Enhancing Automation and Interpretability of Vibrational Spectra Predictions for Water Clusters from Diffusion Monte Carlo.

A method for predicting vibrational spectra of water clusters from wave functions sampled by diffusion Monte Carlo (DMC) is developed to enhance automation, generalizability, and interpretation. This method builds on the established ground-state probability amplitude (GSPA) approach to vibrational spectra predictions and is applied to neutral water clusters, systems defined by high dimensionality and having significant nuclear quantum effects. We develop a chemically informed singular value decomposition (SVD) approach to automate the selection of internal coordinates for vibrational analysis, along with an optimization-based reverse mapping method to visualize vibrational motions in Cartesian space. Both developments are generalized to handle water clusters of varying sizes. The framework is assessed on the q-SPC/Fw potential energy surface, and we find that the chemically informed SVD yields accurate and basis-invariant spectroscopic predictions across all clusters studied, fully addressing the bias observed toward intermolecular overrepresentation of the standard SVD approach. Additionally, we systematically benchmark DMC sampling and descendant weighting convergence to ensure the reliability of the ground-state probability amplitude inputs used in the vibrational analysis. Together, these developments establish an automated and interpretable framework for vibrational spectra prediction from DMC, with potential applicability to a wide range of molecular systems beyond water clusters.