Hello everyone, I'm working on some dielectric metamaterial simulation thing, and I found an article (F Callewaert et al 2016 J. Opt. 18 075006) describing a structure very similar to what I'm simulating.
But when I try repeating the simulation in that article (well, not quitely, since he used FDTD and I'm using FEM, but the model is the same) there's something wrong here. The calculated absorption spectrum has one more peak than the same spectrum in the article (fig 3a black line ), and when I examine the E field distribution at that peak, it shows something clearly shouldn't physically possible.
It seems there are some kind of, I don't know, standing wave? Which shouldn't happen because the model is a 2-D periodic structure plane in 3-D space, and under normal incidence all reflected light should goes to infinity. I added perfect matched layers but they didn't work, the EM field just go through PML like nothing there.
This problem also happens to some other model of mine, all including the similar structure (some kind of pillar standing above some kind of metal), which I can't find any clue to deal with it. When it happens it can skew my simulation result to such an extent that reflectivity goes to 500%, which can't be as the structure contains nothing with gain.
My major is physical chemistry and I learned FEM simulation myself without many optics background, so if someone can help me with that or encountered a similar problem, your reply will help me alot, thanks!
Well I think I figured it out myself after 1 week, just set the PML scaling factor and PML curvature coefficient from 2 to 1000, then the reflectivity spectrum gives the correct result. The E field distribution still have problem though
I am not familiar with COMSOL, but PML is not perfect. Its'no-magic. It is an attempt to treat transition of oblique waves from boundary to infinity, matched to one specific wavelength.
At RF there is a technique of putting any radiating boundary at odd number of quarterwaves away from the model. Furhter it is, better it gets. And make it wider for avoiding false periodicity effects (all bugs will move closer to 0-beta values). Fix the PML wavelength and make it deeper.
Denis, odd number of quarter-waves is a common rule in numeric wave computations. Regardless of radiation boundary type. So called "radiation boundary" in HFSS is much worse than PML. It can treat well only normaly incident plane waves, like most non-PML radiations boundary types. For that reason, it is recommended to place it as far as you can.
I realize it is an old thread, but this might be of use to anyone fiddling with PMLs in Comsol dropping by in the future.
The first thing is to understand what a pml is: you are applying a transformation to the simulation domain that maps the points in space to "infinity" so that you avoid reflections back into your system.
Depending on the kind of transformation, the PML does better with different kinds of fields. In comsol, there are two types of transformations: rational and polynomial, each with their strengths and weaknesses. The two parameters mentioned above also do two different things: the "pml scaling factor" just tells you how much the size of the virtual space is extended. Larger mapped domain means better absorption, but it also means that you need more mesh points to resolve the decay in that space. On the other hand, the "pml curvature" factor changes the distribution of points on the PML. The idea is that in some situations you will have a mixture of fields that the PML attenuate differently, like two plane waves impacting at two angles in the pml. One plane wave impacting perpendicular on the pml boundary will be attenuated much faster that one impacting at grazing angle, and hence it is important to have a denser mesh close to the interface to describe well the attenuation of both waves.
In my experience, what Andrey was saying is important. Take the PMLs away from radiating structures so as to only deal with propagating fields (and to not modify the evanescent fields of the structures).
Generally, the Comsol documentation is quite good, I advise everyone to read it when in doubt.
- Badges
- Science topic