Description of the DIAMOND AiiDA-VASP workflows

Overview

This project brings together Python tools to automate selected steps of ab initio molecular dynamics (AIMD) with VASP, using AiiDA and the aiida-vasp plugin. It was developed within DIAMOND to simplify calculation preparation, submission to computing facilities, and reproducible analysis of the results.

The current version provides two complementary modules: WF1, which extracts inherent structures from a liquid, and WF2, which estimates the equilibrium volume from fixed-volume simulations. A third module, WF3, is planned to combine these steps into a data-generation pipeline for machine-learning interatomic potentials (MLIPs).

VASP calculations are tracked by AiiDA, while some structural analyses use the free Python API of OVITO Basic. Parameters are organized in YAML files so that they can be adapted to a local computer or an HPC cluster.

Schematic of the available AiiDA-VASP workflows

Overview of the available AiiDA-VASP workflows.

Available modules

WF1 — Inherent-structure extraction

Input: a VASP trajectory in XDATCAR format.

WF1 selects configurations at regular intervals, optionally excluding the initial frames corresponding to thermalization. Each selected configuration is then submitted for fixed-cell atomic relaxation through AiiDA’s vasp.relax WorkChain. The relaxed structures represent local minima of the potential energy associated with the sampled configurations.

A post-processing step also makes it possible to:

calculate radial distribution functions (RDFs) from the input trajectory;
estimate species-pair cutoff radii from the RDFs;
perform common-neighbor analysis (CNA) on the relaxed structures using OVITO.

Main outputs: relaxed structures in POSCAR format, an RDF table (rdf.csv), cutoff radii (pair_cutoffs.json), and CNA statistics (cna.csv). Automatically determined cutoffs should be checked before any scientific interpretation.

WF2 — Equilibrium-volume estimation

Input: one or more structures named POSCAR_<temperature> (with the temperature in kelvin, for example POSCAR_650).

For each temperature, WF2 generates several structures with different volumes, then submits NVT AIMD simulations using vasp.vasp. It calculates the average pressure over the final part of each trajectory by reading the OUTCAR files, then fits the pressure–volume relationship using the third-order Birch–Murnaghan equation of state.

The estimated equilibrium volume is the volume at which the fitted pressure becomes zero. This approach avoids systematically running a long NPT simulation when an initial estimate of the equilibrium volume is needed.

Main outputs: average pressures for each volume (pressures.csv), fitting parameters (eos_fits.json), and, if Matplotlib is installed, a pressure–volume plot (pressure_volume.png). The fit requires several appropriately chosen volumes and sufficiently equilibrated pressure data.

WF3 — Future integration

WF3 is under development and is not included in this version. The aim is to combine AIMD sampling, relaxation, equilibrium-volume determination, and calculations across different phases or thermodynamic conditions to produce datasets for materials characterization and, in particular, MLIP training.

Project structure and examples

Reusable code is located in src/diamond_aiida_vasp/, separately from the sample data and configuration files provided in examples/.

Item	Contents
`examples/config/wf1.yml`	Configuration selection, relaxation, and analysis parameters for WF1
`examples/config/wf2.yml`	Temperatures, volumes, and AIMD parameters for WF2
`examples/cluster/`	Example YAML configurations for AiiDA computers and codes
`examples/data/wf1/`	Short example `XDATCAR` trajectory
`examples/data/wf2/`	Example `POSCAR` structures at different temperatures

These examples are intended to introduce the interface and check the preparation steps: they do not, on their own, constitute a scientifically validated dataset. Calculation outputs are stored separately, generally in runs/ and results/.

Requirements and limitations

VASP and its PAW/POTCAR pseudopotential files are not distributed with this project. Users must have the appropriate access rights and install these components in accordance with their licenses.
Cluster YAML files are templates that must be adapted: scheduler, paths, software modules, VASP executable, and computing resources vary between systems. The DIAMOND platform can assist users in adapting these files to their environment.
WF1 and WF2 can be used independently. Automatic chaining of the two modules and autonomous recovery from failed calculations are not yet part of this version.
Running actual calculations requires a working AiiDA profile, a configured VASP code, and registered POTCAR data. Results must be checked for convergence, trajectory representativeness, volume selection, and fit quality.

To set up your environment and run a first example, see the Installation page.

Scientific reference

The methodology is presented in Bissuel et al., Reproducible Container Solutions for Codes and Workflows in Materials Science (arXiv preprint: 2512.13826v1; published article, DOI:10.1002/adem.202503185).