Using Machine Learning to Understand the Causes of Quantum Decoherence in Solution-Phase Bond-Breaking Reactions

Decoherence is a fundamental phenomenon that occurs when an entangled quantum state interacts with its environment, leading to collapse of the wave function. The inevitability of decoherence provides one of the most intrinsic limits of quantum computing. However, there has been little study of the precise chemical motions from the environment that cause decoherence. Here, we use quantum molecular dynamics simulations to explore the photodissociation of Na-2(+) in liquid Ar, in which solvent fluctuations induce decoherence and thus determine the products of chemical bond breaking. We use machine learning to characterize the solute-solvent environment as a high-dimensional feature space that allows us to predict when and onto which photofragment the bonding electron will localize. We find that reaching a requisite photofragment separation and experiencing out-of-phase solvent collisions underlie decoherence during chemical bond breaking. Our work highlights the utility of machine learning for interpreting complex solution-phase chemical processes as well as identifies the molecular underpinnings of decoherence.