Le 8 November 2006, Steven G. Johnson, à bout, prit son clavier pour taper sur son écran: > On Wed, 8 Nov 2006, Loïc Le Guyader wrote: > >In order to have frequency-dependent permittivity in meep, we have to > >decompose it in a sum of Lorentzian. > > > >But it's not always possible. One material I want to use is transparent in > >a > >certain region and I can't fit it with Lorentzian because there feet > >doesn't > >disapears in this region. > > Enough Lorentzians should form a complete basis, so you should be able to > fit any function given enough Lorentzians.
Arf. Well, it also make the memory consuption exploding, and the simulation get slower. > >I guess gaussian would work, at least it has been reported in litterature. > >Is > >it possible to have gaussian sum supported in meep ? > > Currently it doesn't support anything other than Lorentzians. > > I haven't heard of a way to implement Gaussian material dispersion in > time-domain, do you have a reference? No, unfortunately. I allready don't see why we can't use raw data so don't ask me how can we use gaussian term instead of lorentzian ;) The problem currently is that one a the deltaepsilon gets negative in oder to compensate the tail of the two other oscillators. The resulting fit doesn't looks too bad, but I wonder if a deltaepsilon < 0 can create some problem in the simulation. If not, well, I'm allready using it then. Now the problem is that I don't use symmetries to avoid the bug I reported previously and I use two different materials with 3 polarizability each, all this in a 3D simulation. Somehow, I run out of memory :( On Fri, 29 Sep 2006, you said that it was possible to reduce the memory usage due to polarizability. What are the news on this subject ? Best regards. -- _______________________________________________ meep-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
