Machine Learning Driven Advances in Molecular Dynamics of Bulk and Interfacial Aqueous Systems

Area of Science:

• Computational chemistry and materials science.
• Application of artificial intelligence in physical sciences.

Background:

• Molecular dynamics (MD) simulations are vital for studying aqueous and interfacial systems, crucial for energy materials and life sciences.
• Current MD simulations face limitations in force field accuracy, simulation size, and timescale, hindering comprehensive analysis.
• Machine learning (ML) offers a promising avenue to overcome these challenges by improving interaction descriptions and enhancing sampling.

Purpose of the Study:

• To review the principles, implementation, and applications of ML force fields (MLFFs) and ML-enhanced sampling in aqueous and interfacial systems.
• To highlight how ML integration addresses accuracy and computational cost limitations in traditional MD simulations.
• To explore the use of ML-driven data analytics for interpreting complex simulation data.

Main Methods:

• Integration of ML methods, specifically MLFFs, for describing interatomic interactions with quantum-level accuracy.
• Coupling MLFFs with enhanced sampling techniques and ML-driven data analytics, including graph-based approaches.
• Application of these combined methods to diverse systems: bulk water, interfaces, proton transfer, catalysis, phase transitions, and vibrational spectra prediction.

Main Results:

• MLFFs achieve quantum chemistry accuracy at classical computational cost, enabling large-scale simulations (nanoseconds, thousands of atoms).
• ML-enhanced sampling overcomes significant reaction barriers and explores vast configuration spaces, previously computationally prohibitive.
• ML models reveal overlooked factors, like solvent dynamics in phase transitions, and facilitate high-dimensional free energy surface calculations.

Conclusions:

• MLFFs and enhanced sampling significantly advance the study of aqueous and interfacial systems, providing unprecedented physical insights.
• These ML-based approaches make computationally demanding simulations feasible, leading to a better understanding of chemical reactions and material properties.
• Future research should focus on further integrating ML into MD simulations to address current challenges and unlock new scientific discoveries.