The Dataset of Predicting Strain Effects on Adsorption Energy

Strain significantly influences chemical reaction potential energy surface, which is a challenge to research due to the high-dimensional space. Here, we propose an approach to establish the relationship between the geometry, electronic density of states (DOS), and oxygen molecular adsorption energies using convolutional neural network and graph convolution neural network. By examining the adsorption energy of oxygen molecular on alloy substrates in response to surface strain, we identify that volumetric strain is the dominant parameter affecting adsorption energy and energy barriers of the reaction path. Employing features extracted from the electronic DOS, we provide physically insights by predicting adsorption energy responses to external perturbations to the electronic structure. In addition, explainable machine-learning based on moments of DOS identify the physically meaning of DOS feature and reveal the mechanochemical essence is strain regulated surface electronic environment.

应变可显著影响化学反应势能面，由于该问题所处的高维空间特性，相关研究极具挑战。本文提出一种研究方法，借助卷积神经网络（convolutional neural network）与图卷积神经网络（graph convolution neural network），构建几何结构、态密度（electronic density of states, DOS）与氧分子吸附能三者间的关联。通过分析合金基底表面应变对氧分子吸附能的调控规律，本文发现体积应变是影响吸附能与反应路径能垒的核心参数。本文利用从态密度中提取的特征，通过预测吸附能对外界电子结构扰动的响应，给出了物理层面的机理解释。此外，基于态密度矩的可解释机器学习方法，明确了态密度特征的物理意义，并揭示了应变调控表面电子环境这一机械化学本质。