Machine learning reveals complex behaviours in optically trapped particles

摘要: Since their invention in the 1980s, optical tweezers have found a wide range of applications, from biophotonics and mechanobiology to microscopy optomechanics. Simulations motion microscopic particles held by are often required explore complex phenomena interpret experimental data. For sake computational efficiency, these simulations usually model as an harmonic potential. However, more physically-accurate optical-scattering models accurately onerous systems; this is especially true for traps generated with fields. Although accurate, tend be prohibitively slow problems than one or two degrees freedom (DoF), which has limited broad adoption. Here, we demonstrate that machine learning permits combine speed accuracy models. Specifically, show neural network can trained rapidly predict forces acting on particle. We utility approach simulate otherwise: escape dynamics swelling microparticles trap, rotation rates superposition beams opposite orbital angular momenta. Thanks its high accuracy, method greatly enhance efficiently simulated studied.