[ase-users] Efficiency of vibrational frequency using VASP as the calculator
Tristan Maxson
tgmaxson at gmail.com
Mon Jun 5 07:47:33 CEST 2017
This is actually the exact reason why we developed VaspInteractive in the
newer version of ASE. I think it currently requires the development branch
but if you would like to test this, I can help you set it up. This
calculator creates an interactive session of VASP that reports energies and
forces back. I am unsure what version of VASP you are using but it seems
to have been there for a while undocumented.
First install ASE from the development branch. (git clone
https://gitlab.com/ase/ase.git)
Next you need to convert your VASP INCAR etc to ase's format with a normal
vasp calculator, but it seems you have already done this. Ensure you do
not set IBRION as that will be set automatically.
Next, import the interactive version instead and it should work without any
extra configuration. (from ase.calculators.vasp import VaspInteractive)
If this doesn't work please report back with whatever problem or error you
get. I would also be happy to hear what the performance is like for your
system if it does work.
Thank you,
Tristan Maxson
On Sun, Jun 4, 2017 at 11:11 PM, 陳佑任 via ase-users <
ase-users at listserv.fysik.dtu.dk> wrote:
> Dear ASE users,
>
>
>
> I am using ASE’s vibrations package and VASP as a calculator to calculate
> the vibrational frequencies for a benzene molecule on a graphene
> nanoribbon. The calculated frequencies are in good agreement with the
> values computed with the normal mode analysis implemented in VASP itself.
> However, the total CPU time is much longer when using ASE (almost 3 times
> as long as directly using VASP). The reason is that the SCF calculations in
> the VASP-native normal mode analysis require fewer SCF cycles than the
> single points calculations when ASE calls VASP (<10 vs about 20), even if
> we read in the WAVECAR file (the stored wavefunction) of the previous
> calculations. So, seemingly the initial guess for the wavefunction of each
> perturbed geometry is not good enough. We have tried two approaches, 1)
> always reading the WAVECAR saved for the unperturbed equilibrium geometry
> and 2) reading in the WAVECAR from the previous geometry step. Both
> approaches do not differ too much in the CPU time. Does anybody know how
> which settings I have to set so the computation time for the normal mode
> analysis with ASE can be reduced to the computation time of the normal mode
> analysis as implemented in VASP? Below are my settings:
>
>
>
> -----------------------------------
>
>
>
> INCAR from VASP:
>
>
>
> ISMEAR = -1
>
> SIGMA = 0.01
>
> EDIFF = 1.d-6
>
> EDIFFG = -0.01
>
> ISPIN = 2
>
> ALGO = fast
>
> NSW = 600
>
> ISIF = 0
>
> IBRION = 5
>
> LREAL = .F.
>
> IALGO = 38
>
> NELM = 60
>
> ENCUT = 500
>
> ISYM = 0
>
> NPAR = 4
>
> ISTART = 1
>
> IDIPOL = 3
>
> LDIPOL = T
>
> GGA = MK
>
> PARAM1 = 0.1234
>
> PARAM2 = 1.0000
>
> LUSE_VDW = .TRUE.
>
> AGGAC = 0.0000
>
>
>
> -----------------------------------
>
>
>
> INCAR generated from ASE:
>
>
>
> ENCUT = 500.000000
>
> PARAM2 = 1.000000
>
> AGGAC = 0.000000
>
> PARAM1 = 0.123400
>
> SIGMA = 0.010000
>
> POTIM = 0.015000
>
> EDIFF = 1.00e-06
>
> EDIFFG = -1.00e-02
>
> SYSTEM = graphene
>
> GGA = MK
>
> ALGO = fast
>
> ISYM = 0
>
> ISPIN = 2
>
> ISMEAR = -1
>
> ISTART = 1
>
> IALGO = 38
>
> NELM = 60
>
> NSW = 1
>
> IBRION = -1
>
> NPAR = 4
>
> IDIPOL = 3
>
> ISIF = 0
>
> LUSE_VDW = .TRUE.
>
> LDIPOL = .TRUE.
>
> LREAL = .FALSE.
>
>
>
> -----------------------------------
>
>
>
> VASP’s one of the SCF steps:
>
>
>
> N E dE
> d eps ncg rms rms(c)
>
> DAV: 1 -0.781201282540E+03 0.45811E-03 -0.75927E-02 4148
> 0.365E-01 0.929E-02
>
> RMM: 2 -0.781201302856E+03 -0.20316E-04 -0.75765E-04 3552
> 0.346E-02 0.556E-02
>
> RMM: 3 -0.781201306607E+03 -0.37512E-05 -0.11168E-04 3467
> 0.122E-02 0.838E-03
>
> RMM: 4 -0.781201305914E+03 0.69345E-06 -0.58417E-06 2832
> 0.291E-03
>
> 2 F= -.78120131E+03 E0= -.78120113E+03 d E =-.353918E-03 mag=
> 2.5054
>
>
>
> -----------------------------------
>
>
>
> ASE’s one of the SCF steps:
>
>
> N E dE d
> eps ncg rms rms(c)
>
> DAV: 1 -0.781205901599E+03 -0.78121E+03 -0.34936E-01 3944
> 0.790E-01 0.441E-01
>
> RMM: 2 -0.781202464766E+03 0.34368E-02 -0.87841E-03 3849
> 0.121E-01 0.252E-01
>
> RMM: 3 -0.781200989068E+03 0.14757E-02 -0.43555E-03 4124
> 0.666E-02 0.494E-02
>
> RMM: 4 -0.781200905776E+03 0.83293E-04 -0.14996E-04 4230
> 0.150E-02 0.225E-02
>
> RMM: 5 -0.781200879684E+03 0.26091E-04 -0.41327E-05 3953
> 0.778E-03 0.907E-03
>
> RMM: 6 -0.781200886517E+03 -0.68326E-05 -0.71667E-06 3586
> 0.368E-03 0.489E-03
>
> RMM: 7 -0.781200904313E+03 -0.17796E-04 -0.53720E-06 3272
> 0.273E-03 0.230E-03
>
> RMM: 8 -0.781200927253E+03 -0.22940E-04 -0.28785E-06 3218
> 0.179E-03 0.149E-03
>
> RMM: 9 -0.781200963183E+03 -0.35930E-04 -0.49109E-06 3057
> 0.214E-03 0.827E-04
>
> RMM: 10 -0.781200986671E+03 -0.23488E-04 -0.27405E-06 3064
> 0.154E-03 0.595E-04
>
> RMM: 11 -0.781201007939E+03 -0.21268E-04 -0.24989E-06 2888
> 0.143E-03 0.347E-04
>
> RMM: 12 -0.781201019595E+03 -0.11657E-04 -0.15067E-06 2791
> 0.116E-03 0.226E-04
>
> RMM: 13 -0.781201025267E+03 -0.56721E-05 -0.66097E-07 2534
> 0.834E-04 0.139E-04
>
> RMM: 14 -0.781201028437E+03 -0.31692E-05 -0.32406E-07 2352
> 0.677E-04 0.860E-05
>
> RMM: 15 -0.781201029598E+03 -0.11619E-05 -0.10339E-07 2381
> 0.466E-04 0.627E-05
>
> RMM: 16 -0.781201030812E+03 -0.12132E-05 -0.10772E-07 2194
> 0.519E-04 0.344E-05
>
> RMM: 17 -0.781201031253E+03 -0.44139E-06 -0.45790E-08 2116
> 0.374E-04
>
> 2 F= -.78120103E+03 E0= -.78120085E+03 d E =-.353970E-03 mag= 2.5054
>
>
> -----------------------------------
>
>
> Yu-Ren Chen
>
>
> Institute of Atomic and Molecular Sciences, Academia Sinica
>
> _______________________________________________
> ase-users mailing list
> ase-users at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/ase-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.fysik.dtu.dk/pipermail/ase-users/attachments/20170605/6129b55f/attachment-0001.html>
More information about the ase-users
mailing list