ ñò ´ ö÷
ï ï
Ï È ´ È
)
ø«ø ø ø ø«ø ø ø
Ï ø Ï Ï E ´ ´ ´ È
ø
È
» »
ñò
Ï ø Ï ï Ï
È ö÷ Ý ñò ´ ´
ï
´ È ö÷
Ï ï Ï ï« ï Ï ïÈd È
´ È
ï ø ´ È
ï« ï ï
´ ÈdÈ
È È
ï ï
Then
e e UV e m
UV m
LiMjOQ`k*OlR S k*O.n S k*OoRCpZqNr Osrtr U%m nXpZqNr Osrtr V
f]g*h O O
U m
V
Rupvq`_wY<pZq2[0nxpZq2[yY<pvq`_zR|{u}_~{
e e
pvZ
V
f]g*h LNMO%QkOoR{ This is the Cauchy Principle Value of
f
LMOQ`k*O
e e m e V e m
S kOlR S kOlRVpZ
v
S kO'nVpvvz
S kO' (by definition of improper integration)
f O O O O
U%m U%m U%V
which does NOT exist. i.e., the integral does not exist in the proper sense, but it does in the Cauchy Principal Value
sense. However, if an integral exists in the proper sense, then it exists in the Cauchy Principal Value sense and the two
values are the same.
96 L INEAR E LASTICITY
We can discretise the boundary as before and put £ , the singular point, at each node (each
node has ¤ unknowns - ¥ displacements and ¥ tractions - we get ¥ equations per node). The overall
matrix equation ¦ )§¢¨
)¬ ¬ (4.57)
)i® ®
)@¢ ©ª ¢ ©ª
ª ª
ª« ª«
where ).¯D°± and ..¯D°² ±± where ³ is the number nodes.
.. ² ±
± . ± ¦
The diagonal elements of the matrix in Equation (4.57) (for three-dimensions, a ¥ x ¥ matrix)
contains principal value components. If we have a rigid-body displacement of a finite body in any
one direction then we get ¦µ´a¶¢·
´a¶
¹ º»t» ºF» Â
»ÀÁF¿  (no sum on à )
¦
»
i.e., the diagonal entries of (the Ä Â ’s) do not need to be determined explicitly. There is a similar
result for an infinite body.
e.g., a body force arising from a constant gravitational load, or a centrifugal load due to rotation
about a fixed axis or the effect of a steady state thermal Æ load can all be transformed to a boundary
integral.
Firstly, let Ê » Â (the Galerkin Æ tension) be related to » Â by
¢ ¼ Ï
¼CÓÕÔ
»Â Ï
Ê » ÂÌË ÍÎÍ Ï ÐÒÑ Ê » Í*Ë ÍÖÂ
¢ ×Ø
®» ßá àâã
Ù Û'Ü Ï Ï (3D)
¹ » ÊÝÞ
Ê Â ØÚ
»Â
Û'Ü (2D)
Ê Ý Þ
Ýä
Then å Æ ã
¢ ¢ ¼ Ï
¼@ÓÕÔ
» » Â*ç Â È Ï
Ê » ÂÌË ÍdÍ ÐÒÑ Ê » ÍË ÍÖÂ ç Â È
æ æ¢ Ô ä
Ñêé
Under a constant gravitational ¢ë load è Â
åç Â é ã
¢ë  ¼ Ï
¼CÓÕÔ
¹ » é Ï
Â Ê » Í Ë Â ÐÒÑ Ê » *Í Ë ÍÖ È
¢ë æ ¼ Ï ä 8¡
¼ÓîíÔ
é Ï
Â Ê » ÌÂ Ë Í ÐÒÑ ³ Í
ì Ê » Í Ë Â0ï
which is a boundary integral.
Unless the domain integrand is “nice” the above simple application of Green’s theorem won’t
work in general. There has been a considerable amount of research on domain integrals in BEM
which has produced techniques for overcoming some domain methods. The two integrals of note
are the DRM, dual reciprocity method, developed around 1982 and the MRM, multiple reciprocity
method, developed around 1988.
98 L INEAR E LASTICITY
5.1 Introduction
In the previous discussion of steady state boundary value problems the principal advantage of the
finite element method over¢óò the finite difference method has been the greater ease with which com-
plex boundary shapes ò can be modelled. In time-dependent problems the solution proceeds from
an initial solution at ñ and it is almost always convenient to calculate each new solution at a
constant time (ñõô ) throughout the entire spatial domain È . There is, therefore, no need to use
the greater flexibility (and cost) of finite elements to subdivide the time domain: finite difference
approximations of the time derivatives are usually preferred. Finite difference techniques are intro-
duced in Section 5.2 to solve the transient one dimensional heat equation. A combination of finite
elements for the spatial domain and finite differences for the time domain is used in Section 5.3 to
solve the transient advection-diffusion equation - a slight generalization of the heat equation.
ù Ô
Ñ Æ ¯ ® Æ ¯ Æ ¯
about the grid Æ ¯ Ã ³
Æ ¯ point ã ® ã ã
¬ ¢ Ï ® Ï
ö ö ö
» » ø
ã öÕø
Æ ¯ ® Æ ¯ ø ø
Æ ¯ ø
Æ ¯ Æ ¯ ® ã öÕø
» Ð » ¼ ã öø
» (5.3)
¬ ¢ ¼ Ï ® ¤Ï
ö ä ö ä ö
ä
»
» ø Æ ¯ ø ø
ø
Æ ¯ ¬ Æ ¯ ã Õö ø » Ð ® öÕø » öÕø
» (5.4)
¢ ¤
ö ä ä ä
» »
ñ ® ö » ñ (5.5)
Ô Ô ñ
ä
Ñ ø Ñ
where and ñ represent all the remaining terms in the Taylor Series expansions.
®*Æ ¯
Adding Equations (5.3) Æ ¯ and Æ (5.4)¯ gives Æ ¯ ® ã
¬ ¬ ¢ ®
ö
» »
Ð » ø ø
öÕø »
® Æ ¯ Æ ¯ Æ ¯ Æ ¯ ä
or ã ¬ ¼ ¬ ®
® ¢ ® ù
ö » Ð » »
ø
öÕø » ø (5.6)
ä
which is a “central difference” approximation of the second order spatial derivative.
Æ ¯ Æ ¯ ¬ Æ ¯
Rearranging Equation (5.5) gives ã a “difference” ¼ approximation of the first order time derivative
¢ Ô
ö » » Ñ
ö » ûñ (5.7)
ñ ñ
ä
Substituting Equation (5.6) and Equation (5.7) into the transient heat equation Equation (5.1)
Æ ¯ ¬ Æ ¯ Æ ¯ Æ ¯ Æ ¯
gives the finite difference ¼ approximation ¬ ¼ ¬ ®
Ô ¢÷ ®
» » » Ð » »
Ñ ø
ñ Æ ¯ ¬ ø Æ
ûñ
which is rearrangedÆ ¯to give ¬ Æ an¯ expressionÆ for ¯ » Æ in ¯ terms Æ ¯ of the values ® of ® at the ³ th time step
¢ @÷ ® ¬ ¼ ¬ ù
ñ Ð ø
» » Æ ¯ ø ¢óò » » » ñ Æ ¯ ¬ (5.8)
¢óò
¢
» at ³ Ï ù (i.e., ù
ñ
ù »
Given ¢ the ù ù initial values of
Ð
), the values of for the next time step
Ï
are found from Equation (5.8) with à . Applying Equation (5.8) iteratively for time Æ ¯
Æ ¯ Ð ¯ ¬ Figure Æ ¯ 5.1). ¬
steps ³¢ Ô etc. yields the time dependent temperatures at the grid points Æ (see ¬ ¬
ùÏ ù ù
This is an explicit finite difference formula because the value of » depends only on the values of
» ÑÃ Ð
at the previous time step and not on the neighbouring terms » »
ø and at
the latest time step. The accuracy of the solution depends on the chosen values of and ñ and
in fact the stability of the scheme depends on ÷ these satisfying the Courant condition:
® Ï
ñ
ø Ð (5.9)
5 .2 F INITE D IFFERENCES 101
ñ
Ï
³ x
³ x x x
:Ï
ò
ò ¼
Ï Ï Ï ø
Ð
... Ã Ã Ã ...
F IGURE 5.1: A finite difference grid for the solution of the transient 1D heat equation. The
equation is centred at grid point ! shown by the " . The lightly shaded region shows where the
solution is known at time step . With central differences in # and a forward difference in $ an
explicit finite difference formula gives the solution at time step &%(' explicitly in terms of the
solution at the three points below it at step , as indicated by the dark shading.
* » Ñ-A.702143
¯ ¬ 5
If divide Equation (5.11) by, DÍ + we Ô obtain (no sum Ô on E ),
¯ ¢ ¼
Ô
* Ï
Í Ñ Ð ) ) » Ñ -A.5 143 ) 8» Ñ -/.25 143
* + ã +
Í ¢ ¼ Ô ¢ » @ I 8J» I
Ï Ü ü ø
Ð ) Ð ) EH Ñø + +
® ã Ä F,G using Ä FDG Ю Ô (5.12)
¢ ¼ ÔN¢ ¼
Ï Ü ü ø ä Ï
) EK ÑÐ ø Ð Ñø
ð G>ó Ð using Ä F,G G>ó
ä
Equation (5.12) predicts the growth of any component (specified by E ) admitted by the system.
L
102 T RANSIENT H EAT C ONDUCTION
¯ ¬
If all components are to decay,
¯
M* M Ï
M Í M
® M * M for stability (no sum on E ) (5.13)
M Í M ò
Ï
Since the G>Ã
³ term in Equation (5.12) is always between and , we effectively have the stablity
criteria that ¼ Ï ¼ ¼
Ï Ï Ï
) )ON
ð and ò ð ÷ (5.14)
)PN
The first inequality is trivially satisfied, since for positive values of ñ and , and the
second condition will always hold if ¢ ÷ Ï
®
) ñ
ø Ð (5.15)
®
Thus, to ensure stability, the time step should be chosen such that
÷ø
ñ Ð The Courant condition (5.16)
ñ
Ï
³ x x x
³ x x x
:Ï
ò
ò ¼
Ï Ï Ï ø
Ð ... Ã
à à ...
F IGURE 5.2: An implicit finite difference scheme based on central differences in $ , as well as # ,
'
which tie together the 6 points shown by U . The equation is centred at the point (V% W ) shown
by the " . The lightly shaded region shows where the solution is known at time step . The dark
shading shows the region of the coupled equations.
¼ Ô
in whichÏ the spatial second
¢ò derivative of Equation (5.1) has been weighted by at the new time step
¬ Ñ S
and¢ by
® ¢
at the old time step. The original explicit forward difference scheme Equation (5.8)
S
is recovered when Ï
and the implicit central difference (Crank-Nicolson) scheme (5.19) when
. An implicitS backward difference scheme is obtained when .
S S
In the following section the transient heat equation is approximated for numerical analysis
by using finite differences in time and finite elements in space. We also generalize the partial
differential equation to include an advection term and a source term.
Æ Æ
or ã Æ Æ
dXeZ §÷ Z ¢ Ç §÷ 8¡
h ö ö
ö Å Å Å È È ö (5.21)
æ ñ æ k æ ³
ä f fji f f
ö Æ Æ
¢ml2¯ ¯ ¢lon ö
where ö is the normal derivative to the boundary È .
³
Putting and and summing the element contributions to the global equations,
f
Equation (5.21) can be represented C)
by a system of)§ first order
rqc ¢qc )ts ordinary differential equations,
p
q
ñ ) (5.22)
p ) s
t q
where is the global mass matrix, the global stiffness matrix and a vector of global nodal
p
unknowns with steady state values (ñvu w ) . The element contributions to and are
¬
given by x
n2¯ y¢ lzntl2¯|{ 4}
þ
(5.23)
¬ ¬
and lzn l2¯ } } l2¯ }
nw¯~yi¢ ÷ Z { 4} lzn { 4}
þ ö } ö } ö » ö þ ö } ö »
Â
ö » ö ö ø Õ
Õ ö ø Â ö » öø (5.24)
Â Í Í Â
¢ ù ¢ò ù ù ù Ô
Ï
Ñ Ð
If the time domain is now discretized ñ ³Q ñ ³ ~ Equation (5.22) can be re-
¯ ¬ ¯
placed by ) ¼ )
ó ¯ ¬ ¯D
rq ) ¼ ÔX) ¢mq})ts ò
Ï Ï
p Ñ
(5.25)
ñ S S ¢ Ï S
where ® Ô is a weighting factor discussed in Section 5.2. Note that for Ð the method is known
S S
as the Crank-Nicolson-Galerkin method and errors arising from the time domain discretization are
Ñ ¯ ¬ ¯
ñ . Rearranging Equation q
) (5.25)¢ as ¼ ¼ Ô q) q}b
) s
Ï
p p Ñ
ûñ ñ ûñ Ô (5.26)
¯ S S Ï
) Ñ
gives a set of linear algebraic equations to solve at the new time step ) ³ ñ from the known
solution at the previous time step Q ³ ñ.
¬
The stability of the above scheme can be examined by expanding (assumed to be smoothly
¦ ¢ q ) òF» Ô ¢ ¾ » ) s ¢
b
continuous in time) in terms of the eigenvectors (with associated eigenvalues ) of the matrix
p Ñ º» »
. Writing the initial conditions and steady state solution
»
5 .3 T HE T RANSIENT A DVECTION -D IFFUSION E QUATION 105
¾
» »
» ç , the set of ordinary differential equations Equation (5.22) has solution
)@¢ ¾ ¼ Ô
Ñ º» |6
» » »
» ç ç + (5.27)
)
¯
)
The time-difference scheme Equation (5.26) on the other hand, with now replaced by a set
¯ ¬ ¯
of discrete values
at each time¦ ±)step Q
³ ¢ ñ , can ¼ be¼ written Ô ¦ as&) the recursion
¦ b ) s formula
Ï
Ñ
ñ ñ ñ (5.28)
S S ¯
with solution ¼ ¼ Ô
)§¢¾ ¼ Ô Ï Ï
h Ñ »
» Ñ º» » Ï ûñ »
» ç
ç
ñ S » (5.29)
i
¦S ¢
(You can verify that Equation (5.27) and Equation (5.29) are indeed the solutions of Equation (5.22)
» » »
and Equation (5.25), respectively, by substituting and using .)
Comparing Equation (5.27) and Equation (5.29) shows that replacing the ordinary differential
|6 the finite difference approximation Equation
equations (5.22) by ¯ (5.25) is equivalent to replacing
the exponential + in Equation (5.27) by the¼ approximation
¼ Ô
Ï Ï
|6o h
Ï ñ Ñ
»
+ » (5.30)
¢ ñ S
S i
or, with ñ ³ ñ , ¼ ¼ Ô
Ï Ï ¢ ¼
Ï
| Ï ûñ Ñ
» Ï ûñ
»
+ » » (5.31)
ñ S ûñ
S S
The stability of the numerical time integration scheme can now be investigated by examining
the behaviour of this approximation ¼ to the exponential.¼ For stability we require
Ï Ï »
Ï
Ï ñ
» (5.32)
ñ
ù S
since this term appears in Equation (5.29) raised to the power ³ . The right hand inequality in
»
Equation (5.32) is trivially satisfied, since ñ and are all positive, and the left hand inequality
gives S
¼ Ô
» Ï
Ï ñ Ð » Ñ Ð Ð
» or ñ ¬ (5.33)
ñ S ® Ï
¬
ú ® S
A consequence of Equation (5.33) is that the scheme is unconditionally stable if .
For the stability criterion is S
S
ú
¼
» Ï Ð
ûñ Ð (5.34)
S
106 T RANSIENT H EAT C ONDUCTION
»
If the exponential approximation given by Equation (5.31) is negative for any the solution
will contain components which change sign with each time step ³ . This oscillatory noise can be
avoided by choosing ú Ï
¼ Ô ù
Ï
ñ Ñ (5.35)
¦ max
S
where max is the largest eigenvalue in the matrix , but in practice this imposes a limit which
is too severe for ñ and a small amount of oscillatory noise, associated with the high frequency
vibration modes of the system, is tolerated. Alternatively the oscillatory noise can be filtered out
¬
by ¢ averaging.
®
These theoretical results are explored numerically with a Crank-Nicolson-Galerkin scheme
Æ ®Æ
( ) in Figure 5.3, where the one-dimensional diffusion equation
S ¢ ÷ ® ò Ï
ö ö
ø
ö öÕø on Æ
ñ
Æ ù òFÔ ¢ò Æ
Ñ øò ù Ô ¢ò ù Ï ù
Ôz¢
Ï
(5.36)
subject to initial conditions
Ñ Ñ
and boundary conditions ñ ñ
ø
is solved for various timeò increments ò ( ñ ) and element lengths ( ) for both linear and cubic
Hermite elements. Ï
ø Ð4
Decreasing from to with linear elements produces more oscillation ò ò because the
system has more degrees of freedom and leads to greater oscillation. At a sufficiently small ñ the Ï T
oscillations are negligible (bottom right, Figure 5.3). With this value of ñ ( ) the numerical
s ¯
results agree well with theÆ exact solution (top, Figure ¼ Ô 5.3)¯ given by
ù Ô ¢ ¾ Ï Ô
Ð ¯ ¬ Ñ RR A
Ñø ø Ñ Ü ø
ñ Ü + ³ (5.37)
Á ³
Æ
ù Ô
Ñø ¢ò ¢ò ò
ñ Exact solution Ï ù Ï
ò ø
Ï x x Linear CNG with ñ
¢ ò ò
¢ò ò
ñ
ò x
¢ò Ð
Ïñ x x
¢
ñ x x
x
x
ñ w x x
x
ò x
x
x ø
xx x
x
x x Ï
x
µ µ
(b) §©¨ª¬«º°± (c) §©¨eª¯«°6±
x x x x x x
x x x x x x x
x x x
x linear elements x cubic elements
¡·¢_£³¤¥ ¸A¹ ¡·¢®£³¤¥ ¸A¹
¡z²|£³¤¥¦ ¶ ¡z²|£³¤¥¦ ¶
´ ´
ª ª
x
µ (d) §©¨ª¬« µ (e) §©¨eª¯«
x
x
x
x x x x x x x
x x
x
x
x
x
linear elements x linear elements
¡z¢_£v¤6¥¦ x ¡z¢®£¤6¥¦
x
¡o²|£³¤¥¦ ¶ ¡o²|£³¤¥ ¤¦ ¶
´ ´
ª ª
F IGURE 5.3: Analytical and numerical solutions of the transient 1D heat equation showing the
effects of element size »¼# and time step size »¼$ . The top graph shows the exact and approximate
solutions as functions of # at various times. The lower graphs show the solution through time at the
specified # positions and with various choices of »¼# and »½$ as indicated.
108¾ T RANSIENT H EAT C ONDUCTION
matrixx ((5.23)) for a bilinear ® element® (see Figure 1.9 ¼( }'¬^Ô (1.6)).
and ¼À}>®aÔ
¬Ö¬ ¢ ¼ }'¬^Ô
( ¼ }®*Ô }'¬ }>®
¿ ¢ ¼ Ï ¼ Ï ¢ Ï Z Ï ¢ Ï
Ï Ï Ñ
MM þ Ï Ñ
MM þ Ï
x Ñ Ñ M x MM x
® ® M Á
®Ö®.¢ } ¬ ¼ }>®aÔ }¬ }®
À ¢ Ï Z Ï ¢¥ Ï M ¥ M ¥ ¥
Ï
x Ñ ® ã
and .
¬®.¢#  }¬ ¼ }'¬^Ô
( ¼ }'¬^Ô }¬ }® ¢ ¥ ¼ ¥ Ï ¼ Á Ï and
(
similarly
Z Ï ¢ Ï
Ï Ï
x Ñ x Ï x
® Ñ Ð Û
¬ ¢ ¼ }'¬^Ô }>®
( ¼ }>®aÔ }¬ }® ¢ Ï
( ¥ ¥ ®
Ï Ï ä
x Ñ Ñ Ï x
Û and similarly
and .
¬ ¢ }¬ ¼ }'¬^|
( Ô }® ¼ }>®aÔ }¬ }®.¢ Ï
¿ ®
Ï Ï
Ñ Ñ
and similarly
.
¥¤
¬ ¬ ¬ ¬ ¬
¬ Ä ¬ Ä ò ò ò
¬ ¬ ¬ ¬ ¬
¬ Ä ¬ Ä ò ò ò
¢ ©ª à ¬ ¬ ¬
¬Å ¼ ª © ¬
ª« ¬ ¬ Ä ¬ ¬ ¬ Ä ª« ò ò ò
p Ã
¬ ÅÄ ¬ ¬ ¬
¬ Ä mass lumping
u ò ò ò
therefore
Å Ã ²°±± ²°±±
Å Ã
¯ ¬
¢ ò )
The element mass is effectively lumped at the element vertices. Such a scheme has computa-
tional advantages when in Equation (5.26) because each component of the vector is
obtained directly withoutS the need to solve a set of coupled equations. This explicit time integration
scheme, however, is only conditionally stable (see (5.34)) and suffers from phase lag errors - see
below. For evenly spaced elements the finite element scheme with mass lumping is equivalent to
¬ ® ¬
finite differences with central spatial differences. ¢ ÷ ¢ ò ÇC¢ ò
In Figure 5.4, the finite element and finite differences (lumped f.e. mass matrix)Ï ¢ solutions
ò
of the
ÇÆ Æ
one-dimensional advection-diffusion equation (5.20) with ] , ,
ø
are compared for the propogation and dispersion of an initial unit mass pulse at . The length
of the solution domain is sufficient to avoid reflected end effects.®
x ¼ Ô increases with time:
The exact solution is a Gaussian distribution whose variance
Æ
ù ÔN¢ Ñø ÷
]ñ
Ñø
ñ È Ü + ð ñ (5.38)
ð ñ
The finite element solution, using the Crank-Nicolson-Galerkin technique, shows excellent
amplitude and phase characteristics when compared with the exact solution. The finite difference,
or lumped mass, solution also using centered time differences, reproduces the amplitude of the
pulse very well but shows a slight phase lag.
Æ
ù Ô
Ñø
ñ
Ì Ñ ´
¶ ´ ¨[±Í(Ê<ÎÐÏ ¨eª¯« TÍ ÒÊ<ÎÐÏ
¨eª¯« ª Ê
Exact solution
x x Finite element solution
¶ o o Finite difference solution
¨eª¯« ªÓ±Ê
x
¶
x x ¨eª¯«ºÉoÊ
x ¶
¨eª¯« ËzÊ
x x
x x ox
x x ox o x
o
x o x ox
x
x
x ox x o x
o x ø
F IGURE 5.4: Advection-diffusion of a unit mass pulse. The finite element solutions (at $ =Ô4ÕÔÐ'jÖ ,
W
Ô4ÕÔ¯×tÖ , Ô4Õ Ö and Ô4ÕºØTÖ ) and finite difference solutions (at $ =Ô4ÕºØTÖ only) are compared with the exact
solution. »¼# = 0.1, »¼$ = Ô4ÕÔÔÐ'jÖ for 0 ÙÚ$·Ù_Ô4ÕÔÐ'ÛÖ and »¼$ = 0.01 for $ÝÜÞÔ4ÕÔÐ'jÖ .
Chapter 6
Modal Analysis
6.1 Introduction
The system of ordinary differential equations which results from the application of the Galerkin
finite element (or other) discretization of the spatial domain to linear parabolic or hyperbolic equa-
tions can either be integrated directly - as in the last section for parabolic equations - or analysed
by mode superposition. That is, the time-dependent solution is expressed as the superposition of
the natural (or resonant) modes of the system. To find these modes requires the solution of an
eigenvalue problem.
included but does not then result in an uncoupled system unless further simplications are made) is
)
found by solving the free vibration problem Ôrqc) ÔN¢ ·
p ß Ñ Ñ
ñ ñ (6.4)
þ
Ôz¢ of the form
Proof: Consider a solution to Equation (6.4) ¼ Ô ù
A
þ ê Ñ Ñ
ñ ñ ñ (6.5)
f
where and ñ are constants and is a vector of order ³ . Substituting Equation (6.5) into Equa-
®
q
f gives the generalized eigenproblem
tion (6.4) ¢
® ® ® p
¬ ̬ Ô ù ® a® Ô ù ù ¯ ¯Ô q (6.6)
Ñ Ñ Ñ f
having ³ eigensolutions ~ . If is a symmetric matrix (as is the case
when the original partial differential
f f operator is
f self-adjoint) the eigenvectors are orthogonal and
can be “normalized” such that ¢
r ë
è ¢ Ï
ò ¢ ë
r
x » p Ã
 (6.7)
à ì
í
¢ ¬ ù ® ù ù D¯
(the eigenvectors
are said to be -orthonormalised). Combining the ³ eigenvectors into a matrix
î
~
- the modal matrix - rewriting è ¢ Equation (6.7) as
î p î
(6.8)
where q
is the identity matrix, (6.6) becomes ¢
î p îíï
(6.9)
®
where ¬ ò
®
®
¢ ª©
ª
ï ª« f ®
ò ¯ °± (6.10)
f ..
. ² ±±
f
or èq ¢ è ¢ ¢
x ï
î î î p îðï ï
q (6.11)
æ
Thus the modal matrix - whose columns are the -orthonormalised eigenvectors of (i.e.,
satisfying Equation (6.6)) - can be used as the transformation matrix in Equation (6.2) required
to reduce the original systemç of Ô2 èà (6.1)
equations çâ Ôto theç canonical
Ô ¢ èform
ä Ô
ß Ñ î î Ñ ï Ñ î Ñ
ñ ñ ñ ñ (6.12)
6ñ .3 A N A NALYTIC E XAMPLE 113
®
With damping neglected equation Ô Equation Ô ¢ (6.12) Ô becomes ¢ a system
Ï ù ù ù of uncoupled equations
øß » Ñ ø » Ñ » Ñ Ð
ñ ç » ñ ñ Ã ³ èä (6.13)
Ý
ø » f » î
where is the à th component of and is the à th component of the vector . The solution of
Ý
this system is given by the Duhamel integral
à
Ô ¢ Ï Ô ¼ ÔÒ r ô õ
Y
þ A
ø » Ñ » óÑ ò » Ñ ò ò » » » ö
¯ » ò
ñ » ñ ñ (6.14)
ô õ
Ý
f f f
f
» òFÔ÷ þ » ¢ è þ the initial conditions
òÔ¯÷ from
where the constants and are determined
ò Ô÷ þ ¢
øâ » Ñ F » p òÔ÷ ê þ Ñ ¢ Á è òFÔ¯÷ þ è òFÔ÷ þ
Á G
ø » Ñ ø » Ñ » p ê Ñ Á » p ê Ñ Á (6.15)
Á Á G G
î Ñ ¯
with ~
Ô ¢ ç Ôz¢ ¾ Ô
¬
ê Ñ î Ñ »ø » Ñ
ñ ñ »Á ñ (6.16)
Notice that the solution is expressed in Equation (6.16) as the superposition of the natural
modes (eigenvectors) of the homogeneous equations. ø » If the forcing function (load vector) is close
to one of these modes the corresponding coefficient will be large and will dominate the response
- if it coincides then resonance will occur. Very often it is unnecessary to evaluate all ³ eigenvectors
ú
of the system; the higher frequency modes can be ignored and the solution adequately represented
by superposition of the eigenvectors associated with the lowest eigenvalues, where ³ .
nal nature of Equation (6.12) is to approximate the overall energy dissipation of the finite element
system with proportional damping
èà ¢ } ù
} Â Ð » » »
Þ
 (6.17)
» » f
where is a modal damping parameter and  is the Kronecker delta. Equation (6.12) now reduces
Þ ®
to ³ equations of the form Ô2 } â Ô2 ¬ ¬ Ô ¢ Ô
øß » Ñ Ð » »ø » Ñ ø Ñ » Ñ
ñ ñ ñ ñ (6.18)
Ý
f f
with solution (the Duhamel integral)
à
Ô ¢ Ï Ô ¼ Ô ô dõ í
þ ¯ J 7
¯
A
ø » Ñ » Ñóò » Ñ ò » » »ö »
ñ » + ñ ñ + ñ ñ (6.19)
Ý ®
¢ ¼ }
À ô õ f f f
where
»
f
»
Ï
Ô
» . » and » are calculated from the initial conditions Equation (6.15).
ø » Ñ Ô
Once thef componentsf ñ have been found from Equation (6.19) (or alternative time integration
Ñ
methods applied to (6.18)), the solution ê ñ is expressed as a superposition of the mode shapes
»
by Equation (6.16).
operator, is an inhomogeneous source term and is the dependent variable. The fundamental
solution for this equation is a solutionü of ç ù Ô Ôz¢ò
Ñ Ñ
ü (7.1)
Þ
f
where * indicates the adjoint of the operator and is the Dirac delta function. No specific
Þ
boundary conditions are prescribed but in some cases regularity conditions at infinity need to be
satisfied. The fundamental solution is a Green’s function which is not required to satisfy any
boundary conditions and is therefore also commonly termed the free-space Green’s function.
The mathematical theory required to determine the fundamental solution of a constant coef-
ficient PDE is well-developed and has been used successfully to determine the fundamental so-
lutions for a wide range of constant coefficient equations (Brebbia & Walker 1980) (Clements &
Rizzo 1978) (Ortner 1987). Fundamental solutions are known and have been published for many
of the most important equations in engineering such as Laplace’s equation, the diffusion equation
and the wave equation (Brebbia, Telles & Wrobel 1984a). However, by no means can it be guaran-
teed that the fundamental solution to a specific differential equation is known. In particular, PDEs
with variable coefficients do not, in general, have known fundamental solutions. If the fundamental
solution to an operator cannot be found then domain integrals cannot be completely removed from
the integral formulation. Domain integrals will also arise for inhomogeneous equations.
118¾ D OMAIN I NTEGRALS IN THE BEM
Wu (1985) argued that the BEM has several advantages over other numerical methods which
justify its use for many practical problems - even in cases where domain integration is required.
He argued that for problems such as flow problems a wide range of phenomena are described by
the same governing equations. What distinguishes these phenomena is the boundaryÆ conditions of
the problem. For this reason accurate description of the boundary conditions is vital for solution
accuracy. The BEM generates a formulation involving both the dependent variable and the flux
. This allows flux boundary conditions to be applied directly which cannot be achieved in either
the finite element or finite difference methods.
Another advantage of the BEM over other numerical methods is that it allows an explicit ex-
pression for the solution at an internal point. This allows a problem to be subdivided into a number
of zones for which the BEM can be applied individually. This zoning approach is suited to prob-
lems with significantly different length scales or different properties in different areas.
Domain integration can be simply and accurately performed in the BEM. However, the pres-
ence of domain integrals in the BEM formulation negates one of the principal advantages of the
BEM in that the problem dimension is no longer reduced by one. Several methods have been de-
veloped which allow domain integrals to be expressed as equivalent boundary integrals. In this
section these methods will be discussed.
The domain integral in this formulation does not involve any unknowns so domain integration can
be used directly to solve this equation. This requires discretising the domain into internal cells
in much the same way as for the finite element method. As the domain integral does not involve
any unknown values accurate results can generally be achieved using a fairly coarse mesh. This
method is simple and has been shown to produce accurate results (Brebbia et al. 1984a). This
approach, however, requires a domain discretisation and a numerical domain integration procedure
which reduces the attraction of the BEM over domain-based numerical methods.
®
Equation (7.2). Using Green’s second ®
identity ã
¼
Å
Å
ö³ ³
È
¢
ö
(7.3)
¼
ö
ö
9¡
æ ä
® be avoided for certain forms of . If a can be found which satisfies
®
¢
domain integration can ¢ ò
being harmonic (Å ) Green’s second identity
Å
f
, where is the fundamental solution of Laplace’s equation, then for the specific case of
f ã can be reduced to
ö
¢ ¼ 8¡
È ö³ ö³
ö
(7.4)
æ f ä
Therefore if a Galerkin vector can be found and is harmonic the domain integral in Equation (7.2)
can be expressed as equivalent boundary integrals.
Fairweather, Rizzo, Shippy & Wu (1979) determined the Galerkin vector for the two-dimensional
Poisson equation and Monaco & Rangogni (1982) determined the Galerkin vector for the three-
dimensional Poisson equation. Danson (1981) showed how this method can be applied successfully
for a number of physical problems involving linear isotropic problems with body forces. He con-
sidered the practical cases where the body force term arose due to either a constant gravitational
load, rotation about a fixed axis or steady-state thermal loading. In each of these cases the domain
integral can be expressed as equivalent boundary integrals.
This Galerkin vector approach provides a simple method of expressing domain integrals as
equivalent boundary integrals. Unfortunately, it only applies to specific forms of the inhomoge-
neous term (i.e., is required to be harmonic).
7.2.3
Æ Complementary
Æ Function-Particular Integral method Æ
ü ¢
A more general approach can Æ
ü be developed
¢ ò
Æ
using particular solutions. Æ
Consider theü linear
¢ problem
!
the homogeneous equation Æ ¢ Æ , and !
. can be considered as the sum of the complementary function , which is a solution of
Æ a particular solution which satisfies "but is "
! #"
not required to satisfy the boundary conditions of the problem. Applying BEM to the governing
¼ gives¢ ¼
%$
equation using the expansion
Æ ê ê $ (7.6)
"
'&
If a particular solution can be found, all values on the right-hand-side of Equation (7.6) are
¼ ¢
known - reducing the problem to
& ê (7.7)
where is a vector of known values. This linear system can be solved by applying boundary
conditions.
This approach can be applied in a situation where an analytic expression for a particular solu-
tion can be found. Unfortunately particular solutions are generally only known for simple operators
and for simple forms of . Alternatively an approximate particular solution could be calculated nu-
merically. Zheng, Coleman & Phan-Thien (1991) proposed a method where a particular solution
is determined by approximating the inhomogeneous source term using a global interpolation func-
tion. This approach is a special case of a more general method known as the dual reciprocity
boundary element method.
7.3 Æ
Domain Integrals Involving the Dependent Variable
ü ¢ ò
ü ü
Consider the linear homogeneous PDE . For many operators the fundamental solution to
the operator may be unobtainable or may be in an unusable form. This is especially likely if ü
Æ
ü that ¢ itò is particularly difficult to
ü find
(
involves variable coefficients for which case it has been shown
(
a fundamental solution. Instead, a BEM formulation can be derived based on a related operator
with known fundamental solution. A BEM formulation for based on the operator will
be of the form
¼ ¢ ¼
ü ¼Cü
Æ
)( %*
ê È (7.8)
æ f ü
where Æ is the fundamental solution corresponding to the operator . This integral equation is (
similarf to Equation (7.2). However in this case the domain integral term involves the dependent
variable . This problem could be solved using domain integration where the internal nodes are
treated as formal problem unknowns.
7 .3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 121
method with a perturbation method. He considered the two-dimensional generalised Laplace equa-
tion Z ù Ô ù Ô^Ô ¢ò
Å
Ñ <Ñ ø ,+ Å ]
:
Ñø
(7.9)
ù Ôµ¢ Ï Æ
ù Ô
Using the substitution ]
Ñø
Ð Ñø +
Equation (7.9) can be recast as a heterogeneous
® Æ Æ
Helmholtz equation Ç ù Ô ¢ ò
Ñø
Ç Å (7.10)
.- -
Rangogni treated this equation as a perturbation about Laplace’s equation. He considered the
® Æ Æ
class of equations %Ç ù Ô ¢ò ò Ï
Ñø
Å where (7.11)
s
for which he sought a solutionÆ of the
¢
Æîþ form
Æ
¬` #- .- ®Æ
®w ¢ ¾ þ
Æ
-
Â
~
ÂÎÁ
Â
- (7.12)
® Æþ
.-®Æ
Substituting Equation (7.12) into Equation
Æþ
¬`Ç (7.11)
® ® Æ
and grouping
Æ
®wÇ powers
¬ .-
of¢ò gives
Å
Å -
Å
~ - (7.13)
0/
Rangogni (1991) extended this coupled perturbationÆ - boundary
Æ element method to the general
® Æ PDE
second-order variable coefficient §Ç ù Ô
ö
ù Ô
ö
¢ ù Ô
Ñø é Ñø Ñø
Å öÕø ö (7.14)
L
122 D OMAIN I NTEGRALS IN THE BEM
#- '¢ / ù Ô -
Æ Æ
® Æ
He considered the family
of equations
h
Ç ù Ô
ö
ù Ô
ö
ò Ï
Ô
Ñø éÑ ø Ñø Ñ
Å öÕø ö (7.15)
i
Applying the perturbation method to this family of equations allows Equation (7.15) to be ex-
pressed as an infinite series of distinct Poisson equations which can be solved using the boundary
element method. Again Rangogni used an domain mesh to solve these Poisson equations. Ran-
gogni found that in practice convergence was rapid and accurate results were produced.
Gipson, Reible & Savant (1987) considered a class of hyperbolic and elliptic problems which
can be transformed into an inhomogeneous Helmholtz equation. They used the perturbation method
to recast this as an infinite sequence of Poisson equations. They avoided domain discretisation by
using a Monte Carlo integration technique (Gipson 1987) to evaluate the required domain integrals.
Lafe & Cheng (1987) used the perturbation method to solve steady-state groundwater flow
problems in heterogeneous aquifers. They showed the method produced accurate results for sim-
ply varying hydraulic conductivities with convergence after two or three terms. Lafe & Cheng
investigated the convergence of the perturbation method. They found that for rapidly varying hy-
draulic conductivity convergence is not guaranteed. From this investigation they concluded that
accurate results can be obtained so long as the hydraulic conductivity does not vary by more than
one order of magnitude within the solution domain. If the hydraulic conductivity variation is more
significant they recommend using the perturbation method in conjunction with a subregion tech-
nique so that the variation of conductivity within each subregion satisfies their requirements. This
process could become computationally expensive, particularly if convergence is not rapid, as the
solution of multiple subproblems will be required within each subregion.
where is the known fundamental solution to Laplace’s equation applied at point . To avoid
domainf discretisation the domain integral in Equation (7.17) needs to be expressed as equivalent
7 .3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 123
¬
boundary integrals. Using MRM this is achieved by defining a higher-order fundamental solution
® þ
such that ¬z¢
f
Å (7.18)
f f
Using this higher-order fundamental solution the domain integral in Equation (7.17) can be written
as þ þ þ ®
¢ ¬B
ç È ç Å È (7.19)
æ f æ f
Æ
or þ þ ã Æ ¬ ® þ
¢ ¼ ¬ 9¡ ¬
ö ö
ç È ö ö Å ç È (7.20)
æ f³ ³ þ æ
f f ä f
¬
This formulation has generated a new domain integral. ç is a known function so we can introduce
® þ
a new function ç which can be determined analytically
¬N¢ from the relationship
ç Å ç (7.21)
giving ® þ
¬ ¢ ¬ ¬s
Å ç È ç È (7.22)
æ f æ f ®
®
®i¢ higher-order
This process can be repeated by introducing a new ¬ fundamental solution such that
f
Å (7.23)
f f
and continuing until convergence is reached.
This procedure is based on the recurrence relationships
®
¬z¢ ëû¢ò ù Ï ù ù
®
¬¢ ëû¢ò ù Ï ù Ð ù
ç Šç for (7.24)
Ð
Å Â Â for (7.25)
þ f Æ f the boundary
s
Ô
Using theseÆ recurrenceã relationships
Ô
Æ
ö
¼
þ
ö
gives
8¡ ¾ þ
ã integral¬ formulation
ö Â
¼ ¬
ö
8¡ ¢ò
Ñ Ñ çÂ
Ä ö ö ç Â ö Â ö (7.26)
f³ ³ ÂÎÁ f ³ ³
f ä f ä
which is an exact formulation if the infinite series converges. Errors are only introduced at the
stage of boundary discretisation.
g
124 D OMAIN I NTEGRALS IN THE BEM
s
¼ ¢ ¾
Introducing interpolattion functions and discretising the boundary gives the matrix system
¼ Ô #
ê 1 þ
Ñ 1 .243%5687 243%5 7
ÂÎÁ
ÿ (7.27)
.2395 243%5
where and
damental solutions and
687 7
are influence coefficient matrices corresponding to the higher-order fun-
and ÿ contain the nodal values of ç Â and its normal derivative.
The MRM can be applied based on operators other than the Laplace operator. This approach
relies on knowledge of the higher-order fundamental solutions necessary for application of the
method. These solutions have been determined and successfully used for the Laplace operator
in both two and three dimensions but the extension of the method to other equation types needs
further research. Itagaki & Brebbia (1993) have determined the higher order fundamental þ solutions
þ þ Æ Helmholtz equation.
for the two-dimensional¢ ç ù ù Ô
modified
The MRM can be extended to other equations by allowing the forcing function ç to be a general
Ñ ® Æ will ® be
Æ restricted toþ cases where
® Æ
function such that ç ç ñ . The MRM ¢ò ¢¼ the recurrence
 ŠÂ
+ Â
(7.28)
s Æ
Æ
In this case the boundary integral
Ô
formulation® will
Ô ¾ þ
ã Æ be
¼ 9¡ ¢ò
Ä
Ñ Ñ Â
ö
ö
+ Â
 ö
ö
(7.29)
ÂÎÁ f³ ³
f ä
Å ç ¢ çNÔ (7.30)
Æ Æ Ñ
ç ù Ô function ç can be completely ¢ general.
¢ forcing
The ù Ôç
ç ù If ç then ç is a known function of posi-
tion and the differential equation described is simply the Poisson equation. For potential problems
Ñ Ñ
ç ç and for transient problems ç ç ñ . Applying the BEM to Equation (7.30) will
7 .3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 125
give
¼ ¢ ¼
ê ç È (7.31)
æ f
where is the known fundamental solution to Laplace’s equation. The aim of the DR-BEM is to
expressf the domain integral due to the forcing function ç as equivalent boundary integrals.
The DR-BEM uses the idea of approximating ç using interpolation functions. A global approx-
imation to ç of the form
¢ ¾
¬
ô Ç :
ç Â Â (7.32)
ô Ç ÂÎÁ
x
is proposed.  are unknown coefficients and  are Ç approximating functions used in the interpo-
lation and are generally chosen to be functions of the source point and the field point of the fun-
damental solution. The approximating functions  are applied at different collocation points
Æ
- called poles - generally most, but not all, of which are located on ü the ¢ boundary of the problem
domain. Æ
As discussed inÆ the previous chapter Æ
ü
Æ
¢ò
ü the
!
¢ solution to a linear PDE can be constructed as
!
" "
the sum of a complimentary function (which satisfies the homogeneous equation Æ ) and a
(
particular solution to the equation Ç . Instead of using a single particular solution, which
may be difficult to determine, the DR-BEM employs a series x of particular solutions  which are
® Æ Â
related to the approximating functions ¢# as Ç shownëû¢ in Equation
Ï ù ù (7.33).
Å (Â Â ~ (7.33)
By substituting Equations (7.32) and (7.33) into Equation (7.30) the forcing function ç is approxi-
mated by a weighted summation of particular
®Æ
¢ ¾
¬
ô :
solutions ® toÆ the Poisson equation.
Å
ÂÎÁ
 Š (
(7.34)
The DR-BEM essentially constructs an approximate particular solution to the governing PDE as a
summation of localised particular solutions.
With the governing equation rewritten in the form of Equation (7.34) the standard boundary
element approach can be applied. Equation (7.34) is multiplied by a weighting function and
integrated over the domain. Green’s theorem is applied twice and the fundamental solution f of the
Laplacian is used to remove the remaining domain integrals. The name dual reciprocity BEM is
derived from the application of reciprocity relationships to both sides of Equation (7.34). After
applying these steps Equation (7.35) is obtained, where the fundamental solution pole is applied at
126 D OMAIN I NTEGRALS IN THE BEM
Æ
Ô§
point .Æ ãÆ
Ô ¼ 8¡
ö ö
Ñ Ñ
Ä
ö
f³
f
ö
³
¢ ä ¾ :¬ ô Ô Ô
Æ ã Æ
¼
/ Æ
8¡ 8?>
ö ö
Â^Á
<;= Ä Ñ
 Â
Ñ
(Â ö ³ f
ö
º
³
ñ Â
(7.35)
f ä
In implementing a numerical solution of this equation similar steps are taken as for the standard
Æ discretised into elements and interpolation functions are introduced to ap-
BEM. The boundary is Ç
Æ
(
proximate the dependent variable within each element.
The form of each  is known from Equation (7.33) once Æ the Æ approximating functions  have
been defined. It is not necessary to use interpolation functions to approximate each  . However (
Æ
by using the same interpolation
functions to approximate and  the numerical implementation (
(
will generate the same matrices and on both sides of Equation (7.35). The error generated
by approximating each  in this manner has been found to be small and can be justified by the
improved computational efficiency of the method (Partridge et al. 1992).
The application of this method results in the system
¼ ¢ ¾ ô ¼ Ô . A@ CB 7 B 7
x ê
¬
Â
Ñ ê x
(7.36)
ÂÎÁ ¢
where the poles were chosen to be the boundary nodes plus internal points so that
. Although it is not generally necessary to include poles at internal points it has been found
that in general improved accuracy is achieved by doing so (Nowak & Partridge 1992). It has
been shown that for many problems (Partridge et al. 1992) (Huang & Cruse 1993) using boundary
points only in this procedure is insufficient to define the problem. In general using internal points
is likely to improve the solution accuracy as it increases the number of degrees of freedom. No
theory has been developed of how many internal collocation points should be used for optimal
accuracy, or where these points should be positioned within the problem domain. Using internal
poles in this interpolation does not require domain discretisation - it is only necessary to specify
DB EB
the coordinates of the internal collocation points. The internal points can be chosen to be locations
B7 B 7
where the solution is of interest.
The ê and
. *GF
vectors can be treated as columns of the matrices and respectively. This
)
allows Equation (7.36) to be rewritten ¼ as ¢ ¼
F ê DB EB
F F is a vector containing the nodal values of . To solve this system it is necessary to evaluate
(7.37) ô
¢JIKF I
. is defined by Equation (7.32) which,F for the nodal values, can be expressed in matrix form as
where
H . If the matrix is nonsingular this expression can be rearranged to give Equation (7.38)
which provides an explicit expression for F. ¢0I ¬
H (7.38)
7 .3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 127
F
Including this explicit expression for¼. in Equation ¼ * I ¬
) DB EB H
¢ (7.37) gives
ê
(7.39)
The approach taken to solve thisÇ equation will depend on the form of ç .
I
The accuracy of the DR-BEM hinges on the accuracy of the global approximation to the forcing
function ç (defined by Equation (7.32)). Therefore Ç the choice of the approximating functions  is
Æ
a key consideration when implementing the DR-BEM. The only requirement so far prescribed on
the form of the approximating functions  is that the matrix generated should
(
Ç be nonsingular
and that the related particular solutions  can be determined and can be expressed in a practical
into investigating what form of  should be used in Æ
closed form. Some work Ç has been conducted
a given situation to provide the highest accuracy and computational efficiency.
Usually a form of  is defined and this can be used, applying ç ù Ô ¢½¼ Ï
ß
(
( (7.33), to specify
Equation
ç ù Ô ¢
and . The fundamental Ï solution of Laplace’s equation is
Ñ
Ð'Ü in two-dimensional
space and
Ñ ç
Ü in three-dimensional space - where is the Euclidean distance between
f
Ý
ð Ý
f Ç ¢½
the field point Ç theÔ source
and Ý point of the fundamental Ç solution. Due to the dependence of this
fundamental solution only on the approximating function is generally chosen to be some radial
Ñ Ý
function i.e.,   . Several other options for  have been tried (Partridge et al. 1992) but it
Ý
Ç
has been found that in general the most accurate results were generated using some radial function.
For both two and three-dimensional problems Wrobel, Brebbia & Nardini (1986) recommended
® n
choosing  from the series Ç ¢
Ï
   ë ~  (7.40)
Ý Ý Ý
where  is the distance between the field point (node Ç ) and the DR-BEM collocation point (node
Ý
à ). They showed that accurate results can be achieved using Ç some combination of terms from Ç this
 Æ
series. Generally Ç the same approximating function is used at all the collocation points so in this
thesis, for simplicity, the form of approximating Ç ¢ Ç Ô functions  will be referred to by a single .
Choosing to be a function of only one variable simplifies the process of determining and .
Ñ ® Æ ( (
For two-dimensional problems, if then¢the Ç relationship Ô
Ý
Å ( Ñ
Ý
(7.41)
®Æ Æ
can be reduced to the ordinary differential equation
Ï ¢#Ç
®
Ç
( (
Æ (7.42)
Ý
Using defined by Equation (7.40) the corresponding forms of and , for two-dimensional
Ý Ý
( (
128¾ D OMAIN I NTEGRALS IN THE BEM
® n ®
problems, can be shown Æ to be
¢ ®
Ô
( ¢
Ýã
ð
ÝÁ
ÑML ã Ý Ï Ð
n
(7.43)
( öÕø ö
L
Ý ¼ ö ³ ÝON ö ³
I
Ð Ý Ý Ð (7.44)
¢ ¼ ¢ ¥
Ç ä ä
ø ø »
ON
I Â Â »
where and .
Ý Ý
Any combination of terms from Equation (7.40) Çu¢ can be used for specifying . It has been found
Ï
that in general including higher-order terms leads to little improvement in accuracy (Partridge
et al. 1992). The most commonly used form is
Ý Ç
as this approximation will generally give
accurate results with greater computational efficiency than other choices.
Equation (7.40) was recommended as a basis for the approximating function due to Ç the
Ç
particular form of the fundamental solution of Laplace’s equation and its dependence on
Ý
only. If
a different operator is used as the basis of the DR-BEM then it is likely a different form ofÇ will
be more appropriate. The choice of in this case will be discussed in Section 7.3.3.
The performance of the DR-BEM hinges on the choice of the approximating function . The
theory of how to determine the best approximating function is therefore a vital component of
the DR-BEM. Unfortunately the approximating function has generally been chosen and used in a
rather ad-hoc manner. Recently some more formal analysis of the use of approximating functions Ç
has been undertaken.
Golberg & Chen (1994) argued that a formal analysis of the approximating function can be
undertaken using the theory of radial basis functions. Radial basis functions are a generalisation Ç
of cubic splines in multi-dimensions. Cubic splines are known to be optimal for one-dimensional
interpolation. Therefore, rather than being an Ç arbitrary choice, it seems that choosing to be
a radial function is a logical extension for two or three-dimensional problems. Golberg & Chen
showed that, for the Poisson equation, Çó¢ choosing to be a radial basis function ensures convergence
of the DR-BEM. Ï
They also demonstrated that Ç ¢ is a specific member of the group of radial basis
Ý
I
Ç¢
functions. The
theory of using radial basis functions for multi-dimensional approximation is fairly
Ï
advanced. It has been shown that is optimal for three-dimensional problems which justifies
Ý ®
the use of
Ý
when applying the DR-BEM to three-dimensional problems Çý¢ - the constant
is included to ensure a non-zero diagonal for . HoweverÇ for two-dimensional problems it has
é
been shown that optimal approximation is attained using the thin plate spline ó¸ F . This
Ý Ý
observation lead Golberg & Chen to suggest that choosing to be a thin plate spline may improve
the accuracy of the DR-BEM in two dimensions. Recently Golberg (1995) has published a review
of the DR-BEM concentrating on developments since 1990 concerning the numerical evaluation
of particular solutions.
Inhomogeneous Equations
I F
If the forcing function ç is a function
¢ PIQF
of position only then the differential equation under consid-
using singular value decomposition. For large systems they found the computational efficiency
could be improved by employing the conjugate gradient method. Coleman et al. (1991) success-
fully solved inhomogeneous potential and elasticity problems which are governed by operators
other than the Laplacian.
Elliptic Problems F
If ç is a function of the dependent variable then will also be a function of the dependent variable.
® Æ Æ
¢ ò
Consider, for example, the linear second-order differential equation
¢ ¼
Æ
F X¢ I Z¬ Y
Å (7.47)
ê
In this case ç
) D B E B * I
so . Applying the DR-BEM to Equation (7.47), based on the
¬
¼ ¢ ¼ gives
fundamental solution to Laplace’s equation, ¼
ê
ê
(7.48)
Ñ î ê î
]\
R
where ¦ (7.49)
¢
¬
Again, by applying boundary conditions Equation (7.49) can be reduced to a linear system
which can be solved to determine the unknown nodal values.
I
Due to the presence of the fully-populated matrix in Equation (7.49) it is not possible to
solve the boundary problem and internal problem separately. Instead the solution can be treated as
a coupled problem and the solutions at boundary and internal nodes are generated simultaneously.
Derivative Terms The DR-BEM can also be applied for elliptic problems where ç involves
derivatives of the dependent variable (Partridge et al. 1992). Consider, for example, the differ-
130 D OMAIN I NTEGRALS IN THE BEM
Æ
ential equation ®Æ
¢ ò
ö
Å öø (7.50)
¬
. * G I ^
) D B E B %^ _
In this case applying DR-BEM, using the Laplace fundamental solution, gives
¼ ¢¼ ¼
ê ê
(7.51)
`
To solve this problem it is necessary to relate the nodal values of to the nodal values of öÕø . This
ö
`
is achieved by using interpolation functions to approximate ê in a similar manner as was used to
a : dkjMl8m on d
approximate in Equation (7.32). A global approximation function of the form
<` b egfih
c
d
õ
d d
(7.52)
h
õ
can be used to approximate ` where are the chosen interpolation functions and are the un-
known coefficients. In system form this can be expressed as
p bJqGr (7.53)
Although it is not necessary, equating q to s improves the computational efficiency of the method
^ ^ Differentiating Equation (7.53) gives
^%p_ ^9t q
as only one matrix inversion procedure is required.
b r (7.54)
Choosing qubvs and inverting Equation (7.53) to give an explicit expression for r allows Equa-
w w f
w%px w t s p
tion (7.54) to be rewritten as
b s]y (7.55)
j,z {}|~n
Equation (7.39) can now be rewritten as
| z f ww t f
p b~ where b
s]y s sQy (7.56)
By applying boundary conditions Equation (7.56) can be reduced to a linear system which { can be
As mentioned earlier, the approximating function is generally chosen to be b
solved to give the unknown nodal values.
can lead to numerical problems if derivative terms are included in the forcing function a . As shown
. This
in Equation (7.55) derivative terms require derivatives of to be evaluated. For example, evaluating
7 .3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 131
w {.
w ts l
O
bP gives
l l
the matrix requires calculation of . Using the approximating function
b b
(7.57)
This derivative function can become singular, so - as shown by Zhang (1993) - significant numerical
error may result. This will especially be the case in problems where collocation points are located
close together.
Zhang (1993) suggested two possibilities for avoiding this problem. The first suggestion in-
volved using a mapping procedure to map the governing equation to an equation without convec-
tive terms. This method was shown to produce accurate results but is somewhat cumbersome and
{ {{
can only be applied to linear problems. A simpler approach is to choose an approximating func-
bv bP
tion which does not lead to singularities for convective terms. Zhang recommended use of either
or . These approximating functions produce accurate results and can be
simply applied for both linear and nonlinear problems. Zhang recommended the adoption of these
`
approximating functions for all use of the DR-BEM.
The same idea of using Equation (7.53) to allow nodal values of to be associated to its
derivatives can be applied to extend the DR-BEM to cases involving higher-order derivatives or
cross derivatives of the dependent variable. Appropriate approximating functions need to be chosen
to avoid the problem of singularities.
}{ [jx%n
Variable Coefficients The DR-BEM can be readily extended to equations with variable coeffi-
cients. Consider the variable coefficient Helmholtz equation
¥j¦l m n «¬«¬«
i.e.,
«¬«¬«
where µ is the number of collocation points used in applying the DR-BEM.
¶
132 D OMAIN I NTEGRALS IN THE BEM
which is a boundary-only expression for the variable coefficient Helmholtz equation. This method
is general and can easily be extended to accommodate variable coefficient derivative terms and a
sum of variable coefficient terms.
a
Formulating the DR-BEM for a General Elliptic Problem In this section it has been shown
a
how the DR-BEM can be applied for elliptic problems with varying forms of . The DR-BEM can
a a b a a (7.62)
¦j l8m An jMl8m on {ÁAj l8m on l {ÂÃj l8mAn {.ÄÅjMl8m on
Consider a two-dimensional equation of the form
` bÀ ` ` `
(7.63)
¸ z f
b| ¸Ê É
º{X ËÍÌ w t s y {}Î w
where
fÒÑ
ws w9Ïs Ð
(7.65)
b £ s]y
Ì Î Á (7.66)Â
£ , and Ç are diagonal matrices where the diagonalsÄ contain the nodal values of À , and
respectively. is a vector containing the nodal values of .
Ó
BEM can be be applied using essentially any operator of appropriate order with known fundamental
solution. If an appropriate operator can be found the complexity of the forcing function can be
reduced. This should improve the accuracy of the method. The problem with applying the DR-
BEM based on another operator is in choosing the approximating function . A choice of which
ÕÔ
produces accurate results is required but it is also necessary to choose an for which a particular
solution can be determined.
7 .3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 133
Jb
Radial functions have generally been used when applying the DR-BEM. Along the lines of Wrobel
½ { ÕÔ ½
et al. (1986), Zhu chose an approximating function of the form where is a positive
½ Õ Ô ½Õ Ô }{ Õ ×
integer. Determining the particular solution requires solving the ordinary differential equation
b (7.68)
(7.69)
Ù Ú Ø
Ú Ü
where the material parameters , , and À are all assumed to be homogeneous. They applied
lÕ Õ
the DR-BEM based on the steady-state convection-diffusion operator
ÙQ Õ #Ú ÛÚØÜ À Õ b0
(7.70)
Instead of defining a form of the approximating function and solving for Õ Ô Partridge et al. chose
to define Õ Ô and use Equation (7.71) to determine the corresponding approximating function. Al-
though somewhat ad-hoc this approach was found to produce accurate results.
Chapter 8
Õ b
(8.1)
where the diffusivity is a material parameter which can be a constant or a function of position.
Ö× Ö ×kÞgß Ö × {à¯Ö
8.1.1 Coupled Finite Difference - Boundary Element Method
b à¯Ö
This approach discretises the time-domain in a finite difference form. Consider the variation be-
tween a time and a time . The most common approach (Brebbia et al. 1984b) is
Õ Ö b Õ àá Ö Õ
fully implicit finite difference scheme
(8.2)
Using this finite difference approximation the original parabolic equation has been reduced to an
elliptic equation. Using the weighted residuals method an integral equation can be generated from
136ç T HE BEM FOR PARABOLIC PDES
b0
í í (8.5)
applied at some source point ò . The fundamental solution of the modified Helmholtz equation
èäé êë
is known in both two and three dimensions. If an internal solution is required at a specific time
ò bv
this can be determined explicitly from Equation (8.4) where the fundamental solution is applied at
internal point and .
Unfortunately Equation (8.4) contains a domain integral. This integral is generally evaluated by
using a domain mesh (Brebbia et al. 1984b). The domain integral does not include any problem un-
z
knowns so a fairly coarse domain mesh will generally suffice. Applying the BEM to Equation (8.4)
(8.6)
internal solution at b
. A solution at internal nodes can then be calculated. The
time-stepping procedure can be repeated using the z as pseudo-initial
conditions for the next time-step.
÷ x once and stored.
x If a constant time-step is used the matrices , Ö ×äÞgß
The boundary conditions can be applied to form a solution system of the form ü
and can be calculated
óõô%ö bý where
óþô%ö is the vector of unknown nodal values at time and ý is a vector constructed x from
conditions at each time-step it is only necessary to update ý and solve the system for óõô9ö . If a
known nodal values from the previous time-step. For a problem with time-independent boundary
problem has time-dependent boundary conditions the solution system needs to be reformed at each
time-step.
This coupled finite difference - boundary element method (FD-BEM) was first proposed by
Brebbia & Walker (1980) for the diffusion equation. It was implemented and investigated by
Curran, Cross & Lewis (1980). They found that this method will only produce accurate results
if Equation (8.2) accurately approximates the time derivative. This will generally require small
time-steps to be adopted. Curran et al. investigated the use of a higher-order approximation to the
time-derivative. They found that this improved the accuracy of the method. Unfortunately it lead
to a deterioration in convergence behaviour.
é x¢ðÖ ë xVðÖ ×äÞgßã { é ë é Vx ðÖ × ë
Tanaka, Matsimoto & Yang (1994) proposed a generalised version of this time-stepping scheme.
ÿ Õ
They approximated the time variation within an interval as
Õ bÿ Õ<â (8.7)
8 .1 T IME -S TEPPING M ETHODS 137
ÿ
where , termed the time-scheme parameter, is a constant in the range . Substituting ÿ
ä× Þgß { é ë × Õ ×kÞgåß à¯ Ö Õ ×
this approximation and a first-order finite difference approximation of the time derivative into the
diffusion equation gives
ÿ Õ ÿ Õ b (8.8)
If ÿÛbX this approximation of the diffusion equation is equivalent to the standard FD-BEM dis-
ß
cussed earlier. An integral equation can be derived from Equation (8.8). Tanaka et al. implemented
ÿ#b ÿ#b
this method and found it gave accurate results for a range of diffusion problems. They tested the
ÿ]b
accuracy for a Crank-Nicolson scheme ( ), a Galerkin scheme ( ) and a fully implicit
scheme ( ). They found that the best results were achieved using a Crank-Nicolson scheme.
é ë é Vx ðÖ ë Ñ é
treated directly in the time domain by directly integrating over both time and space. The weighted
Õ ò
b
í (8.9)
é ë é ð V
x
ð Ö
ð Ö ë é ë ñ éÖ ë
ð¡xVí ðÖ ðÖ {
{ ñ
ò
oí ò Ö
í ò
b0
(8.11)
This time dependentÖ fundamental solution is known in two and three dimensions. Physically this
x Ö
è9é ë
at a point ò at time
. If an internal solution is required at a specific time this can be determined
fundamental solution represents the effect at a field point at time of a unit point source applied
Öûú
the updated pseudo-initial conditions. The second time-stepping procedure involves only a domain
ú
integral at so, ideally, a domain integral only needs to be calculated once. This, however, will still
Õ bJ ú
require the user to create a domain mesh. As mentioned by Brebbia et al. (1984b), in many practical
cases the domain integral can be avoided. If the initial conditions are
Õ 'b
throughout the body
the domain integral equals zero. If the initial conditions satisfy Laplace’s equation then
a Galerkin vector can be found and the domain integral can be expressed as equivalent boundary
Öûú
integrals. This includes many practical cases such as constant initial temperature or an initial linear
temperature profile. z
Unfortunately, in practice it is not always feasible to restart the integration process at . At
each time-step new and matrices are required so if many time-steps are required the storage
capacity of the computer is likely to be exceeded. This requires the procedure to be restarted
at some time where an internal solution is constructed and used as pseudo-initial conditions to
repeat the process. Therefore, in practice, both time-stepping methods are likely to require domain
integration.
é ¢x ðÖ ë
a solution in the original space can be attained using an inverse transform procedure. The most
é ë
appropriate transform approach for parabolic problems is the Laplace transform.
Consider the diffusion equation
xVðÖ Õ Ö
Õ b é xÈðÖ ë (8.12)
é xÈð ë
with appropriate boundary and initial conditions. The Laplace transform of Õ
and is defined as é
will be sym-
xÈð ë » é xÈðÖ ë ½ Ö
bolised as
b ú y Õ (8.13)
ú x (8.14)
used by other practitioners (Moridis & Reddell 1991) (Zhu, Satravaha & Lu 1994). Liggett & Liu
(1979) compared the Laplace transform method with the time-dependent Green’s function method.
They noted that the direct method is simpler to apply. However, due to its greater efficiency, they
recommended the Laplace transform method for solving diffusion problems.
One limitation of the Laplace transform method is that Equation (8.14) is inhomogeneous so
that applying the standard BEM will generate a domain integral involving the initial conditions.
Traditionally this domain integral has been calculated by using a domain discretisation (Brebbia
et al. 1984b). However, recently Zhu et al. (1994) proposed using the DR-BEM to convert this
domain integral term to equivalent boundary integrals. They chose to apply the DR-BEM based
on the known fundamental solution to the Laplace operator. Considering Equation (8.14) this
Ëz ¸Ð
¸
The DR-BEM can now be applied to Equation (8.14), giving a matrix system of the form
p Æ ]b p ù (8.16)
which can be reduced to a square system by applying boundary conditions. Once the solution
is determined for this elliptic equation in the transform space a solution at a given time can be
é xVðÖ ë é ë
constructed using an inversion process.
é xÈðÖ ë Õ
This Laplace transform dual reciprocity method (LT-DRM) can easily be extended to equations
Ö { Èx ð Õ
of the form
Õ b Ó
(8.17)
ËÍz |
¸ Ð
in which case a matrix expression of the form
¸
p ÆQb p ù (8.18)
is generated. Zhu and his colleagues have successfully extended the LT-DRM to solve diffusion
problems with nonlinear source terms.
(8.19)
140 T HE BEM FOR PARABOLIC PDES
Ó
é x ë é Ö ë
where the thermal diffusivity, , is a constant. In this case the global approximation of implies a
separation of variables such that
Õ Ö b ß
(8.20)
z z ß w w p
p Qb
º
Using Equation (8.20), Equation (7.39) becomes
] sy (8.21)
w {}z
wp z p ß
or
×kÞgß ×
Partridge & Brebbia (1990) recommended using a first-order finite difference approximation to
Õ Ö b Õ àá Ö Õ
the time derivative
Õ ï é ë (8.23)
Ö × Ö ×kÞgß Ö × {ÛàáÖ
(8.25)
where ÿ andÖ ×kÿÞg! ß are weighting parameters with values Ö × in the range $# and the time-step is
b é . ë z*) { é ë
É à¯Ö { z Ñ
between times and . Substituting these approximations into Equation (8.22) an
àáÖ
expression at can be derived in terms of values at
relationship defined by Equation (7.24) becomes
Þgß Ö Õ
Õ b Õ b
(8.27)
{ ú »ì Ä Ö Õ ½î ú »ì Ö ï ½4î
ò Õ ò oí b í
(8.28)
where the matrices ö ö etc are the influence coefficient matrices relating to the higher-order
fundamental solutions. This equation can be solved using a time-integration procedure.
 Ã
 {
and using a time-stepping procedure. This requires some interpolation between the two time-levelsï
The most common approach is to solve this system numerically by discretising the time domain
where ÿ has a value in the range 0 to 1. Differentiating ×kÞgß ×these linear approximations gives
Õ 9 b Õï ×kÞgà¯ß Ö × ï Õ ×
ï 9 à¯Ö × (8.32)
b (8.33)
à ¯ Ö × { b ö ù ÿ ù , b
where ö ÿ ù , b ö ÿg ù , b
ö ÿ . This approach is termed a first-order approach as it removes all but theÕ
ï
first derivatives. A second order approach can be formulated by using quadratic interpolation of
and within z3:8theð
z3time-range.
<Vð : <
and only need to be constructed once outside the time-stepping loop.
Using Equation (8.34) the solution can be advanced in time. If a constant time-step is used the
matrices
If the boundary conditions are not time-dependent the boundary conditions only need to be applied
¶
142 T HE BEM FOR PARABOLIC PDES
once.
Bibliography
Brebbia, C. A. & Nowak, A. (1989), A new approach for transforming domain integrals to the
boundary, in R. Gruber, J. Periaux & R. P. Shaw, eds, ’Proceedings of the Fifth International
Symposium on Numerical Methods in Engineering’, Computational Mechanics Publications,
Springer-Verlag, pp. 73–84.
Brebbia, C. A., Telles, J. C. F. & Wrobel, L. C. (1984a), Boundary Element Techniques, Springer-
Verlag.
Brebbia, C. A., Telles, J. C. F. & Wrobel, L. C. (1984b), Boundary element techniques: Theory
and applications in engineering, Springer-Verlag, New York.
Clements, D. L. & Rizzo, F. (1978), ’A method for the numerical solution of boundary value
problems governed by second-order elliptic systems’, Journal of the Institute of Mathematics
Applications 22, 197–202.
Coleman, C. J., Tullock, D. L. & Phan-Thien, N. (1991), ’An effective boundary element method
for inhomogeneous partial differential equations’, Journal of Applied Mathematics and
Physics (ZAMP) 42, 730–745.
Curran, D. A. S., Cross, M. & Lewis, B. A. (1980), ’Solution of parabolic differential equations by
the boundary element method using discretisation in time’, Applied Mathematical Modelling
4, 398–400.
Fairweather, G., Rizzo, F. J., Shippy, D. J. & Wu, Y. S. (1979), ’On the numerical solution of
two-dimensional problems by an improved boundary integral equation method’, J. Comput.
Phys. 31, 96–112.
Gipson, G. S. (1987), Boundary Element Fundamentals - Basic Concepts And Recent Develop-
ments In The Poisson Equation, Computational Mechanics Publications.
Gipson, G. S., Reible, D. D. & Savant, S. A. (1987), Boundary elements and perturbation theory
for certain classes of hyperbolic and parabolic problems, in C. A. Brebbia, W. L. Wendland &
=
144 BIBLIOGRAPHY
G. Kuhn, eds, ’Boundary Elements IX’, Computational Mechanics Publications and Springer-
Verlag, pp. 115–127.
Golberg, M. A. (1995), ’The numerical evaluation of particular solutions in the BEM - a review’,
Boundary Element Communications .
Golberg, M. A. & Chen, C. S. (1994), ’The theory of radial basis functions applied to the BEM for
inhomogeneous partial differential equations’, BE Communications 5(2), 57–61.
Huang, Q. & Cruse, T. A. (1993), ’Some remarks on particular solution used in BEM formulation’,
Computational Mechanics 13, 68–73.
Itagaki, M. & Brebbia, C. A. (1993), ’Generation of higher order fundamental solutions to the two-
dimensional modified Helmholtz equation’, Engineering Analysis With Boundary Elements
11(1), 87–90.
Lafe, O. E. & Cheng, A. H.-D. (1987), ’A perturbation boundary element code for steady state
groundwater flow in heterogeneous aquifers’, Water Resources Research 23(6), 1079–1084.
Liggett, J. A. & Liu, P. L.-F. (1979), ’Unsteady flow in confined aquifers: A comparison of two
boundary integral methods’, Water Resources Research 15(4), 861–866.
Monaco, A. D. & Rangogni, R. (1982), Boundary element method: Processing of the source term
of the Poisson equation by means of boundary integrals only, in K. P. Holz, U. Meissner,
W. Zielke, C. A. Brebbia, G. Pinder & W. Gray, eds, ’Finite Elements In Water Resources
IV’, Springer-Verlag, pp. 19.29–19.36.
Moridis, G. L. & Reddell, D. L. (1991), The Laplace transform boundary element (LTBE) method
for the solution of diffusion-type problems, in ’Boundary Elements XIII’, Computational
Mechanics Publications and Springer-Verlag, pp. 83–97.
Nowak, A. J. (1987), Solution of transient heat conduction problems using boundary-only formu-
lation, in C. A. Brebbia, W. L. Wendland & G. Kuhn, eds, ’Boundary Elements IX Vol 3’,
Computational Mechanics Publications and Springer-Verlag, pp. 265–276.
Nowak, A. J. & Partridge, P. W. (1992), ’Comparison of the dual reciprocity and the multiple
reciprocity methods’, Engineering Analysis With Boundary Elements 10, 155–160.
Ortner, N. (1987), Construction of fundamental solutions, in C. A. Brebbia, ed., ’Topics In Bound-
ary Elements Research’, Springer-Verlag, Berlin and New York.
Partridge, P. W. & Brebbia, C. A. (1990), The BEM dual reciprocity method for diffusion prob-
lems, in ’Computational Methods In Water Resources VIII’, Computational Mechanics Pub-
lications and Springer-Verlag, pp. 397–403.
Partridge, P. W., Brebbia, C. A. & Wrobel, L. C. (1992), The Dual Reciprocity Boundary Element
Method, Computational Mechanics Publications and Elsevier Applied Science.
Polyzos, D., Dassios, G. & Beskos, D. E. (1994), ’On the equivalence of dual reciprocity and
particular integral approaches in the BEM’, BE Communications 5(6), 285–288.
>BIBLIOGRAPHY 145
Rangogni, R. (1986), ’Numerical solution of the generalized Laplace equation by coupling the
boundary element method and the perturbation method’, Applied Mathematical Modelling
10, 266–270.
Rangogni, R. (1991), Solution of variable coefficients PDEs by means of BEM and perturba-
tion technique, in ’Boundary Elements XIII’, Computational Mechanics Publications and
Springer-Verlag.
Rizzo, F. J. & Shippy, D. J. (1970), ’A method of solution for certain problems of heat conduction’,
AIAA Journal 8(11), 2004–2009.
Tanaka, M., Matsimoto, T. & Yang, Q. F. (1994), ’Time-stepping boundary element method applied
to 2-D transient heat conduction problems’, Applied Mathematical Modelling 18, 569–576.
Wrobel, L. C., Brebbia, C. A. & Nardini, D. (1986), The dual reciprocity boundary element for-
mulation for transient heat conduction, in ’Finite Elements In Water Resources VI’, Compu-
tational Mechanics Publications and Springer-Verlag.
Zhang, Y. (1993), On the dual reciprocity boundary element method, Master’s thesis, The Univer-
sity of Wollongong.
Zheng, R., Coleman, C. J. & Phan-Thien, N. (1991), ’A boundary element approach for non-
homogeneous potential problems’, Computational Mechanics 7, 279–288.
Zhu, S. (1993), ’Particular solutions associated with the Helmholtz operator used in DRBEM’, BE
Abstracts 4(6), 231–233.
Zhu, S., Satravaha, P. & Lu, X. (1994), ’Solving linear diffusion equations with the dual reciprocity
method in Laplace space’, Engineering Analysis With Boundary Elements 13, 1–10.
Index