convergence problem

Message

TAT · #1 Post by **TAT** » Wed Nov 19, 2008 4:27 am

Hi,
I am trying to run Ho2O3 with the PAW potentials that considers F electrons as Valence (Not the standard PAW potential provided with VASP package) that we requested. Unfortunately the run doesn't converge within even 60 electronic SC steps. I tried too change the algorithm with Normal and Very Fast, it didn't work. I set LDIAG to .FALSE, it also didn't work. The energy keeps on oscillating. I didn't touch the mixing parameters as I am not sure about what values to choose so I take the default values. It will be nice to have some suggestions.

prithwish · #2 Post by **prithwish** » Thu Nov 20, 2008 3:32 am

sometimes, the way you have compiled the VASP executable also matters. From my experience, I have seen the very same input is getting converged when VASP is compiled with Intel9 mkl compiler while the its not converging at all with an executable compiled with Intel 10 mkl compiler. Its surprising but a fact.

TAT · #3 Post by **TAT** » Fri Nov 21, 2008 2:31 am

I use Intel mkl I think 9.1.023. Actually I tried PAW potentials for other rare earth elements treating F electrons as valence, but they also don't work. There is no problem at all with PAW potentials treating F electrons as Core.

lcyin · #4 Post by **lcyin** » Mon Nov 24, 2008 3:36 am

According the VASP manual (page 134, Version March 1 2007), PAW potential which only treat the f orbitals as valence orbitals are availabe for La, Ce, Ac, Th, Pa, U, Np and Pu. Pls. check the manual. Hopefully this maybe helpful.

lcyin · #5 Post by **lcyin** » Mon Nov 24, 2008 3:41 am

BTW, I do not think that different MKL or different compiler can results totally different scf convergence performance.

TAT · #6 Post by **TAT** » Tue Nov 25, 2008 5:27 am

One can look into the V_RHFIN file and figure out which electrons are treated as valence electrons and what are core electrons. I agree with you on the point that different MKL or different compiler should not results totally different scf convergence performance. VASP package doesnot contain some of the potentials that I mentioned. We got them on special request.

alex · #7 Post by **alex** » Wed Nov 26, 2008 2:03 pm

Hi TAT,

you may wish to check the posts about Ce2O3, a similar situated challange. One f state ... Looks like a lot of fun. ;-)

Alex

#8 Post by **admin** » Wed Nov 26, 2008 2:33 pm

sorry, if you do not take the PPs from the data set to be downloaded from our ftp server, there is no way to tell anything about convergence,... Please check if your calculation converges using the standard PPs (with the f-electrons either in core or in valence) . V_RHFIN only holds for the PPs which can be downloaded together with this file.
please check whether the f-elecrons in your system are localized or not.
concerning the compilers: please use one of the stable compiler releases of the INTEL compiler (9.1, for example).