Abstract
The article demonstrates that a Method‐of‐Moments solution for large, smooth, two‐dimensional scatterers, requiring matrices that are several hundreds to several thousands in size, can be constructed efficiently by using a combination of transformations based upon FFT and partial QR decomposition of off‐diagonal blocks. The primary contribution of the article is to demonstrate that these techniques, applied in tandem, can typically reduce the active RAM requirements by a factor of 60‐100 for moderately large problems, and by as much as 225 (estimated) for matrix sizes around 100,000. It is also estimated that an effective flop rate of about 8‐35 N2 In N to 0.01N3 can be achieved via this transformation. The reduction factors for RAM requirements and the accuracy of the solution are demonstrated for the problem of a 2D elliptic cylinder with a large aspect ratio, requiring matrix sizes up to 10,000.