Abstract
A new method for the reconstruction of two-dimensional periodic structures from scattered far-field data is presented. The approach is based on the recently developed `methods of variation of boundaries' (MVB) for the solution of forward-scattering problems. Here, the inverse problem is formulated as that of minimization of an appropriate nonlinear least-squares cost functional J. The analytic continuation ideas inherent to MVB are then advanced to enable a global line search at each step of a proposed conjugate gradients algorithm. Through a number of numerical examples it is demonstrated that such globalization may be necessary to overcome the non-convexity of J which, in turn, is shown to persist even when the global minimizer is known to be unique.

This publication has 34 references indexed in Scilit: