You are right. When I symmetrized the eps function the divergences disappeared. I really didnt know that this one last pixel could have such a big influence.
I have also some other kind of divergences in other cases. They seem not to be connected to symmetries but arise from PML layers. I am starting to think if I am not doing any similar mistake there with those last pixels beyond the calculation volume. In general what should appear on those last pixels ? Normally I just extend my photonic crystal there without any special treatment. Is meep expecting something special there ? Thanks for your help Steven. Thomas Jefferson 2008/7/31 Steven G. Johnson <[EMAIL PROTECTED]>: > On Jul 30, 2008, at 9:20 PM, Jefferson Thomas wrote: >> What you are saying is really very helpful. In fact I had no idea I >> have to supply the symmetric >> source. I will try it right now and see how it goes. > > See: http://ab-initio.mit.edu/wiki/index.php/Exploiting_symmetry_in_Meep > > If you don't supply a symmetric source, in practice Meep will > symmetrize it for you, although it is better to have a symmetric > source so that you get the same results with and without symmetry > (this makes things easier to debug). However, the fact that you were > specifying an Ex source at y=0 but also specifying odd symmetry around > y=0 indicates that you either have a typo or were fundamentally > misunderstanding the relationship between symmetry and fields -- > either way, it is very important to correct. > >> On the other hand I dont understand the thing with the symmetric eps. >> Although I dont know how meep works internally with the symmetries I >> have been checking already many times >> what requests it sends to the eps function. From what I have found >> out, if there is any symmetry meep >> asks the eps function only about one relevant half of the volume. > > Yes, it only evaluates the eps function in *about* one half of the > volume. However, it evaluates eps for about one pixel beyond the > mirror plane too, and if your epsilon function isn't symmetric then > terrible things can happen. > >> So lets say I declare the calculation volume : >> volume v = vol3d(10,10,10, resolution); >> And then I specify the X symmetry : >> structure s(v, eps, pml(1.0), meep::mirror(meep::X, v)); >> Then all the vector requests that come to the eps function have x >> coordinate >> between 0 and 5 (or precisely between -0.xx and 5.xx but that is just >> a minor issue). > > The 5.xx is not a minor issue. It's exactly this tiny fraction beyond > the midpoint that is causing your problem since your eps function is > not symmetrical (as proved by the fact that the divergence disappears > when I symmetrized your eps function). > > Hence the divergences you observe starting at the boundary, because > the boundary conditions are internally inconsistent. > > Steven > > > _______________________________________________ > meep-discuss mailing list > [email protected] > http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss > _______________________________________________ meep-discuss mailing list [email protected] http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
