Machine Learning Discovery of Computational Model Efficacy Boundaries.

Computational models are formulated in hierarchies of variable fidelity, often with no quantitative rule for defining the fidelity boundaries. We have constructed a dataset from a wide range of atomistic computational models to reveal the accuracy boundary between higher-fidelity models and a simple, lower-fidelity model. The symbolic decision boundary is discovered by optimizing a support vector machine on the data through iterative feature engineering. This data-driven approach reveals two important results: (i) a symbolic rule emerges that is independent of the algorithm, and (ii) the symbolic rule provides a deeper understanding of the fidelity boundary. Specifically, our dataset is composed of radial distribution functions from seven high-fidelity methods that cover wide ranges in the features (element, density, and temperature); high-fidelity results are compared with a simple pair-potential model to discover the nonlinear combination of the features, and the machine learning approach directly reveals the central role of atomic physics in determining accuracy.

Velocity of a Molecule Evaporated from a Water Nanodroplet: Maxwell-Boltzmann Statistics versus Non-Ergodic Events.

The velocity of a molecule evaporated from a mass-selected protonated water nanodroplet is measured by velocity map imaging in combination with a recently developed mass spectrometry technique. The measured velocity distributions allow probing statistical energy redistribution in ultimately small water nanodroplets after ultrafast electronic excitation. As the droplet size increases, the velocity distribution rapidly approaches the behavior expected for macroscopic droplets. However, a distinct high-velocity contribution provides evidence of molecular evaporation before complete energy redistribution, corresponding to non-ergodic events.

Communication: Nucleation of water on ice nanograins: Size, charge, and quantum effects.

The sticking cross sections of water molecules on cold size-selected water clusters have been simulated using classical and quantum (path-integral) molecular dynamics trajectories under realistic conditions. The integrated cross sections for charged clusters show significant size effects with comparable trends as in experiments, as well as essentially no sign effect. Vibrational delocalization, although it contributes to enlarging the geometric cross sections, leads to a counter-intuitive decrease in the dynamical cross section obtained from the trajectories. These results are interpreted based on the apparent reduction in the effective interaction between the projectile and the target owing to zero-point effects.

Pinning an ion with an intracavity optical lattice.

We report one-dimensional pinning of a single ion by an optical lattice. A standing-wave cavity produces the lattice potential along the rf-field-free axis of a linear Paul trap. The ion's localization is detected by measuring its fluorescence when excited by standing-wave fields with the same period, but different spatial phases. The experiments agree with an analytical model of the localization process, which we test against numerical simulations. For the best localization achieved, the ion's average coupling to the cavity field is enhanced from 50% to 81(3)% of its maximum possible value, and we infer that the ion is bound in a lattice well with over 97% probability.