Dealing with the unconverged problem

[zhuanghl]

Dear all,as is known that it is better to calculae the magnetic system without spin-polarized first,and then based on whose WAVECAR and CHARG to further calculate the spin-polarized case.Right?Here, I still have some questions:
1.Could somebody tell the physical meaning and validity of this operation;
2.In the workshop of the Vasp,it is said the WAVECAR should be removed for further calculation;However,in the manual I guess the WAVECAR should be kept?What is the right one?
3.I try the operation,an error exists:
LDA part: xc-table for Ceperly-Alder, Vosko type interpolation para-ferro
found WAVECAR, reading the header
nup: number of bands has changed, file: 129 present: 153
POSCAR, INCAR and KPOINTS ok, starting setup
FFT: planning ... 1
reading WAVECAR
charge-density read from file:
magnetization density of overlapping atoms calculated
WARNING: random wavefunctions but no delay for mixing, default for NELMDL
Does this waring affect the accuracy,and how to solve it;
PS:I alway use only one CPU;
4.In the case of relaxation,should I relax the system without spin-polarized first,and then relax again from the existed WAVECAR and CHARG?And,what is the input POSCAR?

Many thanks!Any advice are welcomed!

[ericsun]

1. I think this procedure is just for "technical" considerations, rather than "physical". So, it's valid is only because it works in many cases.
2. In most cases, using the WAVECAR from a previous relevant calculation may
speed up your new calculations. However, calculating from scratch should
be always valid.
3. I didn't see the error. For the warning, I think you can manually set
"NELML" as mentioned in the warning itself to add some delay for mixing.
4. For this kind of questions, I think the answer is "to try both and see if
there is difference". If no difference, the question is answered.
<span class='smallblacktext'>[ Edited Sun Jan 08 2006, 04:12AM ]</span>

[admin]

1) this is ONLY meant to speed up the calculations, there is no special physics behind this recipe: it is much faster
to run a non-polarized calculation and usually the pre-converged non-magnetic result is a reasonable starting guess for the spin-polarized system. Therefore it helps to save computing time to switch on spin-polarisation at a late stage of the calculation.
2) do you refer to p15 of handsonI.pdf? for such a small system (spin-polarized O atom ) there is no considerable speedup expected whether you start from a non-magnetic WAVECAR or from scratch.
3) no, this warning does not affect the accuracy; it may be helpful to increase NELMDL if the electronic convergence of the calculation is poor.
4) yes, especially for rather complex systems it may speed up to pre-converge without spin-polarisation first. To continue, use WAVECAR, CHGCAR (as described in the online handbook) and copy CONTCAR of the last ionic step to POSCAR. In any case, please check the convergence of the geometry carefully after magnetization is turned on.