The ABINIT Users Mailing List ( CLOSED )

["Zhang Mike" <mz24cn@hotmail.com>]

I often hear from the forum that the ABINIT failed to converge the 
calculations, which occusionally happens in surface relaxation 
calculations.

Due to my recent work on oxide surface calculations, I find some failures 
on the convergence is associated with the code of the symmetry analysis.

As the most usual setup for parameters in surface calculations, optcell 
equals to 0, allowing the full relaxation of atom positions and to preserve 
bulk cell shape and dimension. Value zero is also the value that tutorial 
files adopt. Another value of optcell, 2, may not be proper. As the manual 
said, "it takes into account the symmetry of the system, so that only the 
effectively relevant degrees of freedom are optimized." 

As we know, SYMMETRY-BREAKING often occurs on crystalline surfaces. For 
example, surface atoms at the edges of unit cells may depart the bulk 
equilibrium positions and move towards one direction, often attracting the 
surface bridging atoms to depart the equilibrium positions, too. That is to 
say, all surface atoms at the edges may move one direction, leading to a 
long-range symmetry breaking, thus occusionally giving rise to surface 
reconstruction to compensate the long-range losing of the symmetry. 

Keeping optcell=0, I have tried many calculations on surface relaxation. I 
find the symmetry-breaking, however, can not be reflected in the converged 
result. This is maybe derived from two reasons. One is optcell=0 also takes 
the symmetry of the surface slab into account, just like optcell=2. The 
other is the method of the molecular dynamics simulation applied in ABINIT 
is stress driving, not the C-P method (simulated anneal method). As for the 
former, by primarily analyzing the code of moldyn.f, I excludeed it. So the 
reason must be the later. Because one symmetry corresponds to an 
equilibrium of the force, so stress-driving method never allows the 
symmetry-breaking result. In fact, the symmetry breaking is often derived 
from the perturbation, which encourages the surface atoms departing the 
equilibrium position (sub-stable) and reaching a lower total energy 
(stable). So, the total energy minimization applied in ABINIT may only 
aquire a local minimum in symmetry-breaking surface slab calculations, NOT 
the global minimum. Namely, the majority of results of the surface 
relaxation calculations may be QUESTIONABLE. Because the original 
(symmetry-preserving) positions is sub-stable for surface atoms, that leads 
to some difficulty to converge the forces and the difference of the forces, 
maybe which account for the failure to converge the calculations. In fact, 
when the number of atoms is larger than 24-64, the convergence is more and 
more difficult. 

As we know, the C-P method is an effective method to acquire a global 
minimum of the total energy. In a word, I wish strongly ABINIT could 
implement C-P algorithm AS SOON AS POSSIBLE.

If something is wrong in what I said, don't hesitate to point out it. Thank 
you.

Good continuation,
Mike Zhang

Tel   : +86-551-5591377(O),5592922(H)                
Fax   : +86-551-5591310                              
Email : mz24cn@hotmail.com                           




_________________________________________________________________
免费下载 MSN Explorer:   http://explorer.msn.com/lccn/