Akin-Ojo Omololu, Song Yang, Wang Feng
Department of Chemistry, Boston University, Boston, Massachusetts 02215, USA.
J Chem Phys. 2008 Aug 14;129(6):064108. doi: 10.1063/1.2965882.
A new method called adaptive force matching (AFM) has been developed that is capable of producing high quality force fields for condensed phase simulations. This procedure involves the parametrization of force fields to reproduce ab initio forces obtained from condensed phase quantum-mechanics/molecular-mechanics (QM/MM) calculations. During the procedure, the MM part of the QM/MM is iteratively improved so as to approach ab initio quality. In this work, the AFM method has been tested to parametrize force fields for liquid water so that the resulting force fields reproduce forces calculated using the ab initio MP2 and the Kohn-Sham density functional theory with the Becke-Lee-Yang-Parr (BLYP) and Becke three-parameter LYP (B3LYP) exchange correlation functionals. The AFM force fields generated in this work are very simple to evaluate and are supported by most molecular dynamics (MD) codes. At the same time, the quality of the forces predicted by the AFM force fields rivals that of very expensive ab initio calculations and are found to successfully reproduce many experimental properties. The site-site radial distribution functions (RDFs) obtained from MD simulations using the force field generated from the BLYP functional through AFM compare favorably with the previously published RDFs from Car-Parrinello MD simulations with the same functional. Technical aspects of AFM such as the optimal QM cluster size, optimal basis set, and optimal QM method to be used with the AFM procedure are discussed in this paper.
一种名为自适应力匹配(AFM)的新方法已被开发出来,它能够为凝聚相模拟生成高质量的力场。该过程涉及对力场进行参数化,以重现从凝聚相量子力学/分子力学(QM/MM)计算中获得的从头算力。在这个过程中,QM/MM的MM部分会被迭代改进,以接近从头算的质量。在这项工作中,已经测试了AFM方法为液态水的力场进行参数化,以便生成的力场能够重现使用从头算MP2以及带有Becke-Lee-Yang-Parr(BLYP)和Becke三参数LYP(B3LYP)交换相关泛函的Kohn-Sham密度泛函理论所计算的力。在这项工作中生成的AFM力场评估起来非常简单,并且得到了大多数分子动力学(MD)代码的支持。同时,AFM力场预测的力的质量可与非常昂贵的从头算计算相媲美,并且发现能够成功重现许多实验性质。使用通过AFM从BLYP泛函生成的力场进行MD模拟得到的位点-位点径向分布函数(RDF)与之前发表的使用相同泛函的Car-Parrinello MD模拟得到的RDF相比具有优势。本文讨论了AFM的技术方面,如最优QM簇大小、最优基组以及与AFM过程一起使用的最优QM方法。