Hi, I'm trying to simulate a semi-infinite substrate (fills half the cell) in the visible (in preparation for later stages).
The characteristic size is 10nm. The block that represents the substrate penetrates all the thickness of PML and is 3000nm wide plus the thickness of PML. (With less penetration the reflections get higher.) The substrate material is a dielectric with a fixed epsilon of 4 (Si3N4). The default medium has a refractive index of 1.5 (SiO2). I'm using a Ez CW source that generates a plane wave across the whole cell (tried a Gaussian source w/ same results). It penetrates the whole thickness of PML on the sides and is attached to one of the medium-PML interfaces as described here: http://www.mail-archive.com/[email protected]/msg03088.html with the fix here http://www.mail-archive.com/[email protected]/msg03089.html. I'm measuring the Poynting field and normalize it to the configuration without the substrate. At wavelength of 450nm I have to make the PML more than 1000nm thick in order to suppress oscillations closer to the substrate-PML borders. Even with a 1000nm thick PML, when I compare the results to those with a 1200nm thick one, then upto 800nm off center there's 0 difference, but thereafter there's a difference of upto 3% that has an oscillating shape. With a 600nm thick PML, the difference relative to the 1200nm one gets as big as 12% and the deviation begins at about 700nm. Changing the length of the simulation doesn't affect the results. Neither does switching to a Ey source. At 800nm wavelength the difference between 1000nm and 1200nm thick PMLs is 3.5%. Is this normal? What could be done to reduce the thickness of PML to save computation time while still getting reliable results? Thx, David
_______________________________________________ meep-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
