|
本帖最后由 北大-陶豫 于 2022-5-17 08:13 编辑
我最近也遇到了同样的问题,但是尝试了楼上的方法(单核运行、使用ALGO=FAST、使用ALGO=VERYFAST等)均未成功。经反复尝试发现,若计算读取了之前的旋轨耦合计算任务产生的WAVECAR,或当前计算量很小(ENCUT极低,K点数极低)时可以正常计算。我的解决方法如下:
(1) 使用极少的K点数(如1x1x1)、极低的ENCUT(如40eV),完成一次成功的旋轨耦合计算。这步计算可以用mpirun多核并行计算。
(2) 读取任务(1)产生的WAVECAR,进行正常的计算。
我猜测题主这一报错可能的原因:
若任务没有读取之前任务的WAVECAR,就会自己猜一个初猜然后进行 SCF 迭代计算,为了减少计算量,迭代的前几步是 non-self-consistent(?)的,之后再切换到正常的 SCF 迭代。对于旋轨耦合计算任务,这步“切换”需要较大量的计算资源(很吃硬盘?很吃内存?我也没搞清楚),所以只有在使用极少的K点数(如1x1x1)、极低的ENCUT(如40eV)的时候才能成功。
下面附上我旋轨耦合计算任务的 INCAR 输入文件模板。
- Global Parameters
- ISTART = 1 (Read existing wavefunction, if there)
- # ISPIN = 1 (Non-Spin polarised DFT)
- # ICHARG = 11 (Non-self-consistent: GGA/LDA band structures)
- LREAL = .FALSE. (Projection operators: automatic)
- ENCUT = 400 (Cut-off energy for plane wave basis set, in eV)
- PREC = Accurate #(Precision level: Normal or Accurate, set Accurate when perform structure lattice relaxation calculation)
- LWAVE = .TRUE. (Write WAVECAR or not)
- LCHARG = .TRUE. (Write CHGCAR or not)
- ADDGRID= .TRUE. (Increase grid, helps GGA convergence)
- # LVTOT = .TRUE. (Write total electrostatic potential into LOCPOT or not)
- # LVHAR = .TRUE. (Write ionic + Hartree electrostatic potential into LOCPOT or not)
- # NELECT = (No. of electrons: charged cells, be careful)
- # LPLANE = .TRUE. (Real space distribution, supercells)
- # NWRITE = 2 (Medium-level output)
- # KPAR = 2 (Divides k-grid into separate groups)
- # NGXF = 300 (FFT grid mesh density for nice charge/potential plots)
- # NGYF = 300 (FFT grid mesh density for nice charge/potential plots)
- # NGZF = 300 (FFT grid mesh density for nice charge/potential plots)
- #ALGO = Fast
- Static Calculation
- ISMEAR = 0 (gaussian smearing method)
- SIGMA = 0.05 (please check the width of the smearing)
- LORBIT = 11 (PAW radii for projected DOS)
- NEDOS = 2001 (DOSCAR points)
- NELM = 60 (Max electronic SCF steps)
- EDIFF = 1E-06 (SCF energy convergence, in eV)
- LSORBIT = .TRUE.
- NCORE = 4
复制代码
请大家批评指正。 |
评分 Rate
-
查看全部评分 View all ratings
|