[CPMD-list] slow geometry optimization - would underrelaxation help?

[Wolfgang Fritsch]

Dear experts,

I am trying to optimize the geometry of a CuGaSe2 cell. It seems to me 
that the convergence is awfully slow and hence wonder whether I am 
missing something. "Slow" means that with the 16 atoms in the cell, 
after a few hundreds of geometry steps convergence is not reached and 
actually, when I check the positions over optimization steps, it appears 
that they move for a while in one direction but then change into 
another, with no clear indication for final convergence yet.

I guess I am missing something like an "under-relaxation factor" which 
would allow to change the coordinates of atoms only a given fraction of 
what happens now - leading to smaller steps but less jiggling forth and 
back. I remember the benevolent effect of such underrelaxation factors 
from codes in fluid dynamics.

I append two sets of input. Maybe someone in the knowing sees my 
blunders right away. The first set is the one I am using now.

The second set I had to abandon as it let the code stall: it runs on, 
say, 3 nodes with 4 cpus each but after a few minutes one of the cpus on 
each node would go idle and no further progress in the calculation would 
ensue (other cpus busy but no further output).

I have tried another method, the steepest descend, but then progress in 
the optimization would be even slower.

Thank you for your attention!

Wolfgang Fritsch

set 1:

&CPMD
   OPTIMIZE GEOMETRY
   RESTART WAVEFUNCTION GEOFILE
   CONVERGENCE GEOMETRY
     0.0001
   CONVERGENCE ORBITALS
     0.00005
   CONVERGENCE CELL
     1.0
   PCG MINIMIZE
   TIMESTEP
     3
   MAXSTEP
   40000
   MEMORY BIG
&END
&DFT
   FUNCTIONAL BLYP
&END
&SYSTEM
   SYMMETRY
    8
   CUTOFF
   110.0
   CELL
   10.6 1.0 1.961 0.0 0.0 0.0
&END

&ATOMS
*Cu_MT_BLYP_NLCC
  LMAX=D SKIP=P LOC=S
  4
  0.0          0.0          0.0
  5.3101       0.0          5.1967
  5.3101       5.3101       10.3935
  0.0          5.3101       15.5902
*Ga_MT_GIA_BLYP.psp
  LMAX=D LOC=S
  4
  5.3101       5.3101       0.0
  0.0          5.3101       5.1967
  0.0          0.0          10.3935
  5.3101       0.0          15.5902
*Se_MT_BLYP.psp
  LMAX=D LOC=D
  8
  2.6752       7.6524       2.6161
......

  CONSTRAINTS
   FIX ATOMS
    1
   1
  END CONSTRAINTS

&END

========================================================

set 2

&CPMD
   OPTIMIZE GEOMETRY
   RESTART WAVEFUNCTION GEOFILE
   CONVERGENCE GEOMETRY
     0.0004
   CONVERGENCE ORBITALS
     0.00005
   CONVERGENCE CELL
     1.0
   CONVERGENCE ADAPT
     0.02
   CONVERGENCE ENERGY
     0.05
   CONVERGENCE CALFOR
     2.5
   CONVERGENCE RELAX
     5
   PCG MINIMIZE
   TIMESTEP
     3
   MAXSTEP
   40000
   MEMORY BIG
&END
rest as above...