[CPMD-list] slow convergence with ODIIS

[Axel Kohlmeyer]

On Wed, 14 Apr 2004 tangxin at student.dlut.edu.cn wrote:

TX> Dear CPMD community
TX>    I want to run an MD to simulate the surface of AlN. So I first optimized the
TX> wavefunction wiht ODIIS + vanderbilt type Psp . But the convergence is too slow.
TX> Should I use some another Keywords in my input ?
TX>    I have read "Slow convergence with surface calculations " in CPMD-list of
TX> 2002-August. But I don't know which keyword I should chooce. Could anyone give
TX> me some advice?

hello,

my guess is, that you are hit by two problems:
1) ODIIS works well (and very fast) in cases where it is comparatively 
easy to converge the wavefunction
2) vanderbilt uspps are a bit more tricky to converge, especially if
you are not in an equilibrium configuration.

the solution to 1) would be to try out

PCG MINIMIZE
TIMESTEP
 20.0

possible countermeasures to 2) are:
- try to improve the initial guess by assigning partial atomic charges.
  e.g. in case of water, adding the following lines in the &ATOMS
  section has improvemed convergence quite a bit for me.

ATOMIC CHARGES
   -0.6
    0.3

- try to make the initial guess worse by using 

INITIALIZE WAVEFUNCTION RANDOM

sometimes the initial guess gets you 'locked' in the wrong state.

- stop the wavefunction optimization with 'touch EXIT', when you have
reached the point, where it does not improve (or gets worse), and then
try to run some steps of geometry optimization or molecular dynamics
with the not fully converged wavefunction and re-optimize the 
wavefunction after that.

- use different pseudopotentials. cpmd v3.7.x with vanderbilt 
pseudopotentials has to be used with some care. not all of the 
functionality of cpmd is supported, sometimes cpmd crashes, 
sometimes it fails silently, thus giving you a wrong answer.
CP/BO molecular dynamics and geometry optimizations work well, 
though. 

hope this helps,
	axel kohlmeyer.

TX>   My input and output file for the AlN surface is as follows :

[...]

-- 
=======================================================================
Dr. Axel Kohlmeyer                        e-mail: axel.kohlmeyer at rub.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================