[Wien] a question on convergence

Stefaan Cottenier Stefaan.Cottenier at fys.kuleuven.ac.be
Tue Jun 24 08:38:17 CEST 2003


> I have a question about the Al. Why the program set the core charge of the 
> Al to 4 and not to 10 in the initialization step. It is normal or should i 
> espect some problems in the calculations?

10 electrons are contained in the "core" which is defined as the largest noble 
gas atom (here Ne) that fits into the atom. But this is NOT the definition of 
core as used by wien2k: all electrons that are so strongly bound that their 
wave functions are confined entirely inside the muffin tin sphere. For Al, 
these are only 1s and 2s, hence 4 electrons. Everything normal.
 
> Could you tell me if the following problem has anything to do with the Al: 

No, it is not related to the above 'problem'.

> I' am working with a 8 atoms primitive cell (7 of Si and 1 of Al). I start a
> first calculation with 200 k points in the kgen program. After 6 iterations 
> the charge convergence test begin to oscilates as well as the :DIS in the 
> SCF file. So I follow your instruction and set the broyden factor to 0.1 in 
> the case.inm fie. 

At this point, make sure you also remove the broyden files (rm *.broy*)

> It certanly improve the convergence and the program was 
> able to make 20 iterations without the oscilation. But the oscilation began 
> after that 20 iterations.
> Then i tried to reduce the number of k-points to 50. After 7 iterations it 
> shows the error 'Segmentation fault' in the lapw2 program. Should I 
> incecrese the size of matrices in the global parameters.

Why did you REDUCE the k-mesh? If we temporarily assume that a mixing factor 
of 0.1 is all right (it probably is, although you have to consider 0.02-0.05 
as well if it still doesn't work), then the next possible reason is a too 
small k-mesh, or a too small basis set. My feeling is that 200 k-points (in 
the FBZ ?) is quite small for such a cell. At least double the number of 
points (in the IBZ) and try again (first save and do dstart again, to get rid 
of traces from your previous trials).

Stefaan




More information about the Wien mailing list