Accurate prediction of phase transition properties—such as entropy of fusion, enthalpy of vaporization, boiling point, melting point, and critical properties—is essential for chemical process design, energy balance calculations, and equipment optimization. However, experimental determination of these properties is time-consuming and resource-intensive. In a study published in ENG. Chem. Eng., researchers at Tianjin University of Science and Technology and Shanghai Jiao Tong University report an enhanced group contribution method called ACGC+ that significantly improves prediction accuracy for nine key phase transition properties.
The ACGC+ method builds on the original atomic connectivity group contribution (ACGC) framework, which uses atomic adjacent groups and shape factors to describe molecular structure. The key innovation is the introduction of ACFs+, which capture not only the global position of each functional group within a molecule (as in the original method) but also the local position by considering the contribution of the core atom and nearby atoms. This dual description enables more precise differentiation of isomers and structural variations.
Comprehensive data sets were compiled from NIST, CRC Handbook, and DIPPR801, with 375 to 6,530 compounds per property. Models were developed using multiple linear regression and validated through external validation, leave-one-out cross-validation, and Y-randomization analysis.
The ACGC+ models demonstrated excellent predictive performance. For phase transition entropy and enthalpy, test R² values ranged from 0.906 (enthalpy of fusion) to 0.992 (enthalpy of vaporization). For boiling point and melting point, test R² values were 0.979 and 0.845, respectively. For critical properties (temperature, pressure, and volume), test R² values exceeded 0.989, with critical volume reaching 0.998. Mean absolute percentage errors were 3.48% for enthalpy of vaporization, 1.60% for boiling point, 0.91% for critical temperature, and 1.30% for critical volume.
Compared to the original ACGC method, ACGC+ reduced mean absolute errors by 1.44% to 7.91% across all properties. Cross-validation Q² values ranged from 0.820 (melting point) to 0.998 (critical volume), and Y-randomization analysis confirmed that the model performance was not due to chance correlation, indicating strong robustness and no overfitting.
Comparative analysis with existing methods—including the Constantinou–Gani (CG), Marrero–Gani (MG), MG+, and Sun–Sahinidis (SS) group contribution methods—showed that ACGC+ maintains competitive or superior predictive performance, particularly for critical pressure and critical volume, despite using larger and more diverse data sets in many cases. For example, the ACGC+ data set for boiling point (5,406 compounds) is 1–7 times larger than those used in other methods, yet the model still delivers high accuracy.
This work demonstrates that incorporating local structural information through ACFs+ provides a practical and effective strategy for improving group contribution predictions. The ACGC+ method offers a robust and reliable tool for estimating critical phase transition properties essential for chemical process engineering.
DOI
10.1007/s11705-026-2675-1