[CP2K:2443] convergence of energy&forces when using periodic decoupling
Teodoro Laino
teodor... at gmail.com
Thu Dec 3 10:38:44 UTC 2009
Hi Csilla,
Csilla Varnai wrote:
> Hi,
>
> I'm trying to get the energy and forces to converge in a periodic QM/
> MM simulation of water when using the periodic decoupling/recoupling
> scheme.
>
> First of all, I've played around with the *bloechl*inp test files in
> the QS directories to see how to set the parameters to get the best
> accuracy. I monitored the energy and the forces when changing the
> density cutoff (150..500 Ry), the number of grids applied (1..7), the
> Ewald precision(1E-3..1E-12) and the RCUT of the MULTIPOLE section,
> and the quantum cell size(3..13A) (please see below an example input
> file I used).
> Blochl suggests 5-7 A separation between periodic images is usually
> enough for the wavefunction separation, when the scheme works, and
> indeed the energy and forces converge beyond about 5 A separation
> (cutoff: 280 Ry, Ewald precision: 1E-8, Rcut: 3A), but the fluctuation
> in the forces is 5E-3 H/b in case of the oxygen (better for H). Also,
> increasing the density cutoff beyond 280 Ry has a 1E-3 H/b fluctuation
> on the oxygen force. Changing the number of grids used may cause 2E-3
> H/b difference in the forces (is this because I didn't reach
> convergence wrt. other parameters or due to inaccuracies in
> interpolation between grids?). The energy fluctuation beyond
> convergence in each case is less than 0.2 mH.
>
changing number of grids is EQUIVALENT to changing the basis set. The
convergence on this point
is less trivial than the convergence with the density cutoff. Try to
play as well with the REL_CUTOFF, but
again don't expect a linear convergence.
> When I go to a system with 1 QM water molecule in a box of 1221 MM
> waters, and I use the decoupling + recoupling scheme (QMMM%PERIODIC
> section) expectedly the energy fluctuation increases up to 0.5 mH, the
> amplitude of the force fluctuation up to 10E-3 H/b (~0.5eV/A), which
> is not acceptable for my calculations.
>
For QMMM periodic you have also the problem of mapping the long-range
part of the electrostatic potential on the coarsest grid level. If this
level is not fine enough you may get troubles. You need to check that
the grid level is accurate enough compared to the number of gaussian you
use in the GEEP expansion (which determines the smoothness of the
long-range component of the electrostatic potential). If you use many
grid levels the coarsest could be too coarse to host the long-range
potential.
Periodic QM/MM (coupling/recoupling) needs a particular attention to
these issues and may not be trivial at the first trial.
Teo
> I would greatly appreciate if anyone could give me suggestions on how
> to improve the force convergence.
>
> Thank you in advance,
> Csilla
>
>
> Here's an example input:
>
> &FORCE_EVAL
> METHOD Quickstep
> &DFT
> BASIS_SET_FILE_NAME ../BASIS_SET
> POTENTIAL_FILE_NAME ../POTENTIAL
> &MGRID
> CUTOFF 280
> &END MGRID
> &POISSON
> PERIODIC NONE
> POISSON_SOLVER MULTIPOLE
> &MULTIPOLE
> EWALD_PRECISION 0.00000001
> RCUT 3.0
> ANALYTICAL_GTERM
> &END MULTIPOLE
> &END POISSON
> &QS
> EPS_DEFAULT 1.0E-12
> &END QS
> &SCF
> EPS_SCF 1.0E-6
> SCF_GUESS ATOMIC
> &END SCF
> &XC
> &XC_FUNCTIONAL Pade
> &END XC_FUNCTIONAL
> &END XC
> &END DFT
> &SUBSYS
> &CELL
> ABC 10.0 10.0 10.0
> &END CELL
> &COORD
> O 4.000000 4.000000 4.000000
> H 4.000000 3.250000 4.520000
> H 4.000000 4.750000 4.520000
> &END COORD
> &TOPOLOGY
> &CENTER_COORDINATES
> &END CENTER_COORDINATES
> &END TOPOLOGY
> &KIND H
> BASIS_SET DZVP-GTH-PADE
> POTENTIAL GTH-PADE-q1
> &END KIND
> &KIND O
> BASIS_SET DZVP-GTH-PADE
> POTENTIAL GTH-PADE-q6
> &END KIND
> &END SUBSYS
> &END FORCE_EVAL
> &GLOBAL
> RUN_TYPE MD
> PROJECT H2O_cell10_rcut3_prec8_cutoff280
> PRINT_LEVEL MEDIUM
> &END GLOBAL
> &MOTION
> &PRINT
> &FORCES
> FORMAT XMOL
> &END FORCES
> &END PRINT
> &MD
> ENSEMBLE NVE
> STEPS 1
> TIMESTEP 0.1
> TEMPERATURE 300.0
> &END MD
> - &END MOTION
>
>
> --
>
> You received this message because you are subscribed to the Google Groups "cp2k" group.
> To post to this group, send email to cp... at googlegroups.com.
> To unsubscribe from this group, send email to cp2k+uns... at googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/cp2k?hl=en.
>
>
>
More information about the CP2K-user
mailing list