Learning the dimension of light’s spatial modes

Abstract Light’s spatial modes describe the distinct intensity and phase patterns of light in the transverse plane. Due to their high dimensionality and complex nature, the spatial modes of light have important applications in quantum optics, optical trapping, optical imaging, and high-dimensional quantum information processing. However, research on the spatial modes of light has primarily focused on their generation and manipulation. In this work, we address the issue of determining the dimensionality of light’s spatial modes. We focus on dimensions 2 to 15 and show that machine learning can accurately identify dimensions from intensity profiles, achieving an averaged accuracy of 99.82% for single-dimension recognition and 96.28% for mixed-dimension recognition. We further perform an independent test and obtain an accuracy of 99.57%. Our work advances the study of light’s spatial modes and fills the gap regarding the dimensionality characterization in this field.