One-Shot Trajectory Learning of Open Quantum Systems Dynamics
In the work published in the Journal of Physical Chemistry Letters, we have proposed a one-shot trajectory learning (OSTL) approach that allows an ultrafast prediction of 10-ps-long quantum dynamics of an open quantum system just in 70 milliseconds. OSTL approach takes simulation parameters such as reorganization energy, temperature etc. as input and predicts the corresponding trajectory in one shot. We demonstrate OSTL application for excitation energy transfer in the well-known Fenna–Matthews–Olson (FMO) complex.
OSTL based on convolutional neural network (CNN) predicts the entire trajectory through multiple output units in the last layer. Thus, the CNN in OSTL does not need to take time as an input function significantly speeding up both training and inference compared to our previously proposed AI-QD approach published in Nature Communications (AI-QD invokes the whole ML architecture for each time step). However, OSTL does not replace AI-QD, as AI-QD has its own applications such as interpolation and extrapolation in the dimension of time, which is not possible with OSTL. The main application of OSTL is its speed which can be exploited in massive quantum dynamics simulations such as interpolating in the simulation parameter space as shown in the Supporting Information of the paper.
- Arif Ullah, Pavlo O. Dral, One-shot trajectory learning of open quantum systems dynamics, J. Phys. Chem. Lett. 2022, 13, 6037–6041. DOI: 10.1021/acs.jpclett.2c01242.