Accelerating the search for carbon cluster isomers via machine learning potential

HD Nguyen and PH Nguyen and GH Bach and OKT Nguyen, JOURNAL OF MOLECULAR MODELING, 32, 28 (2025).

DOI: 10.1007/s00894-025-06613-3

ContextTrivalent C22\documentclass12ptminimal \usepackageamsmath \usepackagewasysym \usepackageamsfonts \usepackageamssymb \usepackageamsbsy \usepackagemathrsfs \usepackageupgreek \setlength\oddsidemargin-69pt \begindocument$$_22$$\enddocument cages are explored with the recent generation of Gaussian approximation potential (GAP-20) and density functional theory (DFT) calculations. Using the GAP-20 to approximate the energy landscape significantly reduces the search time and provides superior starting structures for DFT optimization. The GAP-20, however, fails to capture the Jahn-Teller distortion. The relative GAP-20 energies are overestimated, and the vibrational modes/frequencies are poorly characterized.MethodsVia the CALYPSO package, particle swarm optimization is employed to explore the configuration space. Energies/forces are calculated via the QUIP/LAMMPS module, employing the GAP-20. Structural relaxation follows the conjugate gradient method to a force precision of 10-7eV/\documentclass12ptminimal \usepackageamsmath \usepackagewasysym \usepackageamsfonts \usepackageamssymb \usepackageamsbsy \usepackagemathrsfs \usepackageupgreek \setlength\oddsidemargin-69pt \begindocument$$10<^>-7 \text eV/$$\enddocument& Aring;. Cage isomers are further optimized by DFT, utilizing BFGS optimization scheme, until forces are below 0.05 eV/& Aring;. The B3LYP/6-31+G & lowast;\documentclass12ptminimal \usepackageamsmath \usepackagewasysym \usepackageamsfonts \usepackageamssymb \usepackageamsbsy \usepackagemathrsfs \usepackageupgreek \setlength\oddsidemargin-69pt \begindocument$$\text B3LYP/6-31+G*$$\enddocument level of theory is applied, as implemented in the NWChem software. Vibrational analysis is performed to study the stability/infrared spectra.

Return to Publications page