Improvement of look ahead based on quadratic approximation for crystal structure prediction
Improvement of look ahead based on quadratic approximation for crystal structure prediction
复制标题
DOI：
10.1080/27660400.2022.2059335
复制
发表时间：
2022-04
期刊：
影响因子：
--
通讯作者：
T. Yamashita;Hirotaka Sekine
中科院分区：
文献类型：
--
作者：
T. Yamashita;Hirotaka Sekine
摘要 基于第一性原理计算的晶体结构预测通常非常耗时，因为许多候选结构必须进行局部优化。基于二次近似的 Look Ahead 是我们之前开发的选择型算法之一，可以控制候选的优化优先级。在许多情况下它可以有效地降低计算成本，但是，人们发现它在某些数据集中效率较低。在本研究中，我们提出了一种基于二次近似的改进的Look Ahead分数，其中添加了应力项以克服先前分数的缺点。通过该改进算法进行晶体结构预测模拟，以研究对 Si、Al2O3、NaCl 和 SrCO3 等典型材料的搜索效率。这些结果表明，在大多数情况下，这种改进的算法相对于随机搜索将搜索成本降低到 40% 以下。应力项的引入使得该算法更加稳健和通用。图解摘要
ABSTRACT Crystal structure prediction based on first-principles calculations is usually time-consuming since a lot of candidate structures have to be locally optimized. Look Ahead based on Quadratic Approximation, which is one of the selection-type algorithms we previously developed, can control the optimization priority of the candidates. It can efficiently reduce the computational costs in many cases, however, it has been found to be inefficient in some data sets. In the present study, we proposed an improved score of Look Ahead based on Quadratic Approximation, where the stress term is added to overcome the drawbacks of the previous score. Crystal structure prediction simulations by this improved algorithm are performed to investigate the searching efficiency for typical materials such as Si, Al2O3, NaCl, and SrCO3. These results demonstrate that this improved algorithm reduces the searching cost to less than 40% with respect to random search in most cases. The introduction of the stress term makes this algorithm more robust and versatile. GRAPHICAL ABSTRACT