发表机构
Duke University（杜克大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究推出开源Python库libNLPBE，采用mDINMH方法等技术求解NLPBE，结合密度泛函计算获取Fock矩阵静电校正项，还开发了GPU加速版本，以推进电解质溶液中离子效应的理论研究。
AI 中文摘要
溶质周围的溶剂环境会显著改变其化学性质，在电解质溶液中，这种变化更为明显，其中移动离子通过静电相互作用对溶质产生重大影响。对这些离子效应的理论描述将有益于电解质溶液中化学过程的设计，因此非线性泊松-玻尔兹曼方程（NLPBE）已成为电子结构计算中解决此类效应的高效隐式溶剂模型。然而，适用于分子电子结构计算的NLPBE求解器可用性有限，极大阻碍了对离子效应的理论研究。为提高NLPBE的可及性，我们推出libNLPBE，这是一个开源Python库，可结合密度泛函计算求解分子体系的NLPBE，专门用于获取电解质溶液环境产生的Fock矩阵静电校正项。该库采用密度拟合（DF）近似高效计算溶质的静电势；此外，我们开发了由Holst提出的改进阻尼不精确牛顿多重网格（mDINMH）方法来求解NLPBE，该方法具有用于求解牛顿方程的对称预条件子，可使用专为对称线性算子设计的稳健多重网格方法，还在mDINMH中集成了代数多重网格方法以支持广泛的网格点数；我们还推出了libNLPBE的GPU加速版本，以利用GPU的并行效率。
英文摘要
Solvent environments surrounding a solute can significantly alter its chemical properties. These changes become more prominent in electrolyte solutions, where mobile ions largely influence the solute through electrostatic interactions. A theoretical description of these ionic effects will benefit the design of chemistry performed in electrolyte solutions. As a result, the non-linear Poisson-Boltzmann equation (NLPBE) has emerged as an efficient implicit solvent model for electronic structure calculations to address such effects. However, the limited availability of NLPBE solvers for molecular electronic structure calculations greatly hinders theoretical investigation into the ionic effects. To improve accessibility of the NLPBE, we present libNLPBE, an open-source Python library, that solves the NLPBE for molecular systems in combination with density functional calculations, specifically to obtain electrostatic correction terms arising from the electrolyte solution environment for the Fock matrix. The library employs the density fitting (DF) approximation to efficiently calculate solute electrostatic potentials. Furthermore, we develop the modified Damped Inexact Newton Multigrid developed by Holst (mDINMH) method for solving the NLPBE. The mDINMH features a symmetric preconditioner for solving the Newton equation, which enables the use of robust multigrid methods designed for symmetric linear operators. In addition, an algebraic multigrid method has been incorporated into the mDINMH to support a broad range of grid points. We also present a GPU-accelerated version of libNLPBE to leverage parallelization efficiency of GPUs.
