[CP2K:7883] SCF, MD run-time verses atomic species.
Simiam Ghan
simia... at gmail.com
Wed Aug 17 23:08:12 UTC 2016
Hello Juerg,
Thank you for your reply. I added REFERENCE_C9_TERM .TRUE. to the vdw
section as you suggested and found that my MD simulation became twice (!)
as fast. That is, time per MD step went from ~60 sec to ~30 sec for box
of water with 64 molecules and K,Cl ions. I made a quick comparison of
Energy_Force results with and without this flag and indeed forces and
energies do not change significantly. This is great news. Why is this
not a default? What's the catch? Also, would you say this is now a
correct way to declare a PBE-D3 setup?
&VDW_POTENTIAL
POTENTIAL_TYPE PAIR_POTENTIAL
&PAIR_POTENTIAL
R_CUTOFF 1.5000000000000005E+01
TYPE DFTD3
PARAMETER_FILE_NAME dftd3.dat
REFERENCE_FUNCTIONAL PBE
CALCULATE_C9_TERM T
REFERENCE_C9_TERM T
&END PAIR_POTENTIAL
&END VDW_POTENTIAL
Finally, I mentioned earlier that 30 scf steps were needed to converge the
same box of water with ions. I was then using EPS_SCF = 1E-06. Reducing
this to EPS_SCF = 1E-05 brought the number down to 10-20 scf necessary.
Does that still sound high? Is the definition of EPS_SCF documented
somewhere, as it is apparently not just Energy convergence.
Cheers,
Simiam Ghan
On Friday, July 1, 2016 at 3:10:52 PM UTC+3, jgh wrote:
>
> Hi
>
> no this shouldn't be, but without more information I will have to guess.
> You could also have a look at the timings at the end of the output to
> see if some routines got slower or if all parts of the run were affected.
>
> Two things to consider:
>
> 1) Use REFERENCE_C9_TERM TRUE in order to reduce the time for vdW in MD.
> 2) 30 SCF iterations in MD for such a simple system is pointing to a
> problem
> with your setup.
>
> regards
>
> Juerg
> --------------------------------------------------------------
> Juerg Hutter Phone : ++41 44 635 4491
> Institut für Chemie C FAX : ++41 44 635 6838
> Universität Zürich E-mail: hut... at chem.uzh.ch
> <javascript:>
> Winterthurerstrasse 190
> CH-8057 Zürich, Switzerland
> ---------------------------------------------------------------
>
> -----cp... at googlegroups.com <javascript:> wrote: -----To: cp2k <
> cp... at googlegroups.com <javascript:>>
> From: Simiam Ghan
> Sent by: cp... at googlegroups.com <javascript:>
> Date: 07/01/2016 01:38PM
> Subject: [CP2K:7883] SCF, MD run-time verses atomic species.
>
> Dear all,I am running NVT MD with Quickstep on a box of water with 64
> molecules. If I replace a water molecule with a KCl ion pair, i observe
> that the MD and SCF step times more than double on my setup. (MD from ~20
> sec to ~46 sec). SCF iterations are converging (except the very first MD
> step) in around 30 steps in both cases but each SCF now takes over twice as
> long as before. Is there an explanation of why such a 'small' change in
> system could double the run time? Is KCl really so heavy to calculate
> compared to H2O? The number of electrons in the system increases by 8,
> from 512 to 520. Below my KCl input file.
> Greetings,Simiam
>
>
> &GLOBAL
> PROJECT H2O_KCl RUN_TYPE MD PRINT_LEVEL MEDIUM
> &END GLOBAL
> &FORCE_EVAL
> METHOD Quickstep ! GPW method.
> &SUBSYS ! A subsystem:
> coordinates, topology, molecules and cell.
> &CELL ! Supercell setup.
> ABC [angstrom] 12.414 12.414 12.414 ! Using 64 H2O molecules, we
> thus get a density of 1g/cm^3. PERIODIC XYZ
> ! Use PBC in all dimensions. &END CELL
> &COORD UNIT angstromH -0.567712 -0.469646 -0.645913H 0.626116
> -0.687796 0.308193O 0 0 0(...)K 2.1035 2.1035 4.2735Cl 4.1035 4.1035 2.6035
> ###H 3.73881 3.10388 5.46104
> H 3.65742 2.89924 6.989O 3.1035 3.1035 6.207(...) &END COORD
> &KIND O BASIS_SET DZVP-MOLOPT-GTH-q6 POTENTIAL GTH-PBE-q6
> &END KIND &KIND H BASIS_SET DZVP-MOLOPT-GTH-q1 POTENTIAL
> GTH-PBE-q1 &END KIND &KIND K BASIS_SET DZVP-MOLOPT-SR-GTH-q9
> POTENTIAL GTH-PBE-q9 &END KIND &KIND Cl BASIS_SET
> DZVP-MOLOPT-GTH-q7 POTENTIAL GTH-PBE-q7 &END KIND
>
>
>
>
> &END SUBSYS
> &DFT
> BASIS_SET_FILE_NAME BASIS_MOLOPT POTENTIAL_FILE_NAME
> GTH_POTENTIALS ! SPIN_POLARIZED ! Do spin-polarized
> calculation
> &POISSON PERIODIC XYZ &END POISSON
> &QS METHOD GPW EPS_DEFAULT 1.0E-10 ! Set various epsilons
> for QS to values that will lead ! to
> energy correct up to 1e-10. &END QS
> &MGRID CUTOFF 400 ! This is Ecut of eq. 39 in VandeVondele
> (2005), i.e., plane-wave cutoff ! that
> determines size of finest grid (see caption of Fig. 1). Cutoffs for
> ! the subsequent, coarser grid levels are given by eq. 39.
> NGRIDS 4 ! This is N of eq. 39 in VandeVondele (2005), i.e., number of
> grids used. REL_CUTOFF 40 ! This controls the grid level onto which
> Gaussians will be mapped. &END MGRID
> &XC
> &XC_FUNCTIONAL &PBE PARAMETRIZATION
> ORIG &END PBE &END XC_FUNCTIONAL
> &VDW_POTENTIAL
> POTENTIAL_TYPE PAIR_POTENTIAL
> &PAIR_POTENTIAL TYPE DFTD3
> REFERENCE_FUNCTIONAL PBE CALCULATE_C9_TERM .TRUE.
> PARAMETER_FILE_NAME dftd3.dat R_CUTOFF 15.0 &END
> PAIR_POTENTIAL
> &END VDW_POTENTIAL
> &END XC
> &SCF
> SCF_GUESS RESTART ! Use data from previous run as
> initial guess for wavefunction. EPS_SCF 1.0E-6 !
> Threshold for converged total energy. MAX_SCF 300 !
> Maximum number of SCF iterations performed.
> &OT PRECONDITIONER NONE ! This should be stable with
> respect to the "Cholesky errors" &END OT
> &PRINT &RESTART ON
> BACKUP_COPIES 1
> &EACH MD 1
> &END EACH
> ADD_LAST NUMERIC
> &END RESTART
> &END PRINT
> &END SCF
> &PRINT &E_DENSITY_CUBE
> STRIDE 1 1 1
> &EACH MD 99999999 &END EACH
> ADD_LAST NUMERIC &END E_DENSITY_CUBE
> &PDOS COMPONENTS .FALSE. NLUMO = -1
> FILENAME dosfile LOG_PRINT_KEY TRUE
> &EACH MD 99999999 &END EACH
> ADD_LAST NUMERIC &END PDOS
> &END PRINT
> &END DFT
> &END FORCE_EVAL
> &MOTION
> &MD ENSEMBLE NVT STEPS
> 10000 TEMPERATURE 300.0 ! K
> TIMESTEP 0.5 ! fs
> &THERMOSTAT
> REGION GLOBAL TYPE NOSE
> &NOSE LENGTH 3
> ! Length of Nose-Hoover chain TIMECON
> 20.0 ! Period of typical vibrational motion in system in fs
> &END NOSE
> &END THERMOSTAT &END MD
> &PRINT
> &RESTART
> &EACH MD 1
> &END EACH
> ADD_LAST NUMERIC
> &END RESTART
> &TRAJECTORY ON ADD_LAST NUMERIC
> FILENAME trajectory &END TRAJECTORY
> &END PRINT
> &END MOTION
>
>
>
>
>
> --
>
> You received this message because you are subscribed to the Google Groups
> "cp2k" group.
>
> To unsubscribe from this group and stop receiving emails from it, send an
> email to cp2k+... at googlegroups.com <javascript:>.
>
> To post to this group, send email to cp... at googlegroups.com <javascript:>.
>
>
> Visit this group at https://groups.google.com/group/cp2k.
>
> For more options, visit https://groups.google.com/d/optout.
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20160817/604b857c/attachment.htm>
More information about the CP2K-user
mailing list