Convergence analysis of Left-Right splitting surface scattering method

Abstract

We study the convergence of the Left-Right splitting method (equivalent in key respects to the Method of Multiple Ordered Interactions and Forward-Backward method) for wave scattering by rough surfaces. This is an operator series method primarily designed for low grazing incidence and found in many cases to converge rapidly, often within one or two terms even for large incident angles. However, convergence is not guaranteed and semi-convergence may be observed. Our aims are two-fold: (1) To obtain theoretical and physical insight into the regimes in which rapid convergence occurs and the mechanisms by which it fails, by examining and modifying eigenvalues of the operator; (2) provide a strategy for increasing the speed of convergence or more crucially for overcoming divergence, and providing a stopping criterion. The first is addressed by subtracting successive dominant eigenvectors from the incident field, to examine the impact on divergence and on the incident spectrum. For the second, we apply a generalisation of Shanks' transformation to the operator series; this effectively improves convergence and (unlike eigenvalue subtraction) readily generalises to 3D and composite problems. These results also explain why the method converges so rapidly for much larger incident angles. Finally we ask and give an analytical solution to a key question: For a divergent eigenvector of the iterating operator, what is the exact solution and can it be deduced from the divergent series? We show that the exact solution is well-behaved and can be found from the series in a way which is related to the Shanks transformation.