Molten Salts

The molten salt example demonstrates an ALF workflow for periodic ionic systems. It is intended for workflows where the builder constructs charge-balanced atomic systems and the QM stage labels selected configurations with VASP.

The files needed to run the example are located in examples/molten_salt.

Before You Run

This example is a template for a periodic VASP + HIPPYNN workflow. Before running it on a new machine:

Edit parsl_configs.py for the target Slurm partition, account, QoS, walltime, worker initialization, and CPU/GPU resource requests. See Parsl And HPC Execution.
Edit vasp_config.json so QM_run_command and VASP input settings match the local installation and intended calculation.
Edit hippynn_config.json for the elements, network size, training controls, and device settings for the run. See ML Interfaces.
Confirm that the requested properties in master_config.json match the VASP calculator results and HDF5 keys you want to train on.

Workflow Pattern

This example uses:

Stage	Task
Builder	`alframework.builders.builders.atomic_system_task`
Sampler	`alframework.samplers.mlmd_sampling.simple_mlmd_sampling_task`
QM	`alframework.qm_interfaces.ase_calculator_interface.VASP_ase_calculator_task`
ML	`alframework.ml_interfaces.hippynn_interface.train_HIPPYNN_ensemble_task`

The builder reads builder_config.json and creates charge-neutral random atomic systems from the configured atom_charges. The provided configuration targets 100 atoms, enforces a minimum interatomic distance, and samples a box length from box_length_range. This is the right builder family for molten salts or other ionic atomic systems that are not assembled from molecular fragments.

The sampler reads mlmd_config.json and runs ML-driven MD with a HIPPYNN ASE calculator ensemble. In this example, density fluctuation fields are set to null so the sampler does not rescale the box during sampling.

The QM stage uses ASE’s VASP calculator wrapper. This is the preferred path for periodic DFT engines that ASE already supports. See QM Interfaces for the generic ASE calculator pattern.

Recommended Bring-Up Commands

From examples/molten_salt:

python -m alframework --master master_config.json --test_builder
python -m alframework --master master_config.json --test_qm
python -m alframework --master master_config.json --test_ml
python -m alframework --master master_config.json --test_sampler
python -m alframework --master master_config.json

The test modes use parsl_debug_configuration from master_config.json. Use them to verify the builder, VASP command, HIPPYNN environment, and sampler resource mapping before launching a long active-learning run.

Expected Outputs

During a run, expect the usual ALF output structure:

Path	Purpose
`status.txt`	Restart and progress state.
`h5store/data-*.h5`	VASP-labeled training batches.
`models/model-*`	HIPPYNN model ensemble directories.
`sampling/metadata-*.p`	Sampler metadata and uncertainty records.
`vasp_calculations/`	Per-structure VASP scratch and output directories.
`status_plots/`	Optional progress plots from the configured plotting utility.

Configuration Files

File	Purpose
`master_config.json`	Connects the builder, sampler, VASP QM task, HIPPYNN training task, output paths, queue-depth controls, and Parsl configs.
`builder_config.json`	Defines element charges, target atom count, minimum distance, and box length range for charge-neutral atomic systems.
`mlmd_config.json`	Defines MD time step, maximum time, uncertainty thresholds, temperature schedule, optional density schedule, metadata path, and ASE model loader.
`vasp_config.json`	Defines the VASP command and ASE VASP calculator options.
`hippynn_config.json`	Defines HIPPYNN ensemble, network, species, optimizer, and training options.
`parsl_configs.py`	Generic CPU/GPU Slurm template copied from the water example.

Environment setup for this example now lives in the command strings and worker initialization hooks that users adapt in vasp_config.json and parsl_configs.py.