Feature vector clustering molecular pairs in computer simulations.

Area of Science:

• Computational chemistry and materials science.
• Development of novel analytical frameworks for molecular systems.

Background:

• Understanding microscopic structural organization in liquids and solutions is crucial.
• Existing methods may struggle with the complexity of molecular interactions and symmetry.

Purpose of the Study:

• To introduce a robust clustering framework for analyzing molecular pair organization in liquids.
• To develop a method for representing molecular pairs that accounts for symmetry.
• To apply and validate the framework using molecular dynamics simulations of an ionic liquid.

Main Methods:

• Representation of molecular pairs using representative vectors (RV) derived from key feature vectors (KFV).
• A novel scheme to transform KFV to RV, removing the influence of permutational molecular symmetry.
• Application of clustering analysis techniques (k-means, hierarchical) to classify RVs from molecular dynamics simulations.

Main Results:

• Successful categorization of molecular pairs in an ionic liquid ([P6,6,6,14][BOB]) into physically meaningful clusters.
• Validation of cluster effectiveness using product moment correlation coefficient (PMCC).
• Identification of representative configurations linked to local energy minima from DFT calculations.

Conclusions:

• The proposed framework efficiently reveals local molecular pair structures in complex liquids.
• The method is particularly useful for analyzing liquids and solutions with strong intermolecular interactions.
• Comparison of nonhierarchical and hierarchical clustering algorithms highlights the framework's versatility.