Is the water/Pt(111) interface ordered at room temperature?

Abstract

The structure of the water/Pt(111) interface has been a subject of debate over the past decades. Here, we report the results of a room temperature molecular dynamics study based on neural network potentials, which allow us to access long time scale simulations while retaining ab initio accuracy. We find that the water/Pt(111) interface is characterized by a double layer composed of a primary, strongly bound adsorption layer with a coverage of ∼ 0.15ML, which is coupled to a secondary, weakly bound adsorption layer with a coverage of ∼ 0.58ML. By studying the order of the primary adsorption layer we find that there is an effective repulsion between the adsorbed water molecules, which gives rise to a dynamically-changing, semi-ordered interfacial structure, where the water molecules in the primary adsorption layer are distributed homogeneously across the interface forming frequent hydrogen bonds to water molecules in the secondary adsorption layer. We furthermore show that these conclusions are beyond the time scales accessible to ab initio molecular dynamics.

Structure and energetics of liquid water–hydroxyl layers on Pt(111)

“…31 The architecture of each NNP as well as the training and test errors are reported in Table 1. The errors on the forces and energies are roughly a factor of 2 and 1.3 larger than the training and test errors reported in our previous study of the water/Pt(111) interface, 17 though still comparable in magnitude to other training and test errors reported for similar studies using NNPs. 31 The larger training and test errors can be understood as a result of a more long ranged character due to the presence of adsorbed hydroxyls.…”

“…In the case of no hydroxyl coverage, we obtain the characteristic double-peaked structure of the water/Pt(111) interface, which was analyzed in our previous work. 17 In the presence of hydroxyls the same structure is observed, but it is clear that the presence of co-adsorbed hydroxyls pins the water molecules close to the Pt(111) surface leading to a depletion of the second density peak. This effect is further quantified in Fig.…”

“…where G i denotes a vector of 2-and 3-body symmetry functions, which describe the local environment of each atom up to a cutoff radius, R c . 28 Our motivation for using an ensemble of NNPs rather than a single one are the well known model-and data-based limitations of Behler-Parinello potentials 17,29 (see Section A of the ESI † for more details about this), and throughout the remainder of this paper we therefore use the spread of the ensemble predictions as a lower bound for the accuracy one can expect to achieve with our methodology. Concretely, the NNP ensemble consists of four NNPs all fitted using the RuNNer code, 30 where we employed the same set of symmetry functions as used in a previous NNP study of water on low-index Cu surfaces.…”

“…This can be problematic when modelling properties of the interface such as, e.g., the dynamic structure of the interfacial water layer. [14][15][16] In a recent publication, 17 we have discussed the above issue in detail and presented a possible solution based on fitting an ensemble of neural network potentials (NNPs). Using this approach, we demonstrated that it was possible to bridge the problems of obtaining high accuracy versus proper thermal sampling for the problem of determining the structure of the liquid water-Pt(111) interface.…”

“…In this paper, we extend the methodology of our previous work 17 to model the energetics of hydroxyl formation. Concretely, our main focus is the following OH adsorption reaction on a crystalline Pt(111) surface:…”