A simple technique is described which is capable of providing considerable
reductions in the computational demands of the finite difference time domai
n (FDTD) algorithm. It is shown that, by forming the inner product of the F
DTD time domain data and the transverse field variation of the mode of inte
rest, it is possible to avoid modelling long feed structures in many FDTD m
odels. The precise computational and memory savings resulting from the meth
od depend on the structure of interest, figures of 50% and 15% are achieved
for two example structures.