Answers are grounded in this paper and cite their sources.
Glass formation is one of the most important and fundamental open problems in glass science. Predicting whether a liquid will form a glass appears simple but is considerably complex. In this work, we address glass formation in inorganic nonmetallic liquids using binary classification to predict the probability that a given liquid will form a glass under typical laboratory conditions. Using a dataset of more than 50 000 experimental observations, we trained random forest classifiers that achieved ROC–AUC values of 0.87 and macro‐averaged PR‐AUC of 0.85 on the holdout dataset (i.e., unseen data). Model selection included hyperparameter optimization with Bayesian search and cross‐validation. Two feature representations were evaluated: raw atomic fractions and physicochemical descriptors derived from the compositions. The models are well‐calibrated, enabling users to adjust the decision threshold to balance false‐positive and false‐negative rates according to their specific application. Using SHAP values, we extracted valuable insights from the trained models that are consistent with established glass science and provide new empirical insights. For example, we identified that the weighted sum of DFT bandgap energy of the constituent elements is positively correlated with glass formation, a finding consistent with the electronic character of known network‐forming elements. This result is particularly relevant for composition screening, especially in inverse design problems. The collection of models developed in this work is free software and was named VITRIFY ( Vitrification Inference Tool for Rapid Identification of glass‐Forming abilitY ).
No takes yet. Share an insight, caveat, or question.
Carvalho et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: