Machine Learning Study of the Surface Reconstructions of Cu 2 O(111) Surface
Payal Wadhwa, Michael Schmid, Georg Kresse · 2025
Read Machine Learning Study of the Surface Reconstructions of Cu 2 O(111) Surface by Payal Wadhwa, Michael Schmid, Georg Kresse (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-05026.txt.
Introduction
Payal Wadhwa, Michael Schmid, and Georg Kresse settle a surface dispute with machine learning in a 2025 paper. The top atomic layer of cuprous oxide rearranges into patterns that experts have debated for years. Testing every candidate arrangement with full quantum calculations is far too slow.
The team trained force fields on the fly, learning from quantum data as simulations ran, then used heated parallel sampling to explore arrangements across oxygen-rich and oxygen-poor conditions. Two known structures came out lowest in energy over most conditions: a pyramid-covered surface and a copper-poor flat one. Two more pyramid forms appeared under strongly reducing conditions. Several energy methods agreed on which structures sit on the stability boundary, while others remained method-dependent. Beyond the verdict, the work shows learned force fields can tame surface puzzles cheaply.
Catalog record
| CREATOR | Payal Wadhwa, Michael Schmid, Georg Kresse |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-05026
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Machine Learning Study of the Surface Reconstructions of Cu 2 O(111) Surface Preprint: APS/123-QED Payal Wadhwa, Michael Schmid, Georg Kresse Abstract The atomic structure of the most stable reconstructed surface of cuprous oxide (Cu 2 O)(111) surface has been a longstanding topic of debate. In this study, we develop on-the-fly machine-learned force fields (MLFFs) to systematically investigate the various reconstructions of the Cu 2 O(111) surface under stoichiometric as well as O- and Cu-deficient or rich conditions, focusing on both ( 3 \sqrt{3} × 3 \sqrt{3} )R30° and (2 × \times 2) supercells. By utilizing parallel tempering simulations supported by MLFFs, we confirm that the previously described nanopyramidal and Cu-deficient (1×1) structures are the lowest energy structures from moderately to strong oxidizing conditions. In addition, we identify two promising nanopyramidal reconstructions at highly reducing conditions, a stoichiometric and a Cu-rich one. Surface energy calculations performed using spin-polarized PBE, PBE+U, r 2 SCAN, and HSE06 functionals show that the previously known Cu-deficient configuration and nanopyramidal configurations are at the convex hull (and, thus, equilibrium structures) for all functionals, whereas the stability of the other structures depends on the functional and is therefore uncertain. Our findings demonstrate that on-the-fly trained MLFFs provide a simple, efficient, and rapid approach to explore the complex surface reconstructions commonly encountered in experimental studies, and also enhance our understanding of the stability of Cu 2 O(111) surfaces. I Introduction Cuprous oxide (Cu 2 O), characterized by the space group Pn \bar{3} m, is a p -type semiconducting oxide with a wide range of applications in fields such as photovoltaics, solar cells, gas sensors, photocatalysis, dilute magnetic semiconductors, and transparent electronics Meyer et al. 2012 ; Ito and Masumi 1997 ; Kakuta and Abe 2009 ; Chatterjee et al. 2016 ; Shibasaki et al. 2021 ; Nilius et al. 2024 ; Xu and Xu 2024 ; Gloystein et al. 2023 ; Xu and Xu 2024 ; Gloystein and Nilius 2022 ; Hayashida et al. 2021 ; Soltanmohammadi et al. 2023 ; Chen et al. 2023 ; Ricca et al. 2021 ; Beckord et al. 2023 . Despite its potential for diverse industrial applications, the precise identification of its most favorable surfaces remained a challenge for an extended period, prompting considerable interest in exploring the surface properties of this material Nilius et al. 2024 . Among the three low-index surfaces, the (111) surface has been identified as the thermodynamically most stable. Furthermore, several studies, including those on hydrogen and CO reduction experiments and hydrogen evolution reactions in light-driven experiments to address the catalytic impact of individual (100) and (111) Cu 2 O facets, have highlighted the exceptional catalytic performance of the (111) surface Zheng et al. 2009 ; Bao et al. 2010 ; Chen et al. 2022 . This superior performance is attributed to its potentially non-polar and thermodynamically stable nature. Additionally, pristine Cu 2 O nanocrystals frequently adopt octahedral shapes, which further underscores the dominance of the (111) crystal facets Yin et al. 2005 ; Kuo and Huang 2010 . The crystal structure of bulk Cu 2 O consists of an fcc Cu sublattice with two tetrahedrally coordinated oxygen atoms per unit cube, such that each Cu atom has a linear coordination to two O atoms at opposite sides (O–Cu–O). However, the (111) surface of Cu 2 O is composed of a repeated stacking of electrically neutral O/4Cu/O trilayers Nilius et al. 2024 . For the first time, Schulz and Cox successfully cleaved single crystals in the Cu 2 O(111) plane and identified two distinct surface terminations: a simple hexagonal (1×1) pattern and a ( \sqrt{3} × \sqrt{3} )R30° unit cell Schulz and Cox 1991 . Of these two configurations, the (1×1) pattern was attributed to a stoichiometric (ST) surface, prepared by vacuum annealing at 1000 K. This surface demonstrated associative O 2 adsorption only at low temperatures. Conversely, the ( \sqrt{3} × \sqrt{3} )R30° reconstruction was observed after multiple adsorption-desorption cycles, inducing oxygen removal from the surface. However, x-ray photoelectron spectroscopy did not reveal a notable reduction in its oxygen content in comparison to the (1×1) surface Schulz and Cox 1991 . Nevertheless, the ( \sqrt{3} × \sqrt{3} )R30° reconstruction of the Cu 2 O(111) surface, attributed to oxygen deficiency, prompted extensive experimental and theoretical investigations over the following years Önsten et al. 2009 ; Zhang et al. 2018 ; Pérez León et al. 2012 . Subsequent studies to explore the surface atomic structure of Cu 2 O(111) have predominantly relied on density functional theory (DFT) calculations. Previous studies employing the Generalized Gradient Approximation (GGA) have investigated various reconstructions of Cu 2 O(111) surface, including Cu- and O-rich and deficient states Soon et al. 2007 ; Gloystein et al. 2020 . These studies typically commence with a bulk-terminated stoichiometric (111) surface (ST), which is characterized by an interwoven network of Cu-O six-membered rings. When cleaving the bulk crystal, this non-polar (stoichiometric) surface exposes an equal number of coordinatively unsaturated Cu and O atoms, which are designated as Cu cus and O cus , respectively. The Cu cus atoms are positioned at the center of the Cu 6 O 6 rings and bind to only one subsurface oxygen ion, while O cus are located at the corners of the rings. The findings indicate that the Cu-deficient (CuD) surface, which lacks the highly undercoordinated Cu cus atoms, is thermodynamically the most stable under O-rich conditions. DFT+U calculations indicate that the CuD surface remains the most stable structure across a wide range of oxygen chemical potentials Bendavid and Carter 2013 . The relative stability of the ST versus CuD terminations was further examined using hybrid functionals, which revealed that the CuD termination is less favorable due to the increased cost of polarity compensation Nilius et al. 2016 ; Gloystein et al. 2020 ; Gloystein et al. 2021 . Scanning Tunneling Microscopy (STM) images of the ( \sqrt{3} × \sqrt{3} )R30° reconstruction reveal triangular protrusions in the superstructure, which prompted the proposal of an alternative surface model Gloystein et al. 2020 ; Gloystein et al. 2021 ; Gloystein et al. 2022 . This model entailed the addition of Cu 4 O complexes to the cavities of one-third of the Cu 6 O 6 rings on a Cu-deficient surface. The resulting nanopyramidal reconstruction was found to be energetically favorable when assessed using Heyd-Scuseria-Ernzerhof (HSE) functionals Gloystein et al. 2020 ; Gloystein et al. 2021 . Nevertheless, finding the structural model of these pyramids remains highly reliant on human expertise. The objectives of the present study are summarized below. First, we want to determine whether machine-learned force fields (MLFFs) can enumerate low-energy structures for different stoichiometries. For simplicity, we decided to train our machine-learned force field on stoichiometric structures using PBE only. This is a very pragmatic choice, but easily and conveniently adaptable to other surfaces. Once the MLFFs is trained, we combine it with parallel tempering (PT) to identify candidate structures for different stoichiometries. We consider the well-studied ( \sqrt{3} × \sqrt{3} )R30° unit cell and the slightly larger (2 \times 2) supercell. Subsequently, we incorporated structures from oxygen-deficient and oxygen-rich configurations into the training dataset and generated a second-generation MLFF, and again performed PT simulations for different stoichiometries to evaluate the