摘要
由一名新闻记者-机器人与机器学习每日新闻的工作人员新闻编辑-关于人工智能的新研究结果已经发表。根据来自安徽的新闻报道，由NewsRx记者报道，研究表明：“固态电解质是下一代储能和节能设备的关键成分。”本研究的资助者包括中国国家自然科学基金、中国科学院。新闻记者引用了中国科技大学的一篇文章：“机器学习分子动力学(MLMD)在研究固体电解质中移动离子的扩散率方面有很大的前景，其效率远远高于传统的从头算分子动力学(AIMD)。”我们结合了一种高效的嵌入式原子神经网络(EANN)方法和一种不确定性驱动的主动学习算法,该算法从高温AIMD轨迹中最优地选择数据点来构造固态电解质的ML势,并在Benchmar K系统Li3YCl6上验证了该策略,该系统的MLMD仿真结果存在争议。我们发现通常使用的小型超级电池IN AIMD模拟无法预测临界温度下的超音速转变，导致在室温下Li3Y Cl6中Li+电导率的显著高估。幸运的是，由于EANN Pote Ntial的可扩展性，在足够大的电池中进行扩展的MLMD模拟，在420 K时电导率随温度变化不大，室温电导率明显降低，与实验结果一致，有趣的是，我们的结果都是基于半局部PBE密度泛函，本文分析了文献报道的具有不同ML潜力的MLMD结果不一致的可能原因。
Abstract
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News-New study results on artificial intelligence have been published. According to news reporting originating from Anhui, People's Re public of China, by NewsRx correspondents, research stated, "Solidstate electro lytes are key ingredients in next-generation devices for energy storage and rele ase." Funders for this research include National Natural Science Foundation of China; Chinese Academy of Sciences. The news correspondents obtained a quote from the research from University of Sc ience and Technology of China: "Machine learning molecular dynamics (MLMD) has s hown great promise in studying the diffusivity of mobile ions in solid-state ele ctrolytes, with much higher efficiency than conventional ab initio molecular dyn amics (AIMD). In this work, we combine an efficient embedded atom neural network (EANN) approach and an uncertainty-driven active learning algorithm that optima lly selects data points from high-temperature AIMD trajectories to construct ML potentials for solid-state electrolytes and validate this strategy in a benchmar k system, Li3YCl6, for which several controversy theoretical results exist. Thro ugh systematic MLMD simulations, we find that a typically used small supercell i n AIMD simulations fails to predict the supersonic transition at a critical temp erature, leading to a significant overestimation of the Li+ conductivity in Li3Y Cl6 at room temperature. Fortunately, thanks to the scalability of the EANN pote ntial, extended MLMD simulations in a sufficiently large cell does yield a notab le change of temperature-dependence in conductivity at 420 K and a much lower ro om-temperature conductivity in excellent with experiment. Interestingly, our res ults are all based on a semi-local PBE density functional, which was argued unab le to predict the superionic transition. We analyze possible reasons of the seem ingly inconsistent MLMD results reported in literature with different ML potenti als."
NETL
NSTL