ian, i was doing a lot of supercell calculations last year - at 3d supercell sizes large enough to capture the disorder, and reasonable enough resolution to accurately capture the orientation of your particles, you are going to run up against the memory limit very quickly.
to give you a rough data point, i was calculating states just above the 8th band of the inverse opal structure (plain old spheres, mind you) - i used a 3 x 3 x 3 supercell and set resolution = 16 and num-bands = 217 (since 216 = 8 * 3^3). this calculation used just over 1 GB of memory; i was running this on computers with 2GB of memory each, which is is the maximum addressable memory if you're using a 32 bit os. in reality you are limited to less because of operating system overhead, etc. you might be able to gain something here (factor of 1/2?) by using the inversion symmetry flag at compile time, but even so, you will have a real hard time getting to a larger supercell before you start using too much memory, and the whole thing screeches to a crawl due to constant page faults. redefining your basis vectors doesn't gain you anything, because the actual periodicity of the crystal is still the same. i think all this will do is rescale the "freqs" output by some constant factor, which you will then have to divide back out eventually anyway. you could possibly get larger supercells running in parallel on a shared-memory system, but again the memory requirements scale like n^3 so even then you'll run out very quickly. so, i see a couple of options for you: 1) try to get the target-freq function to work. this has been discussed previously on this list, and i have never had good luck in getting it to converge, but reducing the number of bands you are calculating to a reasonable number is a good thing to be able to do. 2) use a shell script (or perl, or whatever your language of choice is) to auto-generate a bunch of .ctl files that have reasonably-sized structures with a newly randomized orientation at each iteration. run all of these in sequence (or in batch, if you have multiple computers) and apply some statistical techniques to make sense of the output. i would probably do #2. good luck! nate "Ian D. Hosein" <[EMAIL PROTECTED]> wrote: > MPB Users & Steven, > I have a question about calculating bandstructures for supercells. > Starting with an primitive FCC unit cell with a basis of one, I calculate > the freqs for 10 bands and find the gap between the 8th and 9th bands > (inverted FCC using silicon as the interstitial material, air for the > spheres). Now if a double the primitive unit cell dimensions and fill in > the other 7 lattice points in the cell with spheres, I would have to > calculate atleast 65 bands, since the 8th band times n^3 where n=2 is > 8x8=64, in order to see the gap. Obviously if I increase the supercell > size, the number of bands I have to calculate goes up significantly. > Now my specific problem is more complicated. In my supercell I have > asymmetric shaped particles (ellipsoids) all randomly oriented on their > lattice sites. This is called a rotator phase (particles positioned on a > lattice but with orientation disorder) and I'm trying to calculate the band > structure for it. I do this by making a super-"primitve" FCC unit cell of n > x n x n larger than the primitve unit cell of FCC and filling in the lattice > sites with randomly oriented particles. Obviously the larger the supercell > the more "random" the structure is and it will model the actual real > structure more accurately. My question is, how many bands, assuming the gap > still exists, would I have to calculate for this structure? Would it: > > 1) follow the same principle as the spheres? Or, > 2) since the particles are all different, I would just define my supercell > edges as the lattive vectors and the particles inside of it as one huge > basis for this larger crystal? > Would option two allow me to calculate less bands, or would I still have > to calculate many bands in order to see gaps. I'm just trying to figure out > how to handle such a super cell calculation, were the particles are all on > lattice sites, but due to the random orientation it is not a true crystal, > and so I need to define a much larger periodic cell in order to do the > calculations. > > Any thoughts or suggestions could be appreciate. > > Thanks, > > Ian > > > > > > _______________________________________________ > mpb-discuss mailing list > mpb-discuss@ab-initio.mit.edu > http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/mpb-discuss > =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= nate lipkowitz 416 823 8057 [EMAIL PROTECTED] _______________________________________________ mpb-discuss mailing list mpb-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/mpb-discuss
