It is well known in the field of machine learning that committee models improve accuracy, provide generalization error estimates, and enable active learning strategies. In this work, we adapt these concepts to interatomic potentials based on artificial neural networks. Instead of a single model, multiple models that share the same atomic environment descriptors yield an average that outperforms its individual members as well as a measure of the generalization error in the form of the committee disagreement. We not only use this disagreement to identify the most relevant configurations to build up the model’s training set in an active learning procedure but also monitor and bias it during simulations to control the generalization error. This facilitates the adaptive development of committee neural network potentials and their training sets while keeping the number of ab initio calculations to a minimum. To illustrate the benefits of this methodology, we apply it to the development of a committee model for water in the condensed phase. Starting from a single reference ab initio simulation, we use active learning to expand into new state points and to describe the quantum nature of the nuclei. The final model, trained on 814 reference calculations, yields excellent results under a range of conditions, from liquid water at ambient and elevated temperatures and pressures to different phases of ice, and the air–water interface—all including nuclear quantum effects. This approach to committee models will enable the systematic development of robust machine learning models for a broad range of systems.

Experimental studies of the electronic structure of excess electrons in liquids—archetypal quantum solutes—have been largely restricted to very dilute electron concentrations. We overcame this limitation by applying soft x-ray photoelectron spectroscopy to characterize excess electrons originating from steadily increasing amounts of alkali metals dissolved in refrigerated liquid ammonia microjets. As concentration rises, a narrow peak at ~2 electron volts, corresponding to vertical photodetachment of localized solvated electrons and dielectrons, transforms continuously into a band with a sharp Fermi edge accompanied by a plasmon peak, characteristic of delocalized metallic electrons. Through our experimental approach combined with ab initio calculations of localized electrons and dielectrons, we obtain a clear picture of the energetics and density of states of the ammoniated electrons over the gradual transition from dilute blue electrolytes to concentrated bronze metallic solutions.

Photoelectron spectroscopy of microjets expanded into vacuum allows access to orbital energies for solute or solvent molecules in the liquid phase. Microjets of water, acetonitrile and alcohols have previously been studied; however, it has been unclear whether jets of low temperature molecular solvents could be realized. Here we demonstrate a stable 20 μm jet of liquid ammonia (−60 °C) in a vacuum, which we use to record both valence and core-level band photoelectron spectra using soft X-ray synchrotron radiation. Significant shifts from isolated ammonia in the gas-phase are observed, as is the liquid-phase photoelectron angular anisotropy. Comparisons with spectra of ammonia in clusters and the solid phase, as well as spectra for water in various phases potentially reveal how hydrogen bonding is reflected in the condensed phase electronic structure.

The oligomeric state of the storage form of human insulin in the pancreas, which may be affected by several endogenous components of β-cell storage granules such as arginine, is not known. Here, the effect of arginine on insulin oligomerization is investigated independently by protein crystallography, molecular dynamics simulations, and capillary electrophoresis. The combined results point to a strong effect of ionic strength on insulin assembly. Molecular simulations and electrophoretic measurements at low/mM salt concentrations show no significant effect of arginine on insulin aggregation. In contrast, crystallographic data at high/molar ionic strength indicate inhibition of insulin hexamerization by arginine due to its binding at the site relevant for intermolecular contacts, which was also observed in MD simulations. Our results thus bracket the in vivo situation in pancreatic β-cell storage granules, where the ionic strength is estimated to be in the hundreds of millimolar to submolar range. The present findings add to a molecular understanding of in vivo insulin oligomerization and storage, with additional implications for insulin stability in arginine-rich injections.