报告题目Title：Towards Generic Mathematical Descriptors for Structural Analysis and Their Applications in the Property Predictions for Molecular Crystals (建立对于分子晶体结构普适的数学描述符及其在物性预测上的应用)
报 告 人Speaker：Dr Jianliang Jack Yang 杨建樑博士  (School of Chemistry, University of Southampton 英国南安普顿大学化学系)
报告时间Time：2016年10月12日(周三) 9:00
报告地点Venue：校本部E106(金沙威尼斯欢乐娱人城量子与分子结构国际中心SHU ICQMS)
报告摘要：
The steady progress in the field of organic crystal structure predictions (CSP), exemplified by successes in the most recent international CSP blind test, has demonstrated great promise in using this computational tool in advancing our understanding in polymorphism. With global lattice energy exploring algorithms, the number of putative crystal structures generated is typically on the order between 10 to 100K, which poses a significant challenge in extracting structural patterns and thus generalizing useful chemical insights in an automated fashion. A more challenging task is how to correlate this structural information with the predicted physical properties, such as the charge transport parameters for organic crystals. The later is crucial for advancing computer-guided material designs. Unlike molecular conformation samplings in gas/solution phases, where a plethora of dimensionality reduction techniques had been developed to extract intriguing features from the energy landscapes, less had been developed for molecular crystals, due to a lack of generic and robust descriptor to represent molecular crystals ubiquitously.

Dimer synthons are the most widely used term in describing molecular crystals. Starting with the crystal structure landscape of a polyaromatic molecule, I will show how a simple set of Steinhardt bond order parameters can be applied to target -stacking motifs in molecular crystals, and hence to predict nearest-neighbouring transfer integrals for calculating electron mobilities. I will then discuss how a more agnostic fingerprint (SOAP) can be used to describe more extended structural features in molecular crystals, not only that it leads to near-identical structural classification by human inspection, but excellent predictions on charge mobilities. Finally, building on the formalism of many-body expansion for energy evaluation in molecular crystals, it will be demonstrated that how to design different ‘similarity kernels’ for highlighting structural similarities in molecular crystals at different ‘length scales’, which is particularly important for characterising hydrogen-bondings in molecular crystals.

分子晶体结构预测对于帮助我们深入理解分子晶体的多态性有着不可替代的优势，作为一项计算化学工具，其可靠性已经通过周期性的国际盲测得到验证。利用全局性晶体能量优化，一般的晶体结构预测可以生成一万到十万个可能的晶体结构。如何能够利用计算机算法在庞大的数据中找出结构共同点是该领域中略被忽略的一个课题。更为困难的，是如何能够建立这些晶体结构与其物性之间的关联，这对于利用计算机算法指导材料设计有着重要意义。有别于多种成熟的，对于单个分子位于气态或溶液态中结构的降维分析方法，尚未有成熟的数学手段能够很好的分析高维度分子晶体势能面。这很大程度上是由于没有一套完善的"坐标"来给予每一个分子晶体一个独特的"标识"。
 
在传统的结构化学里，所谓的"双子合成子"是用于描述分子晶体基本结构单元的一个常用语言。以此为起点，并利用预测的多笨芳香烃化合物晶体结构为例子，本讲座首先阐述如何运用简单的‘键级数’来描述分子晶体中的双子pi-pi堆叠结构，并以之预测双子间的电子耦合，从而用来加快计算分子晶体电导率的速度。由此，将进一步讲述如何利用更完备的SOAP描述符来描述更为完整的分子晶体结构，包括如何利用它跳过双子耦合来直接预测晶体电导率，以及得到与人为逐个判断相吻合的结构分类。最后，受到利用"多体展开法"在分子晶体结合能计算上应用的启发，本讲座将阐述如何设计不同的"相似核"，从而描述分子晶体在不同尺度上的结构相似性，这对于描述以氢键结合的晶体尤为重要。