Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
2, MARCH 2002
421
I. INTRODUCTION
Fig. 1. Clustering for a simple example with ten collocation points. A large
distance between two collocation points results in a large difference of the
respective equation numbers.
422
Fig. 2. The permuted matrix for the example depicted in Fig. 1 contains rather
large off-diagonal blocks which describe remote interactions and which are
close to some low-rank matrices.
Here,
denotes the Frobenius norm of the matrix . The
solution of this problem is given by the singular-value decomposition (SVD) of the block
Fig. 3. TEAM problem 10. An exciting coil is set between two steel channels,
and a steel plate is inserted between the channels. The surfaces of this geometry
have been discretized by linear triangular elements to obtain an input mesh for
the ACA testing environment.
(1)
and denote the greatest singular triples of the
where
matrix and the rank is chosen so that the required accuracy
of the approximation is fulfilled.
Since the SVD requires the computation of the whole matrix
in advance and since SVD is rather expensive with respect
this analytical solution is not
to numerical work
practicable.
We present now the algorithm of ACA, which allows to generate only few rows and columns of the matrix and approximate the rest of the matrix using only this information.
and for
compute
Let
(3)
The amount of numerical work required by the ACA algorithm
. Thus, if the numerical rank of the approxis
imation remains constant (which is usually the case), then the
total numerical work for the approximation and the memory re.
quirements are both of the order
III. EXAMPLES
A. Application to the Laplace Equation
TABLE I
MEMORY REQUIREMENTS USING THE ACA ALGORITHM
TABLE II
COMPUTATION TIMES USING THE ACA ALGORITHM
The values refer to a 1.2-GHz AMD Athlon PC. Note that the table shows the wallclock
time and not the CPU time. Therefore, it includes the swap time which the computer needed
during the computation for the finest mesh. Still, even the wallclock time does not grow like
423
TABLE III
MEMORY REQUIREMENTS USING THE ACA ALGORITHM
TABLE IV
COMPUTATION TIMES USING THE ACA ALGORITHM
The size of the approximants and their relative size are given
in Table III. The average scaling factors due to the mesh refinements are 5.7 and 4.8, respectively. Thus, we again observe
the asymptotically linear behavior of the memory consumption.
Analogously to the first example we give the time spent generating the approximants as well as the costs of an iteration step
in Table IV.
The numerical examples above show that the memory usage
of the BEM matrices computed by the ACA method grows almost linearly with the number of unknowns on the boundary.
The same behavior is observable with respect to the time of
approximant generation and the matrix vector multiplication.
Hence, by using the ACA method we are able to handle BEM
problems whose solution by application of standard BEM would
be impossible with the same resources.
Fig. 4. Electromechanical relay. The magnetic circuit consists of a pole core,
a magnetic yoke, and a movable armature. Again, the surfaces have been
discretized by linear triangular elements to obtain an input mesh for the ACA
testing environment.
B. Application to a BEM-FEM-Code
Electromagnetic devices can be analyzed by the coupled
BEM-FEM method, where the conducting and magnetic parts
are discretized by finite elements. In contrast, the surrounding
space is described with the BEM. This discretization scheme is
well suited for problems including moving parts and has been
described in detail elsewhere [7][9].
In the air domain, the BEM is applied to solve the equa, where is the Coulomb gauged magnetic
tion
vector potential and an impressed source current density. This
vector equation decouples into three scalar equations for the
Cartesian components of , so that we are left with the same
situation as in the ACA testing environment. We implemented
the ACA algorithm into the BEM-FEM code and performed
computations for the examples depicted in Figs. 3 and 4. However, quadratic six-noded triangles in connection with quadratic
ten-noded tetrahedra have been employed for this analysis.
TEAM problem 10 has been treated as a magnetostatic
problem (for details, see [10]). The symmetry of the problem
has intentionally been disregarded. Some results are collected
in Table V.
The difference of the flux densities with and without ACA
(0.5%) is much smaller than the difference to the measured value
of 1.67 T (3.4%) which is due to the still relatively coarse mesh.
However, the computer resources for ACA-BEM dropped to
about half the amount needed for the standard BEM.
424
TABLE V
MESH AND COMPUTATIONAL DATA FOR TEAM PROBLEM 10
TABLE VI
MESH AND COMPUTATIONAL DATA FOR THE RELAY
tions [5], multipole decomposition [11]) is that only the original entries of the system matrix are used for its approximation.
Thus, the already-developed procedures for the generating of
the BEM matrices can be used after some minor modifications.
The ACA algorithm is not difficult to implement in contrast to
practical implementation of the Taylor series or spherical harmonics used in the multipole method. On the other hand, the
multipole method allows the rapid computation of fields and potentials in the BEM domain once the problem has been solved
[12].
The second advantage of the ACA-BEM is that any arbitrary accuracy of the approximation can easily be reached. In
the worst case, the whole matrix will be generated without any
error. Using the sequence of the less and less accurate approximations of the same coarse discretization we are able to fix the
bound of the acceptable approximation error. Then, an obvious
reduction of this bound due to the increased dimension of the
matrix can be used for the final computations on the fine grid.
REFERENCES
As a final example, the closing process of the electromechanical relay has been studied, where only half of the mesh shown in
Fig. 4 was considered by taking advantage of the symmetry (for
details, see [1] and [7]). Results of this computation are given
in Table VI.
This example requires an enormous amount of CPU time,
because there are many time steps and the BEM matrices have
to be reprocessed frequently due to the motion of the armature.
The ACA implementation for problems with symmetry is not
yet optimized. Despite that fact the memory requirement could
still be reduced to 50% of the previous value.
IV. CONCLUSION
The memory consumption of the standard BEM turns out to
be the limiting factor in many practical applications. The above
results show that the ACA-BEM is a feasible means to overcome
these limitations.
The main advantage of the ACA method over the other fast
BEM techniques (H-Matrices [4], pseudoskeleton approxima-