Hydration Structures and Perturbed Hydrogen Bond Network in Salt Solutions by Advanced Ab Initio Molecular Dynamics and Electronic Structure Simulation Methods

该奖项支持研究和教育活动，旨在促进我们对盐水物理和化学性质的基本理解。盐水在地球上普遍存在，它在许多生理和生物过程中起着重要作用，如活细胞的信号转导以及蛋白质的溶解和折叠。水分子如何排列和围绕盐离子移动的精确图像是理解盐水性质的先决条件。这需要对水和盐分子的电子和原子核进行量子力学建模。然而，由于计算成本高，目前的计算研究只能在短时间内在小模拟箱中模拟盐水，这使得难以提取有关盐分子对水的分子和电子结构的长期影响的准确信息。该项目旨在建立理论工具和计算框架，通过使用机器学习方法来解决这个问题。PI和他的团队将利用计算密集型计算仅对盐水的几种配置进行计算，构建高效的深度学习模型，可以在包含数千个水分子的模拟盒中对盐水进行长时间模拟。由此产生的深度学习模型不仅计算效率高，而且还能够产生准确的预测，其准确度可与先进的完全量子力学计算相媲美。通过这种方法，可以准确预测盐水的微观和宏观性质，并与实验数据进行比较。参与的博士后和研究生将在凝聚态理论和量子化学领域开展基于先进分子动力学模拟和机器学习技术的前沿跨学科研究。盐溶液的精确深度学习模型将被开发并分发给计算材料社区。一个探索水科学的推广活动将在新泽西开发并提交给高中学生。技术总结该奖项支持旨在了解重要盐水溶液的水合结构和探索水合离子如何影响液态水的氢键网络的研究和教育活动。基于密度泛函理论的从头算分子动力学模拟为从第一性原理研究盐水提供了一个理想的框架。然而，盐水性质的准确预测需要高水平的函数近似，并且由于质子的质量小，核量子效应应该被考虑在内。因此，盐水的精确第一原理计算在计算上非常昂贵，并且不能应用于研究水合离子对水结构的长期影响。这位PI和他的团队将通过结合深度学习技术和先进的密度泛函理论计算来克服这些计算挑战。基于混合元GGA SCAN0密度泛函的Feynman路径积分从头算分子动力学模拟，将建立有效的盐溶液深部分子动力学模型。使用深度学习模型，盐水的分子动力学模拟可以在纳秒时间尺度下在包含数千个水分子的模拟盒中进行，其准确度可与直接密度泛函理论计算相媲美。通过这种强大的方法，收敛的热力学性质，如扩散系数和分子结构在盐水中可以准确地预测和研究。利用盐水的分子轨迹，通过深度学习技术对电子性质进行建模，并与现有的实验结果进行比较。参与的博士后和研究生将在凝聚态理论和量子化学领域开展基于先进从头算分子动力学模拟和机器学习技术的前沿跨学科研究。盐溶液的精确深度学习模型将被开发并分发给计算材料社区。一个探索水科学的推广活动将被开发并提交给新泽西的高中生。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的知识价值和更广泛的影响审查标准进行评估来支持。

NONTECHNICAL SUMMARYThis award supports research and educational activities with an aim to advance our fundamental understanding of the physical and chemical properties of salt water. Salt water is ubiquitous on earth, and it plays important roles in many physiological and biological processes, such as signal transduction in living cells as well as protein solubility and folding. The precise picture of how water molecules arrange themselves and move around salt ions is prerequisite to understanding the properties of salt water. This requires a quantum mechanical modeling for both the electrons and nuclei of water and salt molecules. However, due to the high computational cost, current computational studies can only model salt water in small simulation boxes for short time periods, which makes it difficult to extract accurate information about the long-range effects of salt molecules on the molecular and electronic structures of water. This project seeks to build theoretical tools and a computational framework to solve this problem by using machine learning methods. Utilizing computationally intensive calculations for only a few configurations of salt water, the PI and his team will build efficient deep learning models that can model salt water in simulation boxes containing thousands of water molecules for long time scales. The resulting deep learning models will not only be computationally efficient, but they will also be able to yield accurate predictions with comparable accuracy to advanced, fully quantum mechanical calculations. Both the microscopic and macroscopic properties of salt water will be accurately predicted by this approach and compared with experimental data.The participating postdocs and graduate students will carry out cutting-edge, interdisciplinary research based on advanced molecular dynamics simulations and machine learning techniques in the fields of condensed matter theory and quantum chemistry. Accurate deep learning models of salt solutions will be developed and distributed to the computational materials community. An outreach activity that explores the science of water will be developed and presented to high school students in New Jersey.TECHNICAL SUMMARYThis award supports research and educational activities that are aimed at understanding the hydration structure of important salt aqueous solutions and exploring how the hydrated ions will affect the hydrogen-bond network of liquid water. Ab initio molecular dynamics simulations based on density functional theory provide an ideal framework to study salt water from first principles. However, accurate predictions of salt water properties require a high-level functional approximation, and nuclear quantum effect should be taken into account due to the small mass of the proton. Accordingly, accurate first-principles calculations of salt water are computationally very expensive and cannot be applied to study the long-range effect of hydrated ions on the water structure. This PI and his team will overcome these computational challenges by combined deep learning techniques and advanced density functional theory calculations. Efficient deep molecular dynamics models of salt solutions will be built on Feynman path integral ab initio molecular dynamics simulations based on the hybrid meta-GGA SCAN0 density functional. Using deep learning models, molecular dynamics simulations of salt water can be performed in a simulation box containing thousands of water molecules at the nanosecond time scale, with a level of accuracy comparable to direct density functional theory calculations. With this powerful approach, the converged thermodynamic properties such as diffusivities and molecular structure in salt water can be accurately predicted and studied. Using the molecular trajectories of salt water, the electronic properties will also be modeled by deep learning techniques and compared to available experimental results.The participating postdocs and graduate students will carry out cutting-edge, interdisciplinary research based on advanced ab initio molecular dynamics simulations and machine learning techniques in the fields of condensed matter theory and quantum chemistry. Accurate deep learning models of salt solutions will be developed and distributed to the computational materials community. An outreach activity that explores the science of water will be developed and presented to high school students in New Jersey.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.