Atomistic simulation of alumina grain boundary structure and diffusion

含铝或铬合金上的氧化皮提供有效且廉价的腐蚀保护。然而，氧化皮形成的关键原子尺度机制是未知的或有争议的，这限制了洞察力和理解，并阻碍了最佳合金和氧化皮的计算设计。氧化皮形成机制的原子模拟是困难的。首先，需要势能面的精确表示，使得能够模拟晶界中的复杂原子重排，并且包括电荷转移对能量的贡献。其次，模拟单元需要包括数千个原子以采样真实的几何形状，并且模拟需要运行足够长的时间以观察原子重排和扩散。虽然密度泛函理论（DFT）足够精确，但它无法模拟那么多的原子。另一方面，通过采用经典的原子间势可以模拟足够的原子，但这些势通常不够准确或可转移，即使包括电荷转移模型。直到最近，一类精确的原子间相互作用势，通常被称为机器学习势，才变得清晰，它描述了一个接近DFT精度的势能面。在这一类中，我们赞成的方法是原子团簇扩展（ACE），其中我们将电荷转移。它是快速和系统地改进。用这种方法，我们建议开发准确的和可转移的铝-氧系统的原子间相互作用势。我们将采用这些潜力的模拟扩散在晶界的氧化物规模和提取从我们的模拟原子尺度的机制和扩散系数，是决定性的理解氧化物规模的增长速度。我们的项目将是我们所知的第一个解决高温下氧化物晶界原子尺度扩散机制和速率问题的项目，为原子尺度的保护膜设计奠定基础。

Oxide scales on Al- or Cr-containing alloys provide efficient and inexpensive corrosion protection. However, key atomic scale mechanisms in oxide scale formation are unknown or disputed, which limits insight and understanding and blocks the computational design of optimal alloys and oxide scales. Atomistic simulation of mechanisms in oxide scale formation is difficult. Firstly, an accurate representation of the potential energy surface is required that enables the simulation of complex atomic rearrangements in grain boundaries and includes the contribution of charge transfer to the energy. Secondly, the simulation cells need to comprise thousands of atoms to sample realistic geometries and simulations need to be run for sufficiently long times to observe atomic rearrangements and diffusion. While density functional theory (DFT) is sufficiently accurate, it is unable to simulate that many atoms. On the other hand, simulations with sufficient atoms are possible by employing classical interatomic potentials, but these potentials are generally not sufficiently accurate or transferable, even when including a model of charge transfer. Only recently a class of accurate interatomic potentials, often termed machine-learning potentials, became prominant, which describe a potential energy surface with near-DFT precision. Within this category the approach we favour is the Atomic Cluster Expansion (ACE), in which we are incorporating charge transfer. It is fast and systematically improvable. With this approach we propose to develop accurate and transferable interatomic potentials for the aluminium-oxygen system. We will employ these potentials for the simulation of diffusion in grain boundaries of the oxide scale and extract from our simulations the atomic scale mechanisms and diffusion coefficients that are decisive for understanding the growth rates of oxide scales. Our project will be the first to our knowledge which tackles the problem of atomic scale diffusion mechanisms and rates in oxide grain boundaries at high temperature, laying the foundations for atomic scale design of protective scales.