Matrix Completion for Pseudodata Generation of Mixture Thermophysical Properties: A Case Study in Excess Enthalpy

Composition-dependent thermophysical and transport properties are essential for accurate modeling and simulation of chemical processes, but data for parametrization are often sparse and expensive to generate. Matrix completion methods (MCMs), successfully applied to scalar-valued properties, have not been extended to higher-order properties. This work investigates the potential of MCMs to leverage the broad availability of binary mixture data to predict properties with such dependencies. Excess enthalpy in binary liquid mixtures (H^E) is used as a case study. By framing the development as pseudodata generation rather than direct prediction, we emphasize the role of data-driven approaches in improving fundamental models, notably for difficult properties like H^E. Using a simple coherence constraint, we show that parallelized MCMs for 1012 mixtures from 11 functional groups across 7 temperatures outperform the benchmark predictive UNIFAC (Dortmund) in 88% of systems. This study demonstrates the viability of pseudodata generation for higher-order thermophysical and transport properties.