Zheng Jingjing, Frisch Michael J
Gaussian, Inc. , Wallingford, Connecticut 06492, United States.
J Chem Theory Comput. 2017 Dec 12;13(12):6424-6432. doi: 10.1021/acs.jctc.7b00719. Epub 2017 Nov 10.
An efficient geometry optimization algorithm based on interpolated potential energy surfaces with iteratively updated Hessians is presented in this work. At each step of geometry optimization (including both minimization and transition structure search), an interpolated potential energy surface is properly constructed by using the previously calculated information (energies, gradients, and Hessians/updated Hessians), and Hessians of the two latest geometries are updated in an iterative manner. The optimized minimum or transition structure on the interpolated surface is used for the starting geometry of the next geometry optimization step. The cost of searching the minimum or transition structure on the interpolated surface and iteratively updating Hessians is usually negligible compared with most electronic structure single gradient calculations. These interpolated potential energy surfaces are often better representations of the true potential energy surface in a broader range than a local quadratic approximation that is usually used in most geometry optimization algorithms. Tests on a series of large and floppy molecules and transition structures both in gas phase and in solutions show that the new algorithm can significantly improve the optimization efficiency by using the iteratively updated Hessians and optimizations on interpolated surfaces.
本文提出了一种基于具有迭代更新海森矩阵的插值势能面的高效几何优化算法。在几何优化的每一步（包括最小化和过渡结构搜索），通过使用先前计算的信息（能量、梯度和海森矩阵/更新后的海森矩阵）来适当地构建一个插值势能面，并且以迭代方式更新两个最新几何结构的海森矩阵。插值面上优化后的最小值或过渡结构被用作下一个几何优化步骤的起始几何结构。与大多数电子结构单梯度计算相比，在插值面上搜索最小值或过渡结构以及迭代更新海森矩阵的成本通常可以忽略不计。这些插值势能面在比大多数几何优化算法中通常使用的局部二次近似更广泛的范围内，往往能更好地表示真实的势能面。对一系列气相和溶液中的大分子、柔性分子以及过渡结构的测试表明，新算法通过使用迭代更新的海森矩阵和在插值面上进行优化，能够显著提高优化效率。
