Configuration

AmorphGen uses a nested dictionary for pipeline configuration. Most defaults are defined in default_config.py; the calculator and stage functions supply the remaining defaults. Override them with CLI flags, YAML config files, or the Python API.

Default configuration

amorphgen.configs.default_config

Central configuration for the melt-and-quench pipeline.

Override any value by passing cfg_override to MeltQuenchPipeline or to individual stage functions.

Configuration keys

Section	Key	Default	Description
top-level	`model`	`"mace-mpa-0"`	MLIP model name
top-level	`device`	`"auto"`	Compute device (`cpu`, `cuda`, `mps`)
top-level	`default_dtype`	`"auto"`	Calculator precision: `float32` for CHGNet; `float64` for MACE and SevenNet
top-level	`engine`	`"ase"`	Ensemble engine (`ase` or optional `torchsim`; see Calculator backends)
top-level	`seed`	`None`	Seed for random placement and MD random streams
`opt`	`fmax`	`0.01`	Force convergence (eV/A)
`opt`	`max_steps`	`1000`	Max optimisation steps
`opt`	`optimizer`	`"LBFGS"`	ASE optimiser
`opt`	`cell_filter`	`"FrechetCellFilter"`	Cell filter: FrechetCellFilter, UnitCellFilter, ExpCellFilter, StrainFilter, cubic, none
`melt`	`T_start`	`300`	Start temperature (K)
`melt`	`T_end`	`3000`	Melt temperature (K)
`melt`	`T_step`	`100`	Temperature step per segment (K)
`melt`	`rate`	`None`	Heating rate (K/ps), auto-calculates steps_per_T
`melt`	`steps_per_T`	`2000`	MD steps per temperature step
`melt`	`timestep`	`0.5`	MD timestep (fs)
`melt`	`ensemble`	`"NPT"`	Heating ensemble (Berendsen barostat by default)
`melt`	`make_cubic`	`False`	Explicitly reshape the input cell before heating
`eq_premelt`	`ensemble`	`"NVT"`	Pre-melt ensemble
`eq_premelt`	`T`	`300`	Pre-melt equilibration T (K)
`eq_premelt`	`steps`	`100000`	Pre-melt equilibration steps (50 ps)
`eq_high`	`T`	`3000`	High-T equilibration T (K)
`eq_high`	`steps`	`20000`	High-T equilibration steps (10 ps)
`eq_high`	`ensemble`	`"NPT"`	High-T ensemble (MTK barostat by default)
`quench`	`T_start`	`3000`	Quench start T (K)
`quench`	`T_end`	`300`	Quench end T (K)
`quench`	`T_step`	`-100`	Temperature step (K, negative for cooling)
`quench`	`steps_per_T`	`2000`	MD steps per temperature step
`quench`	`rate`	`None`	Cooling rate (K/ps), auto-calculates steps_per_T
`quench`	`ensemble`	`"NVT"`	Cooling ensemble
`eq_low`	`T`	`300`	Low-T equilibration T (K)
`eq_low`	`steps`	`20000`	Low-T equilibration steps (10 ps)
`eq_low`	`ensemble`	`"NVT"`	Low-T ensemble

All MD stages default to a 0.5 fs timestep. The opt section supplies settings for stages 1 and 7; a partial final_opt section overrides individual settings for stage 7. The CLI defaults to cell_filter: cubic for --random-gen --relax, --batch-opt, and --hybrid-ensemble unless a filter is selected explicitly. Classical calculators ignore default_dtype.

See YAML configuration for the safety and repulsive_core blocks, including per-step failure limits and optional reference-model spot checks.

Configuration precedence

CLI arguments > YAML config (--config) > DEFAULT_CONFIG

Override examples

Python API

from amorphgen import MeltQuenchPipeline

pipe = MeltQuenchPipeline(
    input_file="POSCAR",
    work_dir="my_run",
    cfg_override={
        "model": "mace-mpa-0",
        "device": "cuda",
        "opt": {"fmax": 0.05, "cell_filter": "none"},
        "eq_high": {"T": 3000, "steps": 200000, "timestep": 0.5},
        "quench": {"rate": 100, "timestep": 0.5},
    },
)
atoms = pipe.run(stages=[1, 4, 5, 6, 7], resume=True)

YAML config

model: mace-mpa-0
device: cuda

opt:
  fmax: 0.05
  cell_filter: none

eq_high:
  ensemble: NVT
  T: 3000
  steps: 200000
  timestep: 0.5

quench:
  ensemble: NVT
  T_start: 3000
  T_end: 300
  T_step: -100
  rate: 100
  timestep: 0.5