Machine learning-assisted high-throughput virtual screening of novel energetic materials.

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• High-energy density materials (HEDMs) are crucial for defense and industry.
• A key challenge is balancing high energy output with low mechanical sensitivity.
• Traditional methods for HEDM development are inefficient, costly, and risky.

Purpose of the Study:

• To develop an integrated computational strategy using DFT and ML for efficient HEDM discovery.
• To identify novel energetic compounds with an optimal balance of energy and safety.
• To establish a feasible computational paradigm for rapid HEDM design.

Main Methods:

• Constructed a molecular library of over 10,000 derivatives based on a pyrazole scaffold.
• Calculated 10 key properties for 439 representative molecules using DFT (B3LYP/6-31+G(d,p)).
• Evaluated five ML algorithms, selecting XGBoost for high-throughput virtual screening and DFT validation of top candidates.

Main Results:

• The XGBoost model achieved an average test set R² of 0.90, demonstrating high predictive performance.
• High-throughput virtual screening identified 20 promising energetic compounds.
• DFT verification confirmed the reliability and favorable energy-safety balance of the identified candidates.

Conclusions:

• The integrated DFT and ML approach provides a computationally efficient and reliable method for discovering HEDMs.
• This strategy overcomes limitations of traditional methods, enabling rapid identification of high-energy, low-sensitivity materials.
• The study presents a viable computational paradigm for advancing HEDM research and development.