Physical Chemistry Seminar - Prof. Venkat Kapil
Event details
| Date | 06.11.2025 |
| Hour | 17:15 › 18:15 |
| Speaker | Prof. Venkat Kapil, University College London, UK |
| Location | |
| Category | Conferences - Seminars |
| Event Language | English |
An apéro will be served after the seminar!
Machine learning electronic and quantum nuclear effects in condensed phases: Applications to understanding the phase diagram of nanoconfined water
Understanding the microscopic structure and dynamics of condensed phase systems is essential for fields ranging from drug design to catalysis and battery science. While quantum simulations offer predictive accuracy, incorporating quantum effects in condensed phases remains computationally prohibitive. I will present recent progress in developing machine learning (ML) approaches that overcome these challenges. To capture electronic quantum effects, we train foundation models that predict first-principles-quality energies and electronic properties across diverse systems and can be efficiently fine-tuned for specific problems with small volumes of training data—extending the scope of finite-temperature simulations to beyond DFT accuracy. To capture quantum nuclear effects, we develop ML frameworks that learn quantum-effective corrections to Born–Oppenheimer potentials, enabling quantum statistics and approximate dynamics at classical cost. As an application, I will show a fully quantum prediction of the phase diagram of nanoconfined water—a model system relevant to water treatment, catalysis, and battery science, which exhibits several poorly understood anomalous properties. Our approach reveals rich phase behaviour sensitive to the pressure within nanoconfined channels, including a superionic phase of water with ionic conductivity rivalling that of battery materials. Our work demonstrates that ML approaches are maturing toward full quantum accuracy in condensed phases across many classes of systems.
Machine learning electronic and quantum nuclear effects in condensed phases: Applications to understanding the phase diagram of nanoconfined water
Understanding the microscopic structure and dynamics of condensed phase systems is essential for fields ranging from drug design to catalysis and battery science. While quantum simulations offer predictive accuracy, incorporating quantum effects in condensed phases remains computationally prohibitive. I will present recent progress in developing machine learning (ML) approaches that overcome these challenges. To capture electronic quantum effects, we train foundation models that predict first-principles-quality energies and electronic properties across diverse systems and can be efficiently fine-tuned for specific problems with small volumes of training data—extending the scope of finite-temperature simulations to beyond DFT accuracy. To capture quantum nuclear effects, we develop ML frameworks that learn quantum-effective corrections to Born–Oppenheimer potentials, enabling quantum statistics and approximate dynamics at classical cost. As an application, I will show a fully quantum prediction of the phase diagram of nanoconfined water—a model system relevant to water treatment, catalysis, and battery science, which exhibits several poorly understood anomalous properties. Our approach reveals rich phase behaviour sensitive to the pressure within nanoconfined channels, including a superionic phase of water with ionic conductivity rivalling that of battery materials. Our work demonstrates that ML approaches are maturing toward full quantum accuracy in condensed phases across many classes of systems.
Practical information
- Informed public
- Free
Organizer
- Prof. Nikita Kavokine
Contact
- Nikita Kavokine