We describe current approaches to thermodynamic modelling of liquids for the CALPHAD method, the use of available experimental methods and results in this type of modelling, and considerations in the use of atomic-scale simulation methods to inform a CALPHAD approach. We begin with an overview of the formalism currently used in CALPHAD to describe the temperature dependence of the liquid Gibbs free energy and outline opportunities for improvement by reviewing the current physical understanding of the liquid. Brief descriptions of experimental methods for extracting high-temperature data on liquids and the preparation of undercooled liquid samples are presented.Properties of a well-determined substance, B 2 O 3 , including the glass transition, are then discussed in detail to emphasize specific modelling requirements for the liquid. We then examine the two-state model proposed for CALPHAD in detail and compare results with experiment and theory, where available. We further examine the contributions of atomic-scale methods to the understanding of liquids and their potential for supplementing available data. We discuss molecular dynamics (MD) and Monte Carlo methods that employ atomic interactions from classical interatomic potentials, as well as contributions from ab initio MD. We conclude with a summary of our findings.1 Introduction Although few inorganic materials are used in their liquid state, liquids play an important role in the manufacture of many materials. This can be either because they are processed from the liquid or, for fulfilling performance criteria, the formation of the liquid phase must be avoided. Additionally, there are now a number of commercially available amorphous solids, both in the form of oxides and metallic glasses, and accurate descriptions of these sys-tems are necessary over a wide range of temperatures. How-ever, liquid phases of many inorganic substances are only sta-ble at high temperatures where the experimental determination of their properties is more challenging than at lower tem-peratures. Many amorphous phases are, at best, metastable and their properties can be difficult to measure. In addition, the data from glassy phases cannot be directly applied to modelling liquids because they have gone through the glass transition, with the attendant sharp drop in heat capacity. As a result, the experimental data for generating and evaluating models of liquids and glasses are fairly sparse. This has a significant impact on the development and refinement of models that strongly depend on the availability of thermodynamic data, such as those needed in CALPHAD modelling.CALPHAD is one of the major ways to model the thermodynamics of liquids. However, for many systems, especially the unaries (elements and sometimes compounds), the data necessary to develop, parameterize and validate models are limited. Theoretical approaches, e.g. first-principles or classical atomistic simulation, can help provide data
