Efficient Stochastic FDTD Algorithm with Optimized GPU Acceleration

A 3-D curvilinear stochastic finite-difference time-domain (S-FDTD) technique on modern graphics processing units (GPUs) is introduced in this paper for complex media with high levels of statistically-variable heterogeneities. The novel accelerated methodology develops a robust covariant/contravariant dual-grid tessellation and estimates the mean value and standard deviation of field components during only a single run. In this way, notably accurate and stable estimations can be very rapidly and economically obtained, unlike the usual multiple-realization staircase Monte-Carlo FDTD schemes. These merits are successfully verified via realistic microwave setups with highly-varying media uncertainties, where the featured algorithm is shown to overwhelm the typical exceedingly resource consuming approaches.