FEM/BEM NOTES

Professor Peter Hunter
p.hunter@auckland.ac.nz

Associate Professor Andrew Pullan
a.pullan@auckland.ac.nz

Department of Engineering Science The University of Auckland New Zealand February 21, 2001

c Copyright 1997 : Department of Engineering Science

Contents
1 Finite Element Basis Functions 1.1 Representing a One-Dimensional Field . 1.2 Linear Basis Functions . . . . . . . . . 1.3 Basis Functions as Weighting Functions 1.4 Quadratic Basis Functions . . . . . . . 1.5 Two- and Three-Dimensional Elements 1.6 Higher Order Continuity . . . . . . . . 1.7 Triangular Elements . . . . . . . . . . . 1.8 Curvilinear Coordinate Systems . . . . 1.9 CMISS Examples . . . . . . . . . . . . 1 1 2 4 7 7 10 14 16 19 21 21 22 22 23 24 25 27 27 27 28 29 30 32 34 35 37 40 41 41 41 41 43

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

2 Steady-State Heat Conduction 2.1 One-Dimensional Steady-State Heat Conduction . . . . . 2.1.1 Integral equation . . . . . . . . . . . . . . . . . . 2.1.2 Integration by parts . . . . . . . . . . . . . . . . . 2.1.3 Finite element approximation . . . . . . . . . . . 2.1.4 Element integrals . . . . . . . . . . . . . . . . . . 2.1.5 Assembly . . . . . . . . . . . . . . . . . . . . . . 2.1.6 Boundary conditions . . . . . . . . . . . . . . . . 2.1.7 Solution . . . . . . . . . . . . . . . . . . . . . . . 2.1.8 Fluxes . . . . . . . . . . . . . . . . . . . . . . . . 2.2 An x-Dependent Source Term . . . . . . . . . . . . . . . 2.3 The Galerkin Weight Function Revisited . . . . . . . . . . 2.4 Two and Three-Dimensional Steady-State Heat Conduction 2.5 Basis Functions - Element Discretisation . . . . . . . . . . 2.6 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7 Assemble Global Equations . . . . . . . . . . . . . . . . . 2.8 Gaussian Quadrature . . . . . . . . . . . . . . . . . . . . 2.9 CMISS Examples . . . . . . . . . . . . . . . . . . . . . . 3 The Boundary Element Method 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 3.2 The Dirac-Delta Function and Fundamental Solutions 3.2.1 Dirac-Delta function . . . . . . . . . . . . . 3.2.2 Fundamental solutions . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

ii

CONTENTS

3.3 3.4 3.5 3.6 3.7 3.8

3.9 3.10 3.11 3.12

3.13 3.14 3.15 3.16

3.17 4

The Two-Dimensional Boundary Element Method . . . . . . . . . Numerical Solution Procedures for the Boundary Integral Equation Numerical Evaluation of Coefficient Integrals . . . . . . . . . . . The Three-Dimensional Boundary Element Method . . . . . . . . A Comparison of the FE and BE Methods . . . . . . . . . . . . . More on Numerical Integration . . . . . . . . . . . . . . . . . . . 3.8.1 Logarithmic quadrature and other special schemes . . . . 3.8.2 Special solutions . . . . . . . . . . . . . . . . . . . . . . The Boundary Element Method Applied to other Elliptic PDEs . . Solution of Matrix Equations . . . . . . . . . . . . . . . . . . . . Coupling the FE and BE techniques . . . . . . . . . . . . . . . . Other BEM techniques . . . . . . . . . . . . . . . . . . . . . . . 3.12.1 Trefftz method . . . . . . . . . . . . . . . . . . . . . . . 3.12.2 Regular BEM . . . . . . . . . . . . . . . . . . . . . . . . Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Axisymmetric Problems . . . . . . . . . . . . . . . . . . . . . . Infinite Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix: Common Fundamental Solutions . . . . . . . . . . . . 3.16.1 Two-Dimensional equations . . . . . . . . . . . . . . . . 3.16.2 Three-Dimensional equations . . . . . . . . . . . . . . . 3.16.3 Axisymmetric problems . . . . . . . . . . . . . . . . . . CMISS Examples . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

46 51 53 55 56 58 58 59 59 59 60 62 62 62 63 65 67 70 70 70 71 71 73 73 74 77 79 81 83 84 86 86 89 90 93 95 97 97 97 97 99 100 101

Linear Elasticity 4.1 Introduction . . . . . . . . . . . . . . . . . . . 4.2 Truss Elements . . . . . . . . . . . . . . . . . 4.3 Beam Elements . . . . . . . . . . . . . . . . . 4.4 Plane Stress Elements . . . . . . . . . . . . . . 4.5 Navier’s Equation . . . . . . . . . . . . . . . . 4.6 Note on Calculating Nodal Loads . . . . . . . . 4.7 Three-Dimensional Elasticity . . . . . . . . . . 4.8 Integral Equation . . . . . . . . . . . . . . . . 4.9 Linear Elasticity with Boundary Elements . . . 4.10 Fundamental Solutions . . . . . . . . . . . . . 4.11 Boundary Integral Equation . . . . . . . . . . . 4.12 Body Forces (and Domain Integrals in General) 4.13 CMISS Examples . . . . . . . . . . . . . . . . Transient Heat Conduction 5.1 Introduction . . . . . . . . . . . . . . . . . . 5.2 Finite Differences . . . . . . . . . . . . . . . 5.2.1 Explicit Transient Finite Differences . 5.2.2 Von Neumann Stability Analysis . . . 5.2.3 Higher Order Approximations . . . . 5.3 The Transient Advection-Diffusion Equation

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

. . . . . . . . . . . . .

5

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

CONTENTS

iii

5.4 5.5

Mass lumping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 CMISS Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 109 109 109 111 112 113 115 115 116 116 117 118 118 119 120 122 133 133 133 135 136 137 138 141

6 Modal Analysis 6.1 Introduction . . . . . . 6.2 Free Vibration Modes . 6.3 An Analytic Example . 6.4 Proportional Damping 6.5 CMISS Examples . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

7 Domain Integrals in the BEM 7.1 Achieving a Boundary Integral Formulation . . . . . . . . . 7.2 Removing Domain Integrals due to Inhomogeneous Terms . 7.2.1 The Galerkin Vector technique . . . . . . . . . . . . 7.2.2 The Monte Carlo method . . . . . . . . . . . . . . . 7.2.3 Complementary Function-Particular Integral method 7.3 Domain Integrals Involving the Dependent Variable . . . . . 7.3.1 The Perturbation Boundary Element Method . . . . 7.3.2 The Multiple Reciprocity Method . . . . . . . . . . 7.3.3 The Dual Reciprocity Boundary Element Method . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

8 The BEM for Parabolic PDES 8.1 Time-Stepping Methods . . . . . . . . . . . . . . . . . . . . . 8.1.1 Coupled Finite Difference - Boundary Element Method . 8.1.2 Direct Time-Integration Method . . . . . . . . . . . . . 8.2 Laplace Transform Method . . . . . . . . . . . . . . . . . . . . 8.3 The DR-BEM For Transient Problems . . . . . . . . . . . . . . 8.4 The MRM for Transient Problems . . . . . . . . . . . . . . . . Bibliography

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

As the degree of the polynomial is increased the data points are fitted with increasing accuracy and polynomials provide a very convenient form of expression because they can be differentiated and integrated readily. Figure 1.2a. (b) A least-squares polynomial fit to the data. but if the polynomial order is increased further to improve the accuracy of fit a problem arises: the polynomial can be made to fit the data accurately. The points are the measured temperatures. but it oscillates unacceptably between the data points. we divide the bar into three subregions and use low order polynomials over each subregion .1: (a) Temperature distribution u x along a bar. measurements of temperature u against distance x along a bar.g.2b shows three linear polynomials in s fitted by least-squares separately to the data in each element. One approach would be to use a polynomial expression u x = a + bx + cx2 + dx3 + : : : and to estimate the values of the parameters a.Chapter 1 Finite Element Basis Functions 1.1 Representing a One-Dimensional Field Consider the problem of finding a mathematical expression u x to represent a one-dimensional field e. showing the unacceptable oscillation between data points. c and d from a least-squares fit to the data. For low degree polynomials this is a satisfactory approach. as shown in Figure 1.. b. For later generality we also introduce a parameter s which is a measure of distance along the bar. as shown in Figure 1.1a.1b. .called elements. u is plotted as a function of this arclength in Figure 1. To circumvent this. while retaining the advantages of low degree polynomials. u + + + + + + + + ++ + + + + + u + + + + + + + + ++ + + + + + x (b) x (a) F IGURE 1.

2 Linear Basis Functions A new problem has now arisen in Figure 1. We then define a linear variation between these two values by u   = 1 . '1   = 1 .e. This has the advantage that the . The basis functions '1   and '2   are straight lines varying between 0 and 1 as shown in Figure 1. and linear polynomials are independently fitted to the data in each subdomain.2: (a) Temperature measurements replotted against arclength parameter s. (b) The s domain is divided into three subdomains.   u1 + u2 where  0   We define  1 is a normalized measure of distance along the curve. e = global node number of local node n of element e.2b: the piecewise linear polynomials are not continuous in u across the boundaries between elements.2 F INITE E LEMENT BASIS F UNCTIONS u + + + + + + + u + + + + + + + + + + + + + + + + + + + + + + + + + s (b) s (a) F IGURE 1. c etc. b. to ensure continuity of u across the element boundaries. un = Un.  '2   =  such that u   = '1   u1 + '2   u2 and refer to these expressions as the basis functions associated with the nodal parameters u1 and u2. which are the values of u at the two ends of that element. e i.. elements. One solution would be to constrain the parameters a.e where  n. It is convenient always to associate the nodal quantity un with element node n and to map the temperature U defined at global node  onto local node n of element e by using a connectivity matrix  n. 1. but a better solution is to replace the parameters a and b in the first element with parameters u1 and u2 .3.

as shown in Figure 1. since the parameter U2 is shared between the first and second elements .2) = U2 and u2 = U3 . Thus.  and '2   =  .1.1) u   = '1   u1 + '2   u2 with u1 (1. in the first element node 1 global nodes: U1 node 2 U2 node 3 U3 node 4 U4 x element nodes: u1 0 u2 element 1 1  u1 0 u2 element 2 1  u1 0 u2 element 3 1  F IGURE 1. In the second element u is interpolated by (1. u   = '1   u1 + '2   u2 with u1 = U1 and u2 = U2 . interpolation u   = '1   u1 + '2   u2 holds for any element provided that u1 and u2 are correctly identified with their global counterparts.3: Linear basis functions '1   = 1 .2 L INEAR BASIS F UNCTIONS 3 '1   1 '2   1 1 0 1  0 1  F IGURE 1.4: The relationship between global nodes and element nodes.4.

u node 1 + + + + node 2 + + + + + + node 3 + + + + node 4 + + element 1 element 2 element 3 s F IGURE 1.5: Temperature measurements fitted with nodal parameters and linear basis functions.3) with u1 = U3 and u2 = U4 . Figure 1.3 Basis Functions as Weighting Functions It is useful to think of the basis functions as weighting functions on the nodal parameters. The fitted temperature field is now continuous across element boundaries.6 shows the temperature field defined by the three interpolations (1. 0 u1 + 0u2 = u1 which is the value of u at the left hand end of the element and has no dependence on u2 1 1 at  = u 4 = 1 . 1 u1 + 1 u2 = 1 u1 + 1 u2 2 2 2 2 2 which depends equally on u1 and u2 3 at  = 4 . with the parameter U3 being shared between the second and third elements. Similarly.4 F INITE E LEMENT BASIS F UNCTIONS the temperature field u is implicitly continuous. in element 1 at  =0 u 0 = 1 . Thus.3). but is weighted more towards u1 than u2 1 1 at  = u 2 = 1 . 1 u1 + 1 u2 = 3 u1 + 1 u2 4 4 4 4 4 which depends on u1 and u2 . in the third element u is interpolated by u   = '1   u1 + '2   u2 (1.1)–(1. 1.

.

.

.

.

3 .

u 4 = 1 . 3 u1 + 3 u2 = 1 u1 + 3 u2 4 4 4 4 .

Outside the immediately adjacent elements. We now have a continuous piecewise parametric description of the temperature field u   but in order to define u x we need to define the relationship between x and  for each element. in element 1 x   = '1   x1 + '2   x2 (1. w1 (a) s w2 (b) s w3 (c) s w4 (d) s F IGURE 1. where the weighting function wn associated with global node n is constructed from the basis functions in the elements adjacent to that node. these weighting functions can be considered as global functions. The dependence of temperature on x. as shown in Figure 1. A convenient way to do this is to define x as an interpolation of the nodal values of x.4) and similarly for the other two elements. For example. 1 u1 + 1u2 = u2 which is the value of u at the right hand end of the region and has no dependence on u1 . Moreover. u x.3 BASIS F UNCTIONS AS W EIGHTING F UNCTIONS 5 which depends on u1 and u2 but is weighted more towards u2 than u1 at  =1 u 1 = 1 .1. is therefore .6. respectively. Notice the linear fall off in the elements adjacent to a node.6: (a) : : : (d) The weighting functions wn associated with the global nodes n = 1 : : : 4. For example. the weighting functions are defined to be zero. w2 weights the global parameter U2 and the influence of U2 falls off linearly in the elements on either side of node 2.

7.6 F INITE E LEMENT BASIS F UNCTIONS defined by the parametric expressions u   = x  = X X n n 'n   un 'n   xn where summation is taken over all element nodes (in this case only 2) and the parameter  (the “element coordinate”) links temperature u to physical position x. The values of x   and u   are obtained from a linear interpolation of the nodal variables and then plotted as u x.7: Illustrating how x and u are related through the normalized element coordinate  . x   provides the mapping between the mathematical space 0    1 and the physical space x1  x  x2 . . u u1 u2 0  = 0: 2 1  u1 u2 u u x at  = 0:2  x x2 0 x1 x2 x x1 0  = 0: 2 1  F IGURE 1. as illustrated in Figure 1. The points at  = 0:2 are emphasized.

4 QUADRATIC BASIS F UNCTIONS 7 1. 1. 0:5  . '1   must have a factor  . 0:5  F IGURE 1. Notice that since '1   must be zero at  = 0:5 (node 2).8: One-dimensional quadratic basis functions. 1. a quadratic variation of u over an element requires three nodal parameters u1 . another factor is  . For example.4 Quadratic Basis Functions The essential property of the basis functions defined above is that the basis function associated with a particular node takes the value of 1 when evaluated at that node and is zero at every other node in the element (only one other in the case of linear basis functions). This ensures the linear independence of the basis functions. Finally.4  . 1 '3    0 0:5 1 (b) '2   = .8.5) The quadratic basis functions are shown. with their mathematical expressions. u2 and u3 u   = '1   u1 + '2   u2 + '3   u3 (1. 1  1 0 0:5 1 (c) '3   = 2  . It is also the key to establishing the form of the basis functions for higher order interpolation. '1   '2   1 1 0 0:5 1 (a) '1   = 2  . 0:5 and since it is also zero at  = 1 (node 3).5 Two. 1. in Figure 1. Similarly for the other two basis functions. 0:5  .1.and Three-Dimensional Elements Two-dimensional bilinear basis functions are constructed from the products of the above onedimensional linear functions as follows . since '1   is 1 at  = 0 (node 1) we have '1   = 2  .

'1 node 3 '3 2 node 4 2 0 1 node 1 '2 node 2 1 '4 1 2 0 1 0 2 1 1 1 F IGURE 1. Similarly. 2 u2 + '3 1. 2 = '1 1. yn . 2 = 12 (1. 2  receives a contribution from each nodal parameter un weighted by 'n 1 . 2 u1 + '2 1. Notice that 'n 1 . 2 '3 1.8 F INITE E LEMENT BASIS F UNCTIONS Let u 1.9. 2  is 1 at node n and zero at the other three nodes. 2 u3 + '4 1. This ensures that the temperature u 1 . 2 = 1 1 . n = . 2  is evaluated at node n it takes on the value un . '2 1 . 1 1 . 2  and that when u 1 . 2 u4 where '1 1.6) Note that '1 1 . As before the geometry of the element is defined in terms of the node positions xn . These four bilinear basis functions are illustrated in Figure 1. 2 '2 1. 2  = '1 1 '1 2  where '1 1  and '1 2  are the one-dimensional quadratic basis functions.9: Two-dimensional bilinear basis functions. 2  = '2 1 '1 2  : : : etc. 2 = 1 . 2 = 1 . 1 2 '4 1.

2 . 2 = 2 1 . For example. 1 1 . Higher order 2D basis functions can be similarly constructed from products of the appropriate 1D basis functions. 2  (where 0  1 . : : : .1. 2 xn 'n 1. y . a six-noded (see Figure 1. 2 un '1 1.5 T WO - AND T HREE -D IMENSIONAL E LEMENTS 9 1. 2 yn which provide the mapping between the mathematical space 1 . 1 1 . 4 by x= y= X X n n 'n 1. 2  1) and the physical space x.10) quadratic-linear element (quadratic in 1 and linear in 2 ) would have u= where 6 X n=1 'n 1.

1 2 '3 1. 2 = 21 1 . 2 = 41 1 . 1 2 2 1 . 2 1 . 2 '5 1.

1 1 . '2 1. 2 . 2 = 41 1 .

2 = 2 1 . 2 2 . 1 1 . 1 '4 1.

7) (1. Three-dimensional basis functions are formed similarly.9) 0 0 0:5 1 1 F IGURE 1. 2 2 (1.8) (1..g.10: A 6-node quadratic-linear element. 2 = 21 1 . a trilinear element basis has eight . 1 '6 1. e.

6 Higher Order Continuity All the basis functions mentioned so far are Lagrange1 basis functions and provide continuity of u across element boundaries but not higher order continuity. Sometimes it is desirable to use basis functions which also preserve continuity of the derivative of u with respect to  across element boundaries. 3 '3 1. 2 3 '8 1. 3 = 1 1 . 3 = 1 1 . 3 = 1 . 3 = 123 7 8 (1. 2 1 . 2. 2. 1.10 F INITE E LEMENT BASIS F UNCTIONS nodes (see Figure 1. 3 '4 1.11) (1. 3 '6 1.11: An 8-node trilinear element.12) (1. 2 1 . 3 = 1 .10) (1. 1 1 . 1 2 1 . 2. 3 = 1 .13) 3 2 4 1 F IGURE 1. A convenient way to achieve this is by defining two additional nodal parameters . 3 '5 1. 1 23 3 5 6 1 2 '2 1. 3 = 12 1 . 2. 2.11) with basis functions '1 1. 2. 2. 2 3 '7 1. 2. 3 = 1 . 1 1 .

du d n . The basis functions are chosen to ensure that du = du d =0 d .

1 = u01 and du = du d =1 d .

2 = u02 and since un is shared between adjacent elements derivative continuity is ensured. . To derive these cubic 1 Joseph-Louis Lagrange (1736-1813). Since the number of element parameters is 4 the basis functions must be cubic in  .

u   = 0   u1 + 1   u01 + 0   u2 + 1   u02 1 1 2 2 (1. 2u2  3 or. One further step is required to make cubic Hermite basis functions useful in practice. The defined at node n is dependent upon the element  -coordinate in the two adn du where s is jacent elements. c and d are solved to give a = u1 b = u01 c = 3u2 . 2u01 .12.14) where the four cubic Hermite basis functions are drawn in Figure 1. 2u2 Substituting a. It is much more useful to define a global node derivative ds n arclength and then use derivative . b. rearranging. 2u01 .6 H IGHER O RDER C ONTINUITY 11 Hermite2 basis functions let u   = a + b + c 2 + d 3. 3u1 . u02  2 + u01 + u02 + 2u1 . d and impose the constraints u 0 = a = u1 u 1 = a + b + c + d = u2 du 0 = b = u01 d du 1 = b + 2c + 3d = u0 2 d These four equations in the four unknowns a. c and d back into the original cubic then gives u   = u1 + u01 + 3u2 .1. du = b + 2c + 3d 2. 3u1 . b. u02 d = u01 + u02 + 2u1 .

du d .

.

du d where 2 .

ds d n n = ds .

e  d . du n.

15) is an element scale factor which scales the arclength derivative of global node Charles Hermite (1822-1901). . ds n (1.

A two.12 F INITE E LEMENT BASIS F UNCTIONS 1 0   = 1 . @ 1 2 and @2u @1 @2 The need for the second-order cross-derivative term can be explained as follows. @ . Now consider the side 1–3 in Figure 1.  to the  -coordinate derivative of element node n. then @u is cubic in  and quadratic 1 @2 in 2 . 1 2 0 1  0 1  slope = 1 F IGURE 1.dimensional bicubic Hermite basis requires four across element boundaries rather than d derivatives per node @u @u u. The cubic variation of u with 2 is specified by . 2  2 1   =  2 . Thus du is constrained to be continuous ds du . 12 1 0 1  0 1  1 0   =  23 .13. 3 2 + 2 3 1 slope = 1 1   =   .12: Cubic Hermite basis functions. If u is cubic in 1 and cubic in 2 .

@u .

four additional . u3 and . @u @u (the normal derivative) is . But since the four nodal parameters u1 . @2 1 @2 3 @1 also cubic in 2 along that side and is entirely independent of these four parameters.

@u .

@u parameters are required to specify this cubic. . Two of these are specified by @1 1 and @1 3.

@2u .

and @u @1 . is quadratic in 1 and cubic in 2 . @2u and the remaining two by @1 @2 1 and @1 @2 3.

1.6 H IGHER O RDER C ONTINUITY 13 2 .

@u @1 3 node 3 node 4 .

2 = 0 1 0 2 u1 + 0 1 0 2 u2 1 1 2 1 + 0 1 0 2 u3 + 0 1 0 2 u4 1 2 2 . The bicubic interpolation of these nodal parameters is given by u 1. @u @1 1 node 1 F IGURE 1.13: Interpolation of nodal derivative @u @1 node 2 1 along side 1–3.

.

@u 2 1   0   @u 1   0   + 1 1 1 2 @1 1 + 2 1 1 2 .

@1 2 .

@u 1   0   @u + 1 1 2 2 @ + 1 1 0 2 @ 2 2 .

@u1 3 .

@u1 4 + 0 1 1 2 + 0 1 1 2 1 1 1 @2 1 2 .

@2 2 .

@u @u + 0 1 1 2 @ + 0 1 1 2 @ 2 2 1 2 .

@22 u 3 .

2 @42 u 1 1 + 1 1 1 2 1 1 @1 @2 1 + 2 1 1 2 .

@1 @2 2 .

@2u @2u 1   1   + 1 1 2 2 @ @ + 1 1 1 2 @ @ 2 2 1 2 3 1 2 4 (1.16) .

12  2 3 . 3 2 + 2 3   . and s2. 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. 1 (1. As in the one-dimensional case above.14 F INITE E LEMENT BASIS F UNCTIONS where 0   1 1   1 0   2 1   2 = = = = 1 . 2   2  . measuring arclength along the 1-coordinate.e  @1 n . Thus 1 @1 n = @s1 n.12). There are now two arclengths to consider: s1 . measuring arclength along the 2-coordinate.17) are the one-dimensional cubic Hermite basis functions (see Figure 1.

@s .

@u .

@u 2 @2 n = @s2 n.e  @2 n .

@2u .

@2u .

ds .

ds = @s @s  d1  d2 @1 @2 n 1 2 n.e 1 n 2 n 2 .

@u .

@u .

@s (1. Figure 1.14 shows a four element bicubic Hermite surface in 3D space where each node has the following twelve parameters where . The bicubic Hermite basis is a powerful shape descriptor for curvilinear surfaces.18) d1 n d2 n are element scale factors which scale the arclength derivatives of global node  to the  -coordinate derivatives of element node n.

ds 1 and .

15 to the total area of the triangle L1 = Area Area x y P 23 = 1 1 x y = = a + b x + c y = 2 1 1 1 1 123 2 1 x2 y2 3 3 . Consider the ratio of the area formed from the points 2. z. @s @s . @s .and three. y  in Figure 1.7 Triangular Elements Triangular elements cannot use the 1 and 2 coordinates defined above for tensor product elements (i.dimensional elements whose basis functions are formed as the product of onedimensional basis functions). y. @s . 3 and P x.. ds @2z @x @x @ 2 x @y @y @ 2 y @z @z x. The natural coordinates for triangles are based on area ratios and are called Area Coordinates . @s . @s .e. @s @s . @s . @s and @s @s 1 2 1 2 1 2 1 2 1 2 1 2 1. two.

y1 2 x2 .7 T RIANGULAR E LEMENTS 15 z 12 parameters per node 2 y 1 x F IGURE 1.15: Area coordinates for a triangular element. y3 Area P 23 P(x.1. . 3 x3 .14: A surface formed by four bicubic Hermite elements. y2 L1 = 0 2 L1 = 3 L1 = 1 3 L1 = 1 F IGURE 1.y ) 1 x1 .

b2 = y3 . 1 '2 = 2L2 . x2 y1 . c1 = x3 . x2 . y u1 + '2 x. Notice that L1 + L2 + L3 = 1. and a1 = x2 y3 .16 F INITE E LEMENT BASIS F UNCTIONS 1 x1 y1 where  = 1 x2 y2 is the area of the triangle with vertices 123. Similarly. '2 = L2 and '3 = L3 = 1 . Thus. y u2 + '3 x. b1 = 1 x3 y3 y2 . x3 and a3 = x1 y2 . c2 = x1 . L1 . Notice that L1 is linear in x and y . 1 '1 = L1 2L1 . x3 y2 . y3.16: Basis functions for a six node quadratic triangular element. L2 . Six node quadratic triangular elements are constructed as shown in Figure 1. Area coordinate L1 varies linearly from L1 = 0 when P lies at node 2 or 3 to L1 = 1 when P lies at node 1 and can therefore be used directly as the basis function for node 1 for a three node triangle. b3 = y1 . c3 = x2 . area coordinates for the other two triangles containing P and two of the element vertices are 1 2 L2 = Area Area x y P 13 = 1 1 x y = = a + b x + c y = 2 1 2 2 2 123 2 1 x3 y3 1 1 x y P 12 = 1 1 x y = = a + b x + c y = 2 1 3 3 3 123 2 1 x1 y1 2 2 L3 = Area Area where a2 = x3 y1 . 1 '3 = L3 2L3 . x1 . . 1 '4 = 4L1 L2 '5 = 4L2 L3 '6 = 4L3 L1 2 4 6 5 3 F IGURE 1. interpolation over the triangle is given by u x. y2 . x1 y3 . y1 . y u3 where '1 = L1 .16. y = '1 x.

.8 Curvilinear Coordinate Systems It is sometimes convenient to model the geometry of the region (over which a finite element solution is sought) using an orthogonal curvilinear coordinate system. A 2D circular annulus. for example.1. can be modelled geometrically using one element with cylindrical polar r.8 C URVILINEAR C OORDINATE S YSTEMS 17 1.

the annular plate in Figure 1.g.. . the first with r = r1 and the second with r = r2 . e. -coordinates. y 2 1 2 2 3 2 2 3 4 x 0 r1 1 4 r2 (b) r 1 1 (a) (c) F IGURE 1.17: Defining a circular annulus with one cylindrical polar element.17a has two global nodes. Notice that element vertices 1 and 2 in r.

Similarly. y -space in (a). .17a. 2 -space. Global nodes 1 and 2. respectively. y -space in Figure 1. -space or 1 . shown in x. element vertices 3 and 4 map onto global node 2. each map to two element vertices r. as shown in (b) and (c). map onto the single global node 1 in xnode.

and in 1.-space. 2-space. as shown in Figure 1. as shown in Figure 1. The r.17b.17c. .

2 point are given by a bilinear interpolation of the nodal coordinates rn and . coordinates at any 1.

n as in r = 'n 1. 2  rn .

= 'n 1. 2  .

Three orthogonal curvilinear coordinate systems are defined here for use in later sections. . Cylindrical polar r. 2  are given by (1.n where the basis functions 'n 1 .6).

. z  : x = r cos .

y = r sin .

19) . z=z (1.

18 Spherical polar r. .

. F INITE E LEMENT BASIS F UNCTIONS : x = r cos .

cos y = r sin .

. . cos z = r sin Prolate spheroidal .

20) x = d cosh  cos  y = d sinh  sin  cos .  : (1.

z = d sinh  sin  sin .

21) . (1.

r z y  d  x F IGURE 1. . The prolate spheroidal coordinates rae illustrated in Figure 1. . The coordinates .18: Prolate spheroidal coordinates.18 and a single prolate spheroidal element is shown in Figure 1.19.

2 . Only four global nodes are required provided the four global nodes map to eight element nodes as shown in Figure 1. 3 . .19.  are all trilinear in 1 .

8 C URVILINEAR C OORDINATE S YSTEMS 19 (a) 24 y (b) z 2 1 1 3 3 x (c)  90o 0 2 2 1 2 4 1 3 3 4 (d) 2 2 2 1 4 1 4 3 3  1 3 .1.

. F IGURE 1.19: A single prolate spheroidal element. y. . shown (a) in x. (c) in . z -coordinates.

. (b) shows the orientation of the i -coordinates on the prolate spheroid. 2 . 3 -coordinates.-coordinates and (d) in 1 .

6. 7. To define a triangular element mesh run CMISS example 116 (see Figure 1.21. 4. To refine a mesh run the CMISS example 113. 1 2 F IGURE 1.21: 2D bilinear finite element mesh for example 111. After the first refine the mesh should appear like the one shown in Figure 1.9 CMISS Examples 1. The nodes should be positioned as shown in Figure 1.20: Node positions for example 111. To define a 2D cubic Hermite-linear finite element mesh run example 114. After defining elements the mesh should appear like the one shown in Figure 1. To define a 2D bilinear finite element mesh run the CMISS example number 111.23).20 F INITE E LEMENT BASIS F UNCTIONS 1. 2. To define a bilinear mesh in cylindrical polar coordinates run CMISS example 122. 5. To define a quadratic-linear element run the cmiss example 115. 4 2 6 5 1 3 F IGURE 1. .22.20. 3. To define a 3D trilinear element run CMISS example 121.

1.23: Defining a triangular mesh for example 116 .9 CMISS E XAMPLES 21 4 8 5 10 6 1 7 2 9 3 F IGURE 1.22: Refined mesh for example 113 3 4 1 2 F IGURE 1.

Chapter 2 Steady-State Heat Conduction
2.1 One-Dimensional Steady-State Heat Conduction
Our first example of solving a partial differential equation by finite elements is the one-dimensional steady-state heat equation. The equation arises from a simple heat balance over a region of conducting material: Rate of change of heat flux = heat source per unit volume or

d (heat flux) + heat sink per unit volume = 0 dx
or

d ,k du + q u; x = 0 dx dx where u is temperature, q u; x the heat sink and k the thermal conductivity (Watts/m/ Consider the case where q = u

d k du + u = 0 0 x 1 , dx dx subject to boundary conditions: u 0 = 0 and u 1 = 1. This equation (with k = 1) has an exact solution ,  u x = e2 e 1 ex , e,x ,
with which we can compare the approximate finite element solutions. To solve Equation (2.1) by the finite element method requires the following steps: 1. Write down the integral equation form of the heat equation.

(2. C).1) (2. Integrate by parts (in 1D) or use Green’s Theorem (in 2D or 3D) to reduce the order of derivatives. .2) 2.

4. we form the weighted residual Z R!:dx = 0 (2. Integrate over the elements to calculate the element stiffness matrices and RHS vectors.1.24 S TEADY-S TATE H EAT C ONDUCTION 3. 6. 2.1 Integral equation Rather than solving Equation (2. 8. 5.1) directly. dx k du + u dx . 7. Evaluate the fluxes. Introduce the finite element approximation for the temperature field with nodal parameters and element basis functions. Solve the global equations. Assemble the global equations. Apply the boundary conditions.3) where R is the residual d R = .

Substituting Equation (2. given that in real engineering problems this will not be the case.3) gives Z1 0 d du . we try to obtain an approximate solution u for which the residual or error (i.. If u were an exact solution over the whole domain.e.4) into Equation (2. But. dx k dx ! + u! dx = 0 . the amount by which the differential equation is not satisfied exactly at a point) is distributed evenly over the domain.4) for an approximate solution u and ! is a weighting function to be chosen below. (2. the residual R would be zero everywhere.

(2.2 Integration by parts A major advantage of the integral equation is that the order of the derivatives inside the integral can be reduced from two to one by integrating by parts (or. 2. by applying Green’s theorem . substituting f du = ! and g = . ! is chosen such that the residual is kept orthogonal to the space of functions used in the approximation of u (see step 3 below). More precisely.see later).1. equivalently for 2D problems.k dx into the integration by parts . Thus.5) This formulation of the governing equation can be thought of as forcing the residual or error to be zero in a spatially averaged sense.

1 O NE -D IMENSIONAL S TEADY-S TATE H EAT C ONDUCTION 25 formula Z1 dg Z1 df 1 f dx dx = f:g 0 . g dx dx 0 0 gives  .2.

du 1 Z1 .

du d! Z1 d .

. du ! dx .k dx .k dx dx dx 0 0 0 and Equation (2.5) becomes  du 1 Z1 .k dx dx = ! .

In many Russian texts the Galerkin finite element method is known as the Bubnov-Galerkin method. The method was also attributed to I.1. .G. or error.3 Finite element approximation We divide the domain 0 x 1 into 3 equal length elements and replace the continuous field variable u x within each element by the parametric finite element approximation u   = '1   u1 + '2   u2 = 'n   un x   = '1   x1 + '2   x2 = 'n   xn (summation implied by repeated index) where '1   = 1 . 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. This forces the residual R to be orthogonal to the space of functions used to represent the dependent variable u. We also choose ! = 'm (called the Galerkin1 assumption).6) 2. du d! k + u! dx = k ! 0 dx dx dx 0 (2. He published a paper using this idea in 1915. The domain integral in Equation (2. is monotonically reduced as the finite element mesh is refined (see later for a more complete justification of this very important step) . thereby ensuring that the residual. Bubnov in 1913.6) can now be replaced by the sum of integrals taken separately over the three elements Z1 0 1  dx = Z 0 1 3  dx +  dx + 1 3 Z 2 3 Z1 2 3  dx Boris G.  and '2   =  are the linear basis functions for both u and x.

1.26 and each element integral is then taken over  -space S TEADY-S TATE H EAT C ONDUCTION Zx 2 x1 where J  dx = Z1 0 J d = dx d is the Jacobian of the transformation from x coordinates to  coordinates. 2.4 Element integrals The element integrals arising from the LHS of Equation (2.6) have the form Z1 .

Since 'n and 'm are both functions of  the derivatives with respect to x need to be converted to derivatives with respect to  .7) becomes Z1 . du d! + u! J d k 0 dx dx (2.7) where u = 'n un and ! = 'm . Thus Equation (2.

The term 1 Z1 . Emn Notice that un has been taken outside the integral because it is not a function of  .8) d is dx 1  or evaluated by substituting the finite element approximation x   = 'n :Xn . In this case x = 3 d = 3 and the Jacobian is J = dx = 1 . d'm d d'n d + 'm 'n J d un k 0 d dx d dx (2. The term multiplying the nodal parameters u is called n dx d 3 the element stiffness matrix.

d'm d d'n d Z1 .

 '2   =  d'1 = . d'm d'n Emn = k + 'm'n J d = k 3 3 + 'm'n d 0 d dx d dx 0 d d 3 where the indices m and n are 1 or 2. To evaluate Emn we substitute the basis functions 123 '1   = 1 .1 d d'2 = 1 or d or .

! 1 .1: The rows of the global stiffness matrix are generated from the global weight functions.1 O NE -D IMENSIONAL S TEADY-S TATE H EAT C ONDUCTION 27 x Node 1 Node 1 2 3 4 X Node 2 X X X 0 0 0 0 X X U2 U2 = U3 U4 X X X X X Node 3 X X X 0 Node 4 0 F IGURE 2. Thus. The bar is shown at the top divided into three elements.2.

Z1 

d'1 2 1 1 Z ,9k ,12 + 1 ,  2 d = 1 9k + 1 2 9k + '1 d = E11 =
3
0

d

3

0

3

3

and, similarly,

E12 = E21 = 1 ,9k + 1

3 1 6 E22 = 1 9k + 3 3  1 ,9k + 1  1 ,,9k + 1   Emn = 13,,9k +31  31 ,9k + 16
3 6 3 3

is the same for all elements.5 Assembly The three element stiffness matrices (with k = 1) are assembled into one global stiffness matrix. 4 of the global stiffness matrix (shown here multiplied by the vector of global unknowns) are generalised from the weight function associated with nodes 1. Note how each element stiffness matrix (the smaller square brackets in Figure 2. 4. Notice also that the stiffness matrix.1.1) overlaps . 2. This process is illustrated in Figure 2. ::. ::. For simplicity we put k = 1 in the following steps. in this particular case.1 where rows 1. Notice that the element stiffness matrix is symmetric.

6) is  du x=1 . 53 9 28 + 28 .containing many zeros . The RHS of Equation (2. 18 28 + 28 . 53 0 0 9 18 6.28 S TEADY-S TATE H EAT C ONDUCTION with its neighbour because they share a common global node. 53 7 6U27 5 4 35 4 18 9 9 18 53 28 0 0 . 53 0 7 6U17 18 6 053 9.since the basis functions are local to elements. Finite element matrices are always sparse matrices . 18 9 U4 Notice that the first row (generating heat flux at node 1) has zeros multiplying U3 and U4 since nodes 3 and 4 have no direct connection through the basis functions to node 1. The assembly process gives 3 2U 3 2 28 .

du = k dx ! k dx ! x=0 x=1 .

9) for node 1 reduces to  du x=1 . du .9) To evaluate these expressions consider the weighting function ! corresponding to each global node (see Fig. k dx ! x=0 (2. For node 1 !1 is obtained from the basis function '1 associated with the first node of element 1 and therefore !1 jx=0 = 1. Thus Equation (2.6). since !1 is identically zero outside element 1. Also.1. !1 jx=1 = 0.

 du x=1 =0 k dx !n x=0 (nodes 2 and 3) and  du x=1 . Similarly. du = . k dx k dx !1 x=0 x=0 = flux entering node 1.

Putting these global equations together we get 2 28 . 18 6 0 9.918 28 + 28 . Note: k has been left in these expressions to emphasise that they are heat fluxes. 53 0 18 53 53 6. du = k k dx !4 dx x=0 x=1 = flux entering node 4. 53 18 9 2 . 53 9 28 + 28 4 0 0 18 9 .

k dx 7 0 x=0 7 7 6U17 = 6 7 0 7 6 27 6 0 6 7 53 5 4U 5 6 . du 3 3 2U 3 6. 18 3 6 .

10) k dx x=1 or Ku = f . 0 7 7 28 4 du 5 U 9 4 (2.

2.2) the exact solutions at these points are 0:2889 and 0:6102. the vector of unknowns and the global “load” vector. 53 U3 18 9 18 3rd equation . An applied boundary flux of zero. this term would appear . 53 U4 18 9 18 4th equation U4 =0 =0 =0 =1 Note that. the contribution from each element would be different .1. U2 = 0:2855. since effectively no additional constraint is applied to the global equation. Note that if the governing differential equation had included a distributed source term that was independent of u. . The finite element solution is shown in Figure 2. where the flux terms on the RHS are at present unknown 1st equation U1 2nd equation .1 O NE -D IMENSIONAL S TEADY-S TATE H EAT C ONDUCTION 29 where is the global “stiffness” matrix. corresponding to an insulated boundary. rather than an essential boundary condition. At least one essential boundary condition must be applied. 53 U2 + 56 U3 . the known value of flux would enter the appropriate RHS term and the value of U at that node would remain an unknown in the system of equations.2. Moreover.10) rather than on the LHS as here. 2..10) = 0.1. 2.on the RHS of Equation (2. U1 = 0 and U4 = 1.as shown in the next section. K u f 2. if a flux boundary condition had been applied. From Equation (2. if the source term was a function of x.7 Solution Solving these equations gives: U2 = 0:2885 and U3 = 0:6098. respectively.6 Boundary conditions The boundary conditions u 0 = 0 and u 1 = 1 are applied directly to the first and last nodal values: i.1. These so-called essential boundary conditions then replace the first and last rows in the global Equation (2.via its weighted integral .8 Fluxes The fluxes at nodes 1 and 4 are evaluated by substituting the nodal solutions U1 U3 = 0:6089 and U4 = 1 into Equation (2.10). is termed a natural boundary condition. 53 U1 + 56 U2 .e.

0:8496 (k = 1. k dx x=0 = . du flux entering node 1 = . exact solution 0:8509) .

. consistent with heat flow down the temperature gradient.2 as heat entering node 4 and leaving node 1. du flux entering node 4 = k dx x=1 = 1:1357 (k = 1. exact solution 1:3131) These fluxes are shown in Figure 2.

1): d .6) now becomes .2: Finite element solution of one-dimensional heat equation. x = 0 0 x 1 dx Equation (2.30 S TEADY-S TATE H EAT C ONDUCTION u 1:3156 1 0:6098 0:2855 0:8406 0 1 3 2 3 1 x F IGURE 2. dx k du + u . 2.2 An x-Dependent Source Term Consider the addition of a source term dependent on x in Equation (2.

 du 1 Z1 Z1 .

du d! k + u! dx = k ! + x! dx 0 dx dx dx 0 (2.12) . Replacing the domain integral for this source term by the sum of three element integrals Z1 0 x! dx = Z 0 1 3 x! dx + Z 1 3 2 3 x! dx + Z1 2 3 x! dx and putting x in terms of  gives (with Z1 0 x! dx = 1 3 Z1  Z1 1 3 ! d + 3 0 0 dx = 1 for all three elements) d 3 1 +   ! d + 1 3 3 Z1 0 2 +   ! d 3 (2.11) 0 where the x-dependent source term appears on the RHS because it is not dependent on u.

53 18 9 2 .   d = 2 and 9 27 Z1 0 1 1 +    d = 5 gives 9 54 2 27 5 54 and for element 3. where '1 = 1 .   d = 7 and 9 54 Z1 0 1 2 +    d = 5 gives 9 54 7 54 5 54 Assembling these into the global RHS vector. Equation (2.2.953 5618 4 0 0 18 . 27 1 Z1 0 1 1 +   1 . the first term on the RHS of (2. Z1 0 1 2 +   1 . 1 Similarly. the contribution to the element 1 RHS vector from the source term is 54 .   d = 1 9 54 and Z1 0 1  2 d = 1 9 27 Thus. 53 0 6. For example. 53 6 053 . Evaluating these expressions.918 5618 . for element 2.12) corresponding to element 1 is '2 =  .10) becomes 2 28 .  and Z1 1  1 . 0 1 9 Z1 0 'm d .3 T HE G ALERKIN W EIGHT F UNCTION R EVISITED 31 where ! is chosen to be the appropriate basis function within each element.

du 3 3 2U 3 6. k dx 7 2 1 3 0 1 54 x=0 7 6 1 7 + 6 27 + 27 7 0 0 7 6U2 7 = 6 7 65 + 27 76 7 6 53 5 4U 5 6 7 . 18 3 6 .

In Figure 2. This idea is illustrated in Figure 2. 0 7 4 54 54 5 7 28 5 5 4 du U 9 4 k dx x=1 54 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 fails to be exactly zero). The difference between the exact vector e and the approximate vector is the ' u u u u' u .3.3a an exact vector e (lying in 3D space) is approximated by a vector = 1 1 where 1 is a basis vector along the first coordinate axis (representing one degree of freedom in the system).

32 S TEADY-S TATE H EAT C ONDUCTION (a) (b) '3 u3 (c) ue u u1 R '1 u2 '2 u R u u = u1'1 r  '1 = 0 u = u1'1 + u2'2 : : : + r  '2 = 0 u = u1'1 + u2'2 + u3'3 : : : + r  '3 = 0 F IGURE 2. error or residual = e . For a 3D vector space we only need three axes or basis vectors to represent the true vector . The Galerkin technique minimises this residual by making it orthogonal to '1 and hence to the approximating vector .3c) is permitted in the approximation = u1 '1 + u2 '2 + u3 '3 with the result that the residual (now also orthogonal to '3 ) is reduced to zero and = e . but in the infinite dimensional vector space associated with a spatially continuous field u x we need to impose the equivalent orthogonality r u u u u u ' u ' .3: Showing how the Galerkin method maintains orthogonality between the residual vector r and the set of basis vectors 'i as i is increased from (a) 1 to (b) 2 to (c) 3.3b) is added. in Figure 2. Finally. the approximating vector is = u1 1 + u2 2 and the residual is now also made orthogonal to '2 and hence to . (shown by the broken line in Figure 2. If a second degree of freedom (in the form of another coordinate axis in Figure 2.3c.3a). a third degree of freedom (a third axis in Figure 2.

The key point is that in this analogy the residual is made orthogonal to the current set of basis vectors . in finite element analysis. Consider the three-dimensional steady-state heat equation with no source terms: @u @ @ @ . @x kx @x . to the set of basis functions used to represent the dependent variable. @z kz @u = 0 @y @z .or.Z condition u x.4 Two and Three-Dimensional Steady-State Heat Conduction Extending Equation (2. R'dx = 0 u u u for every basis function ' used in the approximate representation of 2. @y ky @u . equivalently.1) to two or three spatial dimensions introduces some additional complexity which we examine here. This ensures that the error or residual is minimal (in a least-squares sense) for the current number of degrees of freedom and that as the number of degrees of freedom is increased (or the mesh refined) the error decreases monotonically.

if k is spatially constant. corresponding to Equation (2. The weighted integral equation. Solution region: F IGURE 2. 7th edition.4: The region and the boundary .kru ! d = Z Z @u kru  r! d . .4).15) (see p553 in Advanced Engineering Mathematics” by E.2.13) and.13).13) over the region . (2. this can be written as y and z axes respectively. (see Figure 2. = . k @n ! d. subject to boundary conditions on .4 T WO AND T HREE -D IMENSIONAL S TEADY-S TATE H EAT C ONDUCTION 33 If kx where kx . This is used (with g = ku and f = ! ) to reduce the derivative order from two to one as follows: Z r  .r  kru ! d = 0 (2. 1993). this reduces to Laplace’s equation k r2 u = 0. Solution region boundary: . ky = kz = k. ky and kz are the thermal diffusivities along the x. .r  kru = 0 (2.16) Z d .14) The multi-dimensional equivalent of integration by parts is the Green-Gauss theorem: Z Z @g f r  rg + rf  rg d = f @n d. Here we consider the solution of Equation (2.. Kreysig. Wiley. (2. is Z .

k dx ! dx = k dx dx dx .16) in Equation (2.14) gives the two-dimensional equivalent of Equation (2. Integration by parts is dx . x x 2 1 Using Equation (2. k dx ! x . du Z du d!  du x cf.6) .

2 plane. The integrand on the LHS of (2. I .1 i = k @xk @i or. @2 7 4 @y @x 5 .5 shows an Let example of this mapping.17) subject to u being given on one part of the boundary and boundary. (2.17) is evaluated using @u being given on another part of the @n (2. for a two-dimensional element. and the geometric terms @xi k are found from the  @   @x . i. @x @y @1 @2 @1 @2 @2 @1 2 @y @x 3 6 @2 . Figure 2..1 6 @x @y 7 = 6 @1 @2 7 = 4 @2 @2 5 4 @y @y 5 @x @y 1 @x @y . as before.34 S TEADY-S TATE H EAT C ONDUCTION (but with no source term): Z Z @u kru  r! d = k @n ! d. . the solution region is the union of the individual elements. 2 @1 @1 3 2 @x @x 3. In each i let i=1 u = 'nun = '1u1 + '2u2 + : : : + 'N uN and map each i to the 1. @ 1 @1 2.e.5 Basis Functions .Element Discretisation = i .18) @u @! @u @ @! @ ru  r! = @x  @x = @ @xi  @ @xj k k i k j k where inverse matrix @ u = 'nun and ! = 'm.

5 BASIS F UNCTIONS .5: Mapping each to the 1 . the basis functions and their derivatives are: '1 = 1 . 11 . 2 @1 @'3 = 1 .   2 @1 @'1 = .26) (2.23) (2.1 .19) (2.1 .27) (2.2.28) (2. 1 @2 @'3 = .  2 @1 @'1 = .24) (2.29) '2 = 11 .22) (2.  1 @2 @'4 =  @1 2 @'4 =  @2 1 (2. 2 '3 = 1 . 12 '4 = 12 . 2  @'1 = .20) (2. 2 plane in a 2  2 element plane.25) (2.21) (2.   1 @2 @'2 = 1 . For each element.E LEMENT D ISCRETISATION 35 2 y 7 3 2 8 3 1 7 8 4 1 8 1 9 4 9 0 4 0 6 5 1 0 5 0 6 1 1 5 4 2 1 2 3 1 4 0 1 0 1 2 5 2 1 1 5 6 2 2 1 1 0 2 0 3 1 1 x F IGURE 2.

e.36 S TEADY-S TATE H EAT C ONDUCTION 2.6 Integration The equation is Z Z @u ru  r! d = k @n ! d. ..30) i. Z . (2.

31) u has already been approximated by 'n un and ! is a weight function but what should this be chosen to be? For a Galerkin formulation choose ! = 'm i..e. (2. @u @! @u @! k + d @x @x @y @y Z @u = k @n ! d. This gives X Z . weight function is one of the basis functions used to approximate the dependent variable. .

4 and Fm is the (element) load vector.6: Considering heat flow in a unit square. : : : . y 2 1 0 1 x 1 F IGURE 2. @n .e.. : : : .. e.6). @'n @'m @'n @'m un k + d i @x @x @y @y Z @u = k 'm d. F = kx where k is the stiffness of spring and F is the force/load. 4 and n = 1. The names originated from earlier finite element applications and extension of spring systems. . (2. This yields the system of equations Emn un = Fm . heat flow in a unit square (see Figure 2. i.32) where the stiffness matrix is Emn where m = 1.g.

1 2 + 2 .1 . 1 3 2 + 2 6 6 6. This is because each element is the same size.1 3 7 6u27 7 6 37 .. 1 .1 76 7 3 6 1 1 2+2 1 5 4u 5 . 6 .g. y  to 1.6 .3 1.1 3 1 . 1 . 1 7 6u 7 = RHS .21 23 . In this case we would have to include the Jacobian of the transformation and @'i . 6 6 1 .1 3 6 6 6. 1 . consider 4 elements (each of unit size) and nine nodes.1 3 .6 3 6 6 .6 . shape and interpolation.1 6 2 3 u9 (2. 1 .31 . Note that if the element was not the unit square we would need to transform from x.3 .3 2 3 2 +3+2+2 3 3 1 .1 . 1 .1 6 6 1 4 .1 . 1 .2. 1 .1 . 1 .1 . 6 7 6u67 3 3 7 6u 7 2 7 6 77 . 3 .34) .7 A SSEMBLE G LOBAL E QUATIONS The first component E11 is calculated as 37 E11 = k Z1 Z1 0 0 1 . 6 6 1 . 1 . However.3 1. 6 7 6u27 k4 15 4 5 3 6 3 3 6 1 1 1 2 .33) u4 Note that the Galerkin formulation generates a symmetric stiffness matrix (this is true for self adjoint operators which are the most common).3 .31 6 3 636 3 6 6 6 1 6 . 2 2 .6 8 . 2. 1 . @'n = @'n @1 + @'n @2 = @'n @i (Refer to also use the chain rule to calculate @xj @x @1 @x @2 @x @i @x Assignment 1) The system of Emn un = Fm becomes 2 2 .6 6 3 2u 3 7 6u17 . For example. Each element has the same element stiffness matrix as that given above. 2 . 1 3 6 6 3 . x2 dxdy = 2k 3 and similarly for the other components of the matrix. typically there is more than one element and so the next step is required.7 Assemble Global Equations Each element stiffness matrix must be assembled into a global stiffness matrix.6 6 . 1 3 2u 3 6 1 6. Given that boundary conditions can be applied and it is possible to solve for unknown nodal temperatures or fluxes. 3 7 6u17 = RHS 6 6.3 1 .1 6 7 6u 7 1 1 7 6 47 . 1 3 21 6 6 6.6 3 3 . y2 + 1 .31 36 . e. 2 coordinates. 1 .6 3 (Right Hand Side) (2.1 6 . 6 6 3 6 6 3 7 6 57 76 7 2+2 1 1 . 1 .1 .

3 1 .31 6 6 6 6.1 .7: Assembling 4 unit sized elements into a global stiffness matrix. It should also be clear that the matrix will be sparse if there is a larger number of elements.1 . To solve.1 .7 and insulating the top (node 8) and the bottom (node 2). The final system to solve is only as big as the number of unknown values of u. . 1 6 6 6. .3 4 3 1 6 . 3 7 6 67 .1 .1 3 3 3 3 4 .1 . we can remove the ith equation and replace it with the known value of ui . . ui is known. boundary conditions can be applied and the equations solved. This means that . firstly boundary conditions are applied to reduce the size of the system.3 8 .6 1 .1 6 3 3 6 1 . This is because the RHS at node i is known but the RHS equation is uncoupled from other equations so the equation can be removed. 1 7 6u67 67 6 7 7 6u 7 76 7 .6 . 1 .1 4 3 3 6 3 1 .1 6 3 3 2 . . 1 .3 1 . 1 6 3 6 6 6 6 6 4 . As an example to illustrate this consider fixing the temperature (u) at the left and right sides of the plate in Figure 2.1 3 6 3 1 1 . If at global node i.1 3 .3 .3 6 2 1 . This yields the system of equations 22 6.1 . Therefore the size of the system is reduced.1 3 6 1 6 1 3 1 3 1 3 1 3 1 3 3 2u13 7 6u27 7 6u37 7 6u 7 7 6 47 7 6u 7 6 7 1 7 6 5 7 = RHS 7 6u 7 .3 .6 3 4 . .1 .38 S TEADY-S TATE H EAT C ONDUCTION y 7 3 4 5 8 4 6 2 2 3 9 global node numbering element numbering 1 1 x F IGURE 2. 1 5 6u77 6 4 85 2 3 u9 Note that the matrix is symmetric. From this system of equations.1 .1 .1 .

and E is the error in the approximation of the integral. We now choose the Gauss points and weights to exactly integrate a polynomial of degree 2I . 1 .3): Z1 0 f   d = I X i=1 Wif i + E Here Wi are the weights associated with sample points i .8).8 Gaussian Quadrature The element integrals arising from two.2. 2 n n F IGURE 2. 2 . The RHS is known at these three nodes (see below).5 and 8). To find out what the RHS is at node 5 Z @u we need to examine the RHS expression @n ! d. For example. Consider first the problem of integrating f   between the limits 0 and 1 by the sum of weighted samples of f   taken at points 1 . with I = 2 we can exactly integrate a polynomial of degree 3: . Other way: . Correct way: on .8 G AUSSIAN Q UADRATURE 39 there are only 3 unknown values of u at nodes (2. This can be explained in two ways. 1 has 2I arbitrary coefficients and there are 2I unknown Gauss points and weights). This is zero as flux is always 0 at internal nodes. therefore there is a 3  3 matrix to solve. does not pass through node 5 and each basis function that is not zero at 5 is zero @u is opposite in neighbouring elements so it cancels (see Figure 2. The RHS is 0 at nodes 2 and 8 because it is insulated. @n 2. = 0 at node 5. Numerical integration or quadrature is therefore required and the most efficient scheme for integrating the expressions that arise in the finite element method is Gauss-Legendre quadrature. : : : .called Gauss points .8: “Cancelling” of flux in internal nodes.or three-dimensional problems can seldom be evaluated analytically. 1 (since a general polynomial of degree 2I . We can then solve the 3  3 matrix and then multiply out the original matrix to find the unknown RHS values. I (see Figure 2.

Z1 Z1 Z 0 0 1 0 d = 1 = W1 :1 + W2:1 (2.38) Z1 0 (2. 0 1 2 ..35) Since a... each integral on the RHS of 2. 2 : : : I : Z1 Let 0 f   d = W1 f 1 + W2 f 2 and choose f   = a + b + c 2 + d 3 . c and d are arbitrary coefficients. f   is sampled at I Gauss points 1 .39) These four equations yield the solution for the two Gauss points and weights as follows: .40 S TEADY-S TATE H EAT C ONDUCTION f   . Thus. b.36) 1  d = 2 = W1 :1 + W2:2 2 2  2 d = 1 = W1 :1 + W2:2 3 3 3  3 d = 1 = W1 :1 + W2:2 4 (2..35 must be integrated exactly.9: Gaussian quadrature.37) (2... I 1  F IGURE 2. Then Z1 0 f   d = a Z1 0 d + b Z1 0  d + c Z1 0  d + d 2 Z1 0  3 d (2.

The number of Gauss points chosen for each i -direction is 4 governed by the complexity of the integrand in the element integral (2.2. r 1 + 1 3. the two Gauss points are given by 1 1 1 = 2 . from (2. In general two. 3 giving 1 1 = 1  p : 2 2 3 Equation (2. 3 = 2 2 5 1 r 5 W1 = 18 W2 = 4 9 5 W3 = 18 (2. for 2x2 Gauss quadrature the 4 weights are all . W1 = W2 = 1 : 2 Then. 1 3 .and three@i terms which come from the dimensional problems the integral is not polynomial (owing to the @xj . 2 3 1+ p . p . 12 = 3 2 21 . substituting in (2. 2 2 5 1 2 = 2 .39) is satisfied identically.or three-dimensional Gaussian quadrature the Gauss point positions are simply the values given above along each i -coordinate with the weights scaled to sum to 1 e.36). Thus.41) 2 For two.37).40) 1 = 1 . 21 + 1 = 0. 2 = 1 .. 2 = 2 1 2 3 1 W1 = W2 = 2 A similar calculation for a 5th degree polynomial using three Gauss points gives (2. 1 and. 2 2 1 + 1 .8).38).g.8 G AUSSIAN Q UADRATURE 41 From symmetry and Equation (2.

The quadrature error must be balanced against the discretization error. 3142.9 CMISS Examples 1. . if the two-dimensional basis is cubic in the 1 -direction and linear in the 2 -direction. To solve for the steady state temperature distribution inside an annulus run CMISS example 312 3. To investigate the convergence of the steady state temperature distribution with mesh refinement run CMISS examples 3141.42 S TEADY-S TATE H EAT C ONDUCTION inverse of the matrix  @x  @j i ) and no attempt is made to achieve exact integration. three Gauss points would be used in the 1-direction and two in the 2-direction. 3143 and 3144. For example. To solve for the steady state temperature distribution inside a plate run CMISS example 311 2. 2.

To gain an intuitive feel for this unusual function. . Fundamental solutions are tied to the Dirac1 Delta function and we deal with both here. 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).Chapter 3 The Boundary Element Method 3. It was rigorously defined as a so-called generalised function by Schwartz in 1950-51.1 Dirac-Delta function What we do here is very non-rigorous.2. and strictly speaking we should talk about the “Dirac Delta Distribution”.1 Introduction Having developed the basic ideas behind the finite element method. consider the following sequence of force distributions applied to a large plate as shown in Figure 3. Dirac (1902-1994) was awarded the Nobel Prize (with Erwin Schrodinger) in 1933 for his work in quantum mechanics. 3. Dirac introduced the idea of the “Dirac Delta” intuitively.M. around 1926-27. 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. There are several key differences between these two methods. 3.2 The Dirac-Delta Function and Fundamental Solutions Before one applies the boundary element method to a particular problem one must obtain a fundamental solution (which is similar to the idea of a particular solution in ordinary differential equations and is the weighting function). we now develop the basic ideas of the boundary element method.1 n 2 wn x = 0 jjxjj x 1 n 1 n 1 Paul A. as we will do here.

1 wn x dx = 1 (i. where it is infinite. the total force applied is unity) but as n increases the area of force application decreases and the force/unit area increases. This is not a function that we are used to dealing with because we have x = 0 if x 6= 0 and “ 0 = 1” i. If we imagine letting n ! 1 we obtain an idealised “point” force of unit strength. As n gets larger we can easily see that the area of application of the force becomes smaller and smaller.1 x h x dx = h 0 (3. given the symbol x.1) . we have The Dirac delta “function” is not a function in the usual sense. Z1 . and it is more correctly referred to as the Dirac delta distribution.. acting at x = 0.. in a nonrigorous sense we have x = nlim wn x !1 the Dirac Delta“function”. the magnitude of the force increases but the total force applied remains unity.e.e. 2 w4 w3 w2 w1 3 2 1 1 2 1 4 1 3 1 2 1 F IGURE 3. Thus. the “function” is zero everywhere except at the origin. It also has the property that for any continuous function h x . Z1 However.1 wn x dx = 1.1: Illustrations of unit force distributions wn .1 x dx = 1 since each Z1 .44 T HE B OUNDARY E LEMENT M ETHOD Each has the property that Z1 .

2 T HE D IRAC -D ELTA F UNCTION AND F UNDAMENTAL S OLUTIONS 45 A rough proof of this is as follows Z1 . where  2 .1 wn x h x dx 1 by definition of x = nlim 2 !1 n nZ 1 .3.1 x h x dx = nlim !1 Z1 .n h x dx by definition of wn x = nlim n h   2 !1 2 n = h 0 1 1 by the Mean Value Theorem. . n n .

n n and as n ! 1. .  ! 0 . 1 1 since  2 .

) (3.e. x. The fundamental solution of a particular equation is the weighting function that is used in the boundary element formulation of that equation.)   . y (i.e. the Dirac Delta function is the slope of the Heaviside2  .3) t t (i. y =  .2)  .1 (Note:  . Most widely known for his integral theorem (the Green-Gauss theorem). Some properties of the Dirac delta are listed below Z1 . . t = step function. but rather its integral behaviour. who pioneered the mathematical study of electrical circuits and helped develop vector analysis. x  . x = H 0  . 3 George Green (1793-1841) was a self-educated miller’s son. x is the Dirac delta distribution centred at x =  instead of x = 0) where H  . The above result (Equation (3. One does not usually talk about the values of the Dirac delta at a particular point.2..4) 3. the two dimensional Dirac delta is just a product of two one-dimensional Dirac deltas. t 0 1 if  if  (3. x h x dx = h   (3..2 Fundamental solutions We develop here the fundamental solution (also called the freespace Green’s3 function) for Laplace’s Equation in two variables. It is therefore important to be able to find the fundamental solution for a particular equation.1)) is often used as the defining property of the Dirac delta in more rigorous derivations. 2 . Most of the common equations Oliver Heaviside (1850-1925) was a British physicist.

. y  is symmetric about this point. . . @ 2 ! + @ 2 ! +  . we solve the above without reference to the original domain or original boundary conditions).16). .46 T HE B OUNDARY E LEMENT M ETHOD have well-known fundamental solutions (see Appendix 3. This is not as difficult as it sounds. Consider solving the Laplace Equation The fundamental solution for this equation (analogous to a particular solution in ODE work) is a solution of @ 2 u + @ 2 u = 0 in some domain 2 @x2 @y2 2. We briefly illustrate here how to find a simple fundamental solution. from Section 1. We expect the solution to be symmetric about the point .  since  . x.. Let r =  . So we adopt a local polar coordinate system about the singular point . x. y2 Then.e.5) in 2 (i. y = 0 @x2 @y2 (3. The method is to try and find solution to r2 ! = 0 in 2 which contains a singularity at the point . x2 +  .8 we have 2 2 @ r ! = 1 @r r @! + r12 @ ! r @r @.

2 q .

@ ! is zero. (3. . y = 0 and owing to symmetry. x.6) For r 0.6) becomes @.  . Thus Equation (3.

2 .

7) Note that this function is singular at r = 0 as required. 1 @ r @! = 0 r @r @r 2 This can be solved by straight (one-dimensional) integration. If D is a small disk of .8) where D is any domain containing r = 0.5) we must have Z D r ! dD = . We choose a simple domain to allow us to evaluate the above integrals. From Equation (3. The solution is ! = A log r + B (3.1 (3. To find A and B we make use of the integral property of the Delta function. 2 Z D dD = .

2: Domain used to evaluate fundamental solution coefficients. 21 :  So we have B remains arbitrary but usually put equal to zero.7).2 T HE D IRAC -D ELTA F UNCTION AND F UNDAMENTAL S OLUTIONS 47 y " . and the fact that D is a disc of radius " Therefore. from Equation (3.2) then from the Green-Gauss theorem @D is the surface of the disk D since D is a disk centred at r = = A 2" " = 2A @D @r dS = 0 so n and r are in the same direction from Equation (3. so that the fundamental solution for the twodimensional Laplace Equation is ! = . 21 log r + B  ! = .  D x F IGURE 3.8) A = . radius " Z D Z @! r ! dD = @n dS @D Z @! 2 0 centred at r = 0 (Figure 3. 21 log r  .3.

= 21 log 1  r (3.9) .

. + ur ! d 2 (3.  2 ).11).e. The basic steps are in fact quite similar to those used for the finite element method (refer Section 2. From Section 2. The fundamental solution for the three-dimensional Laplace Equation can be found by a similar technique. 21 log r  Z ur ! d = . ru:r! d 2 (3.1). ru:r! d @ Z @! Z Z @u = @ @n ! d.11) For the Galerkin FEM we chose ! . . to be 'm . To derive the starting equation for the boundary element method we use the Green-Gauss theorem again on the second integral. . y 2 (singular at the point . We firstly must form an integral equation from the Laplace Equation by using a weighted integral equation and then use the Green-Gauss theorem.48 T HE B OUNDARY E LEMENT M ETHOD where r =  . . 1 ! = 4r 3. 2 Z u  . . the domain integral has been replaced by a point value. Then from Equation (3. .  2 (3. y d = . using the property of the Dirac delta q ! = . one of the basis functions used to approximate u. @ u @n d.12) i.  .4 we have seen that 0= Z Z Z @u r u:! d = @n ! d.3 The Two-Dimensional Boundary Element Method We are now at a point where we can develop the boundary element method for the solution of r2 u = 0 in a two-dimensional domain . x2 +  . where r =  . The result is q where r is now a distance measured in three-dimensions.e.u . This gives Z @u Z 0 = @n ! d.10) @ This was the starting point for the finite element method. For the boundary element method we choose ! to be the fundamental solution of Laplace’s Equation derived in the previous section i.. y 2 (singular at the point . x. the weighting function. ). x2 +  .

u P  + Z . . When . We can see this in a non-rigorous way as 2 follows. so we integrate around one-half of the singularity to get 1 u . a more useful expression would be one relating the value of u at some point on the boundary to boundary integrals.13) the solution domain to integral expressions involving u and @n over the boundary of the solution domain. . We derive such an expression below.  replaced by 1 u . x. 2 Z u  ." . . Rigorous details of where this coefficient 1 comes from are given below. The integral expression we will ultimately obtain is simply Equation (3. We call this enlarged region 0 and let . and we look at each of these in turn. = @n . There are 4 integrals to consider.13).  is on the boundary we only have half of the singularity contained inside the domain." . Equation (3. .. since P is inside the enlarged region 0 .3 T HE T WO -D IMENSIONAL B OUNDARY E LEMENT M ETHOD 49 Thus Equation (3. = @n ! d. In order to be able to evaluate ur2! d in this case we enlarge to include a disk of radius " about P (Figure 3.. It relates the value of u at some point inside @u Z ur ! d = ..  + u @n d.13)) holds if . y d = 0 since the integrand of the second integral is zero at every point except .11) becomes Z @u Z @! u .14) We must now investigate this equation as lim"0 .  2 (i." Z @u @n ! d. This case also happens to be the most important for numerical work as we shall see. If . the singularity of Dirac Delta function is inside the domain).e.  was inside the domain.  in Equation (3.  is on the boundary of the domain .  2 . we integrated around the entire singularity of the Dirac Delta to get u . The previous equation (Equation (3.. The case which needs special consideration is when the singular point .3).  is outside then This equation contains only boundary integrals (and no domain integrals as in Finite Elements) and is referred to as a boundary integral equation. Now. When .  2 @ @ (3." u @! d. 2 2 Z Let P denote the point .e. Rather than having an expression relating the value of u at some point inside the domain to boundary integrals." . (3.3.13) with u .  and this point is outside the region of integration.13) holds for this enlarged domain i.0 = .. ." .

" Z = .  ." .  @ u @r ." @ u @n .. 21 log r d. Firstly consider Z @! Z u @n d.3: Illustration of enlarged domain when singular point is on the boundary.50 T HE B OUNDARY E LEMENT M ETHOD 0 " P ." . = . 21 log r d." ." F IGURE 3.

by definition of ! since .

by the mean value theorem for a surface with a unique tangent at P . = . 2 r @ @ on . " @n @r since r = " on . = ." 1 1 u P  " ! . Thus 1 1 Z u d. 2 " 1 Z u d. 2 " ." .

2 "0 "0 . 2 " " = . = lim .15) By a similar process we obtain . 1 u P  u P  Z @! lim u @n d." (3.

= lim .16) since lim" log "0 as lim"0 . . 2 @n P  " log " = 0 "0 "0 . 1 @u Z @u lim ! @n d." (3.

P F IGURE 3.14)–(3. u P  + u @n d.. then the coefficient 1 is replaced by where 2 2 is the internal angle at P (Figure 3. .  2 @ (i. or 1 u P  + Z u @! d. . = 2 @n . Note: The above is true if the point P is at a smooth point (i. . in the limit and one must deal with Cauchy Principal Values.4: Illustration of internal angle .e. singular point is on the boundary of the region).A = @u ! d. For “nice” integrals (which includes the integrals we are dealing with here) we have 0Z lim @ "0 1 Z (nice integrand) d. Note: If the integrand is too badly behaved we cannot always replace .11) Thus we have 1 Z 0Z @u ! d.e.3 T HE T WO -D IMENSIONAL B OUNDARY E LEMENT M ETHOD 51 It only remains to consider the integrand over . 2 @n @n ... (nice integrand) d. .A = ." ! . lim @ "!0 @n @n .17) (3.A = @! u d.." ... If P happens to lie at some nonsmooth point e. (3. (refer Section 4. = Z @u ! d. lim @ "!0 @n @n ." by ..g. where P = ." . as lim"0 ." 0Z 1 Z @! u d.3. a corner.4).. ." since .18) we get Z @! 1 u P  + Z @u ! d.18) Combining Equations (3. a point with a unique tangent) on the boundary of .

(3. problems with large domains). This can result in huge savings for problems with large volume to surface ratios (i. Z @u Z @! c P  u P  + u @n d. as opposed to volume integrals in a finite element formulation. x2 +  . and . The procedure is thus to use Equation (3.6. not smooth at P if P For three-dimensional problems. and find the volume data as a separate step. = @n ! d. 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.19) where ! = .19) to find the surface distributions of u and @n and then (if required) use Equation (3. smooth at P if P 2 . @u @u u and @n and the value of u at a @u . . 21 log r  c P  = q r =  . Since Equation (3. and . the boundary integral equation expression above is the same. Once the surface distributions of u and @n are known. the value of u at any point P inside can be found since all surface integrals in Equation (3. .19) to find the solution at any point P 2 .19) involves only the surface distributions of point P . and . y2 +  . x2 +  . It also has some disadvantages when compared to the finite element method and these will be discussed in Section 3.19) only involves surface integrals. z2 8 1 if P 2 : inner solid angle 1 2 4 2 .19) are then known. the overall size of the problem has been reduced by one dimension (from volumes to surfaces). smooth at P if P 2 . not smooth at P if P Equation (3. Thus the boundary element method offers some distinct advantages over the finite element method in certain situations. with 1 ! = 4r c P  = q r =  .. Thus we solve for the boundary data first.52 T HE B OUNDARY E LEMENT M ETHOD Thus we get the boundary integral equation.e. We now turn our attention to solving the boundary integral equation given in Equation (3. y2 8 1 if P 2 : internal angle 1 2 2 2 . and .19).

j . .4 Numerical Solution Procedures for the Boundary Integral Equation The first step is to discretise the surface elements).5: Schematic illustration of a boundary element mesh (a) and a finite element mesh (b). qj are values of u and q on element . In general the basis functions used for u and q do not have to be the same (typically they are) and these basis functions can even be different to the basis functions used for the geometry. we only have to interpolate the functions over a one-dimensional element). but are generally taken to be the same (this is termed an isoparametric formulation).20) (a) (b) F IGURE 3.21) Over each element . = @n ! d. Then Equation (3.19) becomes N N X Z @! X Z @u c P  u P  + u @n d.j (3. qj are values of u and q at node on element .j and uj .j j =1 .= N j =1 . j =1 .j (3.j we introduce standard (finite element) basis functions uj = X ' uj and qj @uj = @n X ' qj (3. into some set of elements (hence the name boundary .4 N UMERICAL S OLUTION P ROCEDURES FOR THE B OUNDARY I NTEGRAL E QUATION 53 3.3.22) where uj . . These basis functions for u and q can be any of the standard one-dimensional finite element basis functions (although we are dealing with a two-dimensional problem.

then we have ciui + N XX Z j =1 uj . We can assemble these equations into the matrix system At each node. We now generate one equation per node by putting the point P to be at each node in turn.j N @!i d. = X X q ' @n j j =1 Z ' !i d.25) aij = Z . then we can generate L equations. We therefore have L equations (the number of nodes) and have L unknowns to find. where 2 r is the distance from the singularity point). (3. (3. If P is at node i. As x Cx = f A B A B C K (3. although specialist solvers are required if the problem is large (refer [todo : Section ???]).j where !i is the fundamental solution with the singularity at node i (recall ! is . We can write Equation (3. say. @n and bij = Z . 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). = qj @n j =1 Z .23) This equation holds for any point P on the surface . We need to rearrange the above system of equations to get Au = Bq (3. log r .j ' !i d.27) (compare to the global finite element equations Ku = f ) where the vectors u and q are the vectors of nodal values of u and q .j ' ! d.28) K . while the matrix is dependent on the number of finite element nodes (which include nodes in the domain).54 T HE B OUNDARY E LEMENT M ETHOD This gives c P  u P  + N XX Z j =1 uj . The size of the and matrices are dependent on the number of surface nodes.24) .26) Equation (3. th where is the vector of unknowns. 1 (3.j N XX ' @! d.24) in a more abbreviated form as ciui + where N XX j =1 uj aij = N XX j =1 qj bij (3.25) is for node i and if we have L nodes.j ' @!i d.. Note that the ij component of the A matrix in general is not aij and similarly for B . we must specify either a value of u or q (or some combination of these) to have a well-defined problem. This can be solved using standard linear equation solvers.

The use of the fundamental solution as a weight function ensures that the and matrices are generally well conditioned (see Section 3.j ' @!i d..j .28) can be solved reasonably easily. @n and bij = Z ' !i d.30) (3.29) The right hand side of Equation (3.e.25) with the singular point P located at the required solution point i. it depends on the surface to volume ratio as to which method will generate the smallest and quickest solution. If a solution is then required at a point inside the domain.3. The matrix is therefore also well conditioned and Equation (3. x u q C u q A A B u P  = N XX j =1 qj bPj . all boundary values of and are known. In fact the matrix is diagonally dominant (at least for Laplace’s equation).29) contains no unknowns and only involves evaluating the surface integrals using the fundamental solution with the singular point located at P . then we can use Equation (3. and require far more work and effort than the analogous finite element integrals.31) . 21 log ri  ri = distance measured from node i In terms of a local  coordinate we have bij = aij = Z1 Z 0 '   !i   jJ  j d @!i   jJ  j d = Z '   @!i   dri jJ  j d '   @n @ri dn 1 0 (3.5 Numerical Evaluation of Coefficient Integrals We consider in detail here how one evaluates the aij and bij integrals for two-dimensional problems. Recall that aij = where Z . The vector contains the unknown values of and on the boundary. 3. . N XX j =1 uj aPj (3.j !i = .5 N UMERICAL E VALUATION OF C OEFFICIENT I NTEGRALS 55 mentioned earlier.5 for more on this). Once this has been found. These integrals typically must be evaluated numerically.

56 The Jacobian J   can be found by T HE B OUNDARY E LEMENT M ETHOD s.

2 .

the fact that singular integrands must be accurately integrated. If node i is well removed from element . However. = ds = dx + dy d d d d (3. or also to subdivide the current element into two or more smaller elements and evaluate the integral over each n dri = rr  n i ^ dn (3.33) . 2 Thus every expression in the integrands of the aij and bij integrals can be found at any value of  . For the case when node i is contained in the current element one can use special quadrature schemes which are designed to integrate log-type functions. 21 log ri    q dn where ^ is a unit outward normal vector. The integral still exists. xi 2 + y   . dri we note that To find !i = . we simply rotate the tangent  vector (given by x0   . This is one of the disadvantages of the BEM . The integrals for which node i lies in element .32) s represents the arclength and dx and dy d d interpolation expression for x   and y  .j (or close to it) we see that ri approaches 0 and the fundamental solution !i tends to 1. where The fundamental solution is can be found by straight differentiation of the ri   = x   . yi  are the coordinates of node i. and the integrals can therefore be evaluated numerically using some suitable quadrature schemes.j then standard Gaussian quadrature can be used to evaluate these integrals. if node i is in . However. In such cases special care must be taken . A relatively straightforward way to evaluate all the integrals is simply to use Gaussian quadrature with varying number of quadrature points. It is possible to incorporate adaptive integration schemes that keep adding more quadrature points until some error estimate is small enough. depending on how close or far the singular point is from the current element. 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. 2 J   = d. yi2 where xi . This is not very elegant or efficient. y 0   ) by in the appropriate direction and then normalise. Gaussian quadrature is still the norm. These are to be preferred when one is dealing with Laplace’s equation. but has the benefit that it is relatively easy to implement. these special log-type schemes cannot be so readily used on other types of fundamental solution so for a general purpose implementation. large numbers of Gauss points or other special treatment. but the integrand becomes singular.either by using special quadrature schemes. To find a unit normal vector.j are in general the largest in magnitude and lead to the diagonally dominant matrix equation.

6 The Three-Dimensional Boundary Element Method The three-dimensional boundary element method is very similar to the two-dimensional boundary element method discussed above.3. As in Section 3. subelement. and the expressions given for aij and bij apply equally well to the three-dimensional case. the three-dimensional boundary integral equation is the same as the two-dimensional equation (3. with ! and c P  being defined as in Section 3.6: Illustration of the decrease in ri as node i approaches element .19).11). Details on each of these methods is given in Section 3. and this technique is the norm for elasticity problems (Section 4.6 T HE T HREE -D IMENSIONAL B OUNDARY E LEMENT M ETHOD 57 .j ri ri node i (a) (b) F IGURE 3.j .8.5 we illustrate how to evaluate each of the terms in the integrand of aij and bij . It is also possible to evaluate the “worst” integrals by using simple solutions to the governing equation. . It should be noted that research still continues in an attempt to find more efficient ways of evaluating the boundary element integrals.4. The only real difference between the two procedures is how to numerically evaluate the terms in each integrand of these coefficient integrals. 3.j node i . As noted above. The numerical solution procedure also parallels that given in Section 3.3.

zi 2 dri = rr  ^ to find dri . zi  are the coordinates of node i.58 T HE B OUNDARY E LEMENT M ETHOD The relevant expressions are aij = = Z . 2  . 2j d1 d2 (3. yi 2 + z 1 . 2 !i 1. 2  .j 1 0 ' @!i d.34) Z Z1 0 bij = = ' !i d. 2j d1 d2 @r dn i (3. 2 @!i 1. 2 dri jJ 1.j 1 0 Z .   q i 1 2 where ri 1 . ' 1. 2  . 2 jJ 1 .35) Z Z1 0 The fundamental solution is !i 1. 2 = 4r 1 . @n ' 1. where xi . xi 2 + y 1 . yi . 2  = x 1 . As before we use i dn dn The unit outward normal ^ is found by normalising the cross product of the two tangent vectors .

@x @y @z n t1 = @1 . @1 and t2 . @1 .

BEM: A mesh of the boundary only is required. @ . @ (it relies on the user of any BEM code to 2 2 2 ensure that the elements have been defined with a consistent set of element coordinates 1 and 2 ). This is one of the major pluses of . n 3. Depending on the application some of these differences can either be considered as advantageous or disadvantageous to a particular scheme.7 A Comparison of the FE and BE Methods We comment here on some of the major differences between the two methods. @x @y @z = @ . The Jacobian J 1 . 1. Comment: Because of the reduction in size of the mesh. Note that this is different for the determinant in a two-dimensional finite element code .in that case we are dealing with a two-dimensional surface in two-dimensional space. The integrals are evaluated numerically using some suitable quadrature schemes (see Section 3. whereas here we have a (possibly curved) two-dimensional surface in three-dimensional space. FEM: An entire domain mesh is required. 2  is given by kt1  t2 k (where t1 and t2 are the two tangent vectors). one often hears of people saying that the problem size has been reduced by one dimension.8) (typically a Gauss-type scheme in both the 1 and 2 directions).

and hence an entire domain solution is not required. Comment: There are many problems where the details of interest occur on the boundary. BEM: Only boundary conditions are being approximated. BEM: Both and of the same accuracy. 2. and some contain integrands that become singular. FEM: Sparse symmetric matrix generated. Handles nonlinear problems well. Comment: BEM integrals are far harder to evaluate. Comment: The use of the Green-Gauss theorem and a fundamental solution in the formulation means that the BEM involves no approximations of the differential Equation in the domain .3. 5. virtually any nonhomogeneous equation) for which fundamental solutions are not known. BEM: Cannot even handle all linear problems. . FEM: Differential Equation is being approximated. FEM: Element integrals easy to evaluate. Comment: The need to evaluate integrals involving singular integrands makes the BEM at least an order of magnitude more difficult to implement than a corresponding finite element procedure.it depends partly on the volume to surface ratio. Comment: A fundamental solution must be found (or at least an approximate one) before the BEM can be applied. BEM: Integrals are more difficult to evaluate. FEM: Widely applicable. BEM: Solution on the boundary is calculated first. 8. 6. There are problems where either method can give rise to the smaller system and quickest solution . is a very time consuming exercise. so these integrals need to be evaluated accurately. FEM: Relatively easy to implement. but it can be rather restrictive in its applicability.only in its approximations of the boundary conditions. For problems involving infinite or semi-infinite domains.. There are many linear problems (e. FEM: Reactions on the boundary typically less accurate than the dependent variables. Also the integrals that are the most difficult (those containing singular integrands) have a significant effect on the accuracy of the solution. BEM: Fully populated nonsymmetric matrices generated.g. u q 4. Comment: The matrices are generally of different sizes due to the differences in size of the domain mesh compared to the surface mesh. FEM: Entire domain solution is calculated as part of the solution. BEM: Much more difficult to implement. BEM is to be favoured. particularly in 3D.construction of meshes for complicated objects. 7. 3. There are certain areas in which the BEM is clearly superior.7 A C OMPARISON OF THE FE AND BE M ETHODS 59 the BEM . or are localised to a particular part of the domain. and then the solution at domain points (if required) are found as a separate step.

but that is not the case for BEM. when one begins to use general curved elements and/or solve threedimensional problems then the integrals will not be available as analytic expressions.8. The weights and abscissa for the Gaussian quadrature scheme of order N are chosen so that the above expression will exactly integrate any polynomial of degree 2N .. For the numerical evaluation of two or three-dimensional integrals. which use different fundamental solutions to the log function. log quadrature schemes can be developed which will exactly integrate polynomial functions f  .60 T HE B OUNDARY E LEMENT M ETHOD 3. If one uses constant or linear interpolation for the geometry and dependent variable. The basic tool for evaluation of these integrals is quadrature. but benefit from the fact that they will be exact.each new equation to be used requires its own special quadrature scheme. However. then it is possible to obtain analytic expressions to most (if not all) of the integrals that will appear in the BEM (at least for two-dimensional problems). a Gaussian scheme can be used of each variable of integration if the region of integration is rectangular. As discussed in Section 2. Tables of these are given below It is possible to develop similar quadrature schemes for use in the BEM solution of other PDEs. 1 or less. This is generally not the optimal choice for the weights and abscissae but it allows easy extension to higher order integration. For a two-dimensional BEM solution of Laplace’s equation.8 More on Numerical Integration The BEM involves integrals whose integrands in generally become singular when the source point is contained in the element of integration. the log function has been factored out.8 a one-dimensional integral is approximated by a sum in which the integrand is evaluated at certain discrete points or abscissa Z1 0 f   d  N X i=1 f i wi where wi are the weights and i are the abscissa.e. These are obtained by approximating the integral as Z1 0 log   f   d  N X i=1 f i wi i. . integrals of the form Z1 0 log   f   d will be required. The expressions can become quite lengthy to write down and evaluate. The problem with this approach is the lack of generality .8. It is relatively common to use logarithmic schemes for this.1 Logarithmic quadrature and other special schemes Low order Gaussian schemes are generally sufficient for all FEM integrals. 3. In the same way as Gaussian quadrature schemes were developed in Section 2.

g. can be used. u = x2 .3.766880 0. direct solution methods.wi 0.190435 0. 3. although it is very dependent on the implementation of the BE method. y 2 ) can be used to find the other matrix entries (or just used to check the accuracy of the matrices). one can then use this to solve for some entry in either the or matrix (typically the diagonal entry since this is the most important and difficult to find).1: Abscissas and weight factors for Gaussian integration for integrands with a logarithmic singularity. The conjugate gradient technique is generally regarded as a solution technique for (sparse) symmetric matrix equations. based on LU factorisations. C Cx f . particularly useful if Cauchy principal values are to be found (see Section 4.386875 0.245275 0.. Preconditioners are generally problem dependent . Monte-carlo integration.368997 0. Further solutions to Laplaces equation (e.27) at every node according to the solution u = x.391980 0. For small problems. the time taken for the matrix solution begins to dominate the matrix assembly stage.what works well for one problem may not be so good for another problem.10 Solution of Matrix Equations The standard BEM approach results in a system of equations of the form = (refer (3.718539 0. As mentioned above the matrix is generally well conditioned.556165 0.281461 TO OTHER E LLIPTIC PDE S 61 n 2 n 3 Weight Factors = wi i .wi 0.848982 0.383464 0. modified Helmholtz (CMISS example) Poisson Equation (domain integral and MRM.513405 4 0.wi n i . 3.9 The Boundary Element Method Applied to other Elliptic PDEs Helmholtz. The current technique of favour in the BE community for solution of large BEM matrix equations is a preconditioned Conjugate Gradient solver. Thus if one sets both u and q in Equation (3. fully populated and nonsymmetric.112009 0. DRM. This usually occurs when there is between 500 and 1000 degrees of freedom.039225 TABLE 3. A B 3. For example u = x is a solution to Laplaces’ equation (assuming the boundary conditions are set correctly).063891 0.041448 0.602277 0.2 Special solutions Another approach.094615 0.11) is to use special solutions of the governing equation to find one or more of the more difficult integrals.8.9 T HE B OUNDARY E LEMENT M ETHOD A PPLIED Abscissas = ri i . As the problem size increases.28)).

There are at least two possible methods. 3.I = FEM region = BEM region = FEM boundary = BEM boundary = interface boundary The BEM matrices for B can be written as Au = Bq where is a vector of the nodal values of u and The FEM matrices for F can be written as (3.f .37) Ku = f . in the sense of which part is more dominant FEM or BEM) Consider the region shown in Figure 3.36) u @u q is a vector of the nodal values of @n (3. Often one problem can give rise to a model favouring one method in one region and the other method in another region eg. 1. but to handle the semi-infinite domain (well removed from the foundation).7. in a detailed analysis of stresses around a foundation one needs FEM close to the foundation to handle nonlinearities.62 T HE B OUNDARY E LEMENT M ETHOD B f . Treat the FEM region as an equivalent boundary element and combine with BEM Note that these are essentially equivalent .7: Coupled finite element/boundary element solution domain. BEM is better. where f B . Treat the BEM region as a finite element and combine with FEM 2.I .B .B .11 Coupling the FE and BE techniques There are undoubtably situations which favour FEM over BEM and vice versa.we will only talk about the basic ideas and use Laplace’s Equation to illustrate this. There has been a lot of research on coupling FE and BE procedures .the use of one or the other depends on the problem.f F IGURE 3.

e. as before.of doubtful quality. B = Therefore Equation (3. K T  .e. One must make use of the following uIB = uIF (u is continuous) @ uI = . = .. then we can convert in Equation (3.37) yields = an equivalent BEM system. producing a matrix i. to treat the FEM region as an equivalent BEM region) we firstly note that. Therefore we can assemble this together with original FEM matrix to produce an FEM-type system for the entire region B .g. This approach does not require any matrix inversion and is hence easier (cheaper) to implement 2.I nodal values of u and q are unknown.3.B. treating BEM as an equivalent FEM region) we get (from Equation (3. Existing BEM solvers will not assume symmetric or sparse matrices therefore no new matrix solvers to be implemented .1Au = q q (3. Notes: 1. Attempts have been made to “symmetricise” the K B 1 . K B is in general not symmetric and not sparse. This can be assembled into the existing BEM system (using compatability conditions) and use existing BEM matrix solvers. On .often yields inaccurate matrix . ) B F B F @ nB @ nF Ku Mq To apply method 2 (i. @ uI (q is continuous. To apply method 1 (i.38) q M f Mq M ..38) becomes f B. = . but .. This means that different matrix equation solvers must be used for solving the new combined FEM-type system (most solvers in FEM codes assume sparse and symmetric)..K .1A u = Mq = f KBu = f B where K B = MB .e.11 C OUPLING THE FE AND BE TECHNIQUES 63 where is the stiffness matrix and is the load vector.38) to an equivalent load vector by weighting the nodal values of by the appropriate basis functions.36)) K f If we recall what the elements of in Equation (3. Applying this to (3.37) contained. Notes: f Mq 1.1 A B i. (e. an equivalent stiffness matrix obtain from BEM..e. replace K B by B 2 B results). 2..

g.Green’s functions more widely available that complete systems 3.this yields ! = . Doesn’t introduce singular functions so integrals are easy to evaluate 2. Given a BIE there are other ways of solving the Equation although these are not so widely used. Notes: 1. p @ . = @n ! d. = u @n d. Consider Laplaces Equation in a (bounded) domain Z @u Z @! weighted residuals  ! @n d. Method is not so popular .12 Other BEM techniques What we have mentioned to date is the so-called singular (direct) BEM. 3. Must find a (complete) set of functions (If you just use usual approximations for system is not diagonally dominant so not so good) u matrix 3. This method basically uses a “complete” set of solutions instead of a Fundamental Solution.. Another possibility is to put point P outside of the domain . e.a nonlinear free surface problem).12.2 Regular BEM Consider the BIE for Laplace’s equation Z @! Z @u c P  u P  + u @n d. This leads to singular integrands.creating an equation for each node.12. @ with The usual procedure is to put point P at each solution variable node .1 Trefftz method Trefftz was the first person to perform a BEM calculation (in 1917 .calculated the value (numerical) of the contraction coefficient of a round jet issuing from an infinite tank . = @n d. @ @ if r2 ! =0 The procedure is to express u as a series of (complete) functions satisfying Laplace’s equation with coefficients which need to be numerically determined through utilisation of the boundary conditions.64 T HE B OUNDARY E LEMENT M ETHOD 3. 21 log r  @ @ Z @u Z @!p u @n d.

This method does not involve singular integrands.j N XX ' @!p d. z  = H .. The distance along each node is often found by experimentation . = qj @n j =1 Z .. To overcome this further equations are generated (by placing the singular point P at various locations outside ).an equation involving u and q at each surface node. 3. = X q Z ! d.. The Boundary Element Equation is (with N = 2M (i. : : : . . N is even) constant elements) N N 1 u + X u Z @!i d. There is considerable choice for the location of the point P .various research papers suggesting “ideal” distances (Patterson & Shiekh). 2.e.e. so that integrals are inexpensive to calculate.j i = 1. 3. The system of equations are then overdetermined and are solved in a least squares sense. putting x to .j . z has the property that H x. This method is not very popular. N (3.3.x makes no difference to the problem formulation. z  8x. certain values of k ).j ' !p d.g. Exterior Acoustic Problems.13 Symmetry Consider the problem given in Figure 3. This behaviour can be found in many problems and we can make use of this as follows.x. In practise P is chosen along the unit outward normal of the surface at each solution variable node. Notes: one can generate 1. Both the boundary conditions and the governing Equation exhibit symmetry about the vertical axis. By placing the point P (the singular point) at other distinct points outside as many equations as there are unknowns (or more if required).8 (the domain is outside the circle). 4.39) .. For an acoustic problem (governed by Helmholtz Equation r2 u + k 2 u = 0) in an unbounded region the system of Equations produced by the usual (singular) BEM approach is singular for certain “fictitious” frequencies (i.e. i.13 S YMMETRY 65 Following discretisation as before gives N XX Z j =1 uj . j i 2 i j=1 j @n j =1 . The idea of placing the singularity point P away from the solution variable node is often of use in other situations e. Thus the solution H x. Often the set of Equations generated are ill-conditioned unless P chosen carefully.

. (The Equations for nodes i = M +1. (3. We have N Equations and N unknowns (allowing for the boundary conditions). N are the same as the Equations for nodes i = 1. M ). : : : .j . : : : .j . : : : .44) .8: A problem exhibiting symmetry.j .N . From symmetry we know that (refer to Figure 3.N .N +1. + !i d. (This procedure has halved the number of unknowns. + . + @n d.i So we can write i = 1.j then we can write Equation (3.66 T HE B OUNDARY E LEMENT M ETHOD z H = e. If we define Z @!i Z @!i aij = @n d. : : : . : : : .) (3. . M 2 i j=1 ij j j=1 ij j and solve as before. M (3. = j=1 qj : !i d.j M +1 +1 (3.j . The above M Equations have only M unknowns..sz r2H = s2H x F IGURE 3. ui = un+1.9).41) as M M 1 u + X a u = X b q i = 1.j .42) bij = !i d. M .40) 8 9 8 9 Z @!i Z @!i = X Z Z M = 1u + X u !i d. + @n d.N .41) for nodes i = 1.43) .j Z Z +1 (3. 2 i j=1 j : @n d.

M this means that the integrals over the elements . .e.N . except possibly on the axis of symmetry if linear or higher order elements are used.g.. : : : . Further examples of how symmetry can be used (e. This approach ignores the negative x axis and considers the half plane problem shown.1 F IGURE 3.9: Illustration of a symmetric mesh. An alternative approach to the method above arises from the implied no flux across the z axis.M +1 to .N will never contain a singularity arising from the fundamental solution. Note: Since i = 1. u r.. 3.3.i x .14 A XISYMMETRIC P ROBLEMS 67 z . radial symmetry) are given in the next section.N +1. However now the surface to be discretised extends to infinity in the positive and negative z directions and the resulting systems of equations produced is much larger.i .14 Axisymmetric Problems If a three-dimensional problem exhibits radial or axial symmetry (i.

. z  = u r.1 .

The first step in such a procedure is to write the standard boundary integral equation in terms of cylindrical polars r.2 . . z ) it is possible to reduce the two-dimensional integrals appearing in the standard boundary Equation to one-dimensional line integrals and thus substantially reduce the amount of computer time that would otherwise be required to solve the fully three-dimensional problem.

z  i.e.. 1 Z 0Z2 @!p 1 Z 0Z2 c P  u P  + u @ @n d..

= q @ !p d.q A rq d.

0 . 0 (3.q A rq d. . .45) where rp .

. zp  and rq .p .

zq  are the polar coordinates of P and Q respectively. and .q . and . is the intersection of .

= 0 semi-plane (Refer Figure 3.10).) . (n.b. Q is a point on the surface being integrated over.

11 2 2 2 r1 = rp + rq .68 T HE B OUNDARY E LEMENT M ETHOD z .11: The distance from the source point (P ) to the point of interest (Q) in terms of cylindrical polar coordinates. 2rprq cos . for an axisymmetric problem. . z Q r2 r P rp r1 rq r F IGURE 3.10: Illustration of surface . r F IGURE 3. From Figure 3. For three-dimensional problems governed by Laplace’s equation where rp is the distance from P to Q.

p . .

2rprq cos .q  q 2 = r2 + r2 r 1 2 1 !p = 4r q 2 2 r = rp + rq .

p . .

b cos . zq 2 q 2 = a .q  + zp .

p . .

zq  (3.q  + zp .46) .

3.15 I NFINITE R EGIONS 69 We define 1 Z ! d.

but it also depends upon the distance of these points to the axis of revolution. i. !p is called the axisymmetric fundamental solution and is the Green’s function for a ring source as opposed to a point source.e..49) where p is the dirac delta centered at the point P and r . rP  is the dirac delta centered on the ring r = rp . We also define Z @!p = 1 qp 4 @n d. rp = 0 instead of (3. the axisymmetric fundamental solution cannot be written as simply a function of the distance between two points P and Q.and three-dimensional cases. Unlike the two.48) r2 ! + p = 0 (3. ! p is a solution of r2! + r .47) and K m is the complete elliptic integral of the first kind. K m p !p = 4 p q  a+b 2 0 where m = 2b a+b (3.

3. . (3.47) and Equation (3. zq E m n Q p q p E m . zq   1 zp . b z a.51) where E m is the complete elliptic integral of the second kind.q @! p @n 2 0 (3. . r2 + zp . = q!p d. K m nr Q + a .50) becomes q = p  1 r2 .52) 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. Using Equation (3. However all concepts presented thus far are also .15 Infinite Regions The boundary integral equations we have been using have been derived assuming the domain is bounded (although this was never stated).50) For Laplace’s equation Equation (3.45) as (3.50) we can write Equation (3.b  a + b 2rq Z @!p Z c P  u P  + u @n d.

70 T HE B OUNDARY E LEMENT M ETHOD . Consider the problem of solving r2 u = 0 outside some surface . (see Figure 3. nice) regions provided the solution and its normal derivative behave appropriately as . The boundary integral equations for the bounded domain R can be written as Z @!P Z Z Z @!P c P  u P  + u @n d. = q!P d. . (3. if Z . .53) If we let the radius R ! 1 Equation (3.53) will only be valid for the points on .12: Derivation of infinite domain boundary integral equations.12). + q!P d.. . . ! 1. is the centre of a circle (or sphere in three dimensions) of radius centred at some point x0 on . R R x0 . and surrounding .e. valid for infinite regular (i. F IGURE 3. + u @n d. ..

54) If this is satisfied.. = 0 R!1 . . the boundary integral Equation for Z @!P Z c P  u P  + u @n d. . = q!P d. (3.e. q!P d. will be as expected i.55) . (3. @!P lim u @n .

3.53) behaves at most as O R. In this case Equation (3.1 .1 so that q = O R.2.53) will be satisfied if u behaves at most as O R.e. each term will ! 0 as R ! 1) For two-dimensional problems with !  = O log R we require u to behave as log R so that q = O R. = jJ j d. These are the regularity conditions at infinity and these ensure that each term in the integral Equation (3.. For almost all well posed infinite domain problems the solution behaves appropriately at infinity.15 I NFINITE R EGIONS 71 For three-dimensional problems with !  where jJ j is the Jacobian and O  represents the asymptotic behaviour of the function as R ! 1.1  (i. d.

2 @n 1 = 4r where jJ j = O R2 .R.1 @! = O .  ! = O R.  .d .

1 2 Equation Solution p Laplace @ 2 u + @ 2 u + = 0 @x2 @x2 .72 T HE B OUNDARY E LEMENT M ETHOD 3.1 Two-Dimensional equations Here r = x2 + x2 .16 Appendix: Common Fundamental Solutions 3.16.

2 Helmholtz Equation Solution Wave Equation . 0 1 2 1 log 1 u  = 2 r @ 2 u + @ 2 u + 2u + = 0 0 @x2 @x2 1 2 1 u = 4i H02 r where H is the Hankel funtion.

@2u @2u @2 u c @x2 + @x2 . @t2 + 0 t = 0 H u = . 1 2 Solution Diffusion Equation Solution @ 2 u + @ 2 u .. 1 @u = 0 @x2 @x2 k @t 1 2 where k is the diffusivity. .r 2 c2 r where c is the wave speed. 2c ctt2.

1 exp . pji . 2  = . p = p ej i ji 1  =. r u 4kt 4kt 3 2 Navier’s Equation Solution  @ jk @xj + l = 0 for a point load in direction l.

nir.16.j + 1 . 2  ij + 3r.  2 r2 @n 1 . @r 8 1 . 3. 2  nj r.j  ej for a traction in direction k where  is Poisson’s ratio.2 Three-Dimensional equations Here r = x2 + x2 + x2 . 1 2 3 p .ir.i .

3.17 CMISS E XAMPLES 73 Laplace Equation Solution @ 2 u + @ 2 u + @ 2 u + = 0 @x2 @x2 @x2 0 1 2 3 = 1 u 4r @ 2 u + @ 2 u + @ 2 u + 2 u + = 0 0 @x2 @x2 @x2 1 2 3 1 u = 4r exp .ir Helmholtz Equation Solution Wave Equation .

@2u @2u @2 u @2u c @x2 + @x2 + @x2 . is r 2 t =0 Solution Navier’s Equation Solution u =  @ jk @xj + l = 0 for a isotropic homogenenous Kelvin solution for a point load in direction l. @t2 + 1 2 3 where c the wave speed. u = u el k lk .

4 @r @r 1  = ulk 16G 1 . 2D steady-state heat conduction inside an annulus To determine the steady-state heat conduction inside an annulus run the CMISS example 324. 4. CMISS biopotential problems C4 and C5. 2.47) and for q  see Equation (3. . 3D steady-state heat conduction inside a sphere.3 Axisymmetric problems Laplace For u see Equation (3. 3. c 4r 3. 3 .17 CMISS Examples 1.51) 3.   r lk + @x1 @x2 for a displacement in direction k where  is Poisson’s ratio and G is the shear modulus. To determine the steady-state heat conduction inside a sphere run the CMISS example 328.16. t. CMISS comparison of 2-D FEM and BEM calculations To determine the CMISS comparison of 2-D FEM and BEM calculations run examples 324 and 312.

.

4. 3. 9. Apply the boundary conditions.. Evaluate the components of stress from strain using the elastic material constants. e. In fact. 5. Evaluate the stresses and strains using the nodal displacements and element basis functions. Evaluate the strain energy for each element by integrating the products of stress and strain components over the element volume. Solve the resulting system of equations for the unconstrained nodal displacements. .a technique known as the Rayleigh-Ritz. In all cases the steps are: 1. Minimize the potential energy with respect to the unconstrained nodal displacements. We will first consider one-dimensional truss and beam elements.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 . Evaluate the boundary reaction forces (or moments) at the nodes where displacement is constrained. 2. 8.Chapter 4 Linear Elasticity 4. then two-dimensional plane stress and plane strain elements. and finally three-dimensional elasticity. 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. Evaluate the components of strain in terms of nodal displacements. as we will show later. 6. by fixing nodal displacements.g. 7. Evaluate the potential energy from the sum of total strain energy for all elements together with the work done by applied boundary forces.

0 and x. y  and making an angle .1 with end points 0.2 Truss Elements Consider the one-dimensional truss of undeformed length L in Figure 3.76 L INEAR E LASTICITY 4.

relative to the left hand end. with the x-axis. Y  u l L . Y + v v X. Under the action of forces in the x.and y directions the right hand end of the truss displaces by u in the x-direction and v in the y -direction. y X + u.

1: A truss of initial length L is stretched to a new length l.1 = r L u + sin .and y .directions. Displacements of the right hand end relative to the left hand end are u and v in the x. The new length is l with axial strain 2 2 l . x F IGURE 4. 1 = X + u + Y + v . respectively. 1 p 2 2 e= L + pL2 + 2XXu Y Y v + u2 + v2 + .

1 = 1 + 2 cos .: v + u2 + v2 .

: L L L2 q X = cos .

and Y = sin .

the strain for small displacements u and v is " 2 using u v e  cos .. Neglecting second order terms in the binomial expansion p1 +L  = 1 + 1 " + OL"2.

: L + sin .

: L = L L (4.1) The strain energy associated with this uniaxial stretch is 1Z SE = 2 1A Z e dV = 2 0 1 Z EAe2 dx = 1 ALEe2 e dx = 2 2 0 (4.2) .

u1 + sin . We now substitute for e from Equation (4.4. linearly related to the strain e via Young’s modulus E . v1 and u2. u1 and v = v2 .2) and put u = u2 . v1 .2 T RUSS E LEMENTS 77 where = Ee is the stress in the truss (of cross-sectional area A). v2 are the nodal displacements of the two ends of the truss u2 .1) into Equation (4. where u1.

: v2 . v1 1 SE = ALE cos .

: 2 L L .

The strain in truss 1 (joining nodes 1 and 3) is u1 cos 30 + v1 sin 30 = 3 u1 + 1 v1 L L 2 L 2L The strain in truss 2 (joining nodes 1 and 2) is u1 . u2 cos 90 + v1 sin 90 = v1 L L L The strain in truss 3 (joining nodes 2 and 3) is u2 cos 30 = 3 u2 L 2 L p p . A force of 100 kN is applied in the x-direction at node 1.directions. consider the system of three trusses shown in Figure 4.2: A system of three trusses.and y . node 1 1 30 30 node 3 3 2 node 2 100 kN F IGURE 4. Node 3 is a pivot and therefore has zero displacement in both x. 2 (4. For example.2. The problem is to find all nodal displacements and the stress in the three trusses. Node 2 is a sliding joint and has zero displacement in the y-direction only. The Rayleigh-Ritz approach is to minimize this potential energy with respect to the nodal displacements once all displacement boundary conditions have been applied.3) The potential energy is the combined strain energy from all trusses in the structure minus the work done on the structure by external forces.

Thus the strain in truss p 1 is  23 3:34 .g. 53 u1 Solving these last two equations gives u1 = 3:34 mm and v1 = .1.57:5 kN (compressive) and in truss 3 is zero.0:115 and in truss 3 is zero.4) " p 3u + 1v 1 + v = 0 2 1 2 1 2 1  (4. The nodal reaction forces are shown in Figure 4.78 L INEAR E LASTICITY Since a force of 1 kN acts at node 1 in the x-direction. p . in truss 2 is .3 m2 107 kPa  0:232  10.1:15 mm.  v1 = .2 = 116 kN (tensile).5) gives for two dimensions p . 100u1 2 L 2 2 2 L Minimizing the potential energy with respect to the three unknowns u1 ..3 m2. 100 mm  50 mm timber AE = 5  10. in truss 2 is .3 = 0:232. 100:u1 = 1 AE 4 3 u1 + 1 v1 + 3 u2 + v1 5 .6) gives u2 = 0 Equation (4.6) = 5  10. the potential energy is PE = X1 trusses 2 2 p !2  p !2 23 ALEe2 .4) gives 3u1 + 3v1 = 4  102= 5  104 Equation (4. truss) then L Equation (4. 1 1:15  10.3 m2 107 kPa=m = 5  104 kN m. 100 = 0 2 (4.5) @ PE = AE @u2 L If we choose A p ! p 3 3 2 u2 2 =0 (4. E = 10 GPa and L = 1 m (e. v1 and u2 gives @ PE = AE @u1 L @ PE = AE @v1 L p 3u + 1v 2 1 2 1 !p ! 3 .3. 2 The tension in truss 1 is A = AEe = 5  10.

2. Thus = Z x z dA .4. The axial strain is given by ex = R where z is the lateral distance from the neutral axis in the plane of the bending and R is the radius of curvature in that plane.7) becomes The slope of the beam is dw = .7) M= where I Z E Z z2 dA = EI x z dA = R R (4.3: Reaction forces for the truss system of Figure 4. 4.8) = Z z2 dA is the second moment of area of the beam cross-section. Thus. along the beam (assumed here to be in the x-direction).3 B EAM E LEMENTS 79 100 kN 100 kN 57:7 kN 57:7 kN F IGURE 4.3 Beam Elements Simple beam theory ignores all but axial strain ex and stress x = Eex (E = Young’s modulus) z . where A is the beam x z = Eex = E R (4. E = M R I x and Equation (4. The bending moment is given by M crossectional area.

and the rate of change of slope is the curvature dx d.

2 1 K = dx = d w = R dx2 = Mz I (4.9) (4.10) .

13) .80 L INEAR E LASTICITY Thus the bending moment is 2 M = EI d w = EIw00 dx2 (4. dx2 EIw00  dx (4. dM = . balancing the loading forces p with the axial stresses associated with beam flexure 2 d2 . dx EIw00 dx and the normal force (per unit length of beam) (4. dx2 EI d w = p dx2 d2 p = dV = .11) and a force balance gives the shear force d V = .12) This last equation is the equilibrium equation for the beam.

Z pw dx: PE = 2 0 0 (4. (4. the (flexural) strain energy is 1Z SE = 2 =2 1Z L Z x=0 A L x=0 L Z z 2 1 Z EI w002 dx E R dA dx = 2 A x=0 1 x ex dA dx = 2 ZL Z x=0 E A e2 dA dx x where x is taken along the beam and A is the cross-sectional area of the beam. A linear basis function has a zero second derivative and therefore cannot represent the flexural strain. The boundary conditions associated with the 4th order equilibrium Equation (4.14) The elastic strain energy stored in a bent beam is the sum of flexural strain energy and shear strain energy. The potential energy is therefore L L 1 Z EI w00 2 dx .14) or the equa- . but this latter is ignored in the simple beam theory considered here. 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.15) The finite element approximation for the transverse displacement w must be able to represent the second derivative w 00 . The external work associated with forces p acting normal to the beam and moving through a L transverse displacement w is Z 0 pw dx. Thus.

15) (which contain the square of 2nd 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 (i) (ii) Simply supported zero displacement w = 0 zero moment M = EIw 00 Cantilever zero displacement w = 0 zero slope .4.4 P LANE S TRESS E LEMENTS 81 tions arising from minimum potential energy Equation (4.

d EIw00  = 0 dx zero moment M = EI 00 = 0 shear force V slope . = w 0 = 0 zero shear force V =0 (iii) Free end = .

= w 0  shear force V moment M slope .

we define the displacement vector 2e 3 4 ex 5 and stress vector y exy u = v . displacement w 4. 0 0 3 5. = Ee The strain components are given in terms of displacement ex = @u @x ey = @v @y . The stress-strain relation for two-dimensional plane stress: xy u = E 2 ex + ey  1. 21  E 4 1 2 gradients by 0 0 1.4 Plane Stress Elements For two-dimensional problems. strain vector e = 2 3 x = 4 y 5. 2 y E e  xy = 1 +  xy x can be written in matrix form (4. E e + e  y= x 1 .16) where E = 1.

1 @u + @v exy = 2 @y @x (4.17) .

uT f dA T A Z (4. Following the steps outlined in Section 4.  2 x y x y xy V 1 Z e + e + e e dV = 2 x x y y xy V xy  dV The potential energy is PE = SE . external work 1Z =2 V e Ee dV .18) where represents the external forces acting on the elastic body.1 we approximate the displacement field finite element basis u = 'n un . v = 'n vn and calculate the strains f u with a ex = @u = @'n un @x @x ey = @v = @'n vn @y @y .82 L INEAR E LASTICITY The strain energy is 1Z SE = 2 = 2 1Z V V T e T 1 Z E he2 + e2 + 2ve e + 1 . v2 i dV Ee dV = 2 1 .

.

1 @u + @v = 1 @'n u + @'n v exy = 2 @y @x 2 @y n @x n or (4. Z uT f dA =2 Z uT f dA .18) the potential energy is therefore B EB dV:u . Z PE = 2 1 uT Z = 2 T (4. uT f dA 1 uT Ku .20) From Equation (4.19) 3 2 @'n 2 e 3 6 @x 0 7   6 7 n x e = 4 ey 5 = 6 0 @'n 7 un = Bu 6 6 1 @' 1 @y 7 v exy 4 n @'n 7 5 2 @y 2 @x 1 Z BuT E Bu dV .

5 Navier’s Equation The Galerkin finite element equations for linear elasto-statics can be derived from a physically appealing argument.23) where the internal work done due to the stress field is equated to the work due to internal body forces and external surface forces.i dV Substituting the equilibrium relation tij. Equation (2. in moving through a virtual displacement = uj j to the work done by the stress vector = tj j in moving through a compatible set of virtual diaplcements .15): Z Z @g f r  rg + rf  rg d = f @n d. Ku = f (4. Now. Thus. t t s t s i u i s Z S sj uj dS = Z S tj uj dS = Z S u t i i tij ni uj dS The Green-Gauss theorem. .fj into the right hand integral yields the virtual Z V tij uj. = ).e..i uj + tij uj.4. the principle of virtual work.i = fj (tij are the components of stress) and by Cauchy’s formula. giving Z S sj uj dS = Z V tij uj dV = Z V tij.21) Z We next minimize the potential energy with respect to the nodal parameters un and vn giving 4. Let = e and interpolate the virtual displacements u i S . where ni is a component of the unit normal to S .5 NAVIER ’ S E QUATION 83 where K = BT EB dV is the element stiffness matrix. the principle of virtual work equates the work done by the surface forces = sj j .i work equation = .22) is now used to replace the right hand surface integral by a volume integral. (4. = tij ni j .i dV = Z V fj uj dV + Z S sj uj dS (4. The equilibrium equation in V is . Let be the stress vector acting over the surface S enclosing a volume V of material of mass density and let be the equailibrium external force vector per unit area of surface (i.tij.

e and 4n.k @xj ij 'M. i.m 1 ZZZ . i n i n ij 'n. so u = 'n un i i @'n un u = @x i i.84 L INEAR E LASTICITY from their nodal values.e.A Ui4n.k i @xj e e+ Z @ 1  ti'n d.e  Z @k d A U =@ d ij 'n. e is the global node number of local node n on element e where un  = Ui i This gives 0 X @Z e e 1 0 4n.e Since virtual dispalcements are arbitrary we get XZ e We now have e @k d = X Z b ' d + Z t ' d.24) 4n.j j @ = 'n.k @xj e e @ Z e @k d Z ij 'n.k @xk un i j (4..

.' uN  = 1 @'N @'N uN 2 j. isotropic.k = @'N @'N uN @ @x k n k . ij = ekk ij + 2eij = 1 ui.i we put uj = 'N uN this gives j 2 (4.26) where  and  are Lam´ s constants. the displacement at node N in direction i) j For linear.j + uj. 0 e @m J  d ij 'M.e. the coefficient of uN (i.m @xj @m d @xj (4.i 2 @xi N j 2 @n @xj i (4.27) Note that a coefficient of uN also comes from eji (the first term) j ekk = uk.25) We wish to find EMNij . homogeneous materials we have the generalised Hookes Law. Also eij e 1 u = 1 @ .

6 N OTE ON C ALCULATING N ODAL L OADS 85 So @'N @'N uN + 2 1 @'N @'N uN + 1 @'N @'N uN ij =  ij @n @xk k 2 @n @xj i 2 @n @xi j So the coefficient of uN can be calculated by j .4.

1 ZZZ .

The expression for EMNij can be simplified to give EMNij = where g mn 1 ZZZ . @'N @n @'M @m @'N @n @'M @m + 2 J  d EMNij =  ij 0 @n @xk @m @xk @n @xj @m @xi Note: there is no sum for ij .

4a.   d = 1 pL 2  d = 1 pL 2 . Equation (4. If the element side has a linear basis. @x k k 4.21) has components f fn = Z p'n dx = pL Z 'n d (4..ngmn J  d @xi j (4.m'N.29) gives f1 = pL f2 = pL Z Z '1 d = pL '2 d = pL Z Z 1 . i.29) where  is the normalized element coordinate along the side of length L loaded by the constant stress p kN m. 0 @ 2'M.e.m'N. For example.1 applied to the edge of the plane stress element in Figure 4. the metric tensor resulting from the inner product of basis vectors. consider a uniform load p kN m.6 Note on Calculating Nodal Loads If a normal boundary stress is known it is necessary to compute the equivalent nodal load forces to represent the distributed load.n @m @xn +  ij 'M. The nodal load vector in Equation (4.1 .28) @ = @m @xn .

4b.86 L INEAR E LASTICITY as shown in Figure 4. If the element side has a quadratic basis.29) gives f1 = pL f2 = pL f3 = pL Z '1 d = pL '2 d = pL '3 d = pL Z . Equation (4.

1 Z Z .

j + uj.. p kN m.L 3 . 2 4.30) where ij are the components of the stress tensor ( ij is the component of the traction or stress vector in the ith direction which is acting on the face of a rectangle whose normal is in the j th direction).  1 .7 Three-Dimensional Elasticity Recall that if a body is in equilibrium then we have ij.4: A uniform boundary stress applied to the element side in (a) is equivalent to nodal loads of 1 pL and 1 pL for the linear basis used in (b) and to 1 pL.L 2 2p . and bi is the body force/unit volume (e.L 3 2 2p . A node common to two elements will receive contributions from both elements. 1 d = 1 pL 2 6 as shown in Figure 4. Two adjacent quadratic elements both contribute to a common node in (d).   d = 1 pL 6 4 1 .i 2 i. 2. 2 pL and 1 pL for the quadratic 2 2 6 3 6 basis used in (c).   d = 2 pL 3 2  . where the element length is now L . 3 (4.j + bi =0 i.g.4c. 3 (4.31) . Z Z 2 2 .L 3 2 1p . = ) Recall also that the components of the (small) strain tensor are b g eij = 1 ui. as shown in Figure 4. j = 1. 2.1 L 1 2 pL 1 6 pL 2 3 1 6 pL 1 2 pL (c) 1 6 pL (d) 6 .L 2 1p p 2 (a) (b) F IGURE 4.4d. j = 1.

If we start dealing with materials such as rubber or living tissue then we need to use the exact finite strain tensor).. u u ij = cijklekl where cijkl are the 81 components of a 4th order tensor. the material response is independent of orientation of the material element) then we have the generalized Hook’s Law.e.32) eij = 21   2 3 + 2 kk ij where . This can often work with some smaller grouping of Equation (4. (4.kk +  +  uk. For a linear elastic material we usually propose that ij depend linearly on eij . Currently we have 15 unknowns (6 stresses.e. If the material is isotropic (i..31) into Equation (4. i. In the general three-dimensional case.kj + bj = 0 These are the 3 equations for displacements. is the difference between the final and initial positions of a material point in question). An equation of state (stress-strain relation or constitutive law) is required. The object of solving an elasticity problem is to find the distributions of stress and displacement in an elastic body. uj.  are related to Young’s modules E and Poisson’s ratio  by + E =  3+ 2   = 2   + As long as the displacements are continuous functions of position then Equation (4.4.30) yields Navier’s equation. If all boundary conditions are expressed in terms of displacements Equation (4.32) then into Equation (4. subject to a known set of body forces and prescribed stresses or displacements at the boundaries. (Note: We are assuming here that the displacement gradients are small compared to unity which is the usual situation in solid mechanics.31) and Equation (4. 6 strains and 3 displacements) and only 9 equations..  are Lam´ s constants. ij or inversely = ekk ij + 2eij ij . this means finding 6 stress components ij = ji  and 3 displacements ui each as a function of position in the body. e Note: ..32) e.31) yields strains and Equation (4. Equation (4.30).g.e. Equation (4. . Symmetry reduces the number of unknowns to 21. Equation (4.31) and Equation (4.7 T HREE -D IMENSIONAL E LASTICITY 87 where is the displacement vector (i.32) yields stresses.32) are sufficient to determine the 15 unknown quantities.30).

j d ij ui. This statement is independent of the constitutive law of the material.7). we have essentially used the Green-Gauss theorem once to move from Equation (4.35) is more usually derived from considering virtual work (we use weighted residuals to tie in to Chapter 3). un Z Z Z ij ui.33) and Equation (4.e.  @ ij i j d. .88 L INEAR E LASTICITY 4. In the above derivation.35) (as was done for the derivation of the FEM equation for Laplace’s .j  Z ij. The principle of virtual work equates the internal work due to the stress field (left hand side integral) to the work due to internal body forces and external surface forces.33) to Equation (4. This statement Equation (4.j + bi  ui d =0 (4.34) we have Z ij i. 4. by the divergence theorem ij.35) is the starting point for the general finite element formulation (Section 4..9 Linear Elasticity with Boundary Elements Equation (4. i where ti are the (surface) tractions (i.34) Thus combining Equation (4.35) b u d tiu d.j i u d = = = Z Z Z  ij i .j = i Therefore. ti = ij nj ).j ui + ij ui. (4. The u are usually interpreted as a consistent set of i i virtual displacements (hence we use the notation u instead of w ).j d  (4.j d u = = Z Z b u d i i i i + + @ @ ij nj ui  d.8 Integral Equation Using weighted residuals as before we can write Z  ij. Now u  ij u.j d ij i u .33) where  = u  is a (vector) weighting field. Z Z ij ui.j d  r u d .

To continue. ij.4.i = 1 ij u + 1 jiu 2 i.36) can be written as i i Z .j d u = Z = = = Z Z ij eij d  ekk ekk e e ij ij d kk d + 2 + 2 Z Z eij e d ij Z eij e d ij  eij ij d due to symmetry. Since the body forces. are known functions. ij Using the constitutive law for linearly elastic materials (Equation (4.9 L INEAR E LASTICITY WITH B OUNDARY E LEMENTS 89 equation).37)  ( ij .32)) we have Z ij i. t represents the equilibrium state corresponding to the virtual displacements u ). = i i Z b u @ i id + Z @ tui d. Thus from the virtual work statement.j 2 j.i = 1 ij u + 1 ij u 2 i. e . ij i i Note: What we have essentially done is use integration of parts to get Equation (4.j 2 i.   ij.j (i.36) This is known as Betti’s second reciprical work theorem or the Maxwell-Betti reciprocity relationship between two different elastic problems (the starred and unstarred variables) established on the same domain.37) does not introduce any unknowns into the problem (more about this later). Therefore Equation (4. Equation (4.j 2 j. we firstly note that ij eij  = 1 ij u + 1 ij u 2 i.j = ij u i. bi . . i (4. the second domain integral on the left hand side of Equation (4.35) and the above symmetry we have Z b u d i i + Z t u d.36) above (after noting the reciprocity between ij and eij ). then use it again to get Equation (4.35).j + b = 0). .e. i (4.   Note that b = .j ui d + Z b u d i i = Z t u @ i i d.j where e are the virtual strains corresponding to the virtual displacements. Z @ ti u d.. ij.

x e tj ij i (4.j ui d  =.90 L INEAR E LASTICITY The first domain integral contains unknown displacements in remove. P  eiui d = .38) (or equivalently . x tj x d.b + ei = 0). x ij + Z u P. x represent the displacements and tractions in the j th direction at x ij ij corresponding to a unit point force acting in the ith direction (ei ) applied at P . x uj x d. + Z bj u d j P2 (4. This is known as Somigliana’s 1 identity for displacement. Therefore x Z ij. P .41) where u P. x ..39) If each point load is taken to be independent then u and t can be written as j j u = u P. Therefore. the volume integral is replaced with a point value (as for Laplace’s equation). x and t P.37) becomes ui P  ei = Z @ tj u d. Z @ t P.e. . Equation (4. We can interpret this as the body force components which correspond to a positive unit point load applied at a point P 2 in each of the three orthogonal directions.j + ei  =0 (4. where ei is the ith component of a unit vector in the ith direction i and ei = ei  .39) (and equating components in each ei direction) yields ui P  = Z u @ ij P. . 1 Somigliana was an Italian Mathematician who published this result around 1894-1902.uiP ei i. x bj x d x ij (4.42) where P 2 (see later for P 2 @ ). j Z t u @ j j d. Substituting these into Equation (4. x ei j ij  = t P. Z x . We choose the virtual displacements such that and it is this integral we wish to ij.40) (4.

kk G G u + 1 .   Gr f3 . r.10 Fundamental Solutions  Recall from Equation (4. = 2. 2 r.   r 1 .inj .44) The solutions to the above equation in either two or three dimensions are known as Kelvin 2 ’s fundamental solutions and are given by 1 u P.43) b = ei x . 2  ij + r. 2 u + b = 0 k.38) that ij satisfied ij. the distance between load point (P ) and field point (x).j + or equivalently  x . 4  ij log r .j r.45) for three-dimensions and for two-dimensional plane strain problems.ir.kk (4.4.k g Lord Kelvin (1824-1907) Scottish physicist who made great contributions to the science of thermodynamics .1 @r t P. 2  r. ri = xi x .j g (4.1  Gr f1 . 2  r. P  ei = 0 k. Here r In addition the strains at an point given by = @x@rx = ri .46) . 2 u + x . xi P  x due to a unit point load applied at P in the i jk + th direction are e P.j ik  .k ij + r. r.ki i i.ki i.j ni  (4. x = 8 1. u P. r i r P. x.i jki .ir. x = 16 1 . P  ei = 0 (4. P  i Navier’s equation for the displacements u is i where G = shear Modulus. Thus u satisfy i G G u + 1 .j g ij . 1 .47) ij where = 1.j  @n .1   G f3 .i tively. 4  ij ij + r.ir.ir. r. and (4. x = 8  1. 2. 3 for two-dimensional plane strain and three-dimensional problems respecand r. x = 4  1 .10 F UNDAMENTAL S OLUTIONS 91 4.

 .kg =  1+ (This is a where and are defined above.there are obviously physical differences. r.92 L INEAR E LASTICITY and the stresses are given by ijk P.ir. Note that in three dimensions u and for two dimensions ij =O r . The plane strain expressions are valid for plane stress if  is replaced by  mathematical equivalence of plane stress and plane strain . 2  r.j r.k ij + r.all we have to do is modify the values of the elastic constants).j ki . What the mathematical equivalence allows us to do is to use one program to solve both types of problems .   r f1 .i jk  + r.1 x = 4  1 .

1 t ij = O r2 =O .

1 .

Telles & Wrobel (1984b) pp 190–191. we need to find the equivalent of c P  (in section 3) . one can find the stress at any P 2 firstly by combining derivatives of (4. t P. . Details can be found in Brebbia.42) as P is moved to @ .42) to produce the strains and then substituting into Hooke’s law.) We use the same procedure as for Laplace’s equation but here things are not so easy. x bk x d x ijk Note: One can find internal stress via numerical differentiation as in FE/FD but these are not as accurate as the above expressions. Z u P.e. If P 2 @ we enlarge to 0 as shown. Consequently.11 Boundary Integral Equation Just as we did for Laplace’s equation we need to consider the limiting case of Equation (4.1 u ij = O log r t ij r : Somigliana’s identity (Equation (4. ijk ijk 4.. x tk x d.42)) is a continuous representation of displacements at any point P 2 .called here cij P . Expressions for the new tensors u and t are on page 191 in (Brebbia et al. x . Z ijk P. This yields ij P  = Z u + . (i. x ijk . 255–258. 1984b). x uk x d.

." Z t P.5: Illustration of enlarged domain when singular point is on the boundary.e." x uj x d. x uj x d. x bj x d x ij (4.4." +." +." .48) We need to look at each integral in turn as " 0 (i. Z . . x ij ." | z 0 uj x Z by continuity of Z + uj P  t P." +. x ij (4." . x d." x uj x d.11 B OUNDARY I NTEGRAL E QUATION 93 0 " P . .. Then Equation (4. uj P  d. x ... x = t P.. x ij (4.42) can be written as ui P  = Z u ij P. x ij + .x ij . This can be written as Z t ij P.50) . x uj x d." Z + ." t P." Z 0 u P.49) Z The first integral on the right hand side can be written as t ij P.. " ! 0 from above). x uj x d. The only integral that presents a problem is the second integral." F IGURE 4. x = t P. x tj x d. x uj x .

t u ij j d.) where the integral on the left hand ij side is interpreted in the Cauchy Principal sense. 2 = ." = ln " ." = ." " But if we replace . Thus as " 0 we get the boundary integral equation Z . x tj x d... = ." x Then 1 Z f x dx = Z. Laplace’s Equation).49) as where we interpret this in the Cauchy Principal Value3 sense.2 x 0 Z" 1 dx = @ lim " !0 1 lim . Z (4. x ij = Z .." + ln jxjj2 .2 x dx + Z2 " x 1 dx = ln jxjj. the integral does not exist in the proper sense.2 "  "lim0 ! Z f x dx =0 Z This is the Cauchy Principle Value of . then !0 1 .. x uj x d.94 L INEAR E LASTICITY Let cij P  = ij + lim "0 Z . x ij (4. x + u P. 2 Thus it is What is a Cauchy Principle Value? Consider f x = on . whereas with Laplace’s equation the unknowns were 3 @u u and @n (scalar quantities) here the unknowns are vector quantities. x 1 dx = Z2 . = u tj d. x uj x d. x d.." 1 . if an integral exists in the proper sense. However..52) to represent rigid-body movements.2 x 1 0 Z2 1 dxA + @ lim " !0 2 .52) (or. In practical applications cij and the principal value integral can be found indirectly from using Equation (4." t P. cij uj + Z . then it exists in the Cauchy Principal Value sense and the two values are the same." by Z .2.g. ln " = 0 8" 0 .51) As " 0. . x ij cij P  uj P  + Z .e. i. x bj x d ij Z . t P. . but it does in the Cauchy Principal Value sense. ." ".52) is similar to the numerical implementation of an elliptic equation (e. The numerical implementation of Equation (4. u ij P.2. t P. However... ln 2 + ln 2 . and we write the second integral of Equation (4. in brief (no body force)."2 x 1 1 dxA(by definition of improper integration) which does NOT exist. f x dx ." ! .

we get 3 equations per node). 5 4 . If we have a rigid-body displacement of a finite body in any one direction then we get Au = Bt (4.3 displacements and 3 tractions . use 2u 3 1 u = 4u25 . a 3 x 3 matrix) contains principal value components.e. There is a similar 4.4.53) We can discretise the boundary as before and put P ..the fundamental solution .12 B ODY F ORCES ( AND D OMAIN I NTEGRALS IN G ENERAL ) 95 more convenient to work with matrices instead of indicial notation.. 21 22 23    u31 u32 u33 2t t t 3 11 12 13 t = 4t t t 5 21 22 23    t31 t32 t33 Then (in absence of a body force) we can write Equation (4. 7 and t = 6 . the singular point.54) (for three-dimensions. the diagonal entries of result for an infinite body. tu d. 5 un tn The diagonal elements of the A matrix in Equation (4. (This is because the body force is known . 7 where n is the number nodes.12 Body Forces (and Domain Integrals in General) The body force gives rise to a domain integral although it does not give rise to any further unknowns in the system of equations. = Z . A (the cij ’s) do not need to be determined explicitly.52) as cu + Z .e. 4 . u3 2t 3 1 t = 4t25 t3 2u u u 3 11 12 13 u = 4u u u 5 . (4.. at each node (each node has 6 unknowns . ut d. i..54) ( l = vector defining a rigid body displacement in direction l) i Ail = 0  aii = . The overall matrix equation 2 3 2 3 u1 t 6u2 7 6 t1 7 6 7 6 27 where u = 6 . X i6=j aij (no sum on i) i.

Integrals over the domain containing known functions (eg body force integral) appear in many situations e. ik. In some cases domain integrals must be used but there are techniques developing to avoid many of them. the Poisson equation r2 u = f yields a domain integral involving f .. However. a body force arising from a constant gravitational load. let G (the Galerkin tension) be related to u by ij ij u = G .that of having to prepare only a boundary mesh. a domain mesh nullifies one of the advantages of BEM . Thus Equation (4.96 L INEAR E LASTICITY was chosen so that it removed all unknowns appearing in domain integrals).(n. The question is how do we evaluate domain integrals such as those appearing in the boundary integral forumalation of such equations? Since the functions are known a coarse domain mesh may work.kk .. e. In some standard situations a domain integral can be transformed to a boundary integral. or a centrifugal load due to rotation about a fixed axis or the effect of a steady state thermal load can all be transformed to a boundary integral.g. 2 1 1   G ij ij. Since the integral also contains the fundamental solution and may not be a “nice” region it is unlikely that it can be evaluated analytically).52) is still classed as a Boundary Integral Equation.g.kj 8 1 r ij  Gij = 8G 2 .b. Firstly.

1 1 : 8G r log r ij Then (3D) (2D) Bi = Z u b ij j d = Z .

j .kj bj d Under a constant gravitational load g = gj  G ik.kk . 2 1 . bj = gj  Bi = gj = gj Z . G ij.   1 G ik.

The two integrals of note are the DRM. Z  .kj 1 G . developed around 1988. Unless the domain integrand is “nice” the above simple application of Green’s theorem won’t work in general.  g G nk d. 2 1 .   1 G ik. .k ik. There has been a considerable amount of research on domain integrals in BEM which has produced techniques for overcoming some domain methods. developed around 1982 and the MRM. multiple reciprocity method. ij. dual reciprocity method. 2 1 .j  d which is a boundary integral.

2.2. To solve a truss system run CMISS example 411 This solves the simple three truss system shown in Figure 4.13 CMISS Examples 1. .13 CMISS E XAMPLES 97 4. To solve stresses in a bicycle frame modelled with truss elements run CMISS example 412.4.

.

1 Explicit Transient Finite Differences Consider the transient one-dimensional heat equation @u = D @ 2 u .1) 0 x L. t = u1 and the initial conditions u x. In time-dependent problems the solution proceeds from an initial solution at t = 0 and it is almost always convenient to calculate each new solution at a constant time (t 0) throughout the entire spatial domain . un. Grid points are labelled by the indices i = 0. 5. n is therefore labelled as . There is.1. as shown in Figure 5.2 to solve the transient one dimensional heat equation. (5. I (for the x-direction) and n = 0. The temperature at the grid point i. 1. 0 = 0. : : : . t = u0 and u L. therefore. t 0 @t @x2 where D is the conductivity and u = u x.2 Finite Differences 5. subject to the boundary conditions u 0. 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. t is the temperature. nt = un: i (5. : : : (for the t-direction).2.3 to solve the transient advection-diffusion equation .1 Introduction In the previous discussion of steady state boundary value problems the principal advantage of the finite element method over the finite difference method has been the greater ease with which complex boundary shapes can be modelled.Chapter 5 Transient Heat Conduction 5.a slight generalization of the heat equation. Finite difference techniques are introduced in Section 5. A finite difference approximation of this equation is obtained by defining a grid with spacing x in the x-domain and t in the time domain.1 un+1 i i i .2) Finite difference equations are derived by writing Taylor Series expansions for un+1 . t = u ix.N u x. 1. A combination of finite elements for the spatial domain and finite differences for the time domain is used in Section 5.

 + 1 x2 : @ u + 1 x3 : @ u + O x4 2 3 .100 about the grid point i. n T RANSIENT H EAT C ONDUCTION n n 2 3 .

@x in 2 .

@xu in 6 .

1 = un . x: @u + 2 x2 : @x2 + 1 x3 : @ 3 + O x4 i i @x @x i i 6 . @xu in .  3 1 @2 un.

@u ni .  un+1 = un + t: @t + O t2 i i i un+1 i = un + x: i .

@u n .

.

Adding Equations (5.4) (5.3) and (5.3) (5.5) where O x4  and O t2  represent all the remaining terms in the Taylor Series expansions.4) gives un+1 + un. (5.1 = 2un + x2 : i i i or .

@ 2 u n @x2 i + O x4 .  .

5) gives a “difference” approximation of the first order time derivative .x2  .6) which is a “central difference” approximation of the second order spatial derivative. @2 u n un .1 + O . = i+1 @x2 i x2 (5. 2un + un i i. Rearranging Equation (5.

8) i i i i i Given the initial values of un at n = 0 (i.1 at i i i the latest time step. the values of un+1 for the next time step i i are found from Equation (5.9) . 2. The accuracy of the solution depends on the chosen values of x and t and in fact the stability of the scheme depends on these satisfying the Courant condition:  D xt2  1 : 2 (5.e.1) gives the finite difference approximation un+1 ..7) Substituting Equation (5. un i + O t : = i @t i t (5.6) and Equation (5. : : : . 2un + un.x2  i i i i i 2 t x which is rearranged to give an expression for un+1 in terms of the values of u at the nth time step i  . Applying Equation (5.1 + O .   . x2 : (5.1). yields the time dependent temperatures at the grid points (see Figure 5.1 + O t2 .8) iteratively for time steps n = 1. : : : . t = 0). I . 2. un+1 = un + D xt2 un+1 . un + O t = D un+1 . 2un + un. I  at the previous time step and not on the neighbouring terms un+1 and un. @u n un+1 . : : : etc.8) with i = 1. This is an explicit finite difference formula because the value of un depends only on the values of i +1 +1 un i = 1. 2.7) into the transient heat equation Equation (5.

2 Von Neumann Stability Analysis The concept behind the Von Neumann analysis is that all Fourier components decay as time is advances or as they are processed by an iterative solver.10) An+1ei k kj 4x L  = An ei k j L k h  +1 4x  1 . un+1 = un+1 + 1 .2.14x L  + ei kjL4x  i (5.1 i i+1 . k x (5.. An ei L  we obtain. Considering Equation (5. 2un + un..i kjL4x  k An k . I x F IGURE 5. By subsituting the general Fourier component An ei L  .1 i i i i kj 4x 4t where  = D 2 . as indicated by the dark shading. The lightly shaded region shows where the solution is known at time step n. With central differences in x and a forward difference in t an explicit finite difference formula gives the solution at time step n + 1 explicitly in terms of the solution at the three points below it at step n. k An+1 = 1 .11) by. n shown by the . i.5.2 F INITE D IFFERENCES 101 t n+1 n : x x x x 1 00 1 2 . The equation is centred at grid point i. O 5. we can rearrange this to be of the form. we obtain. 2 ei kj . 2 + ei kjL4x  + e...1: A finite difference grid for the solution of the transient 1D heat equation.8).11) kj 4x If divide Equation (5.

2 + 2cos . k4x = 1 .

12) predicts the growth of any component (specified by (5. 4 + sin2 L x Equation (5.12) k or j ) admitted by the . k4L = 1 .

e..3 Higher Order Approximations An improvement in accuracy and stability can be obtained by using a higher order approximation for the time derivative. For example.  = D 42t  1 x 2 This can be rearranged to be of the form.13) As the sin2 term in Equation (5. 4j This condition will always hold if. If all components are to decay.14) 1  j1 . n + 1 t rather than ix.102 T RANSIENT H EAT C ONDUCTION system.12) is always between criteria that.16) i. we effective have the stablity (5. nt we get 2 . if a central difference approximation is used for . the time step should be at least half the size of the x2 term 2D @u @t 5. (5. An+1  1 k An k for stability (5. 0 and 1.  centering the equation at ix.2.15) 4t  2xD 2 (5.

@u n+ un+1 . un . un = D 1 @ 2 u i i t 2 @x2  .  1 2 by @t i = i t i + O t2 (5.7) and Equation (5.1) is approximated with the “Crank-Nicolson”formula un+1 .17) in place of Equation (5.

n+1 i 1 @2u + 2 @x2 .

n  i (5.e. i i .18) in which the spatial second derivative term is weighted by 1 at the old time step n and by 1 at the 2 2 new time step n + 1.1 . Thus each new time step requires the solution of a system of coupled equations.18) is an implicit scheme .only the time position at which it is centred. Notice that the finite difference time derivative has not changed . The price paid for the better accuracy (for a given t) and unconditional stability (i.. stable for any t) is that Equation (5.the equations for the new time step are now coupled in that un+1 depends on the neighbouring terms i +1 +1 un+1 and un.

un = D ..3 T HE T RANSIENT A DVECTION -D IFFUSION E QUATION 103 t n+1 n : x x x x x x 1 00 1 2 .18) is un+1 . O x A generalization of (5.2: An implicit finite difference scheme based on central differences in t. n + ) shown 2 by the . as well as x. The equation is centred at the point (i. F IGURE 5.. The lightly shaded region shows where the solution is known at time step n.. The dark shading shows the region of the coupled equations.. I x 1 which tie together the 6 points shown by . i.1 i i+1 .5.

@ 2 u i i t @x2 i .

n+1 + 1 . .

 .

19) in which the spatial second derivative of Equation (5. @ 2 u n  @x2 i (5.1) has been weighted by .

. at the new time step and by 1 .

8) is recovered when .  at the old time step. The original explicit forward difference scheme Equation (5.

= 0 and the implicit central difference (Crank-Nicolson) scheme (5.19) when 1 .

An implicit backward difference scheme is obtained when . = 2 .

5.. the advection-diffusion equation. The ratio of advective to diffusive transport is characterised by the Peclet number V L=D where v = k k2 and L is a characteristic length).g. @u + v  ru = Dr2u + f @t (5.  ru then represents advective transport by a velocity field . We also generalize the partial differential equation to include an advection term and a source term.3 The Transient Advection-Diffusion Equation Consider a linear parabolic equation where u is a scalar variable (e. where u is concentration or temperature. = 1. In the following section the transient heat equation is approximated for numerical analysis by using finite differences in time and finite elements in space.20) v v v . D is the diffusivity and f is source term.

20) with weight Z .104 T RANSIENT H EAT C ONDUCTION Applying the Galerkin weighted residual method to Equation (5.

f ! d @t + v  ru . Dr = ! gives =0 or  Z . @u 2u .

The element contributions to M and K are given by Mmne = and Z e 'm'nJ d (5. Putting = 'n un and = 'm and summing the element contributions to the global equations. 1. Equation (5. @u @t + v  ru ! + Dru  r! d @ Z Z @u f! d + @n ! d.23) Z1 @'m @'n @i @j Z1 @ Kmne = D @ @  @x @x J d + vj 'm @'n @xi J d @i j i j k k 0 0 (5. : : : Equation (5. @ (5. (5. 2.24) can be re0.22) dt where M is the global mass matrix. u1 = 0. n = 0.21) can be represented by a system of first order ordinary differential equations. u w M du + K u .21) where @n is the normal derivative to the boundary @ .24) If the time domain is now discretized placed by t = nt. K the global stiffness matrix and u a vector of global nodal unknowns with steady state values (t ! 1) u1 .

u + K .1 1 (5.25) n+1 n M u .

un+1 + 1 . .

 un = Ku1 t where .

Note that for . is a weighting factor discussed in Section 5.2.

25) as M + . Rearranging Equation (5. = the method is known 2 as the Crank-Nicolson-Galerkin method and errors arising from the time domain discretization are O Dt2.

. 1 .tK un+1 = M .

The stability of the above scheme can be examined by expanding (assumed to be smoothly .26) gives a set of linear algebraic equations to solve at the new time step n + 1 t from the known solution n at the previous time step nt. tK un + tKu1 u u (5.

the set of ordinary differential equations Equation (5. bi  ei t s i (5. Writing the initial conditions 0 = ai i and steady state solution 1 = i s u u= u X i bi + ai .3 T HE T RANSIENT A DVECTION -D IFFUSION E QUATION 105 X A M K u s X bi si . can be written as the recursion formula u I + .27) The time-difference scheme Equation (5.26) on the other hand.22) has solution i continuous in time) in terms of the eigenvectors i (with associated eigenvalues i ) of the matrix = .5.1 . with now replaced by a set of discrete values n at each time step nt.

1 .tA with solution un+1 = I . .

.28) u= X i  1 . t 1 . bi  (5. tA un + tAu1 bi + ai .

  n 1 + t.

22) and Equation (5.25).) Comparing Equation (5.29) (You can verify that Equation (5.i i si (5. t 1 . by substituting and using i = i i .   1 . with t = nt.27) and Equation (5.29) are indeed the solutions of Equation (5.27) and Equation (5.27) by the approximation As i s e. respectively. .22) by the finite difference approximation Equation (5.i t in Equation (5.29) shows that replacing the ordinary differential equations (5.25) is equivalent to replacing the exponential e.it or.

  n 1 + t.

1t 1t.i (5.it  1 . ti e.30) .

.

 i = 1 . 1 + t.

1 + t.1  1 . + i i (5.31) The stability of the numerical time integration scheme can now be investigated by examining the behaviour of this approximation to the exponential. For stability we require ti .

i and .29) raised to the power n.  1 i (5.32) since this term appears in Equation (5.32) is trivially satisfied. The right hand inequality in Equation (5. since t.

and the left hand inequality gives ti  2 1 + t. are all positive.

2.i or ti 1 .

33) is that the scheme is unconditionally stable if 1 2  .33) A consequence of Equation (5.  2 (5.

.  1.

106 For .

T RANSIENT H EAT C ONDUCTION 1 2 the stability criterion is ti If the exponential approximation given by Equation (5. 2.31) is negative for any i the solution will contain components which change sign with each time step n. This oscillatory noise can be avoided by choosing 1 .

1 . .34) t 1 . 2 (5.

is tolerated. Alternatively the oscillatory noise can be filtered out by averaging. but in practice this imposes a limit which is too severe for t and a small amount of oscillatory noise. These theoretical results are explored numerically with a Crank-Nicolson-Galerkin scheme 1 (. associated with the high frequency vibration modes of the system.35) where max is the largest eigenvalue in the matrix . max (5.

Decreasing x from 0:25 to 0:1 with linear elements produces more oscillation because the system has moredegrees of freedom and leads to greater oscillation.3). t = 0.36) u x. u 1. where the one-dimensional diffusion equation A @u = D @ 2 u @t @x2 subject to initial conditions and boundary conditions on 0x1 (5. t = x +  n=1 n 2 2 (5. With this value of t (0:01 s) the numerical results agree well with the exact solution (top.37) 5.4 Mass lumping A technique known as mass lumping is sometimes used in which the mass matrix is replaced by a diagonal matrix having diagonal terms equal to the row sums.3. 0 = 0 u 0. At a sufficiently small t the oscillations are negligible (bottom right.n  t sin nx u x. = 2 ) in Figure 5. t = 1 is solved for various time increments (t) and element lengths (x) for both linear and cubic Hermite elements.3) given by 1 2 X . For example. Figure 5.1n e. Figure 5. consider the mass M .

4 M ASS LUMPING 107 ux. The top graph shows the exact and approximate solutions as functions of x at various times. . t x x Exact solution Linear CNG with x = 0:1.3: Analytical and numerical solutions of the transient 1D heat equation showing the effects of element size x and time step size t. t = 0:01 1:0 0:5 t=1 0 u t = 0:02 t = 0:1 x x x x x x x x t = 0:05 x x x x x x x x x x x x 1 u x (b) x = 0:75 x x x x x x x x (c) x = 0:75 x x x x x x x t : x x x x linear elements 0 x u  = 0 25  =01 x t : : cubic elements 1 t 0 u  = 0 25  =01 : 1 t x (d) x = 0:9 x x x x x x x x (e) x = 0:9 x x x x x x x x 0  =01  =01 x : t : linear elements x t x linear elements 1 0  =01  = 0 01 x : t : 1 t F IGURE 5.5. The lower graphs show the solution through time at the specified x positions and with various choices of x and t as indicated.

108 T RANSIENT H EAT C ONDUCTION matrix ((5. 12 1 .6)). 2 2 12 ZZ M13 = ZZ 1 2 1 1 . 22 12 = 3  1 = 1 and similarly M33 and M44 .23)) for a bilinear element (see Figure 1. . M11 = M22 = M12 = M14 = ZZ ZZ ZZ 1 .9 and (1.

1 . 1 . 1 2 1 . 1 . = . 1 1 . 1212 = . 212 = 18 and similarly M34 and M24 .! 60 7 60 5 4 mass lumping 0 0 0 1 7 4 0 07 1 05 0 4 0 0 0 1 4 3 The element mass is effectively lumped at the element vertices. 1 1 1 . 3 9 1 1 1 1 1 . Such a scheme has computational advantages when .32 3 0 1 1= 31 = 1 3 9 21 9 6 18 61 therefore M = 4 1 18 1 36 1 18 1 9 1 136 1 18 1 18 1 36 1 9 1 18 1 36 1 18 1 18 1 9 3 21 4 7 . 3  3 = 18 2 1 1 .31 3 0 1 . 212 = 36 and similarly M23 . 12 2 1 .

To solve for the transient heat flow in a plate run CMISS example 331 2. the finite element and finite differences (lumped f. reproduces the amplitude of the pulse very well but shows a slight phase lag. mass matrix) solutions of the one-dimensional advection-diffusion equation (5. 5. D = 0:1 m2 s. is only conditionally stable (see (5. using the Crank-Nicolson-Galerkin technique.1 . = 0 in Equation (5.e. To investigate the stability of time integration schemes run CMISS examples 3321 and 3322. or lumped mass.34)) and suffers from phase lag errors . however.38) The finite element solution. V t2 M e.see below.26) because each component of the vector n+1 is obtained directly without the need to solve a set of coupled equations. 4Dt u x. This explicit time integration scheme. In Figure 5.20) with V = 5 mps. The finite difference. . solution also using centered time differences. f = 0 are compared for the propogation and dispersion of an initial unit mass pulse at x = 0. The exact solution is a Gaussian distribution whose variance increases with time: u x .5 CMISS Examples 1. t = p 4t (5. For evenly spaced elements the finite element scheme with mass lumping is equivalent to finite differences with central spatial differences. The length of the solution domain is sufficient to avoid reflected end effects.4. shows excellent amplitude and phase characteristics when compared with the exact solution.

1.1 D = 0:1 m2 s.01 for t  0:01 s. t = 0:001 s for 0 t 0:01 s and t = 0. 0:2 s and 0:4 s) and finite difference solutions (at t=0:4 s only) are compared with the exact solution. 0:05 s. x= 0. The finite element solutions (at t=0:01 s. t t = 0:01 s t = 0:05 s x V = 5 m s.5 CMISS E XAMPLES 109 u x.5.4: Advection-diffusion of a unit mass pulse.1 Exact solution x x Finite element solution o o Finite difference solution x x x x x x x t = 0:2 s x x x x t = 0:4 s ox o o x o x oxx x o x o x ox x o x x F IGURE 5. .

.

nodal point accelerations) M . That is.1) successive time steps. This transformation (which is still applicable when damping is P M K .1) can result in the matrices of the transformed system (6. In direct time integration methods u t and _ u t are replaced by finite differences and the resulting system of algebraic equations is solved at f u t = Px t _ P T MP x t + P T CP x t + P T KPx t = P T f _ M u t + C u t + Ku t = f t (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 equations can either be integrated directly . can be chosen to diagonalize and and thereby uncouple the equations entirely. C and K are the mass. if a solution is required over a long time period.. the time-dependent solution is expressed as the superposition of the natural (or resonant) modes of the system. if damping is neglected. a suitable transformation (6. 6.Chapter 6 Modal Analysis 6.3) having a much smaller bandwidth than in the original system and hence being more economical to solve.g. In fact. respectively.or analysed by mode superposition.3) which includes second order time derivatives (e. f t is the external load vector and u t is the vector of n nodal unknowns. To find these modes requires the solution of an eigenvalue problem. damping and stiffness matrices. For a small number of steps this is the most economical method of solution but. or for a large number of different load vectors t.as in the last section for parabolic equations .2 Free Vibration Modes Consider an extension of Equation (6.2) applied to Equation (6.

(6.7) as s I is the identity matrix. Substituting Equation (6.2) required to reduce the original system of equations (6.4) of the form (6.rewriting Equation (6. .7) S s s where (the eigenvectors are said to be M -orthonormalised). n .6) 2 2 2 having n eigensolutons !1 s1  . !2 s2  . 2 . !n sn . Thus the modal matrix .11) (i. : : : .6) becomes KS = MS 2!2 6 1 !22 =6 6 4 0 0 ..5) when the original partial differential operator is self-adjoint) the eigenvectors are orthogonal and can be “normalized” such that Ks = !2Ms (6. : : : . Combining the n eigenvectors into a matrix = 1.10) or ST KS = ST MS  = I  =  P (6..8) (6.e. (6. 2 !n ST MS = I (6. If K is a symmetric matrix (as is the case siT Msj = 1 i=j 0 i 6= j (6.5) into Equation (6.1) to the canonical form K _ x t + ST CS x t + x t = S T f t (6.can be used as the transformation matrix in Equation (6.whose columns are the M -orthonormalised eigenvectors of satisfying Equation (6.the modal matrix . t0 .4) where ! and t0 are constants and is a vector of order n.9) where 3 7 7 7 5 (6.6)) .4) gives the generalized eigenproblem s u t = s sin ! t .12) .112 M ODAL A NALYSIS included but does not then result in an uncoupled system unless further simplications are made) is found by solving the free vibration problem M u t + Ku t = 0 Proof: Consider a solution to Equation (6.

7).! To find the solution by modal analysis we first solve the generalised eigenproblem i. K = . 1  1 01  2  1 2  p and sT = . !2 = 0 or !4 .2 0 M = 0 1 . 2!2 . p p . consider the equilibrium equations M u + Ku = f where 2 0  6 . i. 5 with corresponding eigenvectors T = a 1 1 . To find has solutions ! 1 1 the magnitude of the eigenvectors we use Equation (6. The M -orthognormalised eigenvectors are now sT = p 1 2 3 3 6 6 3 21 1 6 p3 .1 2 2 0 ..e. 2.13) where xi is the ith component of and ri is the ith component of the vector this system is given by the Duhamel integral x S T f .1   (Notice that the orthogonality condition is satisfied: ab 1 1 = 0).3 A N A NALYTIC E XAMPLE 113 With damping neglected equation Equation (6. when used as the transformation matrix..6. T = b .12) becomes a system of uncoupled equations xi t + !i2xi t = ri t i = 1. n (6. has a solution when det .1 = 1  b = p a b 0 1 1 0 1 2 3 3 2 0. p6 7 which.2  2 s=0 . s         1 1 2 1 1 2 0 1 = 1  a = p 2 .  d + sin ! t + cos ! d xi t = ! i i i i i i i t 0 where the constants i and i are determined from the initial conditions 1 xi 0jt=0 = siT M u 0jt=0 xi 0jt=0 = xi 0jt=0 = siT M u 0jt=0 siT M u 0jt=0 _ (6.2 4.15) 6. giving the modal matrix S = 4 1 2 5 s p 3 K M  p 6 .14) 1 Z r   sin ! t . 7!2 + 10 = 0. The solution of (6. : : : .3 An Analytic Example As an example.e.1 2 .2 4 and f = 10 Ks = !2Ms 6 . This characteristic polynomial 2 = 2.

t00 : 4 5 p 6 The solution of the non-homogeneous system. 6.4 Proportional Damping When element damping terms are included in the original dynamic equations (6. One simplification often made in order to retain the diago- S .if it coincides then resonance will occur. subject to given initial conditions.16) as the superposition of the natural modes (eigenvectors) of the homogeneous equations.2) with =  1 .16) Notice that the solution is expressed in Equation (6. Very often it is unnecessary to evaluate all n eigenvectors of the system. p 7 1 0 S T MS = 6 1 23 5 0 1 4 13 2 6 5 = 0 1 = I 4 3 . is found by solving the uncoupled equations x t + P S s s 2 0 0 5 2 1 13 2 10 3 p p 7  0  6p 7 x t = 6 1 23 5 10 = 4 203 5 4 3 . p 7 2 0 ST KS = 6 1 23 5 .p p 6 n X i=1 6 p by means of the Duhamel integral (6. where p n.2 4 4 13 2 6 5 = 0 5 =  4 3 p p . 2 . n s r 6 u t = Sx t = sixi t (6.114 M ODAL A NALYSIS reduces the stiffness matrix to 2 1 13 21 13 p p 7  6 .p p 6 6 3 6 21 2 1 13 13 p p 7 2 0 6 p . the higher frequency modes can be ignored and the solution adequately represented by superposition of the p eigenvectors associated with the p lowest eigenvalues.p p 6 6 and the mass matrix to p 3 p 6 Thus the natural modes of the system are given by 213 p u1 t = 6 13 7 sin p2 t . from Equation (6.14) (in this case with constant) and then. t0 4 5 p 3 and 2 13 .2 6 p . If the forcing function (load vector) is close to one of these modes the corresponding coefficient xi will be large and will dominate the response .1) the transformation to a lower bandwidth system is still based on the model matrix but Equation (6.p u2 t = 6 2 6 7 sin p5 t .12) is then not a system of uncoupled equations. : : : .

19) u 6. To analyse a steel-framed building modal analysis run CMISS example 453 .18)).i!it f sin ! t + cos ! tg xi t = ! i i i i i i i t 0 (6.5 CMISS E XAMPLES 115 nal nature of Equation (6.  dt + e. Once the components xi t have been found from Equation (6. 1 Z r   :ei!i t. To analyse a clamped beam modal analysis run CMISS example 452 3. p where ! i = !i 1 . To analyse a plane stress modal analysis run CMISS example 451 2.6. the solution t is expressed as a superposition of the mode shapes i by Equation (6. Equation (6.5 CMISS Examples 1.12) is to approximate the overall energy dissipation of the finite element system with proportional damping siT Csj = 2!ii ij .19) (or alternative time integration methods applied to (6. i2 . 2 xi t + 2!iixi t + !1 x1 t = ri t _ (6.16).  : sin ! t .18) with solution (the Duhamel integral) s i and i are calculated from the initial conditions Equation (6.12) now reduces to n equations of the form (6.17) where i is a modal damping parameter and ij is the Kronecker delta.15).

116 M ODAL A NALYSIS .

The most common difficulty in applying the BEM is in determining an appropriate fundamental solution. . A linear differential equation can be expressed in operator form as Lu = where L is a linear operator. the diffusion equation and the wave equation (Brebbia.  +  = 0 (7.Chapter 7 Domain Integrals in the BEM 7.1) where * indicates the adjoint of the operator L and is the Dirac delta function. In particular. The fundamental solution for this equation is a solution of L! x. However. If either of these requirements cannot be satisfied then a boundary integral formulation cannot be achieved. Fundamental solutions are known and have been published for many of the most important equations in engineering such as Laplace’s equation. The mathematical theory required to determine the fundamental solution of a constant coefficient PDE is well-developed and has been used successfully to determine the fundamental solutions for a wide range of constant coefficient equations (Brebbia & Walker 1980) (Clements & Rizzo 1978) (Ortner 1987).1 Achieving a Boundary Integral Formulation The principal advantage of the BEM over other numerical methods is the ability to reduce the problem dimension by one. have known fundamental solutions. This reduction of dimension also eases the burden on the engineer as it is only necessary to construct a boundary mesh to implement the BEM. To achieve a boundary integral formulation it is necessary to find an appropriate reciprocity relationship for the problem and to determine an appropriate fundamental solution. PDEs with variable coefficients do not. in general. by no means can it be guaranteed that the fundamental solution to a specific differential equation is known. This property is advantagous as it reduces the size of the solution system leading to improved computational efficiency. Telles & Wrobel 1984a). Domain integrals will also arise for inhomogeneous equations. is an inhomogeneous source term and u is the dependent variable. No specific 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. To achieve this reduction of dimension it is necessary to formulate the governing equation as a boundary integral equation. If the fundamental solution to an operator cannot be found then domain integrals cannot be completely removed from the integral formulation.

the resulting BEM formulation will be Hu . Another advantage of the BEM over other numerical methods is that it allows an explicit expression for the solution at an internal point. Consider the Poisson equation r2 u = . He argued that for problems such as flow problems a wide range of phenomena are described by the same governing equations. requires a domain discretisation and a numerical domain integration procedure which reduces the attraction of the BEM over domain-based numerical methods. Z !d (7. What distinguishes these phenomena is the boundary conditions of the problem. a body force term. An inhomogeneous term may arise due to a number of factors including a source term. However. Applying the BEM gives an equation of the form of . This approach. 7.2 Removing Domain Integrals due to Inhomogeneous Terms Inhomogeneous PDEs occur for a large number of physical problems. Gq = . 7. however. An inhomogeneous linear PDE can be expressed in operator form as Lu = where is a known function of position or a non-zero constant. The BEM generates a formulation involving both the dependent variable u and the flux q. For this reason accurate description of the boundary conditions is vital for solution accuracy. In this section these methods will be discussed. This allows flux boundary conditions to be applied directly which cannot be achieved in either the finite element or finite difference methods. If the fundamental solution is known for the operator L. 1984a).even in cases where domain integration is required. Several methods have been developed which allow domain integrals to be expressed as equivalent boundary integrals. or due to initial conditions in time-dependent problems.1 The Galerkin Vector technique For some particular forms of the inhomogeneous function the domain integral can be transformed directly into boundary integrals. Domain integration can be simply and accurately performed in the BEM. This method is simple and has been shown to produce accurate results (Brebbia et al. As the domain integral does not involve any unknown values accurate results can generally be achieved using a fairly coarse mesh. This requires discretising the domain into internal cells in much the same way as for the finite element method. This allows a problem to be subdivided into a number of zones for which the BEM can be applied individually.2) The domain integral in this formulation does not involve any unknowns so domain integration can be used directly to solve this equation.2. This zoning approach is suited to problems with significantly different length scales or different properties in different areas. the presence 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.118 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 .

2 R EMOVING D OMAIN I NTEGRALS DUE TO I NHOMOGENEOUS T ERMS 119 Equation (7.7. vr2 d Z . Using Green’s second identity Z .2).  r2 v .

where ! is the fundamental solution of Laplace’s equation. then for the specific case of being harmonic (r2 = 0) Green’s second identity can be reduced to domain integration can be avoided for certain forms of . .3) r2v = !. If a v can be found which satisfies Z Z . (7. v @n d. @v @ = @n .

Fairweather. Shippy & Wu (1979) determined the Galerkin vector for the two-dimensional Poisson equation and Monaco & Rangogni (1982) determined the Galerkin vector for the threedimensional Poisson equation. a domain integral I is approximated as N A X f x . In each of these cases the domain integral can be expressed as equivalent boundary integrals. it only applies to specific forms of the inhomogeneous term (i.2 The Monte Carlo method Domain discretisation could be avoided by using a Monte Carlo technique. v @n d. Specifically. This Galerkin vector approach provides a simple method of expressing domain integrals as equivalent boundary integrals. He considered the practical cases where the body force term arose due to either a constant gravitational load. Unfortunately this method often proves to be computationally expensive as a large number of integration points are needed for accurate domain integration.5) where f xi .. y  IN i i i=1 (7. in two dimensions. Gipson has successfully applied this method to a number of Poisson-type problems. . . as this method removes the burden of preparing a domain mesh. This technique approximates a domain integral as a sum of the integrand at a number of random points. The method was proposed by Gipson (1987). Gipson argues however that. yi  is the value of the integrand at random point xi . Unfortunately. Danson (1981) showed how this method can be applied successfully for a number of physical problems involving linear isotropic problems with body forces.2) can be expressed as equivalent boundary integrals. N is the number of random points used and A is the area of the region over which the integration is performed. This method has the secondary advantage of allowing the integration to be performed over a simple geometry enclosing the problem domain . @v @ !d = @n .e. yi .2. 7. is required to be harmonic).if a random point is not in the problem domain its contribution is ignored. This approximation allows a domain integral to be approximated by a summation over a set of random points so domain integration can be performed without requiring a domain mesh. rotation about a fixed axis or steady-state thermal loading.4) Therefore if a Galerkin vector can be found and is harmonic the domain integral in Equation (7. Rizzo. (7.

Gq = Hup . Instead. Zheng. For many operators the fundamental solution to the operator L may be unobtainable or may be in an unusable form.8) ^ where ! is the fundamental solution corresponding to the operator L. a BEM formulation can be derived based on a related operator L ^ will with known fundamental solution. .reducing the problem to (7. Alternatively an approximate particular solution could be calculated numerically. This approach can be applied in a situation where an analytic expression for a particular solution can be found.2.6) If a particular solution up can be found. Z ^ L . 7. Gqp Hu . A BEM formulation for Lu = 0 based on the operator L be of the form Hu . This linear system can be solved by applying boundary conditions. Gq = . However in this case the domain integral term involves the dependent variable u. d 7.6) are known . u can be considered as the sum of the complementary function uc. This integral equation is similar to Equation (7.3 Domain Integrals Involving the Dependent Variable Consider the linear homogeneous PDE Lu = 0. Unfortunately particular solutions are generally only known for simple operators and for simple forms of .120 D OMAIN I NTEGRALS IN THE BEM the extra computational expense is justified. which is a solution of the homogeneous equation Luc = 0. Consider the linear problem Lu = .7) where is a vector of known values. Gq = d (7. L u! d (7. Applying BEM to the governing equation using the expansion u = uc + up gives Hu .3 Complementary Function-Particular Integral method A more general approach can be developed using particular solutions. This approach is a special case of a more general method known as the dual reciprocity boundary element method. Coleman & Phan-Thien (1991) proposed a method where a particular solution is determined by approximating the inhomogeneous source term using a global interpolation function. all values on the right-hand-side of Equation (7.2). This is especially likely if L involves variable coefficients for which case it has been shown that it is particularly difficult to find ^ a fundamental solution. and a particular solution up which satisfies Lup = but is not required to satisfy the boundary conditions of the problem. This problem could be solved using domain integration where the internal nodes are treated as formal problem unknowns.

13) A solution will only exist for all values of " if the terms at each power of " equal zero.7. y = 0 . Rangogni reported that in practice this r2 u + f x.3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 121 7. u0 can be found by solving r2 u0 = 0 which Rangogni assumes will satisfy the boundary conditions of the original problem.3. y u = 0 for which he sought a solution of the form 2 where 0  "  1 (7.  (7. y  Equation (7.13) to be treated as an infinite series of distinct problems which can be solved using the boundary element method.9) can be recast as a heterogeneous r2 u + f x. Rangogni treated this equation as a perturbation about Laplace’s equation. y @u + g x.11) u = u0 + "u1 + " u2 + : : : = 1 X j =0 uj "j (7. He considered the class of equations r2 u + "f x. y u = 0 (7. Equation (7.boundary element method to the general second-order variable coefficient PDE solution to Equation (7. Helmholtz equation (7. Each successive uj can then be found by solving a Poisson equation with homogeneous boundary conditions as uj .10) where f is a known function of position.12) Substituting Equation (7.12) into Equation (7.10) is therefore given by 1 X uj . y @u = h x.1 has been previously determined. The j =0 series converged rapidly and in his numerical examples he achieved accurate results using only u0 and u1 . This allows Equation (7. y @x @y (7.11) and grouping powers of " gives r2 u0 + " r2u1 + fu0 + "2 r2 u2 + fu1 + : : : = 0 .14) .10) is a particular member of this family of equations for which " = 1.1 The Perturbation Boundary Element Method Rangogni (1986) proposed solving variable coefficient PDEs by coupling the boundary element method with a perturbation method. He considered the two-dimensional generalised Laplace equation r   x.9) 1 Using the substitution V x.  . y rV x. Rangogni (1991) extended this coupled perturbation . y  =  2 u x. Rangogni used a domain discretisation to solve these Poisson problems.

gives x Z @!0 Z c  u  + u @n d. Instead of using one higher-order fundamental solution.2 The Multiple Reciprocity Method The multiple reciprocity method (MRM) was initially proposed by Nowak (1987) for the solution of transient heat conduction problems.15) Applying the perturbation method to this family of equations allows Equation (7. y 0  "  1 @x @y   (7. @u !0 @n d. particularly if convergence is not rapid. Gipson.15) to be expressed as an infinite series of distinct Poisson equations which can be solved using the boundary element method. This process could become computationally expensive.16) where b0 = b0   is a known function of position. Lafe & Cheng (1987) used the perturbation method to solve steady-state groundwater flow problems in heterogeneous aquifers. Rangogni found that in practice convergence was rapid and accurate results were produced. (7. + b0 !0 d . Again Rangogni used an domain mesh to solve these Poisson equations. They showed the method produced accurate results for simply varying hydraulic conductivities with convergence after two or three terms. y @u + g x. Reible & Savant (1987) considered a class of hyperbolic and elliptic problems which can be transformed into an inhomogeneous Helmholtz equation. using the fundamental solution to the Laplace operator. as the solution of multiple subproblems will be required within each subregion. They found that for rapidly varying hydraulic conductivity convergence is not guaranteed. the Galerkin vector. Lafe & Cheng investigated the convergence of the perturbation method. To avoid domain discretisation the domain integral in Equation (7.17) needs to be expressed as equivalent  . 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. to convert the remaining domain integrals to equivalent boundary integrals a series of higher-order fundamental solutions is used. Applying BEM to this equation. They avoided domain discretisation by using a Monte Carlo integration technique (Gipson 1987) to evaluate the required domain integrals. Since then the method has been successfully applied to a wide range of problems. The MRM can be viewed as a generalisation of the Galerkin vector approach.17) where !0 is the known fundamental solution to Laplace’s equation applied at point .3.122 D OMAIN I NTEGRALS IN THE BEM He considered the family of equations 2 r u + " f x. If the hydraulic conductivity variation is more significant they recommend using the perturbation method in conjunction with a subregion technique so that the variation of conductivity within each subregion satisfies their requirements. They used the perturbation method to recast this as an infinite sequence of Poisson equations. Consider the Poisson equation r2u = b0 (7. y @u = h x. = Z . 7.

7.3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 123 boundary integrals.19) or Z Z Z .18) Using this higher-order fundamental solution the domain integral in Equation (7.17) can be written as Z b0 !0 d = Z b0 r2 !1 d (7. Using MRM this is achieved by defining a higher-order fundamental solution !1 such that r2!1 = !0 (7.

25) Using these recurrence relationships gives the boundary integral formulation Z .20) This formulation has generated a new domain integral. b1 = r2 b0 (7. 1. This procedure is based on the recurrence relationships (7. + !1r2 b0 d .24) (7. 1. 2. b0 is a known function so we can introduce a new function b1 which can be determined analytically from the relationship (7.21) giving Z !1 r b0 d = 2 Z !1b1 d (7. 2.22) This process can be repeated by introducing a new higher-order fundamental solution !2 such that r2!2 = !1 and continuing until convergence is reached.23) bj+1 = r2 bj r2 !j+1 = !j for j for j = 0. @!1 @u b0 !0 d = u @n . : : : = 0. !1 @n d. : : : (7.

@!0 @u 1 X Z .

!0 @n d. . + bj @n . (7. Errors are only introduced at the stage of boundary discretisation.26) which is an exact formulation if the infinite series converges. !j+1 @n j =0 . = 0 c  u  + u @n . . @!j+1 @bj d.

G0q = H G p r 1 X j =0 HJ +1pj .Equations (7. 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.25) . Brebbia & Nowak (1989) have applied the MRM to the Helmholtz equation r2 u + 2 u = 0 where b0 = .can be employed. The MRM will be restricted to cases where the recurrence relationships .27) where J +1 and J +1 are influence coefficient matrices corresponding to the higher-order fundamental solutions and j and j contain the nodal values of bj and its normal derivative. The MRM can be extended to other equations by allowing the forcing function b0 to be a general function such that b0 = b0  .24) becomes simply x uj+1 = r2 uj = .2 u and the recurrence relationship defined by Equation (7. Itagaki & Brebbia (1993) have determined the higher order fundamental solutions for the two-dimensional modified Helmholtz equation.124 D OMAIN I NTEGRALS IN THE BEM Introducing interpolattion functions and discretising the boundary gives the matrix system H0u . . u.24) and (7.28) c  u  + 1 XZ j =0 . The MRM can be applied based on operators other than the Laplace operator. GJ +1rj  (7.2j u In this case the boundary integral formulation will be (7. t.

30) The forcing function b can be completely general. !j @n 2j (7. Brebbia & Wrobel 1992) and in this chapter the DR-BEM will be illustrated for this choice. For potential problems b = b  . Instead the DR-BEM is applied using an appropriate related operator with known fundamental solution.3 The Dual Reciprocity Boundary Element Method Equation Derivation The dual reciprocity boundary element method (DR-BEM) was developed to avoid the need for domain integration in cases where the fundamental solution of the governing differential equation is unknown or is impractical to apply. @! @u d. The most common choice is the Laplace operator (Partridge.29) 7. Consider a second-order PDE which can be expressed in the form r2u = b (7.30) will x x x . t. If b = b   then b is a known function of position and the differential equation described is simply the Poisson equation. = 0 j  u @n . u.3. Applying the BEM to Equation (7. u and for transient problems b = b  .

34) is multiplied by a weighting function ! and integrated over the domain.33) By substituting Equations (7.33). : : : .31) where ! is the known fundamental solution to Laplace’s equation.generally most. b! d Z (7. M ^ M X j =1 (7. Equation (7. j are unknown coefficients and fj are approximating functions used in the interpolation and are generally chosen to be functions of the source point and the field point of the fundamental solution.3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 125 give Hu .33) into Equation (7. As discussed in the previous chapter the solution to a linear PDE Lu = can be constructed as the sum of a complimentary function uc (which satisfies the homogeneous equation Luc = 0) and a particular solution up to the equation Lup = . Gq = .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. After applying these steps Equation (7. the DR-BEM employs a series of particular solutions uj which are ^ related to the approximating functions fj as shown in Equation (7.35) is obtained.7.34) the standard boundary element approach can be applied. The aim of the DR-BEM is to express the domain integral due to the forcing function b as equivalent boundary integrals. Green’s theorem is applied twice and the fundamental solution of the Laplacian is used to remove the remaining domain integrals. where the fundamental solution pole is applied at .called poles .32) and (7.34). which may be difficult to determine. r2 uj = fj j = 1. The DR-BEM uses the idea of approximating b using interpolation functions. but not all. The name dual reciprocity BEM is derived from the application of reciprocity relationships to both sides of Equation (7.30) the forcing function b is approximated by a weighted summation of particular solutions to the Poisson equation. A global approximation to b of the form b= M X j =1 j fj (7. The approximating functions fj are applied at M different collocation points .32) is proposed. of which are located on the boundary of the problem domain. r2u = ^ j r uj 2 (7. Instead of using a single particular solution.

Z .126 D OMAIN I NTEGRALS IN THE BEM point .

. @! @u c  u  + u @n . 0 Z . ! @n d.

However ^ by using the same interpolation functions to approximate u and uj the numerical implementation ^ will generate the same matrices and on both sides of Equation (7. It is not necessary to use interpolation functions to approximate each uj . 1992). 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). or where these points should be positioned within the problem domain.38) . This allows Equation (7. can be expressed in matrix form as = .it is only necessary to specify the coordinates of the internal collocation points. ! @n d. Gq = H U . is defined by Equation (7.33) once the approximating functions fj have ^ been defined. Using internal poles in this interpolation does not require domain discretisation .36) where the M poles were chosen to be the N boundary nodes plus I internal points so that M = N + I . To solve this system it is necessary to evaluate .38) which provides an explicit expression for . 1992) (Huang & Cruse 1993) using boundary points only in this procedure is insufficient to define the problem.35). The error generated by approximating each uj in this manner has been found to be small and can be justified by the ^ improved computational efficiency of the method (Partridge et al. The ^ j and ^j vectors can be treated as columns of the matrices ^ and ^ respectively. No theory has been developed of how many internal collocation points should be used for optimal accuracy. F = F .37) b F where is a vector containing the nodal values of . @! @hatuj 1 M X @ = uj @n .36) to be rewritten as u q U Q ^ ^ Hu . If the matrix is nonsingular this expression can be rearranged to give Equation (7. In general using internal points is likely to improve the solution accuracy as it increases the number of degrees of freedom.35) In implementing a numerical solution of this equation similar steps are taken as for the standard BEM. The application of this method results in the system H G Hu . Gqj  ^ ^ (7. for the nodal values.32) which.1b (7.  (7. Gq = N +I j =1 X j H uj .A ^ j c  uj  + j =1 . It has been shown that for many problems (Partridge et al. The boundary is discretised into elements and interpolation functions are introduced to approximate the dependent variable within each element. The internal points can be chosen to be locations where the solution is of interest. The form of each uj is known from Equation (7. GQ (7.

where r is the Euclidean distance between the field point and the source point of the fundamental solution. Several other options for fj have been tried (Partridge et al. Some work has been conducted into investigating what form of fj should be used in a given situation to provide the highest accuracy and computational efficiency. fj = fj r . to specify u ^ F and q .39) The approach taken to solve this equation will depend on the form of b. The only requirement so far prescribed on the form of the approximating functions fj is that the matrix generated should be nonsingular and that the related particular solutions uj can be determined and can be expressed in a practical ^ closed form. Therefore the choice of the approximating functions fj is a key consideration when implementing the DR-BEM. They showed that accurate results can be achieved using some combination of terms from this series. ^ ^ for two-dimensional . GQ F ..1b in Equation (7.42) u and q. For both two and three-dimensional problems Wrobel. 21 ln r in two-dimensional  fj = 1 + rj + rj2 + : : : + rjm (7. ^ ^ For two-dimensional problems. Generally the same approximating function fj is used at all the collocation points so in this thesis.  = 4r in three-dimensional space . Due to the dependence of this fundamental solution only on r the approximating function is generally chosen to be some radial function i. Gq = H U . Choosing f to be a function of only one variable simplifies the process of determining u and q .7.e.32)). if f = f r  then the relationship r2 u = f r ^ can be reduced to the ordinary differential equation (7.33). The Approximating Function f The accuracy of the DR-BEM hinges on the accuracy of the global approximation to the forcing function b (defined by Equation (7.40) where rj is the distance between the field point (node j ) and the DR-BEM collocation point (node i).41) Using f d2u + 1 du = f ^ ^ 2 dr r dr defined by Equation (7.37) gives (7.3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 127 Including this explicit expression for ^ ^ Hu . 1992) but it has been found that in general the most accurate results were generated using some radial function. the form of approximating functions fj will be referred to by a single f . applying Equation (7. The fundamental solution of Laplace’s equation is !  ^ space and !  . Usually a form of fj is defined and this can be used. Brebbia & Nardini (1986) recommended choosing fj from the series x x 1 x.  = . for simplicity.40) the corresponding forms of (7.

128 D OMAIN I NTEGRALS IN THE BEM problems. can be shown to be 2 3 m+2 u = r4 + r9 + : : : + r 2 ^ m .

@x @y .

40) was recommended as a basis for the approximating function f due to the particular form of the fundamental solution of Laplace’s equation and its dependence on r only. The theory of how to determine the best approximating function is therefore a vital component of the DR-BEM. It has been shown that f = r is optimal for three-dimensional problems which justifies the use of f = 1 + r when applying the DR-BEM to three-dimensional problems .40) can be used for specifying f . They also demonstrated that f = 1 + r is a specific member of the group of radial basis functions.3. 1+ 2r rm q = rx @n + ry @n 2 + 3 + : : : + m + 2 ^ where rx = xj . It has been found that in general including higher-order terms leads to little improvement in accuracy (Partridge et al. However for two-dimensional problems it has been shown that optimal approximation is attained using the thin plate spline f = r 2 logr .the constant is included to ensure a non-zero diagonal for . This observation lead Golberg & Chen to suggest that choosing f to be a thin plate spline may improve the accuracy of the DR-BEM in two dimensions. Therefore. rather than being an arbitrary choice. Golberg & Chen (1994) argued that a formal analysis of the approximating function f can be undertaken using the theory of radial basis functions. (7. Unfortunately the approximating function has generally been chosen and used in a rather ad-hoc manner.39) can be rewritten b F F . Equation (7. In this case it is not necessary to invert the matrix as can simply be calculated from = using Gaussian elimination. The most commonly used form is f = 1 + r as this approximation will generally give accurate results with greater computational efficiency than other choices.43) (7. Cubic splines are known to be optimal for one-dimensional interpolation. choosing f to be a radial basis function ensures convergence of the DR-BEM. for the Poisson equation.3. F Inhomogeneous Equations If the forcing function b is a function of position only then the differential equation under consideration is simply Poisson’s equation. The performance of the DR-BEM hinges on the choice of the approximating function f . The choice of f in this case will be discussed in Section 7. it seems that choosing f to be a radial function is a logical extension for two or three-dimensional problems. yi . Golberg & Chen showed that. xi and ry = yj . Radial basis functions are a generalisation of cubic splines in multi-dimensions. 1992). If a different operator is used as the basis of the DR-BEM then it is likely a different form of f will be more appropriate.44) Any combination of terms from Equation (7. Recently Golberg (1995) has published a review of the DR-BEM concentrating on developments since 1990 concerning the numerical evaluation of particular solutions. The theory of using radial basis functions for multi-dimensional approximation is fairly advanced. Equation (7. Recently some more formal analysis of the use of approximating functions has been undertaken.

(1991) and Coleman. by applying boundary conditions Equation (7. Zheng et al.45) By applying boundary conditions Equation (7. GQ F F (7. (1991) used a least-squares method where they minimised the sum of squares S= M X m=1 b rm  . 1992). Applying the DR-BEM to Equation (7.3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 129 as ^ ^ Hu . Instead the solution can be treated as a coupled problem and the solutions at boundary and internal nodes are generated simultaneously. gives F . Due to the presence of the fully-populated . Dassios & Beskos (1994) this method is essentially equivalent to the DR-BEM. based on the fundamental solution to Laplace’s equation.1u (7. for example.49) it is not possible to solve the boundary problem and internal problem separately. N X j =1 j fj rm  ! (7. for example. (1991) successfully solved inhomogeneous potential and elasticity problems which are governed by operators other than the Laplacian.1 matrix in Equation (7. As discussed by Polyzos.7.49) Again. Consider. Zheng et al. GQ F . and Coleman et al. Derivative Terms The DR-BEM can also be applied for elliptic problems where b involves derivatives of the dependent variable (Partridge et al.49) can be reduced to a linear system = which can be solved to determine the unknown nodal values.45) can be reduced to a linear system = which can be solved to give the unknown nodal values of u and q .1 . Coleman et al. the differ- AX . Elliptic Problems If b is a function of the dependent variable then will also be a function of the dependent variable.47). the linear second-order differential equation r2 u + u = 0 (7. Gq = d where d = H U .u ^ ^ Hu . For large systems they found the computational efficiency could be improved by employing the conjugate gradient method. GQ Ax (7.48) which can be rearranged to give ^ ^ . suggested several alternative ways of determining the unknown coefficients j for inhomogeneous equations.46) using singular value decomposition. Consider.47) In this case b = . However. Gq = . Zheng et al. Tullock & Phan-Thien (1991) have proposed a method which uses a global shape function to construct an approximate particular solution.u so = .1 H + S  u = Gq where S = H U . H U .

In system form this can be expressed as as only one matrix inversion procedure is required.51) To solve this problem it is necessary to relate the nodal values of u to the nodal values of @x . evaluating .53) to give an explicit expression for tion (7.1 @X F . As mentioned earlier.1 (7. the approximating function f is generally chosen to be f = 1 + r . equating  to F improves the computational efficiency of the method F @u = @  @x @X (7.39) can now be rewritten as (7. gives (7.55) H + R u = Gq where ^ ^ @F R = H U .54) allows Equa- Choosing  = and inverting Equation (7.50) ^ ^ Hu . GQ F .52) can be used to approximate u where j are the chosen interpolation functions and j are the unknown coefficients. As shown in Equation (7.55) derivative terms require derivatives of f to be evaluated. H U .32). This is achieved by using interpolation functions to approximate in a similar manner as was used to approximate b in Equation (7.53) Although it is not necessary.56) By applying boundary conditions Equation (7. For example. y  j (7. GQ F .54) to be rewritten as @u = @F F . using the Laplace fundamental solution.1 @u @x u (7.53) gives u= (7.56) can be reduced to a linear system which can be solved to give the unknown nodal values.1u @x @X Equation (7. A global approximation function of the form @u u= M X j =1 j x.130 D OMAIN I NTEGRALS IN THE BEM ential equation r2u + @u = 0 @x In this case applying DR-BEM. Gq = . Differentiating Equation (7. This can lead to numerical problems if derivative terms are included in the forcing function b.

u. y i. yM  0 0 3 7 7 7 5 (7. Zhang recommended the adoption of these approximating functions for all use of the DR-BEM.53) to allow nodal values of u 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.3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 131 the @F matrix requires calculation of @f . Using the approximating function f = 1 + r gives @x @X @f = @f @r = @f rx @x @r @x @r r (7. . 0 0  . . .58) where  is a function of position .as shown by Zhang (1993) .59) where is a vector of the nodal values of the forcing function b.e. GQ F . Zhang (1993) suggested two possibilities for avoiding this problem. b b 2 x .significant numerical error may result..60) where M is the number of collocation points used in applying the DR-BEM.  xM . =  x.. If the DR-BEM is applied using the known fundamental solution to the Laplace operator then the forcing function is b = . y  in two dimensions. . . This method was shown to produce accurate results but is somewhat cumbersome and can only be applied to linear problems.. Consider the variable coefficient Helmholtz equation r2u +  x u = 0 (7. 4 . The first suggestion involved using a mapping procedure to map the governing equation to an equation without convective terms. y2    K = 6 .. Zhang recommended use of either f = 1 + r3 or f = 1 + r2 + r3. y  0    1 6 0 1  x2. so . where is a diagonal matrix containing the nodal values of  x. Variable Coefficients The DR-BEM can be readily extended to equations with variable coefficients. 6 . Appropriate approximating functions need to be chosen to avoid the problem of singularities. Applying the DR-BEM gives ^ ^ Hu . The relationship b = .u can be written in matrix form as = .1b Ku K (7.57) This derivative function can become singular.7. These approximating functions produce accurate results and can be simply applied for both linear and nonlinear problems. . A simpler approach is to choose an approximating function which does not lead to singularities for convective terms. Gq = H U . The same idea of using Equation (7. This will especially be the case in problems where collocation points are located close together.

61) which is a boundary-only expression for the variable coefficient Helmholtz equation. R u = Gq + Sn where (7. If an appropriate operator can be found the complexity of the forcing function b can be reduced. GQ F . This should improve the accuracy of the method. y @u + n x.59) can be rearranged to give ^ ^ H + SK  u = Gq whereS = H U . y u + l x. GQ F .1 (7. l and m respectively. The problem with applying the DRBEM based on another operator is in choosing the approximating function f .62) Consider a two-dimensional equation of the form r2 u x. ^ ^ ^ S = H U . y @x @y Applying the DR-BEM to this equation gives a matrix system of the form (7.132 D OMAIN I NTEGRALS IN THE BEM Using this matrix expression for b Equation (7. y @u + m x. The DR-BEM Using Other Operators The DR-BEM has been presented in this chapter based on the Laplace operator. However the DRBEM can be be applied using essentially any operator of appropriate order with known fundamental solution. The DR-BEM can be applied in cases where b involves a sum of terms due to the basic property Z b1 + b2  d = Z b1 d + Z b2 d (7. L and M are diagonal matrices where the diagonals contain the nodal values of k.1  . n is a vector containing the nodal values of n.63) H .64) K . y = k x. This method is general and can easily be extended to accommodate variable coefficient derivative terms and a sum of variable coefficient terms. A choice of f which produces accurate results is required but it is also necessary to choose an f for which a particular solution u can be determined. Formulating the DR-BEM for a General Elliptic Problem In this section it has been shown how the DR-BEM can be applied for elliptic problems with varying forms of b.

66) .1 R = S K + L @X + M @Y (7. @F @F F .65) (7.

67) Radial functions have generally been used when applying the DR-BEM. vx . This analysis requires the determination of a particular solution u which satisfies ^ (7. Zhu chose an approximating function of the form f = r m where m is a positive integer. Partridge et al. t (7. vx @u .71) to determine the corresponding approximating function. vx @x .71) Instead of defining a form of the approximating function f and solving for u Partridge et al. (1986). vx @x .7. r2 u + 2 u = b x. y. vy and k are all assumed to be homogeneous. vy @ u . They applied the DR-BEM based on the steady-state convection-diffusion operator (7. chose ^ to define u and use Equation (7. u. ku = 0 @x @y which has a known fundamental solution. Along the lines of Wrobel et al.68) which can be achieved using a variation of coefficients method. vy @u . ku = f ^ @y (7. Determining the particular solution u requires solving the ordinary differential equation ^ d2u + 1 du + 2u = rm ^ ^ 2 dr r dr @u Dr2u .69) where the material parameters D .70) @u ^ ^ ^ Dr2u . (1992) applied the DR-BEM to the transient convection diffusion equation (7. .3 D OMAIN I NTEGRALS I NVOLVING THE D EPENDENT VARIABLE 133 Zhu (1993) has determined the particular solutions necessary for applying the DR-BEM based on the two-dimensional Helmholtz operator. vy @u . Al^ though somewhat ad-hoc this approach was found to produce accurate results. ku = @u @y @t Dr2u .

.

Using the weighted residuals method an integral equation can be generated from . for sufficiently small t.2) 1 1 r2u x. 1984b) is to assume that.1) 8. t =  @u x.1.Chapter 8 The BEM for Parabolic PDES 8. or a semi-analytic technique is used which can directly calculate a solution at a specified time. tm+1  = u x. tm+1 . tm  @t t which allows the diffusion equation in this time-range to be approximated as (8.BEM and the direct time integration method. 1 r2 u x.1 Time-Stepping Methods Several approaches have been proposed for applying the BEM to parabolic problems. These methods can be broadly classified into two main approaches. In this section time-stepping procedures will be considered. tm = 0 .  . The most common approach (Brebbia et al. t u x. The two most commonly used time-stepping methods are the coupled finite difference .3) Using this finite difference approximation the original parabolic equation has been reduced to an elliptic equation.1 Coupled Finite Difference . Either some form of time-stepping procedure is used to advance the solution in time. t @t (8. u x. tm+1 + t u x.Boundary Element Method This approach discretises the time-domain in a finite difference form. tm+1  . These two methods will be outlined in this section for the diffusion equation where the diffusivity  is a material parameter which can be a constant or a function of position. the time derivative can be approximated using a first-order fully implicit finite difference scheme @u x. Consider the variation between a time tm and a time tm+1 = tm + t.  (8. Time-stepping approaches discretise the time domain in some manner and use some form of time marching scheme to advance the solution from one discrete time to the next.

A solution at internal nodes can then be calculated. tm  . Cross & Lewis (1980). + 1 Z um! d t (8. This will generally require small time-steps to be adopted. m+1 = where The boundary conditions can be applied to form a solution system of the form m+1 is the vector of unknown nodal values at time tm+1 and is a vector constructed from known nodal values from the previous time-step.4) gives 1 r2 ! x. They approximated the time variation within an interval as B Hum+1 . They found that this method will only produce accurate results if Equation (8. t ! x. Tanaka. The fundamental solution of the modified Helmholtz equation is known in both two and three dimensions. = @n = u x. The domain integral does not include any problem unknowns so a fairly coarse domain mesh will generally suffice. They found that this improved the accuracy of the method. This coupled finite difference . Curran et al. Using Equa0 is known from the initial conditions so a tion (8. um+1 @! d. If a constant time-step is used the matrices .3). tm+1  and um modified Helmholtz equation x The fundamental solution ! is a solution of the applied at some source point  .  .2) accurately approximates the time derivative.4) contains a domain integral. tm+1 + 1 .boundary element method (FD-BEM) was first proposed by Brebbia & Walker (1980) for the diffusion equation. t = u x. Unfortunately it lead to a deterioration in convergence behaviour. This integral is generally evaluated by using a domain mesh (Brebbia et al. If a problem has time-dependent boundary conditions the solution system needs to be reformed at each time-step.136 T HE BEM FOR PARABOLIC PDES Equation (8.  (8. Gqm+1 = Bum U (8. The time-stepping procedure can be repeated using the internal solution at t = t0 +t as pseudo-initial conditions for the next time-step.6) HG B Ax x u x. solution can be calculated at t = t0 + t. For a problem with time-independent boundary conditions at each time-step it is only necessary to update and solve the system for m+1 . 1984b).5)  x where is a matrix containing the influence coefficients due to the domain integral.4) where um+1 = u  .6) the solution can be advanced in time. Applying the BEM to Equation (8. and can be calculated once and stored.  u x. c   um+1 + Z . It was implemented and investigated by Curran.7) . Unfortunately Equation (8. Z qm+1 !d. investigated the use of a higher-order approximation to the time-derivative. Matsimoto & Yang (1994) proposed a generalised version of this time-stepping scheme.4) where the fundamental solution is applied at internal point  and c   = 1. If an internal solution is required at a specific time this can be determined explicitly from Equation (8. tm.  +  = 0 (8.

11) This time dependent fundamental solution is known in two and three dimensions.8. x. If an internal solution is required at a specific time this can be determined from Equation (8. implemented this method and found it gave accurate results for a range of diffusion problems. t0 .1. (1984b).  r2um = u t u m+1 m (8. Alternatively the time integration process can be restarted at t0 with increasing numbers of intermediate steps used. They tested the 1 accuracy for a Crank-Nicolson scheme ( = 2 ). Two different time-stepping schemes can be used. tF . The variation of u and q with time is unknown so it is still necessary to step in time. . However. They found that the best results were achieved using a Crank-Nicolson scheme. Substituting this approximation and a first-order finite difference approximation of the time derivative into the diffusion equation gives If = 1 this approximation of the diffusion equation is equivalent to the standard FD-BEM discussed earlier.8). Similarly to the FD-BEM.8) 8.2 Direct Time-Integration Method Instead of converting the original parabolic equation to an elliptic equation the problem can be treated directly in the time domain by directly integrating over both time and space.9) Integrating in time once and in space twice gives ZtF Z @! ZtF Z Z c   u .  @u @t .10) with c   = 1. is a constant in the range 0  1. tF  +  u @n d. The weighted residual statement using this approach is ZtF Z  t0 1 x r u x. These two time-stepping approaches are discussed in detail in Brebbia et al. termed the time-scheme parameter. The first method requires a new domain integral to be calculated after each time-step due to . accurate results can be achieved using larger time-steps than with the FD-BEM. each time-step can be treated as a new problem so that an internal solution is constructed at the end of each time-step to be used as pseudo-initial conditions for the next time-step. t0 ! d t0 . Tanaka et al. dt =  q! d. tF . x. t +   tF  = 0 x (8.@t.10) x r2! . a Galerkin scheme ( = 2 ) and a fully implicit 3 scheme ( = 1). as the time dependence is included in the fundamental solution. t . An integral equation can be derived from Equation (8. t ! . r2 um+1 + 1 . t + @! . where the fundamental solution ! satisfies (8. Physically this fundamental solution represents the effect at a field point at time t of a unit point source applied at a point  at time tF . dt + u x.1 T IME -S TEPPING M ETHODS 137 where . tF . t d dt = 0 2  (8.

t dt Applying Laplace transforms to Equation (8. The second time-stepping procedure involves only a domain integral at t0 so. ideally. The most appropriate transform approach for parabolic problems is the Laplace transform. in practice. (1984b). Unfortunately.14) x .t u x. Consider the diffusion equation with appropriate boundary and initial conditions. in practice it is not always feasible to restart the integration process at t0 .  = Z1 0 e. The Laplace transform of u  bolised as U  . 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. in many practical cases the domain integral can be avoided. a domain integral only needs to be calculated once. however. u0 x (8. H G 8. This method was first proposed by Rizzo & Shippy (1970) and has since been successfully 1 r2 U x. t will be sym(8. This.12) gives with transformed boundary conditions.13) U x.2 Laplace Transform Method An alternative approach which avoids time-stepping is to solve the problem in a transform domain which removes the time dependence of the problem. This includes many practical cases such as constant initial temperature or an initial linear temperature profile. t @t (8.12) x x.  and is defined as 1 r2u x. If the initial conditions are u0 = 0 throughout the body the domain integral equals zero. Therefore.  . Equation (8. t =  @u x. Once the solution is determined in Laplace transform space this solution can be inverted to give a solution in the time-domain. will still require the user to create a domain mesh.  =  U x. 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. If the initial conditions satisfy Laplace’s equation r2 u0 = 0 then a Galerkin vector can be found and the domain integral can be expressed as equivalent boundary integrals. This inversion procedure requires solutions to be generated for several values of the transform parameter . u0   is the initial conditions of u.14) is an elliptic PDE which can be readily solved using the boundary element method. The parabolic PDE is thus converted to an elliptic problem for which the boundary element method has been shown to generally produce accurate results. As mentioned by Brebbia et al.138 T HE BEM FOR PARABOLIC PDES the updated pseudo-initial conditions. both time-stepping methods are likely to require domain integration. Once the solution to the elliptic problem is determined in the transform space a solution in the original space can be attained using an inverse transform procedure.

However. 1984b). Traditionally this domain integral has been calculated by using a domain discretisation (Brebbia et al. Liggett & Liu (1979) compared the Laplace transform method with the time-dependent Green’s function method. Therefore the required DR-BEM approximation is X 1 U x.14) is inhomogeneous so that applying the standard BEM will generate a domain integral involving the initial conditions. Considering Equation (8. Satravaha & Lu 1994). u x = N +L f 0 j  j =1 .8.  . 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. (1994) proposed using the DR-BEM to convert this domain integral term to equivalent boundary integrals. However. recently Zhu et al.14) this means that the DR-BEM will be used to convert the right-hand-side to equivalent domain integrals.3 T HE DR-BEM F OR T RANSIENT P ROBLEMS 139 used by other practitioners (Moridis & Reddell 1991) (Zhu. They chose to apply the DR-BEM based on the known fundamental solution to the Laplace operator. They noted that the direct method is simpler to apply.

t + b x. Once the solution is determined for this elliptic equation in the transform space a solution at a given time can be constructed using an inversion process. S  1 u .17) in which case a matrix expression of the form .16) which can be reduced to a square system by applying boundary conditions.  Su0 j (8.15) The DR-BEM can now be applied to Equation (8. t =  @u @t . giving a matrix system of the form (8. u (8.14). H . Gq = . This Laplace transform dual reciprocity method (LT-DRM) can easily be extended to equations of the form 1 x r2 u x.

18) is generated. S 1 u . for example.  Su0 (8.19) . the diffusion equation 1 r2 u =  @u @t (8.3 The DR-BEM For Transient Problems The DR-BEM can also be applied to parabolic problems. Gq = . 8. Consider.  H . Zhu and his colleagues have successfully extended the LT-DRM to solve diffusion problems with nonlinear source terms. R .

21) or C @u + Hu = Gq @t 1 ^ ^ C = .22) Equation (8. Substituting these approximations into Equation (8.25) where u and q are weighting parameters with values in the range 0. GQ F . q = 1:0. u H CH G h i um + 1 . q Gqm+1 = t .1 @u @t where (8.39) becomes (8.22) can be solved using a standard direct time-integration method. 1 t C + uH  c um+1 . .4 The MRM for Transient Problems The MRM can be applied to the diffusion equation r2 u of Laplace’s equation.  H u . q  Gqm (8.1 (8. Gq =  H U . GQ F . um @t t (8.20) 1 ^ ^ Hu . 1 . In this case the global approximation of b implies a separation of variables such that M @u = X f x t j @t j=1 j Using Equation (8.23) u = 1 .140 T HE BEM FOR PARABOLIC PDES where the thermal diffusivity.20). is a constant. and only need to be constructed once. If a constant time-step is used the matrices .22) an expression at tm+1 can be derived in terms of values at tm .24) (8.26) The values of u0 and q 0 are known from the initial conditions so a time-stepping procedure can be used. 1 and the time-step is between times tm and tm+1 = tm + t. Partridge & Brebbia (1990) recommended using a first-order finite difference approximation to the time derivative and linear approximations to u and q within a time-step. 8. 1 =  @u using the fundamental solution @t 1 @u and the recurrence In this case the forcing function becomes b0 =  @t . @u = um+1 . q  qm + q qm+1 (8. Equation (7. u um + uum+1 q = 1 . Using this two-level time-integration scheme the most common choice of time-scheme parameters is u = 0:5.

34) first derivatives. If the boundary conditions are not time-dependent the boundary conditions only need to be applied HL = 1m H1 + H0.8.29) where the matrices H1 .  G . G1 etc are the influence coefficient matrices relating to the higher-order u = 1 .HRum + GRqm where (8. tm 1 0 H H G G . A second order approach can be formulated by using quadratic interpolation of u and q within the time-range. @n @tj j =0 . 1m H1 + 1 .  qm + qm+1 where has a value in the range 0 to 1. If a constant time-step is used the matrices L . In this case the MRM formulation becomes 1 1 X Z @! @j u X Z @j q c   u   + d.27) The higher-order fundamental solutions are known for Laplace’s equation. j =0 .24) becomes j uj+1 = r2 uj = @ u @tj (8.30) (8.29) to be simplified to (8. This equation can be solved using a time-integration procedure.  H0.  um + um+1 q = 1 .32) (8. This most common approach is to use a linear approximation to u and q in this time-range H0u + H1u + H2u + : : : = G0q + G1q_ + G2q + : : : _ (8. This requires some interpolation between the two time-levels marked by m and m + 1. The most common approach is to solve this system numerically by discretising the time domain and using a time-stepping procedure.4 T HE MRM FOR T RANSIENT P ROBLEMS 141 relationship defined by Equation (7. GR = t t t 1 G + 1 . GL = 1m G1 + G0.33) HLum+1 . This gives the matrix system fundamental solutions. GLqm+1 = . HR = . This approach is termed a first-order approach as it removes all but the . L and R only need to be constructed once outside the time-stepping loop. = ! @tj d. q m q_ = q tm and all the other derivatives vanish. (8. u _ m m+1 . R .28) The standard BEM numerical procedure can be applied to this boundary integral equation.31) m+1 m u = u t. Using Equation (8.34) the solution can be advanced in time. Differentiating these linear approximations gives (8. This allows Equation (8.

142 T HE BEM FOR PARABOLIC PDES once. .

Phys.. S. Springer-Verlag.. W. ’Proceedings of the Fifth International Symposium on Numerical Methods in Engineering’. & Wu. L. (1979). C. & Savant. & Phan-Thien. Boundary element techniques: Theory and applications in engineering. Brebbia. ed. Applied Mathematical Modelling 4. Wendland & . 197–202. D. C. L. L. Fairweather. S. N. ’A method for the numerical solution of boundary value problems governed by second-order elliptic systems’. & Rizzo.. Periaux & R. A. ’Solution of parabolic differential equations by the boundary element method using discretisation in time’. SpringerVerlag. D. Telles. J. F. Boundary Element Techniques In Engineering. Brebbia. Tullock. C. Boundary Element Fundamentals . & Lewis. 730–745. in C. A new approach for transforming domain integrals to the boundary. G. P. D. 96–112. F. (1987). Journal of the Institute of Mathematics Applications 22. Shaw. Brebbia. L. Y. D. 73–84. & Wrobel. C. S. Boundary elements and perturbation theory for certain classes of hyperbolic and parabolic problems. A. J. Computational Mechanics Publications.. Springer-Verlag. (1987). ’An effective boundary element method for inhomogeneous partial differential equations’. C. J. A boundary element formulation of problems in linear isotropic elasticity with body forces. (1981). M. A. C. F.. & Nowak. A. S. A. & Walker. Comput. (1991). (1980). F. Curran. 31. eds. A. B.Bibliography Brebbia.. D. G. A. 398–400. D. in R. Newnes– Butterworths. Coleman. S. (1978). Cross. in C. Boundary Element Techniques. C. G. Brebbia. Gipson.. Computational Mechanics Publications. Gruber. 105–122. A. New York. J. Shippy. (1984a). C. pp. J. pp. Brebbia. (1989). C. J. J. J. Journal of Applied Mathematics and Physics (ZAMP) 42. Telles. ’Boundary Element Methods’. Danson. (1980). A..Basic Concepts And Recent Developments In The Poisson Equation. (1984b). S. Gipson. Computational Mechanics Publications and Springer-Verlag. ’On the numerical solution of two-dimensional problems by an improved boundary integral equation method’. Reible. A. & Wrobel. Rizzo. D. L. Clements.

Holz. A. W. 285–288. Golberg. C. A. Boundary Element Communications . ’A perturbation boundary element code for steady state groundwater flow in heterogeneous aquifers’. Moridis. 155–160. (1994). P. Partridge.-F. Springer-Verlag. Golberg. 397–403.. ’On the equivalence of dual reciprocity and particular integral approaches in the BEM’.36. The BEM dual reciprocity method for diffusion problems. Kuhn. (1993).144 BIBLIOGRAPHY G.29–19. Wendland & G. Nowak. (1979). & Rangogni. Computational Mechanics Publications and Springer-Verlag. Lafe. A. J. L. in C. P. BE Communications 5(6). Water Resources Research 15(4). ed. A. A. A. T. (1993). Water Resources Research 23(6). E. pp. Solution of transient heat conduction problems using boundary-only formulation. 861–866. Computational Mechanics 13. Monaco. in ’Boundary Elements XIII’. W. P. Brebbia. C. P. Engineering Analysis With Boundary Elements 10. in K. L. C. 1079–1084. M. (1990). & Brebbia. J. in ’Computational Methods In Water Resources VIII’. (1987). Springer-Verlag. Construction of fundamental solutions. D. The Laplace transform boundary element (LTBE) method for the solution of diffusion-type problems. A. & Cruse. (1992). H. & Wrobel. ’The theory of radial basis functions applied to the BEM for inhomogeneous partial differential equations’. pp. (1987). (1982).-D. Nowak. BE Communications 5(2). . G. 115–127. D. A. (1991). L. 19. & Brebbia. 57–61. D. Q. The Dual Reciprocity Boundary Element Method. 265–276. Brebbia. C. Itagaki. J. ’Unsteady flow in confined aquifers: A comparison of two boundary integral methods’. A. & Partridge. pp. (1994). Liggett. 83–97. & Reddell. ’Boundary Elements IX Vol 3’. Zielke. O. pp. S. W. Berlin and New York. C. Computational Mechanics Publications and Elsevier Applied Science. & Beskos. Brebbia. Engineering Analysis With Boundary Elements 11(1). (1992). C. Kuhn. & Cheng. (1987). Brebbia. Ortner. L. R. D. U.. A. Partridge. 87–90. ’Topics In Boundary Elements Research’. G. A. A. A. eds. Huang. Dassios. W. eds. Boundary element method: Processing of the source term of the Poisson equation by means of boundary integrals only. 68–73. ’Finite Elements In Water Resources IV’. L.. E. ’The numerical evaluation of particular solutions in the BEM . ’Some remarks on particular solution used in BEM formulation’. M. M. Polyzos. G. Gray. Computational Mechanics Publications and Springer-Verlag. ’Generation of higher order fundamental solutions to the twodimensional modified Helmholtz equation’. (1995). ’Comparison of the dual reciprocity and the multiple reciprocity methods’. & Chen. Computational Mechanics Publications and SpringerVerlag. W. N. P. Pinder & W. eds. A. pp. & Liu. in C.a review’. ’Boundary Elements IX’. Meissner. Computational Mechanics Publications and Springer-Verlag.

in C. (1986). F. L. in ’Boundary Elements XIII’. & Shippy. . S. ’A method of solution for certain problems of heat conduction’. X. 266–270. Boundary element methods and inhomogeneous parabolic equations. 569–576. 2004–2009. eds. J. Solution of variable coefficients PDEs by means of BEM and perturbation technique. T. Brebbia & B.BIBLIOGRAPHY 145 Rangogni. (1991). Zhu. Computational Mechanics Publications and Springer-Verlag. BE Abstracts 4(6). The dual reciprocity boundary element formulation for transient heat conduction. Rangogni. (1970). Master’s thesis. Applied Mathematical Modelling 18. (1985). R. AIAA Journal 8(11). Y. pp. Computational Mechanics 7. P. & Phan-Thien. ’BETECH 85’. D. On the dual reciprocity boundary element method. 1–10. J. J. J. Zhu. R. Wrobel. Computational Mechanics Publications and Springer-Verlag. Q. ’A boundary element approach for nonhomogeneous potential problems’. ’Time-stepping boundary element method applied to 2-D transient heat conduction problems’. S.. & Yang. ’Particular solutions associated with the Helmholtz operator used in DRBEM’. (1994). C. C. & Lu. ’Numerical solution of the generalized Laplace equation by coupling the boundary element method and the perturbation method’. Matsimoto. (1994). (1986).. F. ’Solving linear diffusion equations with the dual reciprocity method in Laplace space’. 231–233. Tanaka. Rizzo. D. Brebbia. (1993). Wu. & Nardini. Springer-Verlag.. Satravaha. A. N. M. 19–30. C. 279–288. The University of Wollongong. (1993). Zheng. Coleman.. J. Zhang. (1991). Engineering Analysis With Boundary Elements 13. C. in ’Finite Elements In Water Resources VI’. R. Applied Mathematical Modelling 10. Noye. A.

.

76 Weighted residual. 24 Weighting function. 124. 31 Galerkin vector. 45–48. 72 wave equation. 18 Spherical polar. 24 Isoparametric formulation. 103 Area Coordinates. 17 Prolate spheroidal. 82 Rayleigh-Ritz method. 64 Triangular elements. 77 Regular BEM. 25. 12 Basis functions. 14 cubic. 43–45 Direct time-integration method. 2–16. 56 Global stiffness matrix. 2–4 quadratic. 46 Navier. 17–19 Cylindrical polar. 82 Trefftz method. 129 elliptic problems. 79 Boundary conditions application of. 10–14 bicubic.boundary element method. 26 Fundamental solution. 137 Dual reciprocity BEM approximating function. 91 Laplace. 72 Helmholtz. 24 . 122 Perturbation BEM.Index Advection-diffusion equation. 122–124 diffusion equation. 27 integration by parts. 129 transient problems. 53 Laplace transform method. 7 Beam elements. 72 Kelvin. 43. 10 bilinear. 72 Galerkin formulation. 131 Dual reciprocity BEM. 106 Multiple reciprocity method. 72. 119 Gaussian quadrature. 133 element stiffness matrix. 127 derivative terms. 7–9 linear. 10 Lagrange. 138 Mass lumping. 135 Curvilinear coordinate systems. 53 Hermite. 18 Dirac-Delta function. 14–16 Truss elements. 12. 81–83 Potential energy. 39–42. 14 Basis functions Hermite cubic. 124 Poisson equation. 29. 121 Plane stress elements. 139 variable coefficients. 54 Coupled finite difference . 140 Helmholtz equation. 117 diffusion equation. 64 Strain energy. 75.

Sign up to vote on this title
UsefulNot useful