Accelerating structure search using atomistic graph-based classifiers.

Area of Science:

• Computational materials science
• Chemical physics
• Data analysis

Background:

• Analyzing large datasets of atomic structures is computationally challenging.
• Identifying distinct structural configurations from potential energy surfaces is crucial for materials discovery.
• Global optimization algorithms can stagnate in local minima, hindering efficient exploration of the energy landscape.

Purpose of the Study:

• To introduce a novel atomistic classifier for distinguishing atomic structures.
• To integrate this classifier into a global optimization algorithm to enhance search efficiency.
• To demonstrate the classifier's effectiveness in solving complex global optimization problems.

Main Methods:

• Developed an atomistic classifier combining spectral graph theory and Voronoi tessellation.
• Integrated the classifier as a filtering mechanism within the Global Optimization with First-principles Energy Expressions (GOFEE) algorithm.
• Applied the method to global optimization problems for various systems, including nanoparticles and crystal structures.

Main Results:

• The atomistic classifier successfully discriminates between structures from different potential energy surface minima.
• Incorporation into GOFEE effectively filters out structures from explored regions, reducing search stagnation.
• The method efficiently solved global optimization problems for 2D pyroxene, 3D olivine, Au12, and LJ55/LJ75 nanoparticles.

Conclusions:

• The developed atomistic classifier is a valuable tool for analyzing and sorting large atomic datasets.
• Integrating this classifier into global optimization significantly improves the efficiency and robustness of exploring potential energy landscapes.
• This approach offers a promising strategy for accelerating materials discovery and understanding complex atomic systems.