materialsvirtuallab/snap

Repository for spectral neighbor analysis potential (SNAP) model development.

snap's Issues

Manage DFT training data and making JSON data set

Dear Developers,

Greetings. Last night I watched a demo here, https://www.youtube.com/watch?v=GiebtsjYfgs
This is interesting. I have a very basic query.
Would you kindly help me to know how to extract data from OUTCAR/vasprun.xml file and write a dict contains one structure with data. For multiple structures how can I merge them in a single json file. If you share any script/tool that will be very helpful.
I will wait to hear from you.

Thanks.

Incorrect SNAP parameter error for NbMoTaW potential

Hi,
I am trying to use the SNAP potential for NbMoTaW in LAMMPS (29Oct20). I am getting the following error
ERROR: Incorrect SNAP parameter file (src/SNAP/pair_snap.cpp:732)

If I remove the diagonalstyle 3 from the Ta-W-Nb-Mo.snapparam, it is working.

Is there a workaround for this? Is it safe to omit this line?

Best,
Abu Anand
University of Toronto

VASP Setup

Hi, I met some problem reproducing the VASP result for a Ni slab.

The total energy difference is significant. I hope you can help me figuring out what should be changed.

• Your energy is -33.38981773 eV
• Mine is around -31.66 eV
  results.zip

This is the JSON dict, the 4th item of Ni/Surface.json:

{'_id': '59b4ba43ffc31d171c78afbb',
 'params': {'pp': {'Ni': 'Ni_pv'},
  'env': 'bulk',
  'e_cut': 520,
  'k_prod': 40,
  'bader': False},
 'group': 'Ni',
 'tags': ['Surface'],
 'structure': {'@module': 'pymatgen.core.structure',
  '@class': 'Structure',
  'lattice': {'matrix': [[2.479787, 0.0, 0.0],
    [-1.239893, 2.147558, 0.0],
    [0.0, 0.0, 24.294656]],
   'a': 2.479787,
   'b': 2.479786284100507,
   'c': 24.294656,
   'alpha': 90.0,
   'beta': 90.0,
   'gamma': 119.99999621016981,
   'volume': 129.38086036717198},
  'sites': [{'species': [{'element': 'Ni', 'occu': 1}],
    'abc': [0.333334, 0.666668, 0.082544],
    'xyz': [3.3333399995783e-07, 1.4317081967440002, 2.005378084864],
    'label': 'Ni'},
   {'species': [{'element': 'Ni', 'occu': 1}],
    'abc': [0.0, 0.0, 0.999689],
    'xyz': [0.0, 0.0, 24.287100361984002],
    'label': 'Ni'},
   {'species': [{'element': 'Ni', 'occu': 1}],
    'abc': [0.666666, 0.333332, 0.166332],
    'xyz': [1.2398936666659999, 0.7158498032560001, 4.040978721792],
    'label': 'Ni'},
   {'species': [{'element': 'Ni', 'occu': 1}],
    'abc': [0.333334, 0.666668, 0.334122],
    'xyz': [3.3333399995783e-07, 1.4317081967440002, 8.117379052032],
    'label': 'Ni'},
   {'species': [{'element': 'Ni', 'occu': 1}],
    'abc': [0.0, 0.0, 0.250335],
    'xyz': [0.0, 0.0, 6.081802709759999],
    'label': 'Ni'},
   {'species': [{'element': 'Ni', 'occu': 1}],
    'abc': [0.666666, 0.333332, 0.416978],
    'xyz': [1.2398936666659999, 0.7158498032560001, 10.130337069568],
    'label': 'Ni'}]},
 'num_atoms': 6,
 'outputs': {'energy': -33.38981773,
  'forces': [[-0.0, -0.0, -0.00470706],
   [0.0, 0.0, 0.00549928],
   [0.0, -0.0, -0.00421769],
   [0.0, 0.0, 0.00470706],
   [-0.0, 0.0, 0.00421769],
   [-0.0, -0.0, -0.00549928]],
  'stress': [-22.25152803, -22.25151688, -14.67323629, -5.18e-06, 0.0, 0.0]}}

Here is my INCAR (generated by pymatgen.io.vasp.sets.MVLSlabSet):

ALGO = Fast
EDIFF = 0.0001
EDIFFG = -0.02
ENCUT = 520
IBRION = 2
ICHARG = 1
ISIF = 3
ISMEAR = 0
ISPIN = 2
LORBIT = 11
LREAL = Auto
LWAVE = False
MAGMOM = 6*5.0
NELM = 100
NSW = 99
PREC = Accurate
SIGMA = 0.05

• VASP: 5.4.4
• POTCAR: potpaw_PBE/Ni_pv/POTCAR
• Other files are included in results.zip

MAE of Snap/Mo and EAM/Zhou04

Hi,

From your Snap/Mo paper:

In comparison, the corresponding MAEs in the energies, forces, and stress components for the Mo EAM potential of Zhou et al. [32] are 122 meV/atom, 0.41 eV/A° , and 3.48 GPa, respectively

However, I find that the direct total energy MAE (~4 eV/atom) of the absolute EAM/Zhou04 and DFT is far larger than 122 meV/atom.

But the relative total energy MAE is around this value:

y_eam = array_of_eam_energies / array_of_natoms - eam_bulk
y_dft = array_of_dft_energies / array_of_natoms - dft_bulk
print(mean_absolute_error(y_eam, y_dft)

Is my guess correct?

Example

The total energy of the last structure of Mo.surface.json:

dft_energy = -10.3133037725 * 12

The DFT total energy of bulk bcc Mo is around -10.91973776 eV
The EAM/Zhou04 total energy of this structure is just -75.384721 eV
The EAM/Zhou04 total energy of bulk bcc Mo is -6.8100034 eV

y_dft = -123.75964527
dft_ref = -10.91973776
y_eam = -75.384721
eam_ref = -6.8100034
print(y_dft / 12 - dft_ref - (y_eam / 12 - eam_ref)) # 0.07849067083333416

So the direct diff is 4.03124 eV/atom while the relative diff is 0.078 eV/atom.