Towards Self-Consistent Graph Neural Networks for Predicting the Ideal Gas Heat Capacity, Enthalpy, and Entropy

Ideal gas heat capacity correlations are indispensable for modelling energy systems and evaluating process efficiency. While most correlations are empirical in nature, few are theoretically motivated, where the model parameters reflect physical quantities relating to the molecule. These however are rarely modelled through quantitative structure-property relationships, which hinders extending their applicability to new compounds. This work provides a realisation of a hybrid model that combines data-driven modelling in the form of a graph neural network that outputs a set of parameters used for the ideal gas heat capacity correlation. The study covered over 22,000 data points across 1,909 organic compounds resulting in a mean absolute error of 31.97 J/mol-K, a mean relative error of 11.63% and a correlation coefficient of 0.97.