Boosting Density Embedding with Machine Learning and Nonstandard Workflows

纽瓦克罗格斯大学的Michele Pavanello教授得到了化学部化学理论、模型和计算方法计划的支持，根据第一性原理对分子凝聚相和材料进行建模。帕瓦内洛和他的研究小组将开发编码第一原理量子力学的开源软件，用于预测材料性能及其合理设计。这些方法将利用最先进的机器学习和分而治之的算法来加快计算时间，而不会损害方法的准确性和严密性。帕瓦内洛将通过计算预测反应路径，将新开发的方法应用于催化领域的公开问题，并通过预测电荷迁移率和多态，将其应用于材料工程。与科学进步相辅相成的是一项强有力的推广计划，旨在以罗格斯-纽瓦克和附近学区之间已经建立的计划为基础，培训来自没有代表性的少数族裔学区的高中生计算机编码(巨蟒训练营)。米歇尔·帕瓦内洛教授和他的研究小组将开发密度泛函理论嵌入方法，旨在计算材料和分子凝聚相的电子结构，解决从催化到分子半导体中电荷迁移率的公开问题。该项目将包括：(1)Boost DFT嵌入，使用机器学习的Kohn-Sham单电子约化密度矩阵来访问超过纳米尺度的系统尺寸；(2)开发一种自适应嵌入方法，该方法实时地定义用于模拟的最精确的子系统拓扑，并且能够运行从头算动力学，用于催化和其他化学过程；以及(3)开发“混合”嵌入方案的非标准工作流程，例如嵌入在无轨道DFT中的Kohn-Sham DFT以处理纳米级金属子系统，DFT中的波函数理论用于带电凝聚相系统的应用以及晶体多态的预测。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Professor Michele Pavanello of Rutgers University-Newark is supported by an award from the Chemical Theory, Models and Computational Methods program in the Division of Chemistry to model molecular condensed phases and materials from first principles. Pavanello and his research group will develop open-source software encoding first-principles Quantum Mechanics for the prediction of materials properties and their rational design. The methods will exploit state-of-the-art machine learning and divide-and-conquer algorithms to speed up the computational time without compromising on the accuracy and rigorousness of the approaches. Pavanello will apply the newly developed methods to open problems in catalysis by computationally predicting reaction paths and in materials engineering by predicting charge mobilities and polymorphism. The scientific advances will be complemented by a strong outreach program aimed at training high school students from unrepresented minority school districts in computer coding (Python bootcamps) building on already-established programs between Rutgers-Newark and nearby school districts.Professor Michele Pavanello and his research group will develop density functional theory embedding methods aimed at computing the electronic structure of materials and molecular condensed phases tackling open problems ranging from catalysis to charge mobility in molecular semiconductors. The project will consist of: (1) boost DFT embedding employing machine learned Kohn-Sham one-electron reduced density matrices to access system sizes beyond the nanoscale; (2) develop an adaptive embedding method that defines in real time the most accurate subsystem topology for the simulation and is capable of running ab initio dynamics for applications to catalysis and other chemical processes; and (3) develop nonstandard workflows for “hybrid” embedding schemes, such as Kohn-Sham DFT embedded in orbital-free DFT to tackle nanoscale metallic subsystems, wavefunction theory in DFT for applications to charged condensed-phase systems and prediction of crystal polymorphs.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.