期刊
JOURNAL OF CHEMICAL THEORY AND COMPUTATION
卷 19, 期 15, 页码 5047-5057出版社
AMER CHEMICAL SOC
DOI: 10.1021/acs.jctc.3c00226
关键词
-
Induced dipole models have been limited by the energy conservation issue when utilizing historical data. This study identifies error outliers as the primary factor causing energy conservation failure and proposes a comprehensive scheme to overcome this limitation. The scheme incorporates a multi-order extrapolation method to quicken induction iteration, a preconditioned conjugate gradient with local iterations to refine the iteration process, and a peek step via Jacobi under-relaxation for optimal performance. Simulation evidence suggests that this proposed scheme achieves energy convergence similar to point-charge models with improved efficiency and accuracy.
Induced dipole models have proven to be effective toolsfor simulatingelectronic polarization effects in biochemical processes, yet theirpotential has been constrained by energy conservation issue, particularlywhen historical data is utilized for dipole prediction. This studyidentifies error outliers as the primary factor causing this failureof energy conservation and proposes a comprehensive scheme to overcomethis limitation. Leveraging maximum relative errors as a convergencemetric, our data demonstrates that energy conservation can be upheldeven when using historical information for dipole predictions. Ourstudy introduces the multi-order extrapolation method to quicken inductioniteration and optimize the use of historical data, while also developingthe preconditioned conjugate gradient with local iterations to refinethe iteration process and effectively remove error outliers. Thisscheme further incorporates a peek step via Jacobiunder-relaxation for optimal performance. Simulation evidence suggeststhat our proposed scheme can achieve energy convergence akin to thatof point-charge models within a limited number of iterations, thuspromising significant improvements in efficiency and accuracy.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据