In this section, all the computations are carried out on the same desktop computer with an Intel Core i5-3470 CPU (3.2GHZ) and 4GBmemory. The original FMM is applied in this paper for testing the validity and efficiency of the proposed adaptive expansion technique, although it is seldom used for very low frequency acoustic problems. In addition, the FMM without the adaptive expansion technique is named standard algorithm, and the FMM with the adaptive expansion technique is named adaptive algorithm in this section.
In the numerical examples, the maximum number of the elements in leafs is 60, and p set as 8 in both standard algorithm and in adaptive algorithm. In the GMRES solver, we stop the iteration when the tolerance error in FMM is less than 10^3
The radiation problem on a cylinder domain (Fig. 1) is used to study the performances of the presented adaptive algorithm using the CFMM. The diameter of this cylinder is 2, and the length is 5. The sample point in this example is (0, 5, 0). The pulsating cylinder is formulated by prescribing the normal velocity on a cylinder
surface produced by a pulsating ball with radius a=1, which is shown in Fig. 1. The wave number k=1.0. The pulsating ball is circumscribed by the cylinder.
The results are summarized in Table 1. The CPU time are compared in the 2nd and 3rd columns in the Table 1 for the standard and adaptive algorithms. From the two columns we know that the CPU time used in the adaptive algorithm is much less than that used in the standard algorithm. The bottom row of the Table 1 shows that the adaptive algorithm can solve a problem with 100 thousands nodes using about an hour, the improvement in the CPU time is roughly 60%. The relative errors at sample point obtained by the standard and adaptive algorithms are compared in the 4th and 5th columns in Table 1, The results show that that the relative errors obtained by adaptive algorithm are very much closed to that obtained by standard algorithm in all the cases of DOFs.