Top Page > Browsing
calculation time for a small system
Date: 2020/03/06 18:00
Name: Yongsheng Zhang

Dear devlopers,

The basis sets of Openmx are LCAO. Generally, it seems that it should be more efficient (or faster) than running the plane-wave based code like VASP. However, it is not the case in small systems. I did a test using a bulk TiO2 structure. It turns out that the vASP code is 2-3 times faster than the OpenMX code. They are using the same k-mesh. I am pretty confused about the reasons behind the significant time difference. Would you please explain it to me? Thanks.

CPU-number 2 8
vasp 0.92 0.43
openmx 2.27 0.93
unit: second/scf


DATA.PATH /home/yshzhang/software/openmx3.8/DFT_DATA13
System.CurrrentDirectory ./
System.Name tio2
level.of.stdout 1
level.of.fileout 1

scf.restart off

Species.Number 2
<Definition.of.Atomic.Species
Ti Ti7.0-s2p2 Ti_PBE13
O O5.0-s2p2 O_PBE13
Definition.of.Atomic.Species>

Atoms.Number 6
Atoms.SpeciesAndCoordinates.Unit Ang
<Atoms.SpeciesAndCoordinates
1 Ti 0.00000000 0.00000000 0.00000000 6.00 6.00
2 Ti 2.31484428 2.31484428 1.47137134 6.00 6.00
3 O 1.40975946 1.40975946 0.00000000 3.00 3.00
4 O 3.21992911 3.21992911 0.00000000 3.00 3.00
5 O 0.90508483 3.72460374 1.47137134 3.00 3.00
6 O 3.72460374 0.90508483 1.47137134 3.00 3.00
Atoms.SpeciesAndCoordinates>
Atoms.UnitVectors.Unit Ang
<Atoms.UnitVectors
4.62968857 0.00000000 0.00000000
0.00000000 4.62968857 0.00000000
0.00000000 0.00000000 2.94274268
Atoms.UnitVectors>

scf.XcType GGA-PBE # LDA|LSDA-CA|LSDA-PW
scf.SpinPolarization off # On|Off
scf.ElectronicTemperature 300.0 # default=300 (K)
scf.energycutoff 300.0 # default=150 (Ry)
scf.maxIter 170 # default=40
scf.EigenvalueSolver band # Recursion|Cluster|Band
scf.Kgrid 1 1 1 # means 4x4x4
scf.Mixing.Type rmm-diisk # Simple|Rmm-Diis|Gr-Pulay
scf.Init.Mixing.Weight 0.010 # default=0.30
scf.Min.Mixing.Weight 0.001 # default=0.001
scf.Max.Mixing.Weight 0.200 # default=0.40
scf.Mixing.History 6 # default=5
scf.Mixing.StartPulay 12 # default=6
scf.criterion 1.0e-9 # default=1.0e-6 (Hartree)
scf.system.charge 0.0 # default=0.0
メンテ
Page: [1]

Re: calculation time for a small system ( No.1 )
Date: 2020/03/08 08:24
Name: T. Ozaki

Hi,

In general, OpenMX seems to be slower than VASP for small systems with dense crystal structures,
since OpenMX does not take account of computational reduction by symmertry.

I wonder that OpenMX might be faster than PW codes for systems with low-dimensional structures and
sparse crystal structures.

Regards,

TO
メンテ
Re: calculation time for a small system ( No.2 )
Date: 2020/03/08 22:47
Name: Yongsheng Zhang

Thanks. That makes sense. I agree that OpenMX should be faster for the large low-symmetry system. I am wondering why OpenMX does not include symmetries. This would widen the applications of the code?
メンテ
Re: calculation time for a small system ( No.3 )
Date: 2020/03/08 23:13
Name: Yongsheng Zhang

Dear Prof. Ozaki,

I just realize that I even did the tests using one Gamma-k-point in OpenMX and VASP. In this case, the symmetry operation should have little effects on the computational time. Unfortunately, OpenMX is still slower than VASP. Is that due to the time in building or solving the symmetry dependent interatomic potentials in the two codes? Thanks

Yongsheng
メンテ
Re: calculation time for a small system ( No.4 )
Date: 2020/03/09 09:56
Name: T. Ozaki

Hi,

The elapsed time of each calculation is output in the standard output
and the out file. So, you can figure out which part of calculation
can be a time-consuming part.

Regards,

TO
メンテ

Page: [1]

Thread Title (must) Move the thread to the top
Your Name (must)
E-Mail (must)
URL
Password (used in modification of the submitted text)
Comment (must)

   Save Cookie