DFT reference structures from RMC configurations

Research workflow

Local atomic environments are sampled from RMC configurations constrained by experimental data and represented in small periodic DFT cells. Energies and forces from first-principles calculations provide reference data for ænet ANN interatomic potentials. The resulting potentials can be applied as energetic constraints in large-scale RMC_POT calculations.

The sampling objective is to represent the distribution and variations of local environments. A local environment may be centered on an atom, bond, or multi-atom unit and can include nearby bridges and relative arrangements within a finite observation range.

Local environments and periodic cells

The observation region, preserved geometry, and region subject to perturbations are selected for the local environments of interest. Their sizes are considered together with the DFT cell dimensions, composition, periodic boundary conditions, and the descriptor cutoffs used for ANN training.

Periodic images in small cells can introduce correlations absent from the original configuration. Distances, angles, and relative atomic positions in the selected region are compared before and after constructing the periodic cell.

Sampling reference candidates

The frequencies and diversity of local environments in the RMC configuration guide candidate selection. Representative, rare, and distribution-boundary environments can be sampled using different cell sizes, local-region definitions, displacement amplitudes, and random seeds.

Each candidate is accompanied by provenance and construction metadata: source RMC file, selected local region, periodic cell, perturbation settings, and sampling conditions. Geometric checks cover interatomic distances, periodicity, composition, local structure, and duplicates.

Calculation and validation

VASP calculations supply energy and force labels after convergence and data-integrity checks. ANN models are evaluated against independent DFT references, including energy differences, forces, and errors resolved by local environment. RMC_POT applications additionally examine predicted energy changes for proposed atomic moves.

aenet_env supplies reusable APIs for these stages; study-specific structure-selection and construction methods are developed in their respective examples.