On Thu, 9 Nov 2006, Ashifi wrote:
Thanks for the previous output... I think I'm a step closer to getting
Meep units. I have a few quick questions about geometries and PML.
The code has been attached, or can be seen at:
http://pastebin.us/8034
You are giving all of your distance in units of meters, which is fine.
But then your resolution is 10, which means 10 pixels per meter (i.e.
pixel = 0.1m), which is far too small: it means your wavelength is less
than 1 pixel, and also that your block is less than 1 pixel in size.
You absolutely must have a resolution so that your wavelength is several
pixels long (at least!), otherwise you are going to get garbage.
Is there an issue with specifying blocks with very large (but not
infinite) aspect ratios?
No. But you should think carefully about what resolution you need.
(It is possible to have things with subpixel sizes, because of the
subpixel averaging that Meep does. Whether such subpixel structures are
handled sufficiently accurately depends on the problem, however)
On a different note, what is the difference between using 'no-size' and
zero, when specifying source widths, geometries, flux planes etc?
no-size should only be used for the geometry-lattice. no-size is not zero
(which would cause problems in defining the basis), it is slightly nonzero
(1e-20).
A size of zero in a source width, flux plane, etcetera should be thought
of as a delta function. (e.g. a delta function source, or a flux plane
that interpolates the fields to that point). A size of 1e-20 in the same
context, however, corresponds to a small but finite size (e.g. integration
of the flux over a thickness 1e-20, which will result in a tiny number).
Lastly, is there such a thing as having too much PML if your cell size
is large enough for the wavelengths under consideration (10x lambda
in meep units)? It's not easy to tell when the PML is working when
looking at a sequence of png images (or a gif animation), so I would
prefer to err on the side of caution and put "too much" PML instead of
too little. I recall in a previous thread that pml = 0.1*lambda was too
thin. Is there a general guideline for PML thickness?
It depends on your structure. As a rule of thumb, a PML thickness
comparable to half a wavelength usually works pretty well. Ultimately,
however, the right thing is to keep doubling the PML until your answer
stops changing by more than the accuracy you are interested in.
In your case, however, your PML is going to be messed up simply because
your resolution is far too low, as I mentioned above.
Steven
_______________________________________________
meep-discuss mailing list
[email protected]
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
