Understanding and controlling polymorphism in molecular solids

项目试图解决的研究问题/项目的目标；理解和控制多晶态，即分子结晶成多种固体形式，是结构化学中尚未解决的主要问题。虽然可以通过计算晶格能量获得对多态的一些理论见解，但力场和密度泛函理论并不容易将能量分解为不同类型的相互作用。这严重限制了从晶体结构预测研究中获得的洞察力，否则这些研究可能会指出新的预测规则或“更智能”的筛选方法。一个目标是通过使用一种最新的称为相互作用量子原子（IQA）的能量分配方法来识别多态系统中的相关相互作用。第二个目标是训练一种被称为克里格的机器学习技术，根据这些原子能贡献来推导出一个精确的、可转移的力场，称为FFLUX。我们将使用FFLUX来探索更大的问题，如预测表面终止/反应性和晶体形态。第三个目标是将FFLUX纳入ukri赞助的软件包DL_POLY （Daresbury实验室），并证明FFLUX获得的结果确实比经典力场产生的结果更接近实验。第四个目标是通过使用剑桥结构数据库来选择分子并确定构建力场的主要自由度，从而使这些过程自动化。我们的最终目标是提供一个全面的参数和软件数据库，使晶体工程师能够执行具有DFT精度水平的“点和点击”力场计算。回答这些问题的方法（学生实际会做什么）；该项目建立在EPSRC奖学金“分子组装的可靠计算预测”的基础上。这项工作产生了下一代内部力场，称为FFLUX。这个力场比基于点电荷的力场（如AMBER）更现实。FFLUX还引入了多极矩，这对于精确的静电是必不可少的，它控制了形成晶体的典型极性分子之间相当大一部分的非共价相互作用。现代IQA方法在原子能量分析的严谨性上提供了一个阶级性的变化。IQA是一种无参数的方法，直观，但同时，非常接近原子本身的量子力学特性。当与内部的相对能量梯度（REG）方法相结合时，IQA会返回关于哪些原子控制整个系统行为以及为什么（即通过哪种类型的能量）的强有力的定性陈述。REG方法是由主要导师小组开发的，并带有内部代码REG.py.学生将使用FORTRAN90和Python编程，修改内部代码ICHOR， DL_FFLUX和REG.py，使其与晶体结构预测问题相连接。他将首次使用FFLUX对晶体进行分子动力学模拟。学生将参与细致而系统的测试，从而获得前所未有的洞察力。他将微调FFLUX的参数化。还将与经典力场进行比较。研究的新颖工程和/或物理科学内容（将其置于EPSRC职权范围内的科学）。该项目隶属于化学科学大挑战“定向组装具有目标特性的扩展结构（DAESTP）”。这个项目有很强的机器学习成分，因此与人工智能重叠，这是一个热门的融资话题。相关的科学产品被称为FFLUX，这是一个全新的力场，由新颖的原理设计并编码为软件包。

the research questions the project is trying to address/the objectives of the project; Understanding and controlling polymorphism, where molecules crystallise into multiple solid forms, is a major unsolved problem in structural chemistry. While some theoretical insight into polymorphism can be obtained by calculating lattice energies, force fields and Density Functional Theory do not readily decompose energies into different types of interaction. This severely limits the insight available from crystal-structure prediction studies that could otherwise point to new predictive rules or "smarter" screening approaches. One objective is to identify the pertinent interactions in polymorphic systems by using a recent energy partitioning method called Interacting Quantum Atoms (IQA). The second objective is to train a machine-learning technique called kriging on these atomic energy contributions to derive an accurate and transferable force field called FFLUX. We will use FFLUX to explore larger problems such as predicting surface termination/reactivity and crystal morphology. The third objective is to incorporate FFLUX into the UKRI-sponsored software package DL_POLY (Daresbury lab), and to prove that the results obtained by FFLUX are indeed closer to experiment than those produced by classical force fields. A fourth objective is to automate these processes by using the Cambridge Structural Database to select molecules and identify the major degrees of freedom for building the force field. Our ultimate aim is to provide a comprehensive database of parameters and software to allow crystal engineers to perform "point and click" force field calculations with DFT levels of accuracy.the approach that will be taken to answer these questions (what the student will actually be doing); This project builds on the results of the EPSRC Fellowship entitled "Reliable computational prediction of molecular assembly". This work has led to a next-generation in-house force field called FFLUX. This force field is much more realistic than a point-charge based force field such as AMBER. FFLUX also introduces multipole moments, which are essential for accurate electrostatics governing a considerable part of the non-covalent interactions between the typically polar molecules that form the crystals. The modern IQA method offers a step change in the rigour of atomistic energy analysis. IQA is a parameter-free method that is intuitive but, at the same time, very close to the quantum mechanical character of atoms themselves. When combined with the in-house Relative Energy Gradient (REG) method IQA returns powerful qualitative statements on which atoms govern the behaviour of the overall system and why (i.e. by which type of energy). The REG method has been developed in the main supervisor's group and comes with an in-house code called REG.py.The student will be programming in FORTRAN90 and Python, modifying the in-house codes ICHOR, DL_FFLUX and REG.py to interface them with the problem of crystal structure prediction. He will run molecular dynamics simulations using FFLUX, for the first time on crystals. The student will be involved with careful and systematic testing and thus gather unprecedented insight. He will fine-tune the parameterisation of FFLUX. Comparisons with classical force fields will also be made.the novel engineering and/or physical sciences content of the research (the science that places it within EPSRC's remit).This project resorts under the Chemical Sciences Grand Challenge of "Directed Assembly of Extended Structures with Targeted Properties (DAESTP)". There is a strong Machine Learning component to this project, and thus overlap with Artificial Intelligence, a popular funding topic. The associated scientific product is called FFLUX, which is a completely new force field, designed by novel principles and encoded as a software package.