Modern ab initio methods have rapidly increased our understanding of solid state materials properties, chemical reactions, and the quantum interactions between atoms. However, poor scaling often renders direct ab initio calculations intractable for large or complex systems. There are two obvious avenues through which to remedy this problem: (i) develop new, less expensive methods to calculate system properties, or (ii) make existing methods faster. This paper describes an open source framework designed to pursue both of these avenues. PROPhet (short for PROPerty Prophet) utilizes machine learning techniques to find complex, non-linear mappings between sets of material or system properties. The result is a single code capable of learning analytical potentials, non-linear density functionals, and other structure-property or property-property relationships. These capabilities enable highly accurate mesoscopic simulations, facilitate computation of expensive properties, and enable the development of predictive models for systematic materials design and optimization. This work explores the coupling of machine learning to ab initio methods through means both familiar (e.g., the creation of various potentials and energy functionals) and less familiar (e.g., the creation of density functionals for arbitrary properties), serving both to demonstrate PROPhet’s ability to create exciting post-processing analysis tools and to open the door to improving ab initio methods themselves with these powerful machine learning techniques.

Ab initio molecular dynamics (AIMD) simulations of molecule-surface scattering allow first-principles characterization of the dynamics. However, the large number of density functional theory calculations along the trajectories is very costly, limiting simulations of long-time events and giving rise to poor statistics. To avoid this computational bottleneck, we report here the development of a high-dimensional molecule-surface interaction potential energy surface (PES) with movable surface atoms, using a machine learning approach. With 60 degrees of freedom, this PES allows energy transfer between the energetic impinging molecule and thermal surface atoms. Classical trajectory calculations for the scattering of DCl from Au(111) on this PES are found to agree well with AIMD simulations, with ∼10-fold acceleration. Scattering of HCl from Au(111) is further investigated and compared with available experimental results.

Arsenic is one of the most common heavy metal contaminants found in the environment, particularly in water. We examined the impact of perinatal exposure to relatively low levels of arsenic (50 parts per billion, ppb) on neuroendocrine markers associated with depression and depressive-like behaviors in affected adult C57BL/6J mouse offspring. Whereas most biomedical research on arsenic has focused on its carcinogenic potential, a few studies suggest that arsenic can adversely affect brain development and neural function. Compared to controls, offspring exposed to 50 parts per billion arsenic during the perinatal period had significantly elevated serum corticosterone levels, reduced whole hippocampal CRFR 1 protein level and elevated dorsal hippocampal serotonin 5HT 1A receptor binding and receptor-effector coupling. 5HT 1A receptor binding and receptor-effector coupling were not different in the ventral hippocampal formation, entorhinal or parietal cortices, or inferior colliculus. Perinatal arsenic exposure also significantly increased learned helplessness and measures of immobility in a forced swim task. Taken together, these results suggest that perinatal arsenic exposure may disrupt the regulatory interactions between the hypothalamic-pituitary-adrenal axis and the serotonergic system in the dorsal hippocampal formation in a manner that predisposes affected offspring to depressive-like behavior. These results are the first to demonstrate that relatively low levels of arsenic exposure during development can have long-lasting adverse effects on behavior and neurobiological markers associated with these behavioral changes.

Representation of multidimensional global potential energy surfaces suitable for spectral and dynamical calculations from high-level ab initio calculations remains a challenge. Here, we present a detailed study on constructing potential energy surfaces using a machine learning method, namely, Gaussian process regression. Tests for the 3 A″ state of SH 2 , which facilitates the SH + H ↔ S( 3 P) + H 2 abstraction reaction and the SH + H′ ↔ SH′ + H exchange reaction, suggest that the Gaussian process is capable of providing a reasonable potential energy surface with a small number (∼1 × 10 2 ) of ab initio points, but it needs substantially more points (∼1 × 10 3 ) to converge reaction probabilities. The implications of these observations for construction of potential energy surfaces are discussed.

Use of the non-local correlation functional vdW-DF (from 'van der Waals density functional'; Dion M et al 2004 Phys. Rev. Lett. 92 246401) has become a popular approach for including van der Waals interactions within density functional theory. In this work, we extend the vdW-DF theory and derive the corresponding stress tensor in a fashion similar to the LDA and GGA approach, which allows for a straightforward implementation in any electronic structure code. We then apply our methodology to investigate the structural evolution of amino acid crystals of glycine and l-alanine under pressure up to 10 GPa-with and without van der Waals interactions-and find that for an accurate description of intermolecular interactions and phase transitions in these systems, the inclusion of van der Waals interactions is crucial. For glycine, calculations including the vdW-DF (vdW-DF-c09x) functional are found to systematically overestimate (underestimate) the crystal lattice parameters, yet the stability ordering of the different polymorphs is determined accurately, at variance with the GGA case. In the case of l-alanine, our vdW-DF results agree with recent experiments that question the phase transition reported for this crystal at 2.3 GPa, as the a and c cell parameters happen to become equal but no phase transition is observed.

Scattering and dissociative chemisorption of DCl on Au(111) are investigated using ab initio molecular dynamics with a slab model, in which the top two layers of Au are mobile. Substantial kinetic energy loss in the scattered DCl is found, but the amount of energy transfer is notably smaller than that observed in the experiment. On the other hand, the dissociative chemisorption probability reproduces the experimental trend with respect to the initial kinetic energy, but is about one order of magnitude larger than the reported initial sticking probability. While the theory-experiment agreement is significantly improved from the previous rigid surface model, the remaining discrepancies are still substantial, calling for further scrutiny in both theory and experiment.

The dissociative chemisorption of polyatomic molecules on metal surfaces has attracted much interest in recent years due to their industrial and fundamental importance. Comparing with extensively studied systems such as methane and water, however, dissociative chemisorption of CO2, which is important for CO2 activation, has so far received scant attention. We recently reported vibrational enhancement of the dissociative chemisorption of CO2 on a rigid Ni(100) surface using a nine-dimensional potential energy surface (PES) based on a large number of density functional theory calculations. However, that PES is incapable of describing the lattice motion and energy transfer between this heavy molecule and the surface. To overcome these limitations, we present here ab initio molecular dynamics results for CO2 scattering and dissociation. In addition to formation of adsorbed O and CO, CO2 is found to have a large trapping probability in a chemisorption well, along with a substantial amount of energy loss to surface phonons. The lattice and dynamical steering effects are found to be quite different from what have been observed for direct dissociative chemisorption of methane and water.

The applicability and accuracy of the Behler-Parrinello atomistic neural network method for fitting reactive potential energy surfaces is critically examined in three systems, H + H2 → H2 + H, H + H2O → H2 + OH, and H + CH4 → H2 + CH3. A pragmatic Monte Carlo method is proposed to make efficient choice of the atom-centered mapping functions. The accuracy of the potential energy surfaces is not only tested by fitting errors but also validated by direct comparison in dynamically important regions and by quantum scattering calculations. Our results suggest this method is both accurate and efficient in representing multidimensional potential energy surfaces even when dissociation continua are involved.