Machine learning-based estimator for electron impact ionization fragmentation patterns

Abstract Numerous measurements and calculations exist for total electron impact ionization cross sections. However, knowing electron impact ionization fragmentation patterns is important in various scientific fields such as plasma physics, astrochemistry, and environmental sciences. Partial ionization cross sections can be calculated by multiplying total ionization cross sections with branching ratios for different fragments, which can be deduced from ionization mass spectra. However, the required mass spectrometry data is frequently unavailable. A machine learning-based method to predict mass spectra is presented. This method is used to estimate partial electron impact ionization cross sections using the predicted mass spectra and the appearance thresholds for the ionic fragments. As examples, ammonia and the C 2 F 5 radical are considered: branching ratios derived from the predicted mass spectra and Binary-Encounter Bethe (BEB) total ionization cross sections are used to predict the fragmentation pattern for each species. The machine learning algorithm can also be used to predict mass spectroscopy fragmentation patterns. While effective, the method has key limitations: it does not account for light fragments such as H + , whose peaks are absent in the training data, and its validity is restricted to electron impact energies below 100 eV to minimize the contribution of double ionization, which is not accounted for by the BEB model. Although BEB cross sections are used in this work, the method is not reliant on BEB and can be applied to any set of total ionization cross sections, including experimental measurements.