[Wien] atoms lose and dstart failed in a large unit cell

[Peter Blaha]

I have done a full structural relaxation for 1106 atoms (BN/Rh(111) 
nanomesh) on a 64-node Xeon cluster with infiniband.

For such big calculations you have to "optimize" your calculational 
parameters. Using the default RMT=2.0 and RKMAX=7 will not allow you to 
do such things. One should determine the minimal parameters by eg. a 
force minimization on a smaller cell and determine the "minimal RKMAX", 
for which your positions are converged to eg. 0.02 Ang. In my case 
above, RKMAX=3.5 was sufficient, but this has to be checked out for each 
case.
In addition, iterative diagonalization was used, which is much faster 
(for ALL calculations with a matrix size above 1000 !!! I'm considering 
to make -it the default option, since I've the feeling that most users 
do not adapt to new non-default options ???)

Checking your qsub script it is maybe ok but:
a) It does not use iterative diagonalization (but use version 8.1 for that)
b) it uses all 32 processors in parallel for a single k-point. Do you 
have just ONE k-point in this example ? If not, use a more complicated 
.machines file (see examples on www.wien2k.at/faq)
c) 32 processors are usually not optimal. SCALAPACK may work best for 
"square" meshes like 4x4 (16), 6x6 (36),.... but not 8x4=32 ! But that 
may depend on the implementation of scalapack  and the queue-limits
on your machine.

Naturally the most recent version is better, in particular for "extreme" 
calculations. Check www.wien2k.at for the differences.


> I have got some successful results on different smaller unit cells. 
> Recently, I come to do some big unit cells using parallel mode based on a super computer. The current version of wien2k is WIEN2k_7.3 downloaded in last October. Do you think it is too old to do calculation for a large unit cell? the newest V08-1 will will work on this big unit cell? Do you know what is the biggest case working in wien2k so far?
> 
> I am sorry that I have to use this big unit cell(the unit cell is the relaxed structure from VASP.) to make the physical problem meaningful. 
>