求助：优化Pt（111）晶面时电子步速度

guyingbing

各位老师，大家好！我是一位初学VASP的师弟。我现在想通过计算Pt（111）表面的氢吸附自由能，验证学习的效果。经过前期的优化单胞后。现在我通过切面扩胞了，设置了一个Pt（111）晶面的表面模型。现在在优化的时候，我发现我的进行一个电子步需要369秒，请问这个速度正常吗？我切的面是Pt的（111）晶面，扩胞后是3x3的，切面厚度为5层。我固定了下面3层，优化了最上面的2层原子。计算的时候一共144个原子。截断能是400eV。下面是我使用的INCAR问和KPOINTS文件。
INCAR
##### initial parameters I/O #####
SYSTEM = Pt
#KPAR = 4
NPAR=4
ISTART = 1 # whether or not to read the WAVECAR file.
ICHARG = 1 # how VASP constructs the initial charge density.
LWAVE = .TRUE. # whether the wavefunctions are written to the WAVECAR file
LCHARG = .TRUE. # LCHARG determines whether the charge densities (files CHGCAR and CHG) are written.
LVTOT = .FALSE. # whether the total local potential is written to the LOCPOT file
LVHAR = .FALSE. # whether the electrostatic potential is written to the LOCPOT file
LELF = .TRUE. # whether to create an ELFCAR file or not.
LORBIT = 11 # determines whether the PROCAR or PROOUT files are written.

#### Electronic Relaxation ####
ENCUT = 400 # cutoff energy for the planewave basis set in eV
ISMEAR = 1 # how the partial occupancies are set for each orbital.
SIGMA = 0.2 # the width of the smearing in eV.
EDIFF = 1E-6 # global break condition for the electronic SC-loop
#NELMIN = 5 # specifies the minimum number of electronic SCF steps.
NELM = 300 # maximum number of electronic SC (selfconsistency) steps
#GGA = PE # type of generalized-gradient-approximation one wishes to use.
#LREAL = Auto # whether the projection operators are evaluated in real-space or in reciprocal space.
#LDIPOL = .TRUE.
#IDIPOL = 3 # add dipole correction
#ISYM = 0 # determines the way VASP treats symmetry.
ALGO = Normal # electronic minimisation algorithm
PREC = Normal # specifies the "precision"-mode
# ADDGRID = .TRUE. # whether an additional support grid is used for the evaluation of the augmentation charges
# AMIX = 0.2
# BMIX = 0.0001
# AMIX_MAG = 0.8
# BMIX_MAG = 0.0001
# MAXMIX = 40

#### Magnetic parameters ####
ISPIN = 2
# MAGMOM = 33*0.0 4*2.0 4*-2.0 4*2.0 4*-2.0
# NUPDOWN = 0
# VOSKOWN = 1

#### Geometry opt parameters ####
EDIFFG = -0.01 # EDIFFG defines the break condition for the ionic relaxation loop.
IBRION = 2 # IBRION determines how the ions are updated and moved.
#POTIM = 0.2 # POTIM sets the time step (MD) or step width scaling (ionic relaxations).
NSW = 300 # maximum number of ionic steps.
ISIF = 2 # determines whether the stress tensor is calculated and which principal degrees-of-freedom are allowed to change
# NFREE = 2

#### DFT +U ####
# LDAU = .TRUE.
# LDAUTYPE = 2
# LDAUL =   2  -1   -1   -1  -1
# LDAUU =  4.6 0.0  0.0 0.0 0.0
# LDAUJ =  0.4 0.0  0.0 0.0 0.0
# LDAUALPHA = 0.0 0.5
# LDAUPRINT = 2
# LMAXMIX = 4


KPOINTS


KPT-Resolved Value to Generate K-Mesh: 0.040
0
Gamma
   2   2   1
0.0  0.0  0.0

Penson

几个核？服务器什么配置？

guyingbing

Penson 发表于 2021-3-2 08:28
几个核？服务器什么配置？

双路Intel 2697 V4 36核

ghifi37

吸附氢没必要弄144个原子的体系，零头就差不多了

Penson

guyingbing 发表于 2021-3-3 19:54
双路Intel 2697 V4 36核

那很正常啊，vasp计算超过100个原子没有个64核速度基本都很难受了，何况你这还不是Gamma点计算

xexlalalan

按理说也不该这么慢。我机子24个核，120个原子的体系，同样也是开壳层计算，大概在40秒一个电子步。只不过我只用Gamma点计算，但你这个也不至于这么慢。强烈建议你看看你的服务器内存是否所有的通道都有查内存，VASP吃内存带宽比较厉害。有可能你的计算瓶颈在内存