Research Article

A Fourier Continuation Method for the Solution of Elliptic Eigenvalue Problems in General Domains

Table 2

Number of iterations and computational time required by the FC solver and the finite difference solver to compute to a fixed accuracy for the unit square example (fixed boundaries). (The first 10 eigenvalues are calculated to the tolerance of , but we only show the error in the first eigenvalue for conciseness.) For each given accuracy, the FC(Gram) algorithm outperforms the finite difference algorithm by a large margin. All computations were performed on a 2.67 GHz Intel Xeon processor. (a) Results using centered finite difference; (b) results using FC(Gram) with 5th-order basis.
(a)

Absolute error Time per operation (s) Number of operations

3.00 × 10−4 74725 0.009501 9693
7.57 × 10−5 298800 0.05130 37760

(b)

Absolute error Time per operation (s) Number of operations

3.53 × 10−4 1525 0.001235 728
1.41 × 10−5 6100 0.004211 2245