Detecting Liquid-Liquid Phase Separations Using Molecular Dynamics Simulations and Spectral Clustering.

Area of Science:

• Computational chemistry
• Materials science
• Statistical mechanics

Background:

• Empirical force fields require rigorous testing, particularly for predicting mixture phase diagrams.
• Identifying phase boundaries and critical points in mixtures is challenging due to subtle local environmental changes.
• Finite sampling and size effects complicate the analysis of local order parameters during demixing transitions.

Purpose of the Study:

• To rigorously test the accuracy of empirical force fields by analyzing mixture phase diagrams.
• To develop and apply methods for detecting subtle demixing transitions and critical points.
• To investigate the structural changes associated with demixing in a methanol/hexane mixture.

Main Methods:

• Molecular dynamics simulations of methanol/hexane mixtures at various temperatures.
• Computation of local and global structural properties.
• Application of spectral clustering to analyze the topology of the hydrogen-bond network.
• Testing the method on a Lennard-Jones system for comparison.

Main Results:

• Demixing transitions in methanol/hexane mixtures show abrupt changes in the H-bond network topology, despite continuous macroscopic changes.
• Spectral clustering reveals a fat-tailed distribution of cluster sizes near the critical point, consistent with percolation theory.
• The emergence of system-spanning clusters from smaller aggregates was identified as a key indicator of demixing.
• Spectral clustering successfully detected demixing in a non-hydrogen-bonded Lennard-Jones system.

Conclusions:

• Spectral clustering is a robust method for detecting subtle demixing transitions and critical points in mixtures.
• The topological analysis of H-bond networks provides sensitive insights into phase behavior.
• This approach enhances the validation of empirical force fields for complex mixtures.