(551b) Properties and Phase Behavior of Supercooled Water from First Principles

Water's biological, chemical, and industrial importance has made it an object of intense scrutiny for decades. Apart from the search for an applications-driven understanding of water (e.g., desalination, cloud microphysics), water's anomalous thermophysical properties and complex phase behavior provide fertile ground for advances in fundamental understanding in the general areas of statistical thermodynamics, phase transitions, and metastability. In the last few decades, computer simulations have played a key role in pushing forward our picture of water's properties. Much activity has been focused around developing computational models to predict water's properties, especially for conditions and phenomena that are challenging to study experimentally. Recently, we developed an ab initio deep neural network (DNN) model for water based on density functional theory calculations, and performed enhanced sampling molecular simulations to demonstrate that the properties of this model are consistent with the existence of a metastable liquid-liquid phase transition under deeply supercooled conditions [PNAS, 117 (42), 26040-26046 (2020)]. In this talk, we will briefly review the construction and properties of our DNN model, followed by a discussion of our ongoing work to rigorously probe liquid-liquid coexistence in this model and refine our estimate for the location of the metastable liquid-liquid critical point. We end with a discussion of how our simulation results (obtained entirely from first principles) relate to important experimental observations of supercooled water's properties and metastable phase behavior.