“…In Figure 5, we compare the accuracy of the solvers with three techniques: PAL, PML n (with n = 1, \sigma 0 = (5.16, 2.78, 1.89, 1.43, 1.15, 1.01) for the wavenumber chosen as k = (50,100,150,200,250,300), respectively, which are optimal values as suggested in [10]), and PML \infty (\sigma 0 = 1/k as suggested in [8]) for various k. It can be seen from Figure 5(a) that when the wavenumber k is relatively small, the error curves (with N < 20) of these three methods intertwine with each other, and the error obtained by PML \infty is slightly smaller than that of the other two methods. However, as N increases, the error for PML \infty becomes much larger than that of the PAL and PML n due to the large roundoff errors induced by the Gauss quadrature of singular functions.…”