NIC-CAGE: An open-source software package for predicting optimal control fields in photo-excited chemical systems

Abstract

We present an open-source software package, NIC-CAGE (Novel Implementation of Constrained Calculations for Automated Generation of Excitations), for predicting quantum optimal control fields in photo-excited chemical systems. Our approach utilizes newly derived analytic gradients for maximizing the transition probability (based on a norm-conserving Crank-Nicolson propagation scheme) for driving a system from a known initial quantum state to another desired state. The NIC-CAGE code is written in the MATLAB and Python programming environments to aid in its readability and general accessibility to both users and practitioners. Throughout this work, we provide several examples and outputs on a variety of different potentials, propagation times, and user-defined parameters to demonstrate the robustness of the NIC-CAGE software package. As such, the use of this predictive tool by both experimentalists and theorists could lead to further advances in both understanding and controlling the dynamics of photo-excited systems. File list (2)download file view on ChemRxiv Quantum_Optimal_Control.pdf (0.94 MiB) download file view on ChemRxiv Supplementary_Information.pdf (1.15 MiB)

Harnessing Deep Neural Networks to Solve Inverse Problems in Quantum Dynamics: Machine-Learned Predictions of Time-Dependent Optimal Control Fields

“…46 Finally, we also anticipate that the machine learning techniques used here could be harnessed to predict optimal electric fields for other higherlying transitions, which are known to exhibit more complex patterns in the time and frequency domains. 41 In particular, cross-correlation neural network approaches, which were used to overcome problems associated with the random phase of E(t), could be useful in (1) predicting optimal electric fields for other higher-energy excitations in the time domain or (2) enabling the prediction of the full absorption/emission spectra of molecules since the absorption spectra is merely the Fourier transform of E(t). Taken together, these machine learning techniques show a promising path towards cost-effective statistical approaches for designing control fields that enable desired transitions in quantum dynamical systems.…”

“…To generate the data required for our machine learning approaches, we utilized the NIC-CAGE (Novel Implementation of Constrained Calculations for Automated Generation of Excitations) program developed in our previous work. 41 Given a potential, V (x), this program iteratively calculates a numerical representation of E(t) that enables a ≈100% transition probability between two desired electronic transitions (which, in this work, are the ground and first-excited state, schematically shown in Figs. 1a and 1b.…”

“…While the NIC-CAGE program 41 can obtain transition probabilities with notable accuracy (typically over 97%), it can take hundreds of iterations (or longer) to converge to the final electric field for each potential. Moreover, as mentioned in the Introduction, when a new potential is inputted, the iteration process has to be re-started anew since the program has no prior memory of previously converged cases.…”

Harnessing Deep Reinforcement Learning to Construct Time-Dependent Optimal Fields for Quantum Control Dynamics

“…In this work, we harness new RL techniques to automatically construct optimal control fields, E(t), that enable desired transitions in these continuous/infinitedimensional Hilbert-space dynamical systems. To test the performance of our RL approach, we compare against the NIC-CAGE (Novel Implementation of Constrained Calculations for Automated Generation of Excitations) code [38], which solves the quantum control problem using a traditional gradient-based approach. Specifically, the NIC-CAGE code utilizes analytic gradients based on a Crank-Nicholson propagator, which are computationally more efficient than other matrix exponential approaches (such as those used in the GRAPE [39] or QuTIP [40,41] packages) or higher-order timepropagation methods [42].…”

Inverse design of molecule–metal nanoparticle systems interacting with light for desired photophysical properties

“…In particular, the gradient-based method efficiently tailors incident light for photo-excited systems. 25,26 Combining it with molecular design provides a possibility to further optimize desired properties. Such a design by theoretical means requires the time-domain description of the phenomena involving light such as the LSPR.…”