Methods for efficient eutectic screening in CALPHAD databases

Eutectic points play a critical role in material science due to their low melting points, absence of a solidification range, and formation of unique microstructures that enhance properties. However, identifying eutectics in calculated ternary and higher-order systems remains computationally challenging due to the lack of a well-established screening approach. This work addresses this gap by developing a general, efficient methodology for efficient eutectic screening in multicomponent systems. We first propose and benchmark four CALPHAD-based methods for eutectic identification, evaluating their computational speed and reliability. We then investigate the potential of combining CALPHAD with machine learning to further accelerate the process, leveraging two eutectic properties: (1) eutectics are local minima on the liquidus, and (2) they lie at the intersections of primary crystallization regions. Active learning surrogate models are employed to interpolate CALPHAD data for predicting liquidus and primary crystallization phases, and the performance of each approach is evaluated. By benchmarking multiple strategies and releasing code for three approaches that enable rapid and reliable eutectic screening in high-order systems, this work advances the design of materials such as cast alloys, phase change materials, eutectic high entropy alloys, or metallic glasses.