Trained MLIPs

This page lists the interatomic potentials developed by our group and discover what we’re currently working on.

N2

A machine-learned interatomic potential specifically designed to study the phase diagram of N₂ up to 10 GPa. The model uses pair_style tadah which requires the Tadah!LAMMPS plugin to be installed (Installing LAMMPS Interface). The potential is robust and highly transferable, although its performance above 10 GPa is unknown. At higher pressure, we expect that its predictions will gradually diverge from experimental results. It won’t support high-pressure polymerisation or bond breaking in general due to its rigid nature.

Tadah!N2

Understanding solid nitrogen through molecular dynamics simulations with a machine-learning potential

Kr

A two-body interatomic potential that performs exceptionally well given its simplicity. Two versions are provided. The first model (Tadah!Kr1) was fitted to CCSD(T) dimers, while the second model’s energies were corrected using equilateral Kr trimers (Tadah!Kr2). The latter model is recommended.

Both potentials show very good agreement with experimental data for the equation of state, melting point, and neutron scattering of the fluid. They offer significant improvements over many empirical potentials at higher pressures while retaining applicability at ambient conditions.

Potentials are provided in both the native Tadah! format, which uses pair_style tadah, and the LAMMPS tabulated format, used with pair_style table.

We have re-optimized both potentials using hyperparameter optimization (HPO) to make them compatible with the current version of Tadah!. The updated potentials have been enhanced by incorporating the classical ZBL potential for energies above 1 keV, smoothly joined to the existing CCSD(T)-based model. In other words, the Tadah! potentials are stiffer in the core electron region, where the ZBL screening function is often inadequate.

Below are two original potentials designed for an older version of Ta-dah! (pair_style tadah/tadah), which is discontinued but still available online: Ta-dah old version.

An Accurate Machine-Learned Potential for Krypton under Extreme Conditions