Web tool builds and energy-minimises silver, copper oxide and titania nanoparticles

The work

Technical · from the record

ASCOT is a web tool, hosted on the Enalos Cloud Platform, that builds electrically neutral, stoichiometric spherical nanoparticles of Ag, CuO and TiO2 (anatase and rutile) from selected Crystallography Open Database unit cells, energy-minimises them with a classical force field in LAMMPS, and then computes atomistic descriptors such as average potential energy per atom, average coordination number, common neighbour parameter and hexatic order parameters for the whole particle and separately for its core and shell. The authors ran the tool over particle diameters from 2.5 nm increasing by 0.5 nm up to 7.0 nm and report that average potential energy per atom decreases with size for all four materials over that range, that rutile particles are slightly lower in energy than anatase, and that the choice of force field changes the modelled surface, with a MEAM force field expanding a 3.5 nm anatase particle by 4.5% in x and y and 2.7% in z relative to the geometric construction. No machine learning model is trained or run in the study; the descriptors are presented as intended inputs for later QSAR and machine learning models of nanoparticle toxicity.

How AI was used

AI was not used. The workflow is crystallographic and physics-based throughout: a pre-selected CIF file per material is replicated along the lattice vectors to the smallest box enclosing the requested sphere, atoms outside the target diameter are deleted, excess-species atoms in a 0.02 Å outer shell are removed (with a fixed random seed where more candidates exist than are needed) until the unit-cell stoichiometry is restored, the triclinic box is converted to orthorhombic and extended by 10 Å per edge to suppress periodic self-interaction, and conjugate-gradient energy minimisation is applied with a force field drawn from the OpenKIM database or, for CuO and TiO2 by default, the internally integrated COMB3 potential, halting on energy tolerance, force tolerance, maximum iterations or maximum force evaluations. Descriptors are then computed by fixed formulas with element-specific neighbour cutoffs based on ionic radii, over core and shell regions separated by a constant 4 Å shell depth; that depth was set by passing ASCOT XYZ files to the external Shell Depth Calculator, which locates the maximum-curvature point with the Kneedle algorithm, a rule-based procedure rather than a learned model. The resulting descriptors are offered as inputs for subsequent machine learning or QSAR model development, which lies outside this paper.