Вы находитесь на странице: 1из 2

www.boundary-element-method.

com

Computation of the Discrete Forms of the Laplace Integral Operators


In the development of the boundary element method the original governing partial differential
equation1 which governs the domain of a boundary value problem2 is reformulated as a boundary
integral equation. For Laplaces equation3 the boundary integral equations that arise are made up
of component operators which are termed the Laplace integral operators 4. In order to solve the
boundary integral equation and hence solve the boundary value problem in which the domain
is governed by Laplaces equation - the boundary is normally expressed (approximated) by a set
of panels5 and the boundary integrals can be discretized and solved by an integral equation
method6. The application of the integral equaton method usually results in the formation of a
linear system of equations7. The main computational costs in the boundary element method occur
mainly in two areas; the first is in the formation of the matrices in the linear system of equations
and the second is in its solution. In this document we consider the operatons that are required to
determine the matrix elements in the BEM, all of which are usually the result of numerical
integration.8
The boundary integral equations that arise are Fredholm integral equations. A number of integral
equation methods can be applied to solve the boundary integral equation and most of these
methods are relevant to this algorithmic design and analysis, but this is mainly prompted by the
application of the method of collocation9 .
The discretization of the integal operator to the boundary integral equations that arise in
Laplaces equation require the direct integration over every basis function on every panel for
every collocation point10. If we presume that the boundary is represented by N panels (or N
vertices) and the computational cost of forming the elements of the (usually) square matrices that
arise are approximately equal then the computational cost of forming the linear systems in the
BEM is O(N2) 11. For problems with a large number of elements this is one of the pinch points in
the boundary element method; minimising the cost of the computation of the discrete forms the
matrix components can have a significant impact on the computational requirements and the
computer time taken to process the method.
The determination of each matrix component requires the integration over each panel with
respect to each collocation point. The numerical integation or quadrature over a panel requires
that the value of the integrands are found for all the quadrature points.
Note the the the term integrands is used in the plural form, that is because usually a number of
integral operators need to be evaluated. Again, for the purposes of saving on their computational
cost it is good practice that these are evaluated simultaneously. The integral operators have
similar properties in vector geometry12 to evaluate and so it makes sense to compute each of these
once and use the result in each of the discrete operator evaluations. Often different integal
equation formulations require different operators and hence if a generic routine is required then
it makes computational sence to allow a selection of the operators that are required for each
particular purpose, again to minimise computation time.

Partial Differential Equations


Boundary Value Problems and Boundary Conditions
3 Laplaces Equation
4 The Laplace Integral Operators
5 Boundary Representation in the Boundary Element Method
6 Integral Equations
7 Linear Systems of equations
8 Numerical Integration
9 Solution of Fredholm Integral Equations by Collocation
10 Discretization of the Laplace Integral Operators
11 Big O Notation
12 Vector Arithmetic and Geometry
1
2

www.boundary-element-method.com

The programme of computation listed below is given prominence because it needs to be executed
for each quadrature point and hence it is the key to determining the computational cost of the
overall method. The following computational programme is intended to be optimal. Note that it
is assumed that p (the observation point) and vp (the unit vector through p) are already set and
the notation and methods applied are taken from the documents13 14.
A: Set q, the quadrature point on the panel
B: Set nq, the unit outward normal to the panel at q
C: Compute vp.nq , (scalar or dot product)
D: Compute r, (=q-p)
E: Compute r, r2 , (r=|r|)
F: Compute r3 (required only in the 3D case)
G: Compute

, (

H: Compute

, (

I: Compute

) (required for the M and N operators)

) (required for the Mt and N operators)

(required for the N operator)

J: Compute G, ( =

1
ln
2

in 2D, or =

1 1
in

L: Compure (( =

2
operators)

M: Compute
(
=

N: Compute

O: Compute

2D, or

1
4

in 3D) (accumulate for the L operator)


1 1
4 2

in 3D) (required for the M and Mt

(accumulate for the M operator)


(accumulate for the Mt operator)
1

= 2 2 ( . + 2

) in 2D and

= 2 2 ( . +

) in 3D.

Each operation can be directly costed in terms of floating-point operations., apart from operation
E which needs to compute a square root.
This method was originally pulished as part of the method for the computation of the Helmholtz
integral operators15, which was used for solving acoustics problems by the boundary element
method16.

The Laplace Integral Operators


Discretization of the Laplace Integral Operators
15 Fortran codes for computing the discrete Helmholtz integral operators
16 The Boundary Element Method in Acoustics
13
14

Вам также может понравиться