Setting Up

In order to use this package, you MUST

1. Create a calculations folder where you'd like to run your calculations. Each subfolder of calculations/ should have a unique name and contain a POSCAR. A sample method of creating the calculations folder from a json with names and cifs is available in run_vasp_calculations.py, and an example calculations folder is provided in calculations/.
2. Configure computing_config.json and place it in the calculations/ directory.
3. Make any desired modifications to calc_config.json and place it in the calculations/ directory.

When you're done, your calculation directory should look roughly like this:

graph TD
  A[calculations/] --> B([calc_config.json]) & C([computing_config.json])
  A[calculations/] --> D[Material 1] --> E([POSCAR])
  A[calculations/] --> F[Material 2] --> G([POSCAR])
  A[calculations/] --> H[Material ...] --> I([POSCAR])

Computing Configuration

The computing configuration is set up in computing_config.json.

Here, you can configure settings for each supercomputer you would like to use. Be sure to check that computer at the top matches the supercomputer you're running jobs on.

• As of now, only Perlmutter and Bridges2 at NERSC and QUEST at Northwestern University are supported. Any other SLURM based supercomputers can be easily added, but modifications could be made for other queue management systems.

Supported tags

• user_id (str)
• potcar_dir (str)
• queuetype (str)
• allocation (str)
• constraint (str)
• vasp_module (str)
• ncore (int)
• ncore_per_node (int)
• atoms_per_node (int, optional) — default 32
• rerun_increase (str, optional) — "nodes" or "walltime", default "walltime"
• rerun_increase_factor (int, optional) — default 2

Note

• potcar_dir should be a global path to a folder containing VASP POTCARs.
• constraint is only needed for Perlmutter (to specify cpu nodes).
• ncore is NCORE in the VASP INCAR.
• ncore_per_node should be the number of CPUs on each node.
• For vasp_module, a VASP 6 module is strongly recommended.
• The personal "computer" is only used for internal unit testing, not to run any actual jobs.

Job resource tuning

The following fields are optional — omitting them preserves the previous default behavior.

• atoms_per_node controls how many compute nodes are requested per job (nodes = floor(total atoms / atoms_per_node) + 1).
• rerun_increase chooses which resource to scale when a job times out: "walltime" (increase the walltime) or "nodes" (increase the node count).
• rerun_increase_factor is the multiplier applied to the chosen resource on rerun (e.g., 2 means double).

Warning

Be sure that your settings of KPAR and NCORE are compatible with the computing architecture you're using!

• It's okay if ncore * kpar != ncore_per_node. All cores on the node will still be requested, and some of them will be left empty for extra memory. This can be useful for computing architectures with a weird number of cores (e.g. 52 on Quest).

Example

computing_config.json

{
  "computer": "perlmutter",
  "personal": {
    "user_id": "dwg4898",
    "potcar_dir": "vasp_manager/tests/POTCARS",
    "queuetype": "regular",
    "allocation": "m1673",
    "constraint": "cpu",
    "vasp_module": "vasp/6.4.3-cpu",
    "ncore": 16,
    "ncore_per_node": 128
  },
  "quest": {
    "user_id": "dwg4898",
    "potcar_dir": "/projects/b1004/potpaw_PBE_OQMD",
    "queuetype": "short",
    "allocation": "p31151",
    "vasp_module": "vasp/6.4.3-openmpi-intel-hdf5-cpu-only",
    "ncore": 12,
    "ncore_per_node": 52,
    "atoms_per_node": 16,
    "rerun_increase": "walltime",
    "rerun_increase_factor": 2
  },
  "perlmutter": {
    "user_id": "dwg4898",
    "potcar_dir": "/global/homes/d/dwg4898/vasp_potentials/54/potpaw_pbe",
    "queuetype": "regular",
    "allocation": "m4545",
    "constraint": "cpu",
    "vasp_module": "vasp/6.4.3-cpu",
    "ncore": 16,
    "ncore_per_node": 128,
    "atoms_per_node": 32,
    "rerun_increase": "nodes",
    "rerun_increase_factor": 2
  },
  "bridges2": {
    "user_id": "daleg",
    "potcar_dir": "/jet/home/daleg/vasp_potentials/potpaw_PBE_OQMD",
    "queuetype": "RM",
    "allocation": "dmr160027p",
    "vasp_module": "",
    "ncore": 16,
    "ncore_per_node": 128,
    "atoms_per_node": 32,
    "rerun_increase": "nodes",
    "rerun_increase_factor": 2
  }
}

Calculation Configuration

Tip

See more about VASP INCAR tags here

The calculation configuration is set up in calc_config.json.

For each desired calculation mode, set the INCAR tags in this json.

• Each mode has its own configuration settings with sensible defaults, but these can be easily customized by the user.
• Note: bulkmod does not have its own section in calc_config.json — it uses the static configuration.
• See more about about spin polarization settings ("ispin": "auto") here: Spin Configuration
• See more about DFT+U settings ("hubbards": "wang") here: DFT+U Configuration
• Note: VaspManager uses 1 compute node per atoms_per_node atoms (see Computing Configuration), so don't be surprised if your job requests more than a single node.

Supported tags:

• prec (str): ["Normal" | "Accurate"]
• ispin (str | int): ["auto" | 1]
• hubbards (str | None): ["wang" | null]
• kspacing (float)
• symprec (float)
• nsw (int)
• ibrion (int)
• isif (int)
• lreal (bool)
• potim (float)
• ediffg (float)
• iopt (int)
• nelm (int)
• encut (float)
• ediff (float)
• algo (str): ["Normal" | "Fast" | "VeryFast"]
• ismear (int)
• sigma (float)
• amix (float)
• bmix (float)
• lwave (bool)
• lcharge (bool)
• lvot (bool)
• kpar (int)
• gga (str)
• walltime (str)

Note

• In json files, the equivalent of python's None is null.
• VASP (Fortran) expects bool data to be passed as ".FALSE." or ".TRUE."; this is automatically converted for you.
• walltime should be passed as "hh:mm:ss". You should try to set this such that your job hits NSW before it runs out of time. If your job times out, an archive will be made and the calculation will be resubmitted with increased resources according to rerun_increase and rerun_increase_factor in your computing configuration.
• You can place another calc_config.json file in a specific material's calculation directory (e.g. NaCl/rlx) to use custom settings for only that material. You only need to include settings you want to change from the main calc_config.json.

Example

calc_config.json

{
  "rlx-coarse": {
    "prec": "ACC",
    "ispin": "auto",
    "hubbards": "wang",
    "kspacing": 0.2,
    "symprec": "1e-05",
    "nsw": 60,
    "ibrion": 2,
    "isif": 3,
    "lreal": false,
    "potim": 0.1,
    "ediffg": "1e-04",
    "iopt": 0,
    "nelm": 60,
    "encut": 520,
    "ediff": "1e-05",
    "algo": "Normal",
    "ismear": 0,
    "sigma": 0.05,
    "kpar": 4,
    "gga": "PE",
    "walltime": "01:00:00"
  },
  "rlx": {
    "prec": "ACC",
    "ispin": "auto",
    "hubbards": "wang",
    "kspacing": 0.15,
    "symprec": "1e-05",
    "nsw": 60,
    "ibrion": 3,
    "isif": 3,
    "lreal": false,
    "potim": 0.1,
    "ediffg": "-1e-02",
    "iopt": 0,
    "nelm": 60,
    "encut": 520,
    "ediff": "1e-07",
    "algo": "Normal",
    "ismear": 0,
    "sigma": 0.05,
    "kpar": 4,
    "gga": "PE",
    "walltime": "01:00:00"
  },
  "static": {
    "prec": "ACC",
    "ispin": "auto",
    "hubbards": "wang",
    "kspacing": 0.15,
    "symprec": "1e-05",
    "nsw": 0,
    "ibrion": -1,
    "isif": 3,
    "lreal": false,
    "potim": 0,
    "ediffg": "-1e-02",
    "iopt": 0,
    "nelm": 60,
    "encut": 520,
    "ediff": "1e-07",
    "algo": "Normal",
    "ismear": -5,
    "sigma": 0.05,
    "kpar": 4,
    "gga": "PE",
    "walltime": "01:00:00"
  },
  "elastic": {
    "prec": "ACC",
    "ispin": "auto",
    "hubbards": "wang",
    "kspacing": 0.125,
    "write_kpoints": true,
    "nfree": 4,
    "symprec": "1e-05",
    "nsw": 60,
    "ibrion": 6,
    "isif": 3,
    "lreal": false,
    "potim": 0.005,
    "ediffg": "-1e-02",
    "iopt": 0,
    "nelm": 60,
    "encut": 700,
    "ediff": "1e-07",
    "algo": "Normal",
    "ismear": 0,
    "sigma": 0.05,
    "kpar": 4,
    "gga": "PE",
    "walltime": "04:00:00"
  }
}