Machine learning for predicting control landscape maps of quantum molecular dynamics: Laser-induced three-dimensional alignment of asymmetric-top molecules

Machine learning for predicting control landscape maps of full quantum molecular dynamics is examined through a case study of the laser-induced three-dimensional (3D) alignment of asymmetric-top molecules, an essential technique for observing and/or manipulating molecular dynamics in a molecule-fixed frame. We consider the ``prolate-type'' asymmetric-top molecules with the asymmetry parameters $\ensuremath{-}1<\ensuremath{\kappa}<0$ and the ${C}_{2v}$ symmetry in the low-temperature limiting case, which are aligned by using mutually orthogonal linearly polarized double laser pulses. The landscape map for each molecule consists of 6000 pixels, each pixel of which represents the maximum degree of alignment achieved by each set of control parameters. After examining ways to deal with the markedly different molecular parameters in a unified manner for suitably training a convolutional neural network (CNN) model, we train the CNN model by using 55 training sample molecules to predict the control landscape maps of 35 test sample molecules with reasonably high accuracy. As the predicted landscape maps provide a big picture of the alignment control, we examine, for example, the effectiveness of the double-pulse control scheme in terms of molecular parameters.