From Accurate Quantum Mechanics to Converged Thermodynamics in Solution with Machine Learning Potentials

Add to your list(s) Download to your calendar using vCal

• Niamh O'Neill, University of Cambridge
• Wednesday 26 February 2025, 15:00-15:30
• Unilever Lecture Theatre, Yusuf Hamied Department of Chemistry.

If you have a question about this talk, please contact Lisa Masters.

Obtaining accurate predictions of thermodynamic properties, especially free energies which define the state of a system, is one of the key goals in atomistic simulations. This can enable a direct understanding of atomic-scale processes and provide a direct link to experiment. Achieving this requires converged statistical sampling from accurate wavefunction based potential energy surfaces, which is a formidable challenge due to the very high computational cost of such methods. Here, we leverage advances in machine learning potentials to efficiently obtain converged thermodynamic properties at increasing levels of theory. To showcase the potential of this approach, I will use the ion pairing of CaCO3 as a benchmark system, since it presents a significant challenge from both electronic structure and sampling perspectives. I will show that a machine learning framework based on second order Møller-Plesset Perturbation Theory delivers excellent agreement with experiment for the ion-pair association free energy—a challenging property for first principles atomistic simulations. Furthermore I will show that classical force fields get the right answer for the wrong reasons. Finally, I will discuss steps towards developing CCSD accuracy machine learning models, the ‘gold-standard’ of quantum chemical methods.

This talk is part of the Theory - Chemistry Research Interest Group series.

This talk is included in these lists:

• All Talks (aka the CURE list)
• Department of Chemistry
• Featured lists
• Lennard-Jones Centre external
• School of Physical Sciences
• Unilever Lecture Theatre, Yusuf Hamied Department of Chemistry

Note that ex-directory lists are not shown.