Symbolic AI Improves Vapor-Liquid Equilibrium Prediction

Bongseok Kim, Suman Chakraborty, Gary Huang, Mehek Mathur, Guang Lin, Li Qiao· August 13, 2026 View original

Key takeaways

  • Traditional equations of state struggle with VLE prediction for hydrocarbon-nitrogen mixtures.
  • Symbolic machine learning provides interpretable corrections to the Peng-Robinson equation of state.
  • A two-level strategy generalizes these corrections across different hydrocarbon systems.
  • The approach significantly improves prediction accuracy while maintaining model transparency.

Who benefits

Chemical EngineeringOil & GasPetrochemicalsMaterials ScienceManufacturing

Summary

This research proposes a symbolic machine learning approach to enhance vapor-liquid equilibrium (VLE) predictions for hydrocarbon-nitrogen mixtures. It discovers interpretable symbolic corrections to the Peng-Robinson equation of state, significantly improving accuracy while maintaining interpretability, unlike deep learning models.

This paper addresses the challenge of accurately predicting vapor-liquid equilibrium (VLE) in hydrocarbon-nitrogen mixtures, a task where traditional cubic equations of state often fall short. While deep learning models can offer accuracy, they typically lack interpretability. The researchers introduce a symbolic machine learning method to derive interpretable symbolic corrections for the Peng-Robinson equation of state (PR-EOS). The approach employs a two-level strategy: first, it identifies symbolic expressions for individual hydrocarbon systems. Then, it represents the coefficients of these expressions as functions of the carbon number, enabling accurate predictions across a range of different hydrocarbon systems. The results demonstrate a substantial improvement in prediction accuracy over the original PR-EOS, providing a more transparent and understandable model for VLE prediction.

Why it matters

For chemical engineers and materials scientists, this offers a powerful tool for more accurate and interpretable VLE predictions, crucial for process design, optimization, and safety in industries dealing with hydrocarbon-nitrogen mixtures.

How to implement this in your domain

  1. 1Explore integrating symbolic machine learning techniques to derive interpretable corrections for existing physical models in your domain.
  2. 2Apply a two-level symbolic approach to generalize corrections across different system variations (e.g., varying chain lengths or compositions).
  3. 3Compare the interpretability and accuracy of symbolic models against deep learning alternatives for critical engineering predictions.
  4. 4Utilize the discovered symbolic expressions to gain deeper insights into the underlying physical phenomena.
  5. 5Incorporate these improved prediction models into process simulation and design software.

Original post by Bongseok Kim, Suman Chakraborty, Gary Huang, Mehek Mathur, Guang Lin, Li Qiao

"arXiv:2608.11255v1 Announce Type: new Abstract: Accurate prediction of vapor--liquid equilibrium (VLE) for hydrocarbon-nitrogen mixtures remains challenging for cubic equations of state, particularly across broad ranges of composition and hydrocarbon chain length. While deep lear…"

View on X

Originally posted by Bongseok Kim, Suman Chakraborty, Gary Huang, Mehek Mathur, Guang Lin, Li Qiao on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses