Reactive Force Field Design Guided by Energy Decomposition Analysis

With support from the Chemical Theory, Models and Computational Methods program in the Division of Chemistry, Professors Teresa Head-Gordon and Martin Head-Gordon of the University of California Berkeley will develop next-generation force fields guided by new advances in energy decomposition analysis (EDA). Force fields (FFs) are empirical functions that aim to describe how the potential energy of a system varies with position of the atoms and molecules both accurately and inexpensively. Molecular simulations advance by improving the accuracy of the force field, which is recognized as a difficult yet essential challenge, which can partly be addressed using EDA. An EDA takes advanced quantum mechanical calculations on groups of molecules, and distills their interaction energy into a sum of terms that capture repulsive and attractive physical driving forces. These terms provide valuable first principles data to inform the design and parameterization of new force fields. The research should yield new scientific methods and results on specific chemical systems. Broader impacts will encompass software dissemination, machine learning models, high quality data generation, and education, training, and outreach. Research opportunities will be provided for junior transfer undergraduates who have more difficulty finding research labs during their shorter tenure. Development of a Professional Masters in Molecular Sciences and Software Engineering (MSSE) degree will aid development of a diverse workforce that is highly prepared for programming, data modeling, and machine learning. Specifically, the collaborative research team at UC-Berkeley will formulate a new Gibbs decomposition analysis (GDA) to unravel connections between molecular driving forces and enthalpy-entropy trade-offs. GDA will be adapted to FF simulations in order to probe the molecular interactions controlling interfacial chemistry. For non-covalent interactions, EDA for electrostatic polarization will be advanced to reveal each fragment’s energy lowering and orbital rearrangements. New analysis will be performed to reconcile real space and Hilbert space measures of charge transfer, and to understand Pauli relaxation. These EDA advances have the potential to answer basic questions about non-bonded interactions and help guide the FF development that use charge equilibration to define polarization and charge transfer. The recent emergence of machine learning to represent potential surfaces offers a complementary approach to reactive force field development for chemical bonding. The NewtonNet machine learning model will be trained with EDA data that can be integrated into force fields. In summary, the main objectives of these studies are to provide more powerful EDA tools as well as machine learning models to expand the scope of advanced potential energy surfaces to non-bonded interactions, reactive chemistry, and to better interpretation of observables in condensed phase and interfacial systems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.