Machine learning accelerated topological design of metal-organic frameworks

金属-有机磁体(MOM)是金属-有机骨架(MOF)的一个子类，它们在由有机连接物连接的金属中心之间具有很强的磁性相互作用。MOM因其潜在的高导电性、强的磁性相互作用和低维而被视为新量子技术的构建块。由于金属和有机成分的可调性，这些框架磁体在设计上比传统的无机磁体具有更大的灵活性。因此，挑战是理解MOM中的磁性相互作用与连接体、金属中心和拓扑结构的巨大空间之间的联系。新材料的实验发现是缓慢、困难和昂贵的。必须费力地发现合成路线，然后才能对任何新材料进行表征和分类以确定其用途。基于密度泛函理论(DFT)的第一性原理计算已被证明可以加速发现过程：计算可以筛选出数千种最有希望的材料的候选者，使新材料甚至在昂贵的实验合成集发生之前就可以进行试验和测试。由于可能的建筑单元数量巨大，新的MoM材料的搜索空间是巨大的，远远超过20世纪对无机晶体的研究。在这些复杂的情况下，即使是DFT筛选也可能过于昂贵，因此我们已经展示了如何使用DFT生成的机器学习潜力(MLP)进行预筛选。在这个项目中，我们将开发一种MLP+DFT的组合方法，并发现新的母亲。这些将被用在电子计算机方法中来表征，以向我们的实验同事示意。

Metal-organic magnets (MOMs) are a subclassification of metal-organic frameworks (MOFs) that have strong magnetic interactions between metal centres bridged by organic linkers. MOMs are targeted as the building blocks of new quantum technology for their potential high electrical conductivity, strong magnetic interactions, and low dimensionality. These framework magnets allow for more flexibility in design over their conventional inorganic counterparts due to the tunability of both their metallic and organic components. The challenge then is understanding the connection between the magnetic interactions in MOMs and the vast space of linkers, metals centres, and topologies. The experimental discovery of new materials is slow, difficult, and expensive. Synthesis routes must be laboriously discovered before any new material can be characterised and categorised for its utility. First principles calculations based on density-functional theory (DFT) have been shown to accelerate the discovery process: calculations can screen thousands of candidates for the most promising materials, allowing new materials to be tried-and-tested even before the costly experimental synthesis set takes place. Due to the enormous number of possible building units, the search space of new MOM materials is vast, far greater than the 20th century's investigation of inorganic crystals. In these complex cases, even DFT screening can be too costly, so we have shown how to pre-screen using a DFT-generated machine-learned potential (MLP). In this project we will develop a combined MLP+DFT approach and discover new MOMs. These will be characterised using in silico methods to signpost to our experimental colleagues.