[CPMD-list] Convergence problem with USPP

[Linas Vilciauskas]

Dear CPMD community,

I'm a novice in a CPMD business, so just forgive me if my question
might sound kinda silly for some of you... I'm having a problem when
trying to optimize the geometry of a molecular cluster using the
ultra-soft VDB pseudopotentials. I tried different cut-offs between 20
and 50 Ry. However the geometry optimization converged only with
certain cut-offs (namely 20 and 50 Ry) but didn't with any of
intermediate values (wavefunction doesn't converge in the 1st step)...
Tried different GC-CUTOFF, DUAL values as well as different
CONVERGENCE (ORBITALS, ENERGY, INITIAL) criteria but no luck either.
Just wondering whether I'm doing something incorrectly or there is a
problem with USPPs for certain atoms/combinations? Does anybody have
some ideas?

System info:
CPMD 3.11.1
OS: IBM AIX 5.2 L
FCC: IBM XL 8.1.1 (OpenMP only)

Sample input:
&INFO
&END

&CPMD
 OPTIMIZE GEOMETRY XYZ
 STRUCTURE BONDS ANGLES DIHEDRALS
 PCG MINIMIZE
 ODIIS NO_RESET=30
  10
 MAXSTEP
  1000
&END

&DFT
 FUNCTIONAL BLYP
 GC-CUTOFF
  5.D-6
&END

&SYSTEM
 SYMMETRY
  0
 ISOLATED MOLECULE
 ANGSTROM
 CELL
  8.0 1.0 1.0 0.0 0.0 0.0
 CUTOFF
  25.0
 DUAL
  5.0
&END

&ATOMS
*P_VDB_BLYP.psp FORMATTED
...............
*O_VDB_BLYP.psp FORMATTED
...............
*H_VDB_BLYP.psp FORMATTED
..............
&END


 --
Linas Vilciauskas
Max-Planck-Institut für Festkörperforschung
Heisenbergstraße 1
D-70569 Stuttgart, Germany