RUI: Harnessing Machine Learning Techniques for Atomic and Molecular Collisions

About This Grant

Charged particle scattering processes play a vital role in many fields, including plasma physics, astrophysics, and biomedical physics. In order to advance technologies and develop innovative new techniques in these fields, fundamental charged particle collision data is required for electron and heavy-ion scattering. This collision data often serves as inputs for sophisticated application-based models, and is needed for a wide range of energies, projectiles, target species, and collision processes. Current databases and theoretical models lack data for complex molecular targets, which are particularly relevant for biomedical and plasma physics applications. This research will develop and apply machine learning models to help fill the gaps in available collision data and provide insight into the physics of charged particle collisions with molecules. By merging traditional charged particle collisions physics with the empowering technology of machine learning, it will result in freely accessible, user-friendly models that require only minimal input and expertise to use. Additionally, the research will train the next generation of the science and technology workforce by providing cutting-edge research opportunities to students who will gain necessary career skills through hands-on participation in model development, implementation, and analysis. This research will develop machine learning models for the estimation of electron and heavy-ion collision cross sections for a wide range of molecular target species, energies, and collision processes. Artificial neural network models will be developed to predict cross sections for many different collision processes and paired with convolutional neural network models to include structure and orientation effects through the use of molecular structure images. Cross sections will be calculated for long- and short-chain per- and polyfluoroalkyl substances and biomolecules relevant to ion beam cancer therapies in order to quantify the likelihood of molecular dissociation and the production of reactive secondary electrons that can lead to molecular breakup. This research will also develop a new additivity rule that uses the ‘knowledge’ of an artificial neural network to inform the theoretical rule. This additivity rule will be used to study molecular orientation in collision cross sections, something that is important, but computationally prohibitive in existing theoretical models. The new additivity rule will be used in conjunction with existing theoretical models for atomic collision cross sections to determine how constituent atom properties influence the production of free electrons and provide an improved general method for estimating molecular collision cross sections. 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.