Energy Calculation¶

Each energy calculator has its own config section. Available calculators: maceoff, uma, aimnet, aims, vasp, dftb.

`maceoff`¶

MACE-OFF machine-learned interatomic potential. Requires GPU.

[maceoff]
model_size  = large
save_flag   = True

model_size : str | default = large.: Model size (small, medium, large).
save_flag : bool | default = True.: Checkpoint after each structure.

`uma`¶

UMA (Universal Model for Atoms) from Meta FAIR. Requires GPU.

[uma]
model_name  = uma-s-1p1
task_name   = omc

model_name : str | default = uma-s-1p1.: Pretrained model name. See fairchem docs.
task_name : str | default = omc.: The task specifies the level of theory/DFT calculations to emulate. See fairchem docs.

`aimnet`¶

AIMNet2 neural network potential. Requires GPU.

[aimnet]
model = aimnet2

model : str | default = aimnet2.: AIMNet2 model name or model path.

DFT execution modes¶

DFT calculators (aims, vasp) support two execution modes via the dft_mode option:

parallel (default)

Every Genarris MPI rank launches its own DFT subprocess. This is the original behavior, suitable when each rank has dedicated cores for its DFT job (e.g. via use_slurm host pinning).

serial

Structures from all ranks are gathered to rank 0 via MPI. Rank 0 runs each DFT calculation sequentially, giving the DFT binary the full SLURM / MPI allocation (e.g. srun -n 256 aims.x). Results are broadcast back so every rank has the computed energies. This avoids nested-MPI conflicts and is recommended when you have relatively few structures but need maximum parallelism per DFT job.

Works with any number of Genarris ranks – use many ranks for earlier steps (generation, clustering) and the same job transitions seamlessly into serial DFT when that task is reached.

`aims`¶

FHI-aims all-electron DFT via the ASE calculator.

[aims]
command               = /path/to/aims.x
species_dir           = /path/to/species_defaults/defaults_2020/light
energy_settings_path  = ./aims_settings.json
num_cores             = 128
mpi_launcher          = mpirun
dft_mode              = parallel
use_slurm             = False

command : str.: Path to the FHI-aims binary. Do not include mpirun/srun here; the MPI launcher is controlled by mpi_launcher.
species_dir : str.: Path to the FHI-aims species_defaults directory.
energy_settings_path : str.: Path to a JSON file with FHI-aims settings (passed to ase.calculators.aims.Aims).

num_cores : int.

CPU cores per DFT process. Required when mpi_launcher is not none.

mpi_launcher : str | default = mpirun.

MPI launcher to use for DFT runs. One of:

mpirun – launch with mpirun -np {num_cores}.
srun – launch with srun -n {num_cores}.
ibrun – launch with ibrun.
none – run the binary directly with no MPI wrapper. Use this when the job is already running inside an MPI allocation (e.g. the SLURM job script already uses srun).

dft_mode : str | default = parallel.

DFT execution mode. One of:

parallel – every Genarris MPI rank runs its own DFT subprocess (original behavior).
serial – structures are gathered to rank 0, which runs DFT sequentially. Results are broadcast back. The DFT binary gets the full allocation. Works with any number of Genarris ranks (e.g. mpirun -np 128 gnrs ... for generation, then serial DFT).

use_slurm : bool | default = False.

When mpi_launcher is mpirun and dft_mode is parallel, parse SLURM node list and pin each rank to a specific host. Ignored when mpi_launcher is srun or none, or when dft_mode is serial.

`vasp`¶

VASP plane-wave DFT via the ASE calculator.

[vasp]
command               = /path/to/vasp_std
energy_settings_path  = ./vasp_settings.json
num_cores             = 128
mpi_launcher          = mpirun
dft_mode              = parallel
use_slurm             = False

command : str.: Path to the VASP binary. Do not include mpirun/srun here; the MPI launcher is controlled by mpi_launcher.
energy_settings_path : str.: Path to a JSON file with VASP settings (passed to ase.calculators.vasp.Vasp).
num_cores : int.: CPU cores per DFT process. Required when mpi_launcher is not none.
mpi_launcher : str | default = mpirun.: MPI launcher for DFT runs. One of mpirun, srun, ibrun, or none. See the aims section above for details.
dft_mode : str | default = parallel.: DFT execution mode (serial or parallel). See the aims section above for details.
use_slurm : bool | default = False.: When mpi_launcher is mpirun and dft_mode is parallel, pin each rank to a SLURM host. Ignored for srun, none, or serial.

`dftb`¶

DFTB+ density functional tight-binding via the ASE calculator.

[dftb]
command               = dftb+ > dftb.out
sk_files              = ./3ob-3-1/
energy_settings_path  = ./dftb_settings.json

command : str.: Shell command to run DFTB+.
sk_files : str.: Path to the Slater-Koster parameter files.
energy_settings_path : str.: Path to a JSON file with DFTB+ settings (passed to ase.calculators.dftb.Dftb).

Energy Calculation¶

maceoff¶

uma¶

aimnet¶