[Pw_forum] Total energy diverges in cell relaxation

[Axel Kohlmeyer]

On Thu, 16 Oct 2008, Hui Tang wrote:

hi hui,

your cell is expanding quite a bit and your cutoff is not 
that large. with an expanding cell, your _effective_ cutoff 
becomes much smaller as the number of plane waves remains
constant during a variable cell calculation. have you tested 
the stess tensor convergence vs. the cutoff? you basically
have to figure out how many plane waves you need for a well
converged stress tensor at the largest expansion of your cell 
and then you have to set the cutoff at the beginning so that 
it results in the same number of plane waves...

that would explain why it works (better) if you restart.

also, did you test the pseudopotentials for ghost states?

cheers,
   axel.

p.s.: please provide an affiliation. thanks.

HT> Hi Nic,
HT> 
HT> 1)  I have tested the pseudo-potentials (psps) with small atomic
HT> separations. The pseudo-potentials give reasonable till two atoms are as
HT> close as 0.8 A. When the total energy starts to diverge, which is about
HT> Ionic iteration 17, the smallest inter-atomic distance is about 1.4 A. So I
HT> do not think the breakdown of the psps is the trigger of this divergence of
HT> the total energy, although it should be the reason for the enormous large
HT> negative energies.
HT> 
HT> 2) I have tried to restart the calculation from where the divergence starts.
HT> Once I did that, the first scf calculation would give a reasonable energy
HT> around -41.7 Ry, and the calculation could run smoothly till the total
HT> energy converges most of the times. So what I deal with this problem right
HT> now is to restart calculations from where they start to have diverged total
HT> energies and it's been working well. However, I hope this is not the optimal
HT> way to solve this problem. :)
HT> 
HT> Thank you very much,
HT> Hui
HT> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.