The ABINIT Users Mailing List ( CLOSED )

[Thierry Deutsch <tdeutsch@cea.fr>]

Mike,  Matthieu,

        I give some comments from my "numerical experience".

        The convergence of surface relaxations is a difficult problem.
Even with Si, it is sometimes difficult to have a good convergence (defects, 
surfaces).
I have tried to calculate some systems with ABINIT and CPMD.
Some systems do not converge with ABINIT and did with CPMD.
But the inverse is also true. So I dont give up hope to have a "perfect"  
program ;-).

CPMD has both schemes:
- diagonalisation+mixing as ABINIT which is the only method possible if there 
is no gap.
- direct minimisation of energy (which you call CP).

The last scheme is faster for semiconductor or insulator. Nevertheless, there 
are also some problems of convergence. I think it is easier to have a robust 
method to converge (DIIS+some controls) because the scheme is more simple.

So the solution is ti tune by hand, except if you want to discover and 
implement an efficient scheme.