To simplify the problem, I tried to record the incident field only, without introducing the scatterer. Here are a few animations. In all three tests, I have a 3D simulation where the z direction has a periodic boundary condition and boundary layers in the x y directions and I show the y=center slice over time. In the first simulation, I have pml boundary layers and simulate a source that contains only frequencies with well defined mapped incident directions; the second simulation also has pml boundary layers but the source is defined such that it covers a larger range of frequencies and some small frequencies don't have well-defined incident angles; the third simulation has thick absorber boundary layers and the source is defined the same as in the second simulation. I observed that only in the first simulation the incident field looks like it is propagating in a fixed direction whereas the other two do not. I was wondering if this is what I should expect when there are evanescent waves and whether this could be the source of the error in my scattering tests.
1. https://wavefiber3d.s3.amazonaws.com/subfreq_modeindex-4.gif 2. https://wavefiber3d.s3.amazonaws.com/fullfreq_modeindex-4.gif 3. https://wavefiber3d.s3.amazonaws.com/fullfreq_modeindex-4_absorber.gif Thank you! On Sun, Dec 6, 2020 at 12:10 PM Mandy Xia <[email protected]> wrote: > Thank you Ardavan! I will try this and ineffective pml might be why I > still have an inaccurate result when the incident angle is large. But I was > also wondering when I have the source covering the whole frequency range, > including those don't have well defined mapped incident angle (evanescent > waves), whether you think the ineffective pml was also the cause for the > inaccurate result on two ends of the frequency range. On the large > frequency end, the incident angle is not very grazing (143 degree) but > there is still huge difference and not converging. > > On Sun, Dec 6, 2020 at 11:52 AM Ardavan Oskooi <[email protected]> > wrote: > >> Note that PMLs become less effective for incident waves at large oblique >> angles. This effect is described in: >> >> https://meep.readthedocs.io/en/latest/FAQ/#why-are-the-fields-not-being-absorbed-by-the-pml. >> >> Rather than increase the PML thickness as you seem to be doing, a better >> solution is to replace the PML with an Absorber. >> >> On 12/6/20 08:16, Mandy Xia wrote: >> > incident angles, especially at the two ends of the frequency range >> > (i.e. the largest frequency and the smallest frequency that has a well >> > defined mapped incident angle). I tried to increase the resolution and >> > pml thickness but they didn't help. I tried to construct the source >> > with only frequencies that have well-defined mapped incident angles >> > and the matching to the ground truth significantly improved (although >> > at the small frequency end there is still a relatively large >> > difference and I ran out of memory when increasing the resolution and >> > couldn't make a conclusion if it was converging or not). I'm trying to >> > figure out why this is the case and what is the impact on the >> > simulation from the evanescent waves. I have a periodic boundary in >> > the z direction and pml layers in the x and y directions. My source is >> >
_______________________________________________ meep-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
