We have developed a high-dimensional neural network potential (NNP) to accurately describe the structural and energetic properties of borophene deposited on silver. This NNP matches the accuracy of density functional theory (DFT) calculations while achieving computational speedups of several orders of magnitude, enabling the study of extensive structures that may reveal intriguing moiré patterns or surface corrugations.
Our approach involves constructing the training dataset efficiently using an iterative technique known as the “adaptive learning approach.” This method ensures comprehensive coverage of structural variations and energetics encountered in borophene-silver systems.
The developed NNP potential demonstrates excellent agreement in predicting the structure, energy, and forces compared to DFT calculations. Moreover, we have investigated the stability of various borophene polymorphs, including those not initially included in the training dataset. Our results show enhanced stabilization for specific hole densities, notably for the α allotrope (ν = 1/9).
Furthermore, we find that the stability of borophene on the metal surface is orientation-dependent, suggesting structural corrugation patterns that require long-time simulations on extended systems to observe fully.
Additionally, the NNP is capable of simulating vibrational densities of states and producing realistic structures, with simulated scanning tunneling microscopy (STM) images closely matching experimental observations.
In summary, our study showcases the effectiveness of the developed NNP in accurately describing and predicting the behavior of borophene on silver surfaces, offering valuable insights into its stability, structure-property relationships, and potential applications.
The research introduces a sophisticated neural network potential (NNP), an artificial intelligence model designed to predict the behavior of borophene when placed on top of silver. This NNP is as precise as the best physics simulations currently used (Density Functional Theory - DFT calculations) but operates much faster, enabling the study of larger and more complex structures. These structures may exhibit interesting patterns like moiré patterns (large-scale interference patterns seen when two grids are overlaid at an angle) or surface corrugations (surface waves).
To train this neural network, researchers used an adaptive learning approach, allowing the model to learn iteratively and improve its predictions over time. The developed NNP demonstrated excellent agreement with DFT calculations, accurately predicting the structure, energy, and forces within borophene.
Furthermore, the study revealed that different forms of borophene (polymorphs) exhibit varying levels of stability, influenced by specific patterns of empty spaces (hole density). The orientation of borophene on silver was shown to affect its stability, suggesting potential formation of wavy patterns on the surface, which could only be observed through long-term simulations on large systems enabled by the NNP.
In addition to structural properties, the NNP successfully simulated vibrational densities of states, describing how atoms in the material vibrate, and generated realistic structures. Simulated images from the Scanning Tunneling Microscope (STM), depicting atomic arrangements on the surface, closely resembled actual experimental observations.
In summary, this research developed a powerful tool for predicting the properties of borophene on silver, offering valuable insights for material design and detailed property studies. The NNP's capability to efficiently simulate complex behaviors opens avenues for exploring novel materials and their applications.
