Classification of the HCN isomerization reaction dynamics in Ar buffer gas via machine learning

Area of Science:

• Chemical Physics
• Computational Chemistry
• Machine Learning Applications

Background:

• The isomerization of hydrogen cyanide (HCN) to cyanamide (CNH) is a fundamental chemical reaction.
• Understanding reaction dynamics, especially in the presence of inert atoms like Argon (Ar), is crucial for chemical kinetics.
• Traditional methods for studying reaction dynamics can be computationally intensive.

Purpose of the Study:

• To investigate the effect of Argon (Ar) on the HCN ⇄ CNH isomerization reaction dynamics.
• To apply machine learning techniques for analyzing reaction phase space and predicting reactivity.
• To elucidate how the Ar atom influences the reaction pathway and transition state.

Main Methods:

• Ab initio calculations at the CCSD(T)/aug-cc-pVQZ level to develop the potential energy surface.
• Classical trajectory simulations to generate reaction data.
• Machine learning models trained on initial conditions (positions and momenta) to predict reactivity.
• Analysis of machine learning model predictions to understand the role of the Ar atom.

Main Results:

• Machine learning models achieved over 95% prediction accuracy for reaction occurrence.
• The models successfully identified reactivity boundaries without prior knowledge of reaction dynamics theory.
• The presence of an Ar atom was found to displace the effective saddle point of the reaction.
• Specific positioning of the Ar atom near the N or C atom was shown to hinder forward or backward reactions, respectively.

Conclusions:

• Machine learning is a powerful tool for analyzing complex reaction dynamics and predicting chemical reactivity.
• The Ar atom significantly influences the HCN ⇄ CNH isomerization by modifying the reaction's transition state.
• ML-aided analyses offer a promising avenue for deeper insights into chemical reaction mechanisms.