New developments in evolutionary structure prediction algorithm USPEX
New developments in evolutionary structure prediction algorithm USPEX
复制标题
DOI：
10.1016/j.cpc.2012.12.009
复制
发表时间：
2013-04-01
影响因子：
6.3
通讯作者：
Zhu, Qiang
中科院分区：
文献类型：
--
作者：
Lyakhov, Andriy O.;Oganov, Artem R.;Zhu, Qiang
本文介绍了用于晶体结构预测的进化算法USPEX的最新进展及其对团簇结构预测的适应性。我们展示了如何生成随机对称结构，以及如何引入“智能”变异算子，学习更好的局部环境。这些和其他方面的发展大大提高了算法的效率，并允许可靠地预测单元单元中最多具有200个类似原子的结构。我们证明了粒子群优化算法(PSO)的高级版本可以在我们的方法的基础上被创建，但PSO的性能明显优于USPEX。我们还展示了如何将元动力学的想法用于进化结构预测，以摆脱局部极小值。我们的团簇结构预测算法，使用了最初为晶体开发的思想，也显示出优异的性能，并优于其他最先进的算法。(C)2012爱思唯尔B.V.保留所有权利。
We present new developments of the evolutionary algorithm USPEX for crystal structure prediction and its adaptation to cluster structure prediction. We show how to generate randomly symmetric structures, and how to introduce 'smart' variation operators, learning about preferable local environments. These and other developments substantially improve the efficiency of the algorithm and allow reliable prediction of structures with up to similar to 200 atoms in the unit cell. We show that an advanced version of the Particle Swarm Optimization (PSO) can be created on the basis of our method, but PSO is strongly outperformed by USPEX. We also show how ideas from metadynamics can be used in the context of evolutionary structure prediction for escaping from local minima. Our cluster structure prediction algorithm, using the ideas initially developed for crystals, also shows excellent performance and outperforms other state-of-the-art algorithms. (c) 2012 Elsevier B.V. All rights reserved.