[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Spin-dependent calculations and DOS resolution



Hi Dave,

We have been doing some spin polarized calculations recently for clusters with a couple thousand atoms and, to tell you the truth, I haven't noticed that broadening effect you talk about. In fact, if I'm correct, you control the broadening for each point of the LDOS with the eimag parameter. In your case that is set to 0.1 eV, so you can expect to have very little detail in your LDOS. The problem with the fine structure of the LDOS, according to what John explained to me once, is that it has to be consistent with the finite size of your cluster and that is related to the electron mean free path, if I remember correctly. You can decrease that parameter, but you need to go to larger clusters so the LDOS still makes sense. What we did for our calculations is to try several calculations for clusters of different sizes and different values of eimag ranging from 0.01 eV to 0.15 eV. We took the value of eimag for which the LDOS is the same for all the different clusters. I know that there is a formula (a very simple expression) to estimate this value beforehand (I think it is simply that the electron mean free path has to be much smaller than the size of your cluster, or something like that), but I never seem to find it when I need it. Maybe John could shed some ligth here :-)

Also, to perform the calculations I was talking about before, we've used the parallel MPI FEFF. It works magnificently. I've noticed though, that when we use a lot of processors (128 or 256), sometimes FEFF crashes. However, I'm not sure if this is a bug in FEFF or a problem with RMS, which is the queuing system used in this machine. Also, the MPI libraries used in this cluster are not the standard MPICH, but some proprietary libs by Compaq, so that could also be part of the problem. I don't even know if FEFF has been ever tested with so many processes and for such a demand for memory (like 50 Gb or something like that for each calculation). In general, though, it is very stable, it compiles without any problem, and runs very fast so, definitely, it's a worthy investment.

Hope this helps,

Angel
On Thursday, February 13, 2003, at 08:38  AM, David Eustace wrote:

Hi all,

I have just started doing spin dependent LDOS calculations on CrN, a simple rock salt structure. I get good correlation with published data, but a noticeable loss of fine structure compared with non-spin dependent calculations of the same size. I have already established the parameters which give the level of fine structure i'm after with the non-spin dependent case, and I'm simply repeating the calculation with the SPIN card added (repeating twice, of course, with SPIN 2 then SPIN -2). The "resolution" of the plot decreases.

Is this a natural manifestation of the SPIN card?
Is it simply a case of making the calculation bigger to get better resolution?

I should add I'm only interested in obtaining the spin up and spin down LDOS and XANES.

I include an example of one of my feff.inp files. Please point out if i'm being too simplistic in my approach to these calculations (I have a suspicion there must be more to it than this).

Also, I had sent out a question in an earlier email about the modular feff 8.2 version which can run in parallel, and i didn't get any response. The manual says that its still 'experimental'. Is there serious problems with it or does it currently run reasonably reliably? We're interested in obtaining the modular version but only if it is non-troublesome!! . Any comments at all would be appreciated.

Thanks for any response,

Dave Eustace <feff.inp>