I think the main issue in your simulation is the proximity of the PML to the structure. It is a good practice to leave a distance of at least half the maximum wavelength between the structure and the PML; the reason is that we want to avoid evanescent fields at the PML, which can be difficult to absorb. After increasing the simulation size I got a scattering cross section positive everywhere:
The simulation I used is attached: D=9nm_simplified.fsp (259.1 KB). For simplicity I centered the sphere at the origin. Some additional changes that helped with the simulation are:
- Used smaller mesh accuracy (level 2) and used the mesh override “mesh_particles” to refine the region of the TFSF source and the cross-section monitor (dx = dy = dz = 0.5 nm).
- Used symmetric and antisymmetric boundary conditions to speed up the simulation.
If the results for the cross section have converged (if the mesh is fine enough, etc.), they should not depend on the position of the analysis group as long as it encloses the sphere and part of the substrate. The reason is that the cross section is calculated from the transmission through the box monitor, which should be independent of the size of the box (note that the substrate is lossless).