WITHDRAWN: Kolmogorov-Arnold Networks for Predicting Dense Gas Thermodynamics

Accurate and efficient modeling of dense-gas thermodynamics is critical for high-fidelity CFD simulations in applications involving supercritical fluids, compressible flows with real-gas effects, and advanced propulsion systems. Traditional equation-of-state evaluations in such regimes are often expensive and can become a bottleneck in large-scale simulations. In this work, we investigate Kolmogorov–Arnold Networks (KANs) as machine learning surrogates for dense-gas thermodynamic relationships, using a dataset spanning a wide range of thermodynamic states. We systematically compare KAN-based surrogates against a suite of standard machine learning models, including multilayer perceptrons, gradient-boosted trees, Nearest Neighbor methods, and other competitive approaches. Our results show that KANs can achieve equal or better fit while providing improved interpretability through their structured functional representation. Moreover, KANs exhibit computational efficiency in inference, making them attractive for integration into CFD solvers where thermodynamic properties must be queried repeatedly. The proposed approach highlights the potential of KANs as practical, physically meaningful, and scalable surrogates for dense-gas dynamics in next-generation CFD applications.