Phase diagrams are important for understanding a material’s behavior and thermodynamic stability under different conditions, but they can be difficult to predict via simulations due to the challenge of estimating chemical potentials. Chew and Reinhardt review the history of methods aimed to compute phase diagrams, focusing on the thermodynamic phase behavior of materials.

The authors provide an overview of computational methods to determine phase equilibria that scientists have progressively developed since the 1980s along with their advantages and disadvantages for various applications. They discuss the modern machine learning approach to predict phase diagrams without running first principle simulations.

“We explain why one might care about thermodynamics and phase behavior to begin with, why it is challenging and what insights we can gain by doing this,” author Aleks Reinhardt said. “We provide a number of tips and tricks concerning many of these methods we describe, including outlining when and why things can go wrong.”

Phase diagrams can provide a useful check on new theoretical models.

“Since phase behavior is such a stringent test of a potential, working it out can give us confidence that a model is good, or indeed that it is less than good,” Reinhardt said.

Reinhardt added that many significant advances in the field have occurred only recently.

“Huge strides have been made in developing ever-better computationally tractable potentials for many systems of interest. It is only in the last two years, for example, that the phase behavior of water has been explored at a quantum-mechanical level of description,” Reinhardt said.

The team hopes their work will help others understand the most suitable methods for their applications while avoiding common pitfalls.

Source: “Phase diagrams-Why they matter and how to predict them,” by Pin Yu Chew and Aleks Reinhardt, Journal of Chemical Physics (2023). The article can be accessed at