UC Merced

Liquid water, a ubiquitous and vital substance, exhibits a range of fascinating and anomalous properties. Understanding its molecular behaviors is crucial for elucidating its unique roles in chemistry, biology, and materials science. This dissertation presents a series of theoretical investigations aimed at improving the theoretical modeling of liquid water and advancing our knowledge of its structure, dynamics, and thermodynamic properties.

First, we systematically scrutinize various electronic structure methods and charge models, evaluating their performance in predicting dipole moments of isolated water, water clusters, and liquid water, as well as charge transfer in water dimer and liquid water. We identify the Iterative Hirshfeld method as the best performing charge model to assign partial atomic charges for liquid water. Our final pragmatic quantum-chemical charge assigning protocol for liquid water is the Iterative Hirshfeld method and a quantum region with a cutoff radius of 5.5 Å.

With the training data from our quantum-chemical charge-assigning protocol, we develop a machine-learning (ML) model trained to assign point charges for water in a post-MD manner. The resulting model improves the predictions of the dielectric constant and low-frequency IR spectrum of liquid water. Our analysis reveals that polarization dominates the enhancement of the dielectric constant of liquid water, and charge transfer is primarily responsible for the hydrogen-bond stretch peak at 200 cm$^{-1}$ in the IR spectrum.

Furthermore, we develop an ML potential for liquid water, including long-range interactions. This model reproduces the potential energy of a water dimer as a function of the separation between the water molecules, and reliably predicts the dipole moment of the dimer simultaneously, allowing a consistent treatment of potential energy and dipole moment surfaces.

Finally, we examine the behaviors of an antifreeze protein (AFP), DAFP-1, in two rigid non-polarizable water models. DAFP-1, an amphiphilic protein, remains at the water/air interface in the experiment. However, it sinks in the simulation with TIP3P water model, whereas staying at the interface with TIP4P/2005. Using umbrella sampling, we calculate the free energy profiles for the sinking process for each water model and investigate how different water models influence the behaviors of AFPs.