Machine learning model for non-equilibrium structures and energies of simple molecules.

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning Applications

Background:

• Machine learning (ML) offers a promising approach to predict molecular properties with high accuracy and computational efficiency.
• Traditional methods like Density Functional Theory (DFT) provide accurate results but are computationally expensive for large-scale simulations.

Purpose of the Study:

• To train kernel-based ML models using Bag of Bonds and many-body tensor representations for predicting molecular properties.
• To utilize trained ML models for optimizing molecular structures via Metropolis Monte Carlo (MMC) simulations with simulated annealing.

Main Methods:

• Training kernel-based ML models on datasets of non-equilibrium molecular structures for six small molecules.
• Employing Bag of Bonds and many-body tensor representations for feature engineering.
• Performing Metropolis Monte Carlo (MMC) simulations with simulated annealing using the trained ML models.

Main Results:

• ML models accurately predicted atomization energies for the studied molecules.
• Optimized molecular structures and energies obtained via ML-driven MMC simulations showed strong agreement with DFT results.
• The ML approach achieved quantum chemical accuracy at molecular mechanics speed.

Conclusions:

• Trained ML models can effectively perform classical simulations like MMC without relying on conventional force fields.
• This ML-driven methodology enhances simulation accuracy while significantly reducing computational cost.
• The study demonstrates the potential of ML for accelerating molecular property prediction and structure optimization in computational chemistry.