[CPMD-list] path integral: normal & staging modes; electron thermostat

[Lan-Feng Yuan (lyuan@Princeton.EDU)]

  Dear Axel and CPMD users,

> LY>    There are 2 keywords "NORMAL MODES" and "STAGING", 
> corresponding to
> LY> using normal modes and staging transforms instead of primitive
> LY> coordinates. But when I try to use either of them (with a 
> "1.0" in the
> LY> line below to determine masses), the calculation always 
> crashes after
> LY> the 1st step:

> i cannot tell. i am no expert on path-integrals (there is one
> a few offices down the hall, but he is not available currently).
> would you mind posting your input, so we can check whether there
> is a technical problem, i.e. it runs fine on a different machine.

  Thanks a lot. The input is:

&CPMD
 CENTER MOLECULE
 ISOLATED MOLECULE
 MOLECULAR DYNAMICS
 PATH INTEGRAL
 CONVERGENCE ORBITALS
  1.d-6
 EMASS
  340.0
 TIMESTEP
  5.
 NOSE IONS MASSIVE
  300.0  2000.0
 NOSE PARAMETER
  4 0 0 6. 7 4
 MAXSTEP
  2000
 TRAJECTORY XYZ
 RESTART COORDINATES CELL WAVEFUNCTIONS VELOCITIES ACCUMULATORS NOSEP NOSEE LATEST
&END

&PIMD
 TROTTER DIMENSION
  16
 STAGING
  1.
&END

&DFT
 NEWCODE
 FUNCTIONAL PBE
&END

&SYSTEM
 ANGSTROM
 SYMMETRY
  1
 CELL
  10.000 1. 1.    0.0 0.0 0.0
 CUTOFF
  70.00
&END

&ATOMS
 ISOTOPE
  15.994915
  1.007825
  1.007825
*O_MT_PBE             KLEINMAN-BYLANDER
  LMAX=P
1
4.962387678251      4.942387678251      5.000000000000
*H_MT_PBE.psp              KLEINMAN-BYLANDER
  LMAX=S
1
5.920940797096      4.834804493524      5.000000000000
*H_MT_PBE.psp              KLEINMAN-BYLANDER
  LMAX=S
1
 4.834804493524      5.920940797096      5.000000000000
&END

  And the 16 initial RESTART files are generated by:

&CPMD
 CENTER MOLECULE
 ISOLATED MOLECULE
 OPTIMIZE WAVEFUNCTION
 CONVERGENCE ORBITALS
 1.d-6
 PATH INTEGRAL
&END

&PIMD
 INITIALIZATION
 GENERATE REPLICAS
 DEBROGLIE CENTROID
  300.0
 TROTTER DIMENSION
  16
&END

  (and the same sections of DFT, SYSTEMS, and ATOMS)

> LY>   The other question is whether I should use electronic 
> thermostat for
> LY> insulators. For a normal Car-Parrinello MD, there is no need. 
> Is there
> LY> some reason that it is needed for PIMD?
> 
> yes, there is. the movement of your replica is in the same frequency
> region than that of your fictitious electron dynamics, so you need
> careful thermostatting to maintain adiabacity. the only alternative
> would be a path integral BO-MD...

  Aha ... So how to do it? First let the system go without electron thermostat for several tens to 100 steps, then pick the average EKINC/P? And the best choice for its frequency is ~ 3 times of the highest vibrational frequency of ions, right?

  Thank you very much! Best regards,

  Lan-Feng