On 11/30/2017 02:21 PM, Da Silva, Jaime wrote:
I'm designing a large ring resonator (200µm diameter) in the visible
spectrum (640nm). My 2D simulations have been running for nearly 20h
(meep time 11,000) on HPC, using 12 nodes (36 cores/node). I believe
There are three things you can try to speed your computation.
1) Since the index contrast of your ring resonator is relatively small,
the Q of the resonant mode will likely also be small. Also, regardless
of whether the Q is small or not, harminv is still able to resolve Qs of
1e6 given only a few periods of the field data (see Section 5.2 and Fig.
9 of the Meep paper). This means that you don't need to run your
simulation for too long after the sources have turned off for harminv to
obtain an accurate result. You can facilitate harminv's performance by
using a narrow-band source as has been discussed previously on this list.
2) Your ring structure has no sharp corners/edges (in 2d, at least) and
is a lossless dielectric (i.e., refractive index is purely real). In
this case, Meep's subpixel smoothing enables you to reduce the
resolution while still ensuring good accuracy.
3) Due to the four-fold symmetry of the ring structure, you can use two
mirror symmetry planes to reduce the size of the computation by a factor
of 4. To use this feature, you will need to use 4 separate point/line
sources. Note that both the structure and the sources/fields must
respect the symmetries that are defined. It might be worthwhile to
review this topic in the documentation
(http://meep.readthedocs.io/en/latest/Exploiting_Symmetry/).
Finally, if this may be relevant, you should be setting (set-param!
force-complex-fields? true) in order to obtain time-average values,
rather than snapshots, for the energy. This is because Meep uses real
fields by default.
_______________________________________________
meep-discuss mailing list
meep-discuss@ab-initio.mit.edu
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
