Neural network atomistic potentials for global energy minima search in carbon clusters

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• Global energy optimization is crucial for determining compound properties, especially for atomic clusters.
• The complexity of energy landscapes increases exponentially with cluster size, necessitating efficient search methods.
• Neural network (NN) potentials offer a computationally efficient approach to molecular structure optimization.

Purpose of the Study:

• To evaluate the applicability of ANI-1ccx and ANI-nr NN potentials for global minima optimization in carbon clusters.
• To determine if NN potentials trained on specific datasets can accurately represent potential energy surfaces (PES) for related domains.
• To assess the robustness of NN potentials in navigating complex energy landscapes for carbon clusters.

Main Methods:

• Testing ANI-1ccx and ANI-nr NN potentials for global minima searches on carbon clusters (C_n, n=3-10).
• Incorporating cluster connectivity restrictions to guide the optimization process.
• Utilizing Density Functional Theory (DFT) or ab initio calculations in conjunction with NN potentials.

Main Results:

• ANI-1ccx and ANI-nr potentials demonstrated effectiveness in identifying global minimum structures for small carbon clusters.
• The NN potentials, when combined with connectivity constraints and further calculations, proved capable of capturing the global minimum for larger clusters like C₂₀.
• These NN potentials function as reliable pre-samplers of the potential energy surface.

Conclusions:

• ANI-1ccx and ANI-nr NN potentials are robust tools for pre-sampling potential energy surfaces in carbon cluster global minima optimization.
• The integration of connectivity restrictions enhances the reliability of NN potentials for complex systems.
• NN potentials show significant promise for accelerating the discovery of stable atomic cluster structures.