an accurate potential energy surface incorporating the six adsorbate degrees of freedom of HCl on Au(111) by using the neural network approach. In our molecular dynamics simulations, we model the experimental beam conditions to study the influence of the rovibrational state population distribution of HCl in the molecular beam on reaction. Likewise, molecular dynamics with electronic friction calculations based on the parameterfree local density friction approximation in the independent atom approximation (LDFA-IAA) are performed to get first insights into how electron hole pair excitation might affect the reactivity of HCl. Although satisfying agreement with the experiment could not yet be achieved by our simulations, we find that i) RPBE yields larger reaction barriers than PBE and lower computed reaction probabilities, improving the agreement with experiment, ii) the reactivity strongly depends on the rovibrational state population, iii) surface atom motion and electronically non-adiabatic effects modeled with the efficient to use but approximate LDFA influence the reaction only modestly, and iv) a moderate amount of charge is transferred from the surface to the dissociating molecule at the transition state. We suggest that the reported experimental reaction probabilities could be too low by a factor of about 2-3, due to potential problems with calibrating coverage of Au by Cl using an ill-defined external standard in the form of Auger peak ratios. However, taking this factor into account would not yet resolve the discrepancy between the theoretical reaction probabilities presented here and the experimental sticking probabilities.

In this work, we analyzed a data set formed by 566 donor/acceptor pairs, which are part of organic solar cells recently reported. We explored the effect of different descriptors in machine learning (ML) models to predict the power conversion efficiency (PCE) of these cells. The investigated descriptors are classified into two main categories: structural (topology properties) and physical descriptors (energy levels, molecular size, light absorption, and mixing properties). In line with previous observations, ML predictions are more accurate when using both structural and physical descriptors, as opposed to only using one of them. We observed that ML predictions are also improved by using larger and more varied data sets. Importantly, the structural descriptors are the ones contributing the most to the ML models. Some physical properties are highly correlated with PCE, although they do not improve notably the ML prediction accuracy as they carry information already encoded in the structural descriptors. Given that various descriptors have significantly different computational costs, the analysis presented here can be used as a guide to construct ML models that maximize predictive power and minimize computational costs for screening large sets of candidates.

X-doped graphene surfaces, where X is a heteroatom, are interesting for electrocatalytic applications in fuel cells because active sites are generated on the surface because of the presence of these heteroatoms. In this work, a comparative study of the oxygen reduction reaction (ORR) on three graphene surfaces doped with nitrogen, boron, and phosphorus was made by using density functional theory. Our simulation reveals that the ORR via a four-electron transfer mechanism is energetically more favorable than the two-electron transfer mechanism, where the latter pathway would lead to the unwanted oxygen peroxide formation. In addition, the energies calculated for each ORR step show that the P-doped surface is the one that favors the oxygen reduction reaction the most.

CO oxidation on transition metal surfaces is not only a prototype for studying surface chemistry but also of critical importance in applications such as pollution control and fuel cells. The reverse process, the dissociation of CO2, is also key in the sequestration of this greenhouse gas. However, our understanding of the dynamics involved in these processes is still incomplete. Theoretical studies of surface dynamics have so far been largely hindered by the high computational costs of on-the-fly calculations. To overcome this bottleneck, we report here a high-dimensional potential energy surface (PES) for both the dissociative chemisorption and recombinative desorption of CO2 on Pt(111), by using a machine learning method. Trained with a large number of density functional points in a large configuration space, the multipurpose neural network PES accurately reproduces the geometry and energy of the stationary points along the CO2 reaction path on Pt(111), as well as the dynamical results obtained using the ab initio molecular dynamics method. In addition, we propose a new perspective on the chemical shape of the surface, which reveals the site specificity of the chemical barrier. This approach opens the door to accurate studies of these relevant reactions on surfaces, with a low computational cost, granting access to a more in-depth description of the chemical processes taking place in these systems.

We present a review of the field of high throughput virtual screening for organic electronics materials focusing on the sequence of methodological choices that determine each virtual screening protocol. These...