Search for H-Bonded Motifs in Liquid Ethylene Glycol Using a Machine Learning Strategy

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Atomic position trajectories from ab initio molecular dynamics (AIMD) simulations contain rich structural information.
• Analyzing these trajectories for recurrent patterns in hydrogen-bonded liquids is computationally challenging.

Purpose of the Study:

• To apply a machine learning strategy to identify recurrent hydrogen-bonded fragments in liquid ethylene glycol (EG).
• To investigate the utility of self-organizing maps (SOM) for analyzing AIMD trajectories.

Main Methods:

• Utilized a self-organizing maps (SOM) machine learning approach, a type of artificial neural network.
• Analyzed AIMD simulation trajectories of liquid ethylene glycol (EG).
• Clustered hydrogen-bonded fragments based on molecular conformation and hydrogen-bond connectivity.

Main Results:

• The SOM successfully mapped high-dimensional trajectory data onto a 2D grid, preserving topological properties.
• Identified a hydrogen-bonded cyclic dimer as a recurring motif.
• Discovered a bifurcated hydrogen-bonded structure as another recurring motif in liquid EG.

Conclusions:

• SOM-based machine learning provides an efficient method for identifying structural motifs in AIMD simulations.
• Recurrent hydrogen-bonded structures, including cyclic dimers and bifurcated motifs, are present in liquid ethylene glycol.