Investigating the bimodality of liquid water via chemical bond descriptors and k-means clustering analysis

Liquid water exhibits anomalous behavior, with several of its physical and chemical properties deviating from those of ordinary liquids. Some of these anomalies stem from a bimodality observed in macroscopic properties such as density, isothermal compressibility, and specific heat, as well as the radial distribution function and vibrational spectra, which reflect the collective structural dynamics of liquid water. The infrared spectrum is particularly sensitive to hydrogen-bonded arrangements, thus it could be a crucial link between macroscopic properties and microscopic interactions. This study utilizes structural models that align with the experimental infrared spectrum of liquid water, offering a novel proof of concept. It bridges a critical gap by demonstrating how the bimodality of liquid water can be revealed through fully localized chemical bond descriptors. These descriptors are derived from local vibrational mode theory, the chemical bond overlap model and its topological extension, as well as the quantum theory of atoms in molecules. Key findings reveal that the local force constant, intra-overlap Coulomb repulsion, and Cremer and Kraka’s energy density associated with hydrogen bonds exhibit bimodal behavior. Notably, by employing the unsupervised machine learning algorithm K-means for clustering analysis, we identified distinct regimes of hydrogen-bond strength, revealing systematic patterns that differentiate the low-density liquid and high-density liquid networks. Overall, this study reinforces the concept that liquid water consists of two distinct types of local structures, providing new insights into its bimodal nature and advances in this field. • Local bond descriptors reveal bimodality in liquid water’s hydrogen bonds. • HDL and LDL show distinct, wide-ranging distributions of hydrogen-bond properties. • K-means clustering distinguishes LDL and HDL hydrogen-bonding regimes.