Accelerating the search for carbon cluster isomers via machine learning potential.

Area of Science:

• Computational materials science
• Quantum chemistry

Background:

• Exploration of trivalent C22 cages using Gaussian approximation potential (GAP-20) and density functional theory (DFT).
• GAP-20 accelerates energy landscape approximation and provides initial structures for DFT optimization.
• GAP-20 limitations include failure to capture Jahn-Teller distortion and inaccurate energy/vibrational predictions.

Purpose of the Study:

• To investigate the structural and vibrational properties of trivalent C22 cages.
• To evaluate the performance of GAP-20 in approximating the energy landscape for C22 cages.
• To refine structural and vibrational characterization using DFT.

Main Methods:

• Particle swarm optimization via CALYPSO package to explore configuration space.
• Energy and force calculations using QUIP/LAMMPS with GAP-20, followed by conjugate gradient relaxation.
• DFT optimization (B3LYP/6-31+G*) and vibrational analysis using NWChem for accurate characterization.

Main Results:

• GAP-20 provides efficient initial structural exploration but overestimates relative energies.
• GAP-20 fails to accurately model Jahn-Teller distortions in C22 cages.
• DFT calculations are essential for precise structural and vibrational frequency determination.

Conclusions:

• While GAP-20 is a useful tool for initial screening, DFT is indispensable for accurate analysis of C22 cage properties.
• The Jahn-Teller distortion is a critical factor not captured by GAP-20.
• Accurate vibrational analysis confirms the necessity of DFT for predicting stability and infrared spectra.