Thermophysical property data are of fundamental importance for all chemical engineering processes. Although large databases of experimental thermophysical property data exist, the relevant properties have been measured for only a fraction of known substances. This is due to the considerable time and cost involved in generating experimental data, as well as the large number of substances of interest and the even greater number of mixtures that can be formed from them. Therefore, predictive methods for the thermophysical properties of mixtures are essential. Alongside established physics-based methods, machine learning (ML) approaches are increasingly being applied.

Developing accurate ML models requires large amounts of experimental data. In addition to publicly available literature data, substantial amounts of experimental data are stored in the databases of chemical companies. However, these data are generally not publicly accessible. My research therefore focuses on methods that enable the training and further development of ML models without requiring full disclosure of the underlying data. One focus is Federated Learning, in which multiple data holders collaboratively train a model without exchanging their experimental data. Another focus is Continual Learning, in which previously trained models are updated with new data without requiring access to the original training data. A particular challenge is preventing the models from losing previously acquired knowledge during this process.

Lecture Support

• Computerlab Moleculare Simulations (SS 2024)
• Thermodynamics 1 (WS 2024/25)
• Thermodynamics of Mixtures (WS 2025/26)

Office Hours

By arrangement.

Publications

• P. Zittlau, S. Mross, D. Gond, M. Kohns: Molecular Modeling and Simulation of Organic Electrolyte Solutions for Lithium Ion Batteries, J. Chem. Phys. 161 (2024) 124118. [DOI]
• M. Kohns, P. Zittlau, F. Jirasek: Prediction of the density of aqueous electrolyte solutions with matrix completion methods, Fluid Phase Equilib. 597 (2025) 114454. [DOI]
• R. Loubet, P. Zittlau, L. Vollmer, M. Hoffmann, S. Fellenz, F. Jirasek, H. Leitte, H. Hasse: Using Large Language Models for Solving Textbook-Style Thermodynamic Problems, Comput. Chem. Eng. 204, 109333 (2026). [DOI]

Presentations

• P. Zittlau, N. Hayer, S. Roos, H. Hasse, F. Jirasek: Confidentiality-preserving training of thermodynamic models with federated learning, Thermodynamik-Kolloquium 2025, Hamburg, 22.-24.10.2025.
• P. Zittlau, N. Hayer, S. Roos, H. Hasse, F. Jirasek: Confidentiality-preserving training of thermodynamic models with federated learning, ESAT 2026, Lissabon, 10.-13.05.2026.

Posters

• P. Zittlau, M. Kohns, F. Jirasek: Prediction of the Density of Electrolyte Solutions with Matrix Completion Methods, Thermodynamik-Kolloquium, Chemnitz, 26.-28.09.2022.
• P. Zittlau, S. Mross, D. Gond, M. Kohns: Molecular Models for Lithium Ion Battery Electrolytes: Cyclic Carbonate Solvents and Lithium Salts, Thermodynamik-Kolloquium, Stuttgart, 25.-27.09.2024.