xnn.common.data.hub.lode_dimers#

LODE non-bonded interactions dataset builder (molecular dimers & toy systems).

The dataset accompanies the extension of the long-distance equivariant (LODE) framework to diverse long-range interactions, and bundles several sub-datasets: biomolecular sidechain dimers with DFT energies & forces, their isolated monomers, two point-charge toy systems (pure Coulomb and pure 1/r**6 dispersion), and Xenon dimers/trimers.

All files are ASE-readable extxyz in eV / eV.A units (positions in angstrom), carrying rich per-frame info – most usefully label on the biomolecular dimers, tagging each pair by the polarity of its two fragments: CC (charged-charged), CP (charged-polar), PP (polar-polar), plus AA / CA / PA involving apolar fragments.

One additional subset, bio_scan, ships inside the repository (under datasets/lode_dimers/, no download): a curated distance scan of three charged/polar dimers (one CC, one CP, one PP, ~13 separations each), used by the long-range example notebooks (LES, BAMBOO electrostatics).

Reference#

K. K. Huguenin-Dumittan, P. Loche, H. Ni, M. Ceriotti, “Physics-inspired equivariant descriptors of non-bonded interactions”, J. Phys. Chem. Lett. (2023). Data: https://archive.materialscloud.org/records/405an-d8183 (DOI 10.24435/materialscloud:23-99).

Notes

Frames are periodic only nominally – each system sits in a large (>=30 A) cubic box so it is effectively isolated – so cell / pbc are populated from the file.
Reading extxyz requires the ase extra (pip install xnn[ase]).

Classes

LODEDimersBuilder()

Builder for the LODE non-bonded interactions dataset.