Dear MEEP users, first I am sorry for stating that no material should be put in PML. On the contrary, it can even help with the problem we discussed few weeks ago:
I am convinced the problem is really in PML being able to amplificate the evanescent waves in the near field of the oscillator. After a daylong experimenting, I ended up with putting a thin highly lossy medium into part of the PML volume. The propagating waves are attenuated enough by PML and do not reflect. The evanescent waves which freely pervade the PML are now attenuated in the lossy medium and the simulation is eventually stable. There is one caveat: when the evanescent waves get attenuated, most of the resonances in the structure become damped. With the lossy medium or without it, one has to provide enough space around the structure if the narrow resonances are to be simulated properly! Otherwise it makes no sense to run a long simulation anyway. I have also observed the case when losses and amplification nearly cancelled out and the resonances became narrow even in a small volume. However, generally it is much more practical to allocate a great simulation volume than to fine tune the losses in PML! It would be interesting if a evanescent-wave-friendly (e. g. amplification-damping neutral) PML could be coded in MEEP in the future. I would be glad to read you thoughs! F. D. 2012/12/19, Filip Dominec <[email protected]>: > Hi, Ali, > as I suggested, try to make a BIG spacing between your structure and the > PML. I recommend to give it as big spacing from PML, as the transverse > dimensions are. I would be glad if somebody gives us some better suggestions > on what to try next. > > It is also possible that it is the silver itself which is unstable. This > would manifest as oscillations with highest possible spatial frequency. > Check what happens when the silver is substituted e. g. with silicon. Such a > problem may be fixed by reducing the conductivity and/or decreasing Courant > factor. > > I believe that MEEP may well be technically fine, but the PML mathematics > are generally really tricky. PML is not an area of an absorbing material, > but a sophisticated representation of infinite volume in finite (and small) > number of points. > > Filip > > On Wed, 19 Dec 2012 18:18:11 +0100 > Ali Naqavi <[email protected]> wrote: > >> Hi Filip, >> >> Do you think the problem is related to the theory of PML or just a >> mistake >> in implementation in MEEP? >> >> I tried in the meanwhile different thickness of the silver layer and in >> all >> cases there was something like a resonance that I really do not know >> where >> it comes from and I do not know how to get rid of it. I have to run the >> simulation for a long time so it seems that this reflection effect is >> inevitable. As you mentioned this phenomenon is observed only at low >> frequency but I need to let the field decay which means taking into >> consideration the low frequencies. >> >> >> Best, >> Ali >> >> >> On Mon, Dec 17, 2012 at 1:47 PM, Filip Dominec >> <[email protected]>wrote: >> >> > Hi, Ali, and all the MEEP users, >> > first, I believe one should not place any material into the volume >> > occupied by PML. However, what you report actually seems to be a much >> > deeper problem in the PML implementation in MEEP. I already wanted to >> > re-report it. >> > >> > In my simulations, I encounter similar weird transverse oscillations >> > in the vicinity of the PML layer quite often. A typical triggering >> > simulation is (any non-dispersive) structure in the center of volume, >> > Bloch-periodicity on the X-Y faces and PMLs on the Z face. An example >> > is documented e. g. at the bottom of this page: >> > http://fzu.cz/~dominecf/misc/meep/index.html#pml >> > >> > The oscillations have quite a low frequency, probably given by >> > f=(width/speed_of_light). Their Poynting vector seems to be transverse >> > to Z axis and they have similar axes of symmetry as the simulated >> > structure. They seem to build up from low-frequency numerical error, >> > and then grow exponentially. >> > >> > There is always a workaround in making quite a big distance between >> > the structure and PMLs. However, this substantially increases the >> > simulated volume. I already started reading some theory on PML in the >> > Taflove's superbible, but I would be very happy if somebody more >> > experienced provided their opinion on this instability! >> > >> > Filip >> > >> > 2012/12/17, Ali Naqavi <[email protected]>: >> > > Dear MEEP users, >> > > >> > > The structure I want to simulate using the meep c++ is so simple: a >> > > multilayer solar cell with a silver layer at the backside. So the >> > structure >> > > looks like this: >> > > >> > > PML >> > > ----------- >> > > Air >> > > ----------- >> > > Silicon >> > > ------------ >> > > silver >> > > ------------ >> > > PML >> > > >> > > >> > > >> > > As far as I realized, one should try not to put dispersive materials >> > > next >> > > to PML; otherwise the field might blow up inside PML. Do you know any >> > > method to get rid of this field blow up? >> > > >> > > To resolve the mentioned problem, I put a nondispersive dielectric >> > > layer >> > > under silver so that adjacent to the PML, I do not have silver now but >> > > a >> > > dielectric. Still, there is a non-negligible amount of reflection >> > > from >> > the >> > > PML. Do you know how to solve this problem? I guess it might be due >> > > to >> > the >> > > strength of PML but I do not know how to modify it in meep c++. >> > > >> > > Best, >> > > Ali >> > > >> > >> >> >> >> -- >> Ali Naqavi >> Institute of Microengineering >> EPFL | Ecole Polytechnique Fédérale de Lausanne > _______________________________________________ meep-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
