Machine learning study of universal electronic stopping cross-sections of ions in matter

Abstract

Accurate electronic stopping cross-section (ESCS) database of ions in matter is crucial for precise simulation of radiation damage. Based on the experimental-cleaned database of SRIM, binary theory and unitary convolution approximation as well as the descriptor pool extracted from these models, we developed a universal machine learning ESCS database using the least absolute shrinkage and selection operator (LASSO) algorithm. This method allows for predictions for ion-target combinations with atomic numbers from 1 to 92, within the energy range from 1 keV/u to 1 GeV/u, addressing the limitations of machine learning on training dataset. The database exhibits remarkable accuracy in predicting ESCS and ion depth distribution/range, along with robust reciprocity performance. Key descriptors are also determined, which closely mimic the Lindhard-Scharff-Schiott and Bohr-Bethe-Bloch formulations, achieved through precise adjustments of the exponent of individual elements. The proposed universal ESCS database surpasses the accuracy of existing databases, supporting related applications across a wide range of energies and systems.

Keywords

• Electronic stopping cross-section
• Machine learning
• Least absolute shrinkage and selection operator (LASSO)