Machine-learning approach for constructing control landscape maps of three-dimensional alignment of asymmetric-top molecules

Abstract

A machine-learning approach to draw landscape maps in a low-dimensional control-parameter space is examined through a case study of three-dimensional alignment control of the asymmetric-top molecule SO2. As a minimal model, we consider the control by using a set of mutually orthogonal, linearly polarized laser pulses that are parameterized by the time delay and fluence ratio. The parameters are represented either by points in the parameter space or by time- and frequency-resolved spectra. Machine-learning models based on convolutional neural networks together with two considerably different representations, which are trained by using a reasonably small number of training samples, construct maps that have sufficient accuracy to predict temperature-dependent control mechanisms.