Empirical valence bond models for reactive potential energy surfaces: a parallel multilevel genetic program approach

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Method Development

Background:

• Empirical valence bond (EVB) potential energy surfaces (PES) are crucial for molecular simulations.
• Accurate PES construction requires efficient exploration of function and parameter spaces.
• Ab initio electronic structure calculations provide high-fidelity energy data.

Purpose of the Study:

• To develop a novel computational method for constructing accurate EVB PES.
• To improve the efficiency and accuracy of genetic programming for PES fitting.
• To apply the new method to model proton transfer reactions.

Main Methods:

• Development of a parallel multilevel genetic program (PMLGP).
• Utilizing a hierarchical genetic programming approach with coevolving populations.
• Optimization of genetic operator probabilities by a higher-level genetic program (HLGP).
• Testing against standard parallel genetic programming methods.

Main Results:

• The PMLGP demonstrated significant improvements in accuracy and efficiency over standard methods.
• The method successfully generated an accurate EVB model for proton transfer.
• The HLGP dynamically adapted genetic operators for better population fitness.

Conclusions:

• PMLGP offers a powerful and efficient approach for constructing EVB PES.
• The developed method advances computational modeling of chemical reactions.
• Accurate EVB models are achievable for complex systems like proton transfer in different environments.